BIODEVICES 2009

Proceedings of the
International Conference on
Biomedical Electronics and Devices
Porto - Portugal
J anuary 14 17, 2009
Organized by
INSTICC Institute for Systems and Technologies of Information,
Control and Communication
Technical Co-sponsorship by
IEEE EMB Engineering in Medicine and Biology Society
In Cooperation with
AAAI Association for the Advancement of Artificial Intelligence
Endorsed by
IFMBE International Federation for Medical and Biological Engineering

I

Copyright 2009 INSTICC Institute for Systems and Technologies of
Information, Control and Communication
All rights reserved
Edited by Teodiano Freire Bastos Filho and Hugo Gamboa
Printed in Portugal
ISBN: 978-989-8111- 64-7
Depsito Legal: 285421/08
http://www.biodevices.org
[email protected]
BIODEVICES is part of BIOSTEC International J oint Conference on Biomedical Engineering Systems and Technologies
http://www.biostec.org/
II
BRIEF CONTENTS

INVITED SPEAKERS ........................................................................................................................ IV
SPECIAL SESSION CHAIRS ............................................................................................................. IV
ORGANIZING AND STEERING COMMITTEES .................................................................................... V
PROGRAM COMMITTEE ................................................................................................................. VI
AUXILIARY REVIEWERS .............................................................................................................. VII
SELECTED PAPERS BOOK ........................................................................................................... VIII
OFFICIAL CARRIER ..................................................................................................................... VIII
FOREWORD .................................................................................................................................... IX
CONTENTS ..................................................................................................................................... XI
III
INVITED SPEAKERS

Edward H. Shortliffe
Arizona State University
U.S.A.
Vimla L. Patel
Arizona State University
U.S.A.
Pier Luigi Emiliani
Institute of Applied Physics Nello Carrara (IFAC) of the Italian National Research Council (CNR)
Italy
Maciej Ogorzalek
J agiellonian University
Poland
Egon L. van den Broek
University of Twente
The Netherlands


SPECIAL SESSION CHAIRS

SPECIAL SESSION ON ACTIVE MATERIALS FOR MEDICAL DEVICES
Andres Diaz Lantada, Universidad Politecnica de Madrid, Spain
IV
ORGANIZING AND STEERING COMMITTEES

CONFERENCE CO-CHAIRS
Ana Fred, IST- Technical University of Lisbon, Portugal
J oaquim Filipe, INSTICC / Polytechnic Institute of Setbal, Portugal
Hugo Gamboa, Instituto de Telecomunicaes, Portugal
PROGRAM CO-CHAIRS
Teodiano Freire Bastos Filho, Federal University of Esprito Santo, Brazil
Hugo Gamboa, Instituto de Telecomunicaes, Portugal
PROCEEDINGS PRODUCTION
Srgio Brissos, INSTICC, Portugal
Marina Carvalho, INSTICC, Portugal
Helder Coelhas, INSTICC, Portugal
Vera Coelho, INSTICC, Portugal
Andreia Costa, INSTICC, Portugal
Bruno Encarnao, INSTICC, Portugal
Brbara Lima, INSTICC, Portugal
Raquel Martins, INSTICC, Portugal
Carla Mota, INSTICC, Portugal
Vitor Pedrosa, INSTICC, Portugal
Vera Rosrio, INSTICC, Portugal
J os Varela, INSTICC, Portugal
CD-ROM PRODUCTION
Elton Mendes, INSTICC, Portugal
GRAPHICS PRODUCTION AND WEB DESIGNER
Marina Carvalho, INSTICC, Portugal
SECRETARIAT AND WEBMASTER
Marina Carvalho, INSTICC, Portugal
V
PROGRAM COMMITTEE
Oliver Amft, ETH Zurich, Switzerland Jose Luis Pons, Instituto de Automatica Industrial,
Spain
Rodrigo Varejo Andreo, CEFETES, Brazil
Alejandro Ramirez-Serrano, University of Calgary,
Canada
Luciano Boquete, Alcala University, Spain
Susana Borromeo, Universidad Rey J uan Carlos,
Spain
Adriana Mara Rios Rincn, Universidad del
Rosario, Colombia
Enrique A. Vargas Cabral, Facultad de Ciencias y
Tecnologia - Universidad Catlica, Paraguay
Joaquin Roca-Dorda, Polytechnic University of
Cartagena, Spain
Ramn Ceres, IAIA-CSIC, Spain
Mario Sarcinelli-Filho, Federal University of Espirito
Santo, Brazil
Fernando Cruz, College of Technology of
Setubal/Polytechnic Institute of Setubal, Portugal
Mohamad Sawan, Ecole Polytechnique de Montreal,
Canada
Pedro Pablo Escobar, Faculty of Engineering,
Universidad Nacional del Centro, Argentina
Fernando di Sciascio, Institute of Automatics
National University of San J uan, Argentina
Marcos Formica, Faculty of Bioengineering,
Argentina
Wouter Serdijn, Delft University of Technology,
The Netherlands
Juan Carlos Garcia Garcia, Universidad de Alcala,
Spain
Jorge Vicente Lopes da Silva, Center For
Information Technology Renato Archer, Brazil
Gerd Hirzinger, DLR, Germany
Amir M. Sodagar, University of Michigan, U.S.A.
Jongin Hong, Imperial Collge London, U.K.
Ioan G. Tarnovan, Technical University of
Cluj-Napoca, Romania
Giacomo Indiveri, UNI - ETH Zurich, Switzerland
Bozena Kaminska, Simon Fraser University, Canada
Alexandre Terrier, Ecole Polytechnique Federale de
Lausanne (EPFL), Switzerland
Rui Lima, ESTiG, Braganca Polytechnic Institute
(IPB), Portugal
Mrio Vaz, Inegi Lome, FEUP, Portugal
Ratko Magjarevic, Faculty of Electrical Engineering
and Computing, Croatia
Chua-Chin Wang, National Sun Yat-Sen University,
Taiwan
Dan Mandru, Technical University of Cluj Napoca,
Romania


Manuel Mazo, University of Alcala, Spain

Paulo Mendes, University of Minho, Portugal

Joseph Mizrahi, Technion, Israel Institute of
Technology, Israel

Raimes Moraes, Universidade Federal de Santa
Catarina, Brazil


Pedro Noritomi, Centro de Tecnologia da Informao
Renato Archer, Brazil
Kazuhiro Oiwa, National Institute of Information and
Communications Technology, J apan


Evangelos Papadopoulos, NTUA, Greece

Laura Papaleo, University of Genova, Italy
Jos Luis Martnez Prez, Grupo de Robotica y
Cibernetica, Universidad Politecnica de Madrid, Spain
VI
AUXILIARY REVIEWERS


Getlio Igrejas, Instituto Politcnico de Bragana,
Portugal
Susana Palma, PLUX, Lda, Portugal
Hugo Silva, PLUX, Lda, Portugal

VII
SELECTED PAPERS BOOK

A number of selected papers presented at BIODEVICES 2009 will be published by Springer-Verlag in a CCIS
Series book. This selection will be done by the Conference Co-chairs and Program Co-chairs, among the papers
actually presented at the conference, based on a rigorous review by the BIOSTEC 2009 Program Committee
members.



OFFICIAL CARRIER


VIII
FOREWORD


This volume contains the proceedings of the Second International Conference on Biomedical Electronics and
Devices (BIODEVICES 2009), organized by the Institute for Systems and Technologies of
Information Control and Communication (INSTICC), technically co-sponsored by the IEEE
Engineering in Medicine and Biology Society (EMB), in cooperation with AAAI and endorsed by
IFMBE.
The purpose of the International Conference on Biomedical Electronics and Devices is to bring together
researchers and practitioners from electronics and mechanical engineering, interested in studying
and using models, equipments and materials inspired from biological systems and/or addressing
biological requirements. Monitoring devices, instrumentation sensors and systems, biorobotics,
micro-nanotechnologies and biomaterials are some of the technologies addressed at this
conference.
BIODEVICES is one of three integrated conferences that are co-located and constitute the
International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC).
The other two component conferences are HEALTHINF (International Conference on Health
Informatics) and BIOSIGNALS (International Conference on Bio-inspired Systems and Signal
Processing).
The joint conference, BIOSTEC, has received 380 paper submissions from more than 55 countries
in all continents. 57 papers were published and presented as full papers, i.e. completed work (8
pages/30 oral presentation), 126 papers reflecting work-in-progress or position papers were
accepted for short presentation, and another 63 contributions were accepted for poster
presentation. These numbers, leading to a full-paper acceptance ratio below 15% and a total oral
paper presentations acceptance ratio below 49%, show the intention of preserving a high quality
forum for the next editions of this conference.
The conference included a panel and five invited talks delivered by internationally distinguished
speakers, namely: Egon L. van den Broek, Pier Luigi Emiliani, Maciej Ogorzalek, Vimla L. Patel
and Edward H. Shortliffe. Their participation has positively contributed to reinforce the overall
quality of the Conference and to provide a deeper understanding of the field of Biomedical
Engineering Systems and Technologies.
The proceedings of the conference will be indexed by several major indices including DBLP,
INSPEC and ISI-Proceedings and it will also be submitted for indexing to EI. A book with the
revised versions of a short list of selected papers from the conference will be published by
Springer-Verlag in Communications in Computer and Information Science (CCIS).
The program for this conference required the dedicated effort of many people. Firstly, we must
thank the authors, whose research and development efforts are recorded here. Secondly, we thank
the members of the program committee and the additional reviewers for their diligence and expert
reviewing. Thirdly, we thank the keynote speakers for their invaluable contribution and for taking
the time to synthesise and prepare their talks. Fourthly, we thank the program chairs, Teodiano
IX
X
Freire Bastos Filho and Hugo Gamboa, whose collaboration was much appreciated. Finally, special
thanks to all the members of the INSTICC team, especially Marina Carvalho at the conference
secretariat whose collaboration was fundamental for the success of this conference.
This year, the organization will distribute two paper awards at the conference closing session: the
best paper award and the best student paper award. The decision was mainly based on the paper
classifications provided by the Program Committee.
We wish you all an exciting conference and an unforgettable stay in the lovely and historic city of
Porto. We hope to meet you again next year for the 3
rd
BIODEVICES, details of which will soon
be available at http://www.biodevices.org.


Ana Fred, IST IST / Technical University of Lisbon, Portugal
Joaquim Filipe, Polytechnic Institute of Setbal / INSTICC, Portugal
Hugo Gamboa, PLUX Biosignal Acquisition and Processing, Portugal

CONTENTS

INVITED SPEAKERS
KEYNOTE LECTURES
BIOMEDICAL INFORMATICS - Its Scientific Evolution and Future Promise IS-5
Edward H. Shortliffe
COGNITIVE SCIENCE APPROACH TO UNDERSTANDING HUMAN-COMPUTER
INTERACTION IN MEDICINE IS-7
Vimla L. Patel
TECHNOLOGY FOR THE INDEPENDENT LIVING OF PEOPLE WITH ACTIVITY
LIMITATIONS IS-9
Pier Luigi Emiliani
PATTERN RECOGNITION AND STATISTICAL LEARNING TECHNIQUES FOR APPLICATIONS
IN SKIN CANCER DIAGNOSIS IS-13
Maciej Ogorzalek
BIOSIGNALS AS AN ADVANCED MAN-MACHINE INTERFACE IS-15
Egon L. van den Broek, Viliam Lis, Joyce H. D. M. Westerink, Marleen H. Schut and Kees Tuinenbreijer
PAPERS
FULL PAPERS
LABEL FREE BIO SENSING METHOD USING RADIO FREQUENCIES SPECTROSCOPY FOR
CELL DETECTION AND DISCRIMINATION
Claire Dalmay, Arnaud Pothier, Pierre Blondy, Fabrice Lalloue and Marie-Odile Jauberteau 3
EVALUATION OF PSD COMPONENTS AND AAR PARAMETERS AS INPUT FEATURES
FOR A SVM CLASSIFIER APPLIED TO A ROBOTIC WHEELCHAIR
Andr Ferreira, Teodiano Freire Bastos-Filho, Mrio Sarcinelli-Filho, Jos Luis Martn Snchez, Juan Carlos Garca Garca
and Manuel Mazo Quintas 7
INVESTIGATION OF OPERATING PARAMETERS FOR A SEMEN QUALITY ANALYSIS SYSTEM
S. Atherton, C. R. Evans, P. Roach, D. C. Hughes, G. McHale and M. I. Newton 13
MULTIFOCAL ELECTRORETINOGRAPHY - Early Detection of Glaucoma based on Wavelets
and Morphological Analysis
J. M. Miguel, S. Ortega, I. Artacho, L. Boquete, J. M. Rodrguez, P. De La Villa and R. Blanco 17
STUDY OF THE PROPERTIES OF BIOTIN-STREPTAVIDIN SENSITIVE BIOFETS
Thomas Windbacher, Viktor Sverdlov, Siegfried Selberherr, Clemens Heitzinger, Norbert Mauser and Christian Ringhofer 24
AVALANCHE PHOTODIODES FOR HIGH-RESOLUTION PET IMAGING SYSTEMS
R. Bugalho, B. Carrio, C. S. Ferreira, M. Ferreira, R. Moura, C. Ortigo, J. Pinheiro, P. Rodrigues, J. C. Silva,
A. Trindade and J. Varela 31
2.4GHZ WIRELESS ELECTROMYOGRAPH SYSTEM WITH STATISTICALLY OPTIMAL
AUTOMATIC GAIN CONTROL - Design and Performance Analysis
Andrea Morici, Giorgio Biagetti and Claudio Turchetti 39
XI
XII
AN ASYNCHRONOUS PROGRAMMABLE PARALLEL 2-D IMAGE FILTER CMOS IC BASED
ON THE GILBERT VECTOR MULTIPLIER
Rafa Dugosz and Vincent Gaudet 46
SYNCHRONIZING AN X-RAY AND ANESTHESIA MACHINE VENTILATOR - A Medical Device
Interoperability Case Study
David Arney, Julian M. Goldman, Susan F. Whitehead and Insup Lee 52
AUTOFLUORESCENCE SPECTROSCOPY OF A HUMAN GASTROINTESTINAL CARCINOMA
CELL LINE - Design of Optical Sensors for the Detection of Early Stage Cancer
D. S. Ferreira, M. Henriques, R. Oliveira, J. H. Correia and G. Minas 61
A LOW-POWER INTEGRATED CIRCUIT FOR ANALOG SPIKE DETECTION AND SORTING
IN NEURAL PROSTHESIS SYSTEMS
A. Bonfanti, T. Borghi, R. Gusmeroli, G. Zambra, A. S. Spinelli, A. Oliynyk, L. Fadiga and G. Baranauskas 67
A BIOLOGICAL MONITORING MODULE BASED ON A CERAMIC MICROFLUIDIC PLATFORM
Walter Smetana, Bruno Balluch, Ibrahim Atassi, Khatuna Elizbarowna Gvichiya, Erwin Gaubitzer, Michael Edetsberger
and Gottfried Khler 75
MINIATURIZED ELECTROCHEMICAL SENSING SYSTEMS FOR IN VITRO AND IN VIVO
BIOMEDICAL APPLICATIONS
V. I. Ogurtsov, K. Twomey, N. V. Bakounine, C. McCaffrey, J. Doyle, V. Beni and D. W. M. Arrigan 83
IMPLEMENTATION OF AN AUTOMATED ECG-BASED DIAGNOSIS ALGORITHM
FOR A WIRELESS BODY SENSOR PLATAFORM
Francisco J. Rincn, Laura Gutirrez, Mnica Jimnez, Vctor Daz, Nadia Khaled, David Atienza, Marcos Snchez-lez,
Joaqun Recas and Giovanni De Micheli 88
DEVELOPMENT OF A MYOELECTRIC CONTROLLER BASED ON KNEE ANGLE
ESTIMATION
Alberto Lpez Delis, Joo Luiz Azevedo de Carvalho, Adson Ferreira da Rocha, Francisco Assis de Oliveira Nascimento
and Geovany Arajo Borges 97
CATECHOL DETETION USING AN OPTICAL MEMS SENSOR
Peter H. Dykstra, Stephan T. Koev, Reza Ghodssi, Gregory F. Payne and Liangli Yu 104
CHITOSAN FOR MEMS - Demonstration of Micromechanical and Optical Biosensors
Stephan T. Koev, Peter H. Dykstra, Reza Ghodssi, Gary W. Rubloff, William E. Bentley and Gregory F. Payne 109
DROPLET MANIPULATION ON HIGH ADHESION SUPERHYDROPHOBIC SURFACES
Daisuke Ishii, Masatsugu Shimomura and Hiroshi Yabu 113
POLYISOPRENE NANOSTRUCTURED CARBON COMPOSITE (PNCC) MATERIAL FOR
VOLATILE ORGANIC COMPOUND DETECTION
Gita Sakale, Maris Knite, Valdis Teteris and Velta Tupureina 117
SHORT PAPERS
BIOSIGNALS WITH A FLOOR SENSOR - Near Field Imaging Floor Sensor Measures Impedance
Changes in the Torso
Henry Rimminen and Raimo Sepponen 125
VIBRATIONAL SPECTROSCOPY (FTIR-ATR AND FT-RAMAN) - A Rapid and Useful Tool
for Phycocolloid Analysis
Leonel Pereira, Ana M. Amado, Paulo J. A. Ribeiro-Claro and Fred van de Velde 131
XIII
BIODEVICES BASED ON SHAPE-MEMORY POLYMERS - Current Capabilities and Challenges
Andrs Daz Lantada, Pilar Lafont Morgado, Hctor Lorenzo-Yustos, Vicente Lorenzo Esteban, Julio Muoz-Garca,
Jos Luis Muoz Sanz, Javier Echavarri Otero and Juan Manuel Munoz-Guijosa 137
INFRARED THERMOGRAPHY AS A SUPPORT TOOL FOR DEVELOPING SHAPE-MEMORY
POLYMER BIODEVICES
Andrs Daz Lantada, Pilar Lafont Morgado, Hctor Lorenzo-Yustos, Julio Muoz-Garca, Jos Luis Muoz Sanz,
Javier Echavarri Otero and Juan Manuel Munoz-Guijosa 145
AN EXPLOITATION OF THE SELF-ORGANIZING MAP FOR HUMAN MOTION ANALYSIS
W. Kurdthongmee and P. Kurdthongmee 151
SURFACE MODIFICATION OF DENTAL DEVICES - Surface Analysis of Plasma-based Fluorine
and Silver Ion Implanted & Deposited Acrylic Resin
Yukari Shinonaga, Kenji Arita and Milanita E. Lucas 157
MICROCOMPUTERIZED SYSTEM TO ASSESS THE PERFORMANCE OF LUNG VENTILATORS
Daniel Marinho Silva, Maurcio Campelo Tavares and Raimes Moraes 161
MODELING OF MERIDIAN CHANNELS
Zimin Wang, Yonghong Tan and Miyong Su 167
MAGNETOMETRY USING ELECTROMAGNETICALLY INDUCED TRANSPARENCY IN A
ROOM TEMPERATURE VAPOUR CELL - Developing an Optical Magnetometer that Utilises the Steep
Dispersion Curve Observed in EIT to Detect Ti
Melody R. Blackman and Benjamin T. H. Varcoe 173
DEVELOPMENT OF AN ELECTRICAL STIMULATION DEVICE
FOR OSSEOINTEGRATED AMPUTEES - A Novel Approach for Expediting Skeletal Attachment and
Rehabilitation
Brad Isaacson, Jeroen Stinstra, Rob MacLeod and Roy Bloebaum 178
BRAIN COMPUTER INTERFACE - Feedback Effect Analysis by Comparison of Discrimination
Capability of On-line and Off-line Experimental Procedures based on LDA
Jos Luis Martnez Prez and Antonio Barrientos Cruz 186
DATA ACQUISITION ELECTRONICS FOR PET MAMMOGRAPHY IMAGING
Carlos Leong, Pedro Machado, Vasco Bexiga, J. Paulo Teixeira, Isabel C. Teixeira, Joel Rego, Pedro Neves,
Fernando Piedade, Pedro Lous, Pedro Rodrigues, Andreia Trindade, R. Bugalho, J. F. Pinheiro, M. Ferreira
and Joo Varela 192
DEVELOPMENT OF STRATHCLYDE UNIVERSITY DATA LOGGING SYSTEM (SUDALS)
FOR USE WITH FLEXIBLE ELECTROGONIOMETERS
Vivek Padmanaabhan Indra Mohan, G. Valsan and P. J. Rowe 198
ELECTRONIC DEVICE FOR SEISMOCARDIOGRAPHY - Noninvasive Examination and Signal
Evaluation
Zdenek Trefny, Milan Stork and Martin Trefny 204
AN ASIC SOLUTION FOR INTELLIGENT ELECTRODES AND ACTIVE-CABLE USDED
IN A EARABLE ECG MONITORING SYSTEM
Geng Yang, Jian Chen, Fredrik Jonsson, Hannu Tenhunen and Li-Rong Zheng 209
ROBUST EAR LOCATED HEART RATE MONITOR
L. Rossini, R. Vetter, C. Verjus, P. Theurillat, P. Renevey, M. Bertschi and J. Krauss 214
A WIRELESS EMBEDDED DEVICE FOR PERSONALIZED ULTRAVIOLET MONITORING
Navid Amini, Jerrid E. Matthews, Foad Dabiri, Alireza Vahdatpour, Hyduke Noshadi and Majid Sarrafzadeh 220
XIV
IMMUNOSENSORS FOR ATRAZINE DETECTION IN RED WINE SAMPLES
Enrique Valera, ngel Rodrguez, Javier Ramn-Azcn, Francisco J. Sanchez and M.-Pilar Marco 226
A LOW COST LED BASED BILIRUBIN METER - Description and Evaluation of a Low Cost
Spectrophotometer Bilirubin Analyzer
L. A. L. Azeka and M. S. V. de Paiva 231
CORRECTION OF ACOUSTIC LENS ERROR IN SPATIAL COMPOUNDING OF ULTRASONIC
DIAGNOSTIC IMAGES
Myoung H. Choi 235
AN AUTOMATED ATHLETE PERFORMANCE EVALUATION SYSTEM - From Theory to Practice
Hugo Silva, Gonalo Martins, Susana Palma, Pedro Mil-Homens and Maria Valamatos 239
SIMULATION AND EXPERIMENTAL DESIGN OF A SYMMETRY CONTROLLER FOR FES
CYCLING OPTIMISED ON STROKE PATIENTS
Emilia Ambrosini, Simona Ferrante, Thomas Schauer, Alessandra Pedrocchi and Giancarlo Ferrigno 245
EXPERIMENTAL DIGITAL BPSK MODULATOR DESIGN WITH VHDL CODE
FOR BIODIVECES APPLICATIONS
Gihad Elamary, Graeme Chester and Jeffrey Neasham 251
MESOTHERAPY DEVICE FOR ESTHETIC APPLICATIONS
M. S. Martins, V. M. G. Correia, J. G. Rocha and J. M. Cabral 256
A RECONFIGURABLE ARRAY FOR BLIND SOURCE-SEPARATION ON AN FPGA
Ricardo Escalona, Daniel Herrera and Miguel Figueroa 262
IMPROVING SURFACE ENERGY AND HYDROPHILIZATION OF POLY(ETHYLENE
TEREPHTHALATE) BY ENZYMATIC TREATMENTS
Isabel C. Gouveia, Laura C. Antunes and Joo A. Queiroz 268
USING MULTI-AGENT SYSTEMS TO STUDY PARACRINIENNE CELLS INTERACTION
Lynda Dib 276
A RF TRANSCEIVER FOR WIRELESS MONITORING SYSTEMS OF THE VERTEBRAL COLUMN
BEHAVIOUR
J. P. Carmo and J. H. Correia 281
A MULTI-LAYERED MICROFLUIDIC DEVICE FOR MAGNETOPHORETIC CELL SEPARATION
Hye-Lyn Lee, Suk-Heung Song, Hee-Taek Lim, Hyung-Joon Kim, Min-Suk Park and Hyo-Il Jung 286
PAIN AND EFFICIENCY IN NEONATAL BLOOD SAMPLE SCREENINGS - New Devices
for Reducing Pain and Improving Blood Sample Quality
Bruno Wacogne, Christian Pieralli, Gonzalo Cabodevila, Nolwenn Baron, Sandrine Marioli and Lionel Pazart 290
THE DESIGN AND FABRICATION OF IMPLANTED INTRACRANIAL PRESSURE SENSOR
Tian Bian, Zhao Yulong and Jiang Zhuangde 296
INERTIAL SENSOR BASED IDENTIFICATION OF HUMAN MOVEMENTS
Ivo Stancic, Josip Music, Ana Kuzmanic Skelin, Tea Marasovic, Norberto Salgado, Tamara Supuk and Vlasta Zanchi 300
ADAPTIVE AURICULAR ELECTRICAL STIMULATION CONTROLLED BY VITAL BIOSIGNALS -
Transition from Fixed to Adaptive and Synchronized Electrical Stimulation Controlled by Heart Rate
Variability and Blood Perfusion
Eugenijus Kaniusas, Jozsef Constantin Szeles, Tilo Materna and Giedrius Varoneckas 304
DEVELOPMENT OF A MECHANICAL INSTRUMENT TO EVALUATE BIOMECHANICALLY
THE SPINAL COLUMN IN PREGNANT WOMEN
Cludia Quaresma, Mrio Forjaz Secca, Joo ONeill and Jorge Branco 310
XV
DEVELOPING A PUPILLOMETER
Gonalo Leal, Pedro Vieira and Carlos Neves 314
QUALITY ASSESSMENT IN COLONOSCOPY - New Challenges Through Computer Vision-based
Systems
Fernando Vilario and Gerard Lacey 320
DEVELOPMENT OF A SLEEP MONITORING SYSTEM WITH WEARABLE VITAL SENSOR FOR
HOME USE
Takuji Suzuki, Kazushige Ouchi, Ken-Ichi Kameyama and Masaya Takahashi 326
POSTERS
A WAY FOR PREDICTING AND MANAGING THE GLYCAEMIC INSTABILITY
OF THE DIABETIC PATIENT
Farida Benmakrouha, Christiane Hespel, Mikhail V. Foursov and Jean-Pierre Hespel 335
DEVICE FOR SYNCHRONIZED ROTATION
Shuh Jing Ying, Rufael Berhane and Rajiv Dubey 339
HAND-HELD LUMINOMETER WITH ECL-BASED BIOSENSOR FOR LACTATE
DETERMINATION
A. Martnez-Olmos, A. J. Palma, J. Ballesta-Claver, M. C. Valencia-Miron and L. F. Capitan-Vallvey 343
SELECTIVE OSTEOBLASTIC CELL MICRO-ARRAYS ON DIAMOND FILMS
Bohuslav Rezek, Lenka Michalkov, Egor Ukraintsev, Alexander Kromka and Marie Kalbacova 347
ON-DETECTOR ELECTRONICS OF THE CLEAR PEM SCANNER
E. Albuquerque, V. Bexiga, R. Bugalho, B. Carrio, C. S. Ferreira, M. Ferreira, J. Godinho, F. Gonalves, C. Leong,
P. Lous, P. Machado, R. Moura, P. Neves, C. Ortigo, F. Piedade, J. F. Pinheiro, P. Relvas, A. Rivetti, P. Rodrigues,
J. C. Silva, M. M. Silva, I. C. Teixeira, J. P. Teixeira, A. Trindade and J. Varela 355
HARDWARE IMPLEMENTATION FOR EDGE DETECTION IN CDNA MICROARRAY IMAGES
Bogdan Belean, Monica Borda and Albert Fazakas 359
NEW FAST TRAINING ALGORITHM SUITABLE FOR HARDWARE KOHONEN NEURAL
NETWORKS DESIGNED FOR ANALYSIS OF BIOMEDICAL SIGNALS
Rafa Dugosz and Marta Kolasa 364
WRIST-WORN FALL DETECTION DEVICE - Development and Preliminary Evaluation
Mattia Bertschi and Leopoldo Rossini 368
A NEW LINEAR ARRAY IMAGING SYSTEM OF ELECTRICAL AND ULTRASONIC PROPERTIES
IN A LIVING BODY
Akira Kimoto, Yuuta Taninaka and Katsunori Shida 372
MODELLING OF SAW BIOSENSORS
Marija Hribek, Slavica Risti, Zdravko ivkovi and Dejan Toi 376
A WIRELESS EEG ACQUISITION SYSTEM WITH THERMOELECTRIC SCAVENGING
MICRODEVICE
J. P. Carmo, L. M. Goncalves, R. P. Rocha and J. H. Correia 380
A NOVEL MOBILE MONITORING SYSTEM FOR FAST AND AUTOMATED BACTERIA
DETECTION IN WATER
Christoph Heller, Ulrich Reidt, Andreas Helwig, Florian Klettner, Gerhard Mller, Alois Friedberger, Leonhard Meixner,
Karl Neumeier, Petra Lindner, Ramona Molz and Hans Wolf 384
XVI
CONTROL OF CELL ADHESION AND FUNCTIONS USING SELF-ORGANIZED HONEY
COMB-PATTERNED POLYMER FILMS
Masaru Tanaka, Akinori Tsuruma, Sadaaki Yamamoto and Masatsugu Shimomura 390
ACOUSTIC THERMOAGITATION BASED ON PIEZOELECTRIC -PVDF POLYMER FILMS -
Potential Evaluation in Lab-on-a-Chip Applications
V. F. Cardoso, G. Minas, P. Martins, J. Serrado Nunes, L. Rebouta, S. Lanceros-Mndez and G. Botelho 394
CONTACT LESS RADIO-FREQUENCIES BIOSENSOR FOR BIOLOGICAL PARAMETERS
ANALYSIS
K. Grenier, D. Dubuc, M. Kumemura, H. Toshiyoshi and H. Fujita 398
WEARABLE TECHNOLOGY - Development of Polypyrrole Textile Electrodes for Electromyography
S. Rodrigues, R. Miguel, J. Lucas, C. Gaiolas, P. Arajo and N. Reis 402
A SCALABLE AND OPEN SOURCE LINEAR POSITIONING SYSTEM CONTROLLER
M. C. Medeiros, A. J. A. Fernandes, C. A. Teixeira and M. Graa Ruano 410
DESIGN OF A BIO-INSPIRED WEARABLE EXOSKELETON FOR APPLICATIONS IN ROBOTICS
Michele Folgheraiter, Bertold Bongardt, Jan Albiez and Frank Kirchner 414
EXTENDED HEALTH VISIBILITY IN THE HOSPITAL ENVIRONMENT
H. Fernndez Lpez, J. A. Afonso, J. H. Correia and Ricardo Simes 422
MICROFLUIDIC CELL STIMULATOR USING BEAD IMPACT
Young-Hun Kim, Tae-Jin Kim, Hyung-Joon Kim, Min-Suk Park and Hyo-Il Jung 426
A LOW-COST EEG STAND-ALONE DEVICE FOR BRAIN COMPUTER INTERFACE
Alexandre Ribeiro, Antnio Sirgado, Joo Aperta, Ana Lopes, Jorge Guilherme, Pedro Correia, Gabriel Pires
and Urbano Nunes 430
SPECIAL SESSION ON ACTIVE MATERIALS FOR MEDICAL DEVICES
CHARACTERISATION AND MEDICAL APPLICATIONS OF MAGNETORHEOLOGICAL FLUIDS
Javier Echavarri Otero, Andrs Daz Lantada, Pilar Lafont Morgado, Juan Manuel Munoz-Guijosa, Jos Luis Muoz
Sanz, Hctor Lorenzo-Yustos and Julio Muoz-Garca 437
COLLIMATION OF X-RAY DIAGNOSTIC BUNDLE BY MEANS OF STEERING FERROFLUID
Andrzej Dyszkiewicz, Pawe Poe, Jakub Zajdel, Bartomiej Pawlus, Damian Chachulski and Pawe Kpiski 441
NOVEL COMBINED TEMPLATE FOR AMPEROMETRIC BIOSENSORS WITH CHANGEABLE
SELECTIVITY
Julija Razumiene, Vidute Gureviciene, Jurgis Barkauskas, Virginijus Bukauskas and Arunas Setkus 448
MODELLING AND TRIALS OF PYROELECTRIC SENSORS FOR IMPROVING ITS
APPLICATION FOR BIODEVICES
Andrs Daz Lantada, Pilar Lafont Morgado, Hctor Hugo del Olmo, Hctor Lorenzo-Yustos, Javier Echavarri Otero,
Juan Manuel Munoz-Guijosa, Julio Muoz-Garca and Jos Luis Muoz Sanz 453
POLYMERIC FILM SENSORS BASED ON PAH-PAZO IONIC SELF-ASSEMBLED
MULTI-NANOLAYERS
Celso Ribeiro, Paulo J. Gomes, Paulo A. Ribeiro, Maria Raposo, Hugo guas, Pedro Santos, Beatriz Borges
and Pedro Brogueira 458

AUTHOR INDEX 463
INVITED
SPEAKERS

KEYNOTE
LECTURES

IS-5
BIOMEDICAL INFORMATICS
Its Scientific Evolution and Future Promise
Edward H. Shortliffe
Arizona State University, U.S.A.
Abstract: Biomedical informatics is the scientific field that deals with biomedical information, data, and knowledge
their storage, retrieval and optimal use for problem solving and decision making. The field has broad
applications across all of biomedicine, ranging from molecular and cellular processes (bioinformatics) and
the management of structural or visual information about tissues and organs (imaging informatics) to
patient-oriented tasks (clinical informatics) and population-based policy and analysis (public health
informatics). All these areas of application draw upon core methods from the discipline that define its
scientific base.
In this presentation, Dr. Shortliffe will review the nature of the field, emphasizing its relationship to the
worlds of engineering and computer science. Practical issues in the use of informatics techniques will be
summarized, ranging from the need for more individuals trained at the intersection of computer science and
biomedicine to the recurring barriers to systems implementation. Almost forty years in the development of
clinical decision-support systems have taught us several lessons about what kinds of tools are most likely to
be effective and accepted by clinical users. This talk will discuss some of the lessons of our experience in
this field, debunking some of the early myths that developed and suggesting key notions that should guide
researchers and system developers in the years ahead.
BRIEF BIOGRAPHY
Edward H. Shortliffe is Professor of Basic Medical
Sciences and Professor of Medicine at the University
of Arizona College of Medicine and Professor of
Biomedical Informatics at Arizona State University.
Until May 2008 he served as the founding dean of
the Phoenix campus of the University of Arizonas
College of Medicine. Previously he was the Rolf A.
Scholdager Professor and Chair of the Department
of Biomedical Informatics at Columbia College of
Physicians and Surgeons in New York City (2000-
2007) and Professor of Medicine and of Computer
Science at Stanford University (1979-2000).
Effective J uly 1, 2009, he will become the President
and Chief Executive Officer of the American
Medical Informatics Association, based in Bethesda,
MD.
Dr. Shortliffe has spearheaded the formation and
evolution of graduate degree programs in biomedical
informatics at Stanford, Columbia, and now at
Arizona State University. His research interests
include the broad range of issues related to
integrated decision-support systems, their effective
implementation, and the role of the Internet in health
care. He is an elected member of the Institute of
Medicine of the National Academy of Sciences, the
American Society for Clinical Investigation, the
Association of American Physicians, and the
American Clinical and Climatological Association.
He has also been elected to fellowship in the
American College of Medical Informatics and the
American Association for Artificial Intelligence. He
is a Master of the American College of Physicians
and was a member of that organizations Board of
Regents from 1996-2002. He is Editor-in-Chief of
the J ournal of Biomedical Informatics, and serves on
the editorial boards for several other biomedical
informatics publications. In addition, he received the
Grace Murray Hopper Award of the Association for
Computing Machinery in 1976, the Morris F. Collen
Award of the American College of Medical
Informatics in 2006, and has been a Henry J . Kaiser
Family Foundation Faculty Scholar in General
Internal Medicine. Dr. Shortliffe has authored over
300 articles and books in the fields of biomedical
computing and artificial intelligence.



IS-7
COGNITIVE SCIENCE APPROACH TO UNDERSTANDING
HUMAN-COMPUTER INTERACTION IN MEDICINE
Vilma L. Patel
Arizona State University, U.S.A.
Abstract: Given the complexities of modern medicine, delivery of safe and timely care is a huge challenge. Errors,
misunderstandings, and inaccuracies, large and small, are routine occurrences in our everyday activities.
Health information technology (HIT) has undoubtedly reduced the risk of serious injury for patients during
hospital stays. However, its true potential for preventing medical errors remains only partially realized and,
paradoxically, systems may even give rise to hazards of their own. There is a growing recognition that many
errors are neither solely attributable to lapses in human performance nor to flawed technology. Rather they
develop as a product of the interaction between human beings and technology. In our view, errors are the
product of cognitive activity in human adaptation to complex physical, social, and cultural environments.
How well the design of health IT complements its intended setting and purpose is critically important for
safe and effective performance. It is the flawed design and poor integration with clinical work, rather than
any person or the technology itself, that is at the root of its suboptimal performance. Attention to the design
principles of humancomputer interaction (HCI) in clinical software design is needed as a critical safety
hazard. In this presentation, I argue for a place of prominence for cognitive science. Cognitive science
provides a framework for the analysis and modeling of complex human performance and has considerable
applicability to a range of issues in informatics. I will discuss how cognitive science principles are
applicable to understanding HCI concerns that make the integration of computing and clinical practice a
difficult task.
BRIEF BIOGRAPHY
Vimla L. Patel, PhD, DSc, FRSC: Patel is Professor
and Vice-Chair of the Arizona State Universitys
Department of Biomedical Informatics and the
Director of the Center for Decision Making and
Cognition in the Ira A. Fulton School of
Engineering. She came to Arizona from Columbia
University in New York in March 2007. She is also
a professor of Basic Medical Sciences in the
University of Arizona, College of Medicine in
Phoenix. She has Bachelor of Science degree in
Biochemistry from Otago University in New
Zealand and her MA and PhD in Cognitive and
Educational Psychology from McGill University in
Montreal, Canada.
Dr. Patel is recognized as a leader in applied
cognitive science research for models of decision-
making and studies of human-computer interaction
in health care. An elected fellow of the Royal
Society of Canada (Academy of Social Sciences)
and also of the American College of Medical
Informatics, she was a recipient of the annual
Swedish Woman of Science award in 1999. Her
research interests include human performance,
decision-making, medical errors, assessment of
human-computer interaction in healthcare domains.
She is a prolifric writer with an extensive
publication record, and is an associate editor of
J ournal of Biomedical Informatics and on the
editorial board of J ournal of AI in Medicine and
Advances in Health Scince Education. Her research
is funded by National Intitute of Mental Health,
National Library of Medicine (NIH), USArmy
(TARTC) and the J ames S McDonnell Foundation.
http://www.fulton.asu.edu/~patel


See: Patel, V.L., & Kaufman, D.R., (2006).
Cognitive science and biomedical informatics. In
E.H. Shortliffe & J .J . Cimino (Eds.), Biomedical
informatics: Computer applications in health care
and biomedicine (3rd ed., pp. 133-185). New York:
Springer-Verlag.




IS-9
TECHNOLOGY FOR THE INDEPENDENT LIVING OF PEOPLE
WITH ACTIVITY LIMITATIONS
Pier Luigi Emiliani
Italian National Research Council, Italy
[email protected]
EXTENDED ABSTRACT
According to the eEurope 2005 and i2010 Action
Plans, eInclusion, i.e. the access and use of
information, telecommunication, and environmental
control systems and services, is considered crucial
for the independent living of all European citizens. It
is also claimed that, in order to reach this goal a
Design for All (DfA) approach should be used. This
implies that, in the specification and implementation
of new products, the needs, requirements and
preferences of all users must be taken into account,
to the greatest extent possible, with a special
attention to all groups at risk of exclusion, including
people with activity limitations and elderly people.
The official definition of eInclusion has been
published in the 2006 Riga ministerial declaration
eInclusion means both inclusive ICT and the use of
ICT to achieve wider inclusion objectives. It focuses
on participation of all individuals and communities
in all aspects of the information society. eInclusion
policy, therefore, aims at reducing gaps in ICT
usage and promoting the use of ICT to overcome
exclusion, and improve economic performance,
employment opportunities, quality of life, social
participation and cohesion. Therefore eInclusion
implies not only accessibility of information,
telecommunications, and environmental control
facilities by anyone at any place and at any time, but
also the proactive support to inclusion in all living
contexts. It aims to enable equitable access and
active participation of potentially all people in
existing and emerging human activities, by
developing universally accessible and usable
products and services and suitable support
functionalities in the environment. These products
and services must be capable of accommodating
individual user requirements in different contexts of
use, independent of location and used technology.
Therefore approaches aiming to grant the use of
equipment or services have to be generalized,
seeking to give access to the Information Society as
such.
This approach is also in line with the one at the basis
of the preparation of the new WHO International
Classification of Functioning, Disability and Health
(ICF)
1
, where a balance is sought between a purely
medical and a purely social approach to the
identifications of problems and opportunities for
people in their social inclusion. When dealing with
the problems of people who experience some degree
of activity limitation or participation restrictions,
ICF uses the term disability to denote a
multidimensional phenomenon resulting from the
interaction between people and their physical and
social environment. This is very important, because
it allows grouping and analysis of limitations that
are not only due to impairments. People may have
impairments, activity limitations or participation
restrictions that characterise their ability (capacity)
to execute a task or an action (activity), but their
performance is influenced by the current
environment. The latter can increase the
performance level over the capacity level (and
therefore is considered a facilitator) or can reduce
the performance below the capacity level (thus being
considered as a barrier).
According to observatories around the world, the
Information Society is foreseen to emerge and
evolve as some form of intelligent environment,
according to the Ambient Intelligence (AmI)
paradigm (Ducatel et al., 2001). The future
Information Society is not seen as being
characterised by an increased diffusion and use of
present-day computers and telecommunication
terminals, but as the emergence of an environment in
which people are surrounded by fixed and mobile
intelligent objects, interconnected through fixed and
mobile networks, and capable of recognising and
responding to the presence of different individuals.
The AmI environment is supposed to be populated
by a multitude of hand-held and wearable micro-
devices, while computational power and interaction
peripherals (e.g., embedded screens and speakers,

1
See http://www3.who.int/icf/icftemplate.cfm

IS-10
ambient displays) are distributed in the environment.
Devices will range from personal, carrying
individual and possibly private information (e.g.,
wrist-watches, bracelets, personal mobile displays
and notification systems, health monitors embedded
in clothing), to public, available in the surrounding
environment. Interaction with the objects and with
the intelligence in the environment will allow access
to information, interpersonal communication and
environmental control. Objects will allow
cooperative activities and offer important
functionalities to support people. A variety of new
products and services will be made possible by the
emerging technological environment, including
home networking and automation, mobile health
management, interpersonal communication, and
personalised information services. These
applications will be characterised by increasing
ubiquity, nomadism and personalisation, and are
likely to pervade all daily human activities.
The intelligent environment is supposed to develop
according to the two components foreseen in the
above reported definition for eInclusion. It will
make available an accessible ICT environment
where intelligent objects offer functionalities useful
for access to information, interpersonal
communications and environmental control.
Moreover the environment is supposed to be
connected with a control centre and with the external
world to contribute with more complex
functionalities (services), thus explicitly supporting
people.
In this dynamically evolving technological
environment, accessibility and usability of such
complex systems by users with different
characteristics and requirements cannot be addressed
through ad hoc Assistive Technology solutions
introduced after the main building components of
the new environment are in place. Instead, there is a
need for more proactive approaches, based on a
Design for All approach, defined as The design
of products and environments to be usable by all
people, to the greatest extent possible, without the
need for adaptation or specialized design
2
along
with the requirement of redefining the role and
scope of assistive technologies in the new
environment. In such a context, the concept of
Design for All acquires critical importance in
facilitating the incorporation of accessibility in the

2
See http://trace.wisc.edu/world/gen_ud.html

new technological environment through generic
solutions.
Conceptually, the Design for All approach is a well-
defined body of knowledge, which in architecture
and industrial design has brought about very
important results. In the Information Society, the
adoption and practice of Design for All, although
advocated by many actors in the field, still presents
significant challenges, due to the inherent
characteristics of the sector, and mainly the
consolidate industry practice of designing
mainstream product targeted to the so called
typical user. Design for All methodology in the
context of Information Society Technologies is not
to be conceived as an effort to advance a single
solution for everybody, but as a user centred
approach to providing products that can
automatically address the possible range of human
abilities, skills, requirements and preferences.
Consequently, the outcome of the design process is
not intended to be a singular design, but a design
space populated with appropriate alternatives,
allowing for automatic and intelligent adaptation and
personalisation (Emiliani and Stephanidis, 2005).
The last step is to devise a technical approach,
according to which this implementation strategy can
become a reality. As shown in several projects
funded by the European Commission, a working
technical approach is the one based on adaptability
and adaptivity (Stephanidis ed. 2001). This is based
on the fact that any system and service should be
intelligent enough to be able to automatically adapt
its functionalities and interface to each single user,
according to her/his known characteristics or an
assumed stereotype (different privacy levels) when
s/he starts to interact, and modify them in real time
as a function of the usage and context (if allowed to
monitor the user behaviour).
The change of paradigm from the Assistive
Technology approach to Design for All is considered
by many people working in the sector to be not only
too ambitious but even dangerous. This is because
they fear that it could jeopardise, at least in the short
term, advances toward integration made by groups
of people with disabilities. However this is not the
case, because the two approaches must be
considered as complementary: they should converge
towards the creation of a more accessible society
through the continuous redefinition of the problem
of producing a barrier-free technology. The
emerging situation can thus be addressed through an
evolutionary approach. In the short term, the
development of the Information Society can be

IS-11
supported by a technology which enhances the
possibilities offered by Assistive Technology,
merging in the medium term into accessible systems
and services and, in the long term, into an intelligent
environment, which has the potential of being usable
by most users if their needs are taken into account
proactively during the design phase.
In the near future, ICT will continue to develop with
a Design for All approach, therefore producing more
accessible mainstream technology. This will cause
not only the emergence of intelligent objects but also
their inclusion into AmI-like environments, i.e.
environments that incorporate partially and in
interconnected islands AmI concepts. AmI will
materialise when the individual AmI-like islands
will merge and when enough intelligence will be
available to guarantee functionality and security of
the infrastructure and the corresponding services
throughout the entire society. Therefore, transition
toward AmI is undergoing in an incremental way.
The main conceptual change is supposed to be the
use of the concept of integrated support. So far,
assistive technology has supported the augmentation
of the capabilities of the individuals and the
adaptation of single artefacts for accessibility. The
new approach is based on the assumption that the
artefacts in the environment are interconnected and
integrated in an intelligent control system, in order
to support people with services and applications that
offer useful functionalities. Therefore, the main
emphasis is not on technology itself and its
adaptation, but on useful functionalities the
environments could or should offer irrespective of
their technical implementation. Services in the
environment will reduce the level of capacity needed
to carry out the required activities.
Finally, in the far future a complete implementation
of the AmI environment is assumed to emerge, as
described in the ISTAG scenarios. The possible
impact on some groups of people with activity
limitations is briefly analysed with suggestions of
possible new opportunities and additional problems.
The main conceptual change is that e-accessibility
has dealt with tasks to be carried out to use
equipment and services, while the AmI environment
is dealing with goals of users to be automatically
identified and proactively facilitated.
REFERENCES
Ducatel, K.; Bogdanowicz, M.; Scapolo, F.; Leijten, J . &
Burgelman, J . C. (2001), Scenarios for Ambient
Intelligence in 2010, Technical report, Information
Society Technologies Programme of the European
Union Commission (IST).
Emiliani, P. L. & Stephanidis, C. (2005), Universal
access to ambient intelligence environments:
opportunities and challenges for people with
disabilities, IBM Systems Journal 44(3), 605--619.
Stephanidis, C. (ed.) (2001), User Interfaces for All
Concepts, Methods and Tools, Mahwah, NJ :
Lawrence Erlbaum Associates.
BRIEF BIOGRAPHY
Personal Data
Education: Physics, University of Florence.
Current position: Director of Institute of Applied
Physics Nello Carrara (IFAC) of the Italian
National Research Council (CNR), President of the
CNR Research Area in Florence.
Work Experience
Research in the theory and applications of digital
signal processing and information technology.
Management of research projects. Lecturer on signal
processing at the University of Florence (Electronics
Engineering Department). Applications in
telecommunications problems.
Specific Experience
Applications of digital signal processing in aids for
disabled persons.
EC Funded Projects
Project Leader of the Concerted Action on
Rehabilitation of the Blind - DG XII, Biomedical
Engineering Committee (1988-91). Vice-Chairman
of the projects COST 219 and COST 219 bis.
Project Manager of the RACE Project R1066 -
IPSNI, Integration of People with Special Needs in
the Integrated Broadband Telecommunication
Network (1989-1991). Cooperation with Handynet,
DG V (1988-1991). Project Manager of the TIDE
Project 103 - GUIB Textual and Graphical User
Interfaces for Blind People (1992-1993). Project
Manager of the RACE Project 2009 - IPSNI II
Access to Broadband Services and Applications by
People with Special Needs (1992-1995). Project
Manager of the TIDE Project 215 - GUIB 2 Textual
and Graphical User Interfaces for Blind People
(1993-1994). Responsible for Line F Emerging

IS-12
Areas of Potential Rehabilitation Technology
Research and Development of the TIDE study
HEART Horizontal European Activities in
Rehabilitation Technology (1993-1994). Project
Manager of the TIDE Project 1001 ACCESS
Development Platform for Unified ACCESS to
Enabling Environments (1994-1996). Technical
Manager of the ACTS Project 042 AVANTI
Adaptive and Adaptable Interactions for
Multimedia Telecommunications Applications
(1995-1998). Technical Manager of the IST Project
PALIO Personalised Access to Local Information
and services for tOurists (2000-2003). Partner of
the Thematic Network IS4ALL Information
Society for All (2000-2003). National Contact
Centre of EDeAN European Design for All e-
Accessibility Network (2003-). European
representative in the Advisory Committee of the
Web Accessibility Initiative (WAI) (1998-). Vice-
chairman of the project Cost 219ter Accessibility
for All to Services and Terminals for Next
Generation Networks (2002-2007). Responsible of
the Secretariat of the European Design for All
eAccessibility Network (EDeAN) (2007). Project
manager of the Coordination Action ICT 033838
DfA@eInclusion Design for All for eInclusion
(2007-2009).
Publications
(Co-)author of over 160 technical papers published
in scientific archival journals, books, and
international conferences on speech processing,
signal processing and communication aids.
IS-13
PATTERN RECOGNITION AND STATISTICAL LEARNING
TECHNIQUES FOR APPLICATIONS
IN SKIN CANCER DIAGNOSIS
Maciej Ogorzalek
Jagiellonian University, Poland
Abstract: Digital photography provides new powerful tools for computer-assisted diagnosis systems in dermatology.
Dermoscopy is a special photography technique which enable taking photos of skin laisions in chosen
lighting conditions. Digital photography allows for seeing details of the skin changes under various
enlargments and colouring. Computer-assisted techniques are used for feature extraction and pattern
recognition in the selected images. Special techniques used in skin-image processing are discussed in detail.
Feature extraction and classification techniques based on statistical learning and model ensembling
techniques provide very powerful tools which can assist the doctors in taking decisions. Performance of
classifiers will be discussed in specific case of melanoma cancer diagnosis. The techniques have been tested
on a data set of more then six thousand images.
BRIEF BIOGRAPHY
Maciej Ogorzalek is professor and Head of the
Department of Information Technologies,
J agiellonian University, Krakow Poland (oldest
university in Poland funded 1364), Poland and a
Chair Professor of Department of Electronic and
Information Engineering, Hong Kong Polytechnic
University. He is an IEEE Fellow since 1997,
Recipient of the IEEE Guillemin-Cauer Award in
2002, Recipient of the IEEE-CAS Golden Jubilee
Award in 2004, IEEE Distinguished Lecturer from
2001 to 2003. Author of over 280 technical papers
and one book (Chaos and Complexity in Nonlinear
Electronic Circuits - World Scientific). He served as
an Associate Editor of IEEE Transactions on
Circuits and Systems, Part I 1993-1995 and 1999-
2001, the elected member of the Editorial Board of
the Proceedings of the IEEE since 2004, the Editor-
in-Chief of the IEEE Circuits and Systems Magazine
from 2004-2007, an Associate Editor Int. J. Circuit
Theory and Applications since 1999, an Associate
Editor of J ournal of The Franklin Institute since
1997, and the Member of the Editorial Board of the
International J ournal of Bifurcation and Chaos. He
was the Chairman of the Technical Committee of
Nonlinear Circuits and Systems of IEEE CASS from
1997 to 1998, the Founding member of the IEEE
CASS Technical Committee on Biomedical Circuits
and Systems, the General Chairman of European
Conference on Circuit Theory and Design in 2003.
He was keynote speaker at numerous international
conferences and guest editor of several special issues
of journals - most recently Special Issue on
Computational System Biology for the Proceedings
of the IEEE (August 2008). He has served on CAS
Society Executive Committee since 2002 - currently
he is the President of IEEE CASS.



BIOSIGNALS AS AN ADVANCED MAN-MACHINE INTERFACE
Egon L. van den Broek
Center for Telematics and Information Technology, University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands
[email protected]
Viliam Lis y
Agent Technology Center, Dept. of Cybernetics, FEE, Czech Technical University
Technick a 2, 16627 Praha 6, Czech Republic
[email protected]
Joyce H. D. M. Westerink
User Experience Group, Philips Research Europe, High Tech Campus 34, 5656 AE Eindhoven, The Netherlands
[email protected]
Marleen H. Schut, Kees Tuinenbreijer
Philips Consumer Lifestyle Advanced Technology, High Tech Campus 37, 5656 AE Eindhoven, The Netherlands
{marleen.schut,kees.tuinenbreijer}@philips.com
Keywords:
Emotion, BioSignals, Man-Machine Interface, Automatic classication.
Abstract:
As is known for centuries, humans exhibit an electrical prole. This prole is altered through various phys-
iological processes, which can be measured through biosignals; e.g., electromyography (EMG) and electro-
dermal activity (EDA). These biosignals can reveal our emotions and, as such, can serve as an advanced
man-machine interface (MMI) for empathic consumer products. However, such an MMI requires the correct
classication of biosignals to emotion classes. This paper explores the use of EDA and three facial EMG
signals to determine neutral, positive, negative, and mixed emotions, using recordings of 24 people. A range
of techniques is tested, which resulted in a generic framework for automated emotion classication with up to
61.31% correct classication of the four emotion classes, without the need of personal proles. Among vari-
ous other directives for future research, the results emphasize the need for both personalized biosignal-proles
and the recording of multiple biosignals in parallel.
That men are machines (whatever else they may be)
has long been suspected; but not till our generation
have men fairly felt in concrete just what wonderful
psycho-neuro-physical mechanisms they are.
William James (1842 1910)
1 INTRODUCTION
Despite the early work of William James and others,
it still took until the last two decades before emo-
tions were widely acknowledged and embraced by
engineering. But, now it is generally accepted that
emotions cannot be ignored; they inuence us, with
or without being aware, in all possible ways (Picard,
1997). Let us briey denote three issues on how emo-
tions inuence our lives: 1) our long term physical
well-being; e.g., Repetitive Strain Injury (RSI) (van
Tulder et al., 2007), cardiovascular issues (Schuler
and OBrien, 1997; Frederickson et al., 2000), and
our immune system (Ader et al., 1995; Solomon et al.,
1974); 2) our physiological reactions/signals (Fair-
clough, 2009; Picard et al., 2001; van den Broek et al.,
2009); and 3) our cognitive processes; e.g., perceiv-
ing, memory, reasoning (Critchley et al., 2000).
As is illustrated by the previous three issues,
we are (indeed) psycho-neuro-physical mecha-
nisms (James, 1893; Marwitz and Stemmler, 1998),
who both send and perceive biosignals; e.g., elec-
tromyography (EMG), electrocardiography (ECG),
and electrodermal activity (EDA). These biosignals
can reveal a broad plethora of peoples characteristics;
e.g., workload, attention, and emotions. In this paper,
IS-15
we will focus on biosignals that reveal peoples emo-
tional state. Such biosignals can act as a very useful
interface between man and machine; e.g., computers
or consumer products such as an MP3-player. Such an
advanced Man-Machine Interface (MMI) would pro-
vide machines with empathic characteristics, capable
of coping with the denoted issues.
For research on biosignals as an advanced
MMI, traditional emotion research using interviews,
questionnaires, and expert opinions are not suf-
cient (Fairclough, 2009; van den Broek et al., 2009).
The recent progress in brain imaging techniques en-
ables the inspection of brain activity while experienc-
ing emotions; e.g., EEG and fMRI (Critchley et al.,
2000; Grandjean and Scherer, 2008). The former re-
search methods have as disadvantages that the mea-
surements tend to be subjective, are very limited in
explaining, and do not allow real time measurements:
they can only be used before or after emotions are
experienced. Although EEG techniques are slowly
brought to practice; e.g., Brain Computer Interfac-
ing (Bimber, 2008), these techniques are still very ob-
trusive. Hence, they are not usable in real world situ-
ations; e.g., for the integration in consumer products.
As sort of a way between these two research methods,
psychophysiological (or bio)signals can be used (Fair-
clough, 2009; Marwitz and Stemmler, 1998; van den
Broek et al., 2009). These are not, or at least less, ob-
trusive, can be recorded and processed real time, are
rich sources of information, and are relatively cheap
to apply.
The traditional methods (e.g., questionnaires),
brain imaging techniques, and biosignal measures
used to infer peoples emotional state, all share one
thing: the problem of a lack of ground truth; i.e., a
theoretically grounded, observable, operational def-
inition of the construct(s) of interest (Fairclough,
2009; van den Broek et al., 2009). In addition, a
range of other prerequisites should be taken into ac-
count when using such methods. In van den Broek et
al. (2009), these are denoted for affective signal pro-
cessing; however, most of them also hold for brain
imaging techniques and traditional methods. The pre-
requisites include: 1) the validity of the research em-
ployed, 2) triangulation, 3) omitting the inference
of emotion from the signals, if possible, and 4) in-
clusion and exploitation of signal processing knowl-
edge. For a discussion on these topics, we refer to
van den Broek et al. (2009). Let us now assume
that all prerequisites are satised. Then, it is feasi-
ble to classify the biosignals in terms of emotions.
In bringing biosignals-based emotion recognition to
products, self-calibrating, automatic classication is
essential to make it useful for Articial Intelligence
(AI) (Picard, 1997; Minsky, 2006), Ambient Intel-
ligence (AmI) (Aarts, 2004), and MMI (Fairclough,
2009; Kim and Andr e, 2008).
In the pursuit toward empathic technology for AI,
AmI, and MMI purposes, we will discuss the work on
classifying four biosignals signals: three facial EMGs
and EDA. The research in which the data was gath-
ered is discussed in both (van den Broek et al., 2006)
and (Westerink et al., 2008). Therefore, we will now
only provide a brief summary of it in the next section.
After that, in Sections 3 and 4, we will briey intro-
duce the classication and preprocessing techniques
employed. This is followed by Section 5 in which
the classication results are presented. In Section 6,
we reect on our work, critically review it, and draw
some nal conclusions.
2 RECORDING EMOTIONS
An experiment was conducted in which the subjects
emotions were elicited using lm fragments that are
known to be powerful in eliciting emotions in labora-
tory settings; see also (Rottenberg et al., 2007). The
physiological signals used, facial EMG and EDA, are
commonly known to reect emotions (Kreibig et al.,
2007).
2.1 Participants
In the experiment, 24 subjects (20 females) partici-
pated (average age 43 years). The relative small num-
ber of males is due to an expected better facial emo-
tion expression of females (Kring and Gordon, 1998).
2.2 Equipment and Materials
We selected 8 lm fragments (duration: 120 sec.
each) for their emotional content. For specications
of these lm fragments, see (van den Broek et al.,
2006; Westerink et al., 2008). The lm fragments
were categorized as being neutral or triggering pos-
itive, negative, or mixed (i.e., simultaneous nega-
tive and positive; (Carrera and Oceja, 2007)) emo-
tions. This categorization was founded on Russells
valence-arousal model (Russell, 1980).
A TMS International Porti5-16/ASD system was
used for the biosignals recordings, which was con-
nected to a PC with TMS Portilab software (http://
www.tmsi.com/). Three facial EMGs were recorded:
the right corrugator supercilii, the left zygomaticus
major, and the left frontalis muscle. The EMG sig-
nals were high-pass ltered at 20 Hz, rectied by tak-
ing the absolute difference of the two electrodes, and
IS-16
average ltered with a time constant of 0.2 sec. The
EDA was recorded using two active skin conductivity
electrodes and average ltering with a time constant
of about 2 sec.
2.3 Procedure
After the subject was seated, the electrodes were at-
tached and the recording equipment was checked.
The 8 lm fragments were presented to the subject
in pseudo-random order. A plain blue screen was
shown between the fragments for 120 seconds; so, the
biosignals returned to their baseline level for the next
stimulus.
After the viewing session, the electrodes were re-
moved. Next, the subjects answered a few questions
regarding the lm fragments viewed. To jog their
memory, representative print-outs of each fragment
were provided.
3 CLASSIFICATION
TECHNIQUES
In this section, we briey introduce the techniques
used for those readers who are not familiar with (all
of) them. First, ANalysis Of VAriance (ANOVA) and
Principal Component Analysis (PCA) are briey in-
troduced, which will be both applied for preprocess-
ing purposes. Second, the three classication tech-
niques k-Nearest Neighbors (k-NN), Support Vector
Machines (SVM), and Neural Networks (NN) are in-
troduced. Third and last, the Leave-one-out cross val-
idation (LOOCV) technique is introduced, which is
used for the evaluation of the classiers.
3.1 ANalysis Of VAriance (ANOVA)
ANalysis Of VAriance (ANOVA) is a statistical test to
determine whether or not there is a signicant differ-
ence between means of several populations. We will
sketch the main idea here. For a more detailed expla-
nation, see for example (King and Minium, 2007).
ANOVA assumes that the data of each population
is independent and randomly chosen from a normal
distribution. Moreover, it assumes that all the pop-
ulations have the same variance. These assumptions
usually hold with empirical data and the test is fairly
robust against limited violations.
ANOVA examines the variance of population
means compared to within class variance of the popu-
lations themselves. The result of the test is the proba-
bility p that all the populations were chosen from dis-
tributions with the same mean and variance. Hence,
the smaller p, the higher the chance that there is a real
difference between the populations.
3.2 Principal Component Analysis
(PCA)
This linear transformation derives from an input data
space a rst base vector in the direction of the biggest
variance in the data. Every next base vector is in-
dependent from the previous ones and represents the
highest possible variance of the data with the indepen-
dence constraint; see also (Everitt and Dunn, 2001).
Formally, if we have a data set
_

x
s
_
sSubj
of n-
dimensional vectors then the principal components of
vector

x
s
from the data set are a sequence of compo-
nents of vector

y
s
that are linear combinations of the
components of vectors

x
s
,
y
s
i
= a
i1
x
s
1
+a
i2
x
s
2
+ +a
in
x
s
n
= a
i

x
s
such that
i N 1 i n : a
i

a
i
= 1
i, j N 1 i < j n : a
j

a
i
= 0
and subsequently, each y
i
=
_

y
s
i
_
sSubj
has the maxi-
mal possible variance with respect to the constraints.
Variance covered by y
i
is dened as
Var(y
i
) = a
1

S a
1
where S is the covariance matrix of the original data
set.
At this point, we have to nd vectors a
i
that maxi-
mize the variance with respect to the constraints. This
kind of optimization problems can be solved using the
method of Lagrange multipliers.
In this case, the result is that a
i
is the eigenvec-
tor of S corresponding to the i-th largest eigenvalue.
Once we have the vectors a
i
, we can perform the
transformation by mapping all the data vectors to its
principal components.

y
s
=
_
_
_
a
1

.
.
.
a
n

_
_
_

x
s
The principal components computed this way are
very sensitive to scaling. In order to deal with the dif-
ferent scaling and capture the underlying structure of
the data set, the components can be derived from the
correlation matrix instead of the covariance matrix. It
is equivalent to extracting the principal components
in the described way after normalizing all the compo-
nents of the original data set to have unit variance.
PCA is also a powerful tool for data inspection
through visualization. For this purpose, often plots
IS-17
1 0 1
1
0
1


1 0 1
1
0.5
0
0.5
1 0 1
1
0
1
1 0 1
1
0.5
0
0.5
1 0 1
1
0
1
1 0 1
1
0.5
0
0.5
Figure 1: Visualization of the rst two principle components of all six possible combinations of two emotion classes. The
emotion classes are plotted per two to facilitate the visual inspection. The plots illustrate how difcult it is to separate even
two emotion classes, where separating four emotion classes is the aim.
are made with the principle components on the axis.
Figure 1 presents such a visualization. It presents for
each set of two emotion classes, of the total of four, a
plot denoting the rst two principle components. The
six resulting plots illustrate the complexity of separat-
ing the emotion classes perfectly.
3.3 k-Nearest Neighbors (k-NN)
We have decided to use this technique because it is a
very intuitive and simple machine learning algorithm.
The main idea is that each new feature vector, which
is close to some of the vectors from the training
set, will probably belong to the same class as most
of these vectors. The training phase of the classier is
simply storing all (or a suitable subset) of the training
samples with the correct classication category in a
database.
A metric (e.g., Euclidean distance) is selected that
assigns a non-negative real number to each pair of
input vectors. The number represents how close the
input vectors are to each other. When a new vector
has to be classied, the metric is used to count the
distance of the new sample from all the samples in
the database. After this, the number of representa-
tives of each class among the closest k samples are
considered. If there is a class with a higher number
of representatives than all the other classes; then, the
new sample is classied to this class. If there is a tie
of two or more classes; then, the sample is classied
randomly to one of these classes.
k-NN is often applied. Consequently, various tu-
torials and introductions have been written. We refer
to (Bishop, 2006), who provides an excellent intro-
duction.
3.4 Support Vector Machine (SVM)
A Support Vector Machine (SVM) ensures the opti-
mal division of a set of data to two classes with respect
to the shape of the classier and misclassication of
the training samples. Using a suitable kernel function,
it can create an optimally shaped classier.
The main ideas of this classier can be best ex-
plained through the example of a binary linear classi-
er; i.e., a separating hyperplane w.x +b, formally:
y
i
(w.x
i
+b) 1
i
, for i = 1, 2, . . . , N
where x
i
are the data samples, y
i
{1, +1} is the
corresponding class of the i-th data sample, w is the
normal vector of the hyperplane and b is the shift of
the hyperplane. To make the plane optimal, the size
of w and the sum of
i
must be minimized. It can be
proved that minimization of these parameters can be
solved by maximization of:
W() =
N

i=1

1
2
N

i, j=1

j
y
i
y
j
(x
i
.x
j
)
with constraints
0
i
C, for i = 1, ..., N
and
N

i=1

i
y
i
= 0.
IS-18
where C is a constant determining the trade-off be-
tween minimizing of the size of w and the sum of
i
.
This is a problem that can be solved using methods of
quadratic programming.
After we have the Lagrange multipliers
i
, the
classication is already easy:
f (x) = sgn
_
N

i=1
y
i

i
.(x.x
i
) +b
_
It is possible to see from the derivation of this method
that most of the
i
s are usually equal to 0. The re-
maining relevant subset of the training data (x
i
) is
called support vectors.
For a non-linear classication problem, we can
transform the input space with an appropriate non-
linear function into a higher-dimensional feature
space. For this, we only need the dot product of
the transformed vectors. It is the only part of the
computation where we have to work with the higher-
dimensional space.
k(x,y) =
_
(x).(y)
_
,
which results in a scalar. The function k is called
the kernel function. Please note that the SVMs intro-
duced above classify samples into two classes. How-
ever, usually we want to distinguish between multiple
classes. This can be done using a separate binary clas-
sier for each target class.
For more information on SVM, we refer
to (Burges, 1998). This paper provides a gentle in-
troduction on SVM for pattern recognition. Alterna-
tively, (Bishop, 2006) and (Vapnik, 1999) can be con-
sulted.
3.5 Neural Network (NN)
Neural Networks (NN) are the least intuitive group
of approaches used. It has a solid theoretical basis;
e.g., (Bishop, 2006). Nevertheless, their performance
is not always satisfying in practice. Each neuron in a
NN can perform only a trivial task, but after connect-
ing more of them to a network, they can approximate
any function (Leshno et al., 1993).
For our task, we will use the multilayer percep-
tron. Its perceptrons count weighed w
i
sum of val-
ues on its inputs (x
i
), subtracts a bias (b) and apply a
sigmoid-shaped function () to the result, producing
single number as the output of the neuron.
y =
_
n

i=1
w
i
x
i
b
_
Its neurons are divided into several layers. Inputs of
one layer are all the outputs of the neurons in the pre-
vious layer. Its transfer function is crucial for the
NNs functioning. For this, often gradient descent is
used, which tries to minimize an overall error of the
network expressed by the error function:
E(w) =

kX

jY
_
y
j
(w,x
k
) d
k j
_
2
where X is the set of indexes of training samples, Y
the set of output neurons, y
j
(w,x
k
) is the output of the
j-th output neuron for input x
k
and weight vector w,
and d
k j
is the desired output on the j-th neuron for the
k-th training sample.
Using theorems about derivation of a composite
function, it is possible to derive the gradient of the er-
ror function. Following this vector with subsequent
small adaptations of the weights, the algorithm will
nd a minimum of the error function. However, the
risk remains that the error function reaches a local
minimumand the learning process stops even thou the
error is far from the global optimum. Although vari-
ous methods exist that improve the gradient descend,
none of them guarantees an optimal solution.
The strongest point of using neural network for
our problem is its natural capability of incremental
learning. Most of the algorithms used for neural net-
work learning are intrinsically incremental.
For more information on NN, we refer to (Bishop,
2006). However, various other introductions have
been published that differ with respect to both the pro-
vided details and their length.
3.6 Leave-one-out Cross Validation
(LOOCV)
Leave-one-out Cross Validation (LOOCV) is a
method to determine how good a classier is. If we
have some inputs with known correct classications,
for each of the data samples we:
Make the classier learn fromthe data without the
selected sample.
Classify the omitted sample.
Compute the ratio of wrong classications.
A little modication of this method in our case is that
we do not leave out only one data sample, but we do
not consider all data from one subject in the learning
process. It is more accurate estimation of the classi-
cation error on an unknown subject.
The results reported in this paper are determined
by this method if it is not specied another way. For
more information on LOOCV, we refer to (Bishop,
2006).
IS-19
Table 1: The best feature subsets for k-Nearest Neighbor (k-NN) classier determined by ANalysis Of VAriance (ANOVA),
using normalization per signal per participant.
electrodermal activity (EDA) facial electromyography (EMG)
Frontalis Corrugator Zygomaticus
Mean o
Absolute Deviation o
Feature Standard Deviation o o
Variance o o
Skewness o o o
Kurtosis o
4 PREPROCESSING
The quest towards self-calibrating algorithms for con-
sumer products and for AmI and AI purposes gave
some constraints to processing the signals. For ex-
ample, no advanced lters should be needed and the
algorithms should be able to handle noisy and prefer-
ably also corrupt data. Therefore, we chose to refrain
fromadvanced preprocessing schemes and only apply
some basic preprocessing.
4.1 Normalization
Humans are known for their rich variety in all as-
pects, this is no different for their emotional reac-
tions and their physiological derivatives. In develop-
ing generic classiers, this required the normalization
of the signals. This could boost its performance sig-
nicantly (Rani et al., 2006).
For each person, for all his signals, and for all their
features separately, two linear normalizations were
applied:
x
n
=
x min
max min
,
where x
n
is the normalized value, x the recorded
value, and max and min the global maximumand min-
imum and
x

n
=
x x

,
where x

n
is the normalized value, x the recorded
value, and x and the global mean and standard de-
viation.
The linear normalization of datasets (e.g., signals)
has been broadly discussed, this resulted in a variety
of normalization functions; e.g., see (Boucsein, 1992;
Iglewicz, 1983).
4.2 Baseline Matrix
In their standard work, Picard et al. (2001) introduce
a baseline matrix for processing biosignals for emo-
tion recognition. This could tackle problems due to
variation both within (e.g., inter day differences) and
between participants. Regrettably, Picard et al. (2001)
do not provide evidence for its working. However,
their idea is appealing and, hence, was judged as
worth trying.
The baseline matrix requires biosignals recordings
while people are in a neutral state. Such recordings
were, however, not available. Alternatively, one of
the two neutral lm fragments was chosen (van den
Broek et al., 2006; Westerink et al., 2008).
In line with Picard et al. (2001), the input data
was augmented with the baseline values of the same
dataset. The results of some initial tests, using various
weights, were far from convincing. A maximum per-
formance improvement was achieved of 1.5%, using
a kNN classier. Therefore, the baseline matrix was
excluded in the nal processing pipeline.
4.3 Feature Selection
To achieve good classication results, the set of input
features is crucial. This is no different with classify-
ing emotions (Fairclough, 2009; van den Broek et al.,
2009). To dene an optimal set, a criterion function
should be dened. However, no such criterion func-
tion is available in our case. Then, an exhaustive
search in all possible subsets of input features is re-
quired to guarantee an optimal set (Cover and Camp-
enhout, 1977). To limit this enormous search space,
an ANOVA-based heuristic search was applied.
For both the normalizations, we performed
feature-selection based on ANOVAs. We selected the
features with ANOVA p values below 0.0013, as this
led to the best precision. The features selected are in
Table 1.
The last step of preprocessing is PCA. The im-
provement of the PCA is not that big compared to
feature selection solely; but, it is positive for both nor-
malizations; see also Table 2. Figure 1 presents for
each set of two emotion classes, of the total of four, a
plot denoting the rst two principle components. As
IS-20
such, the six resulting plots illustrate the complexity
of separating the emotion classes.
5 CLASSIFICATION RESULTS
This section denotes the results achieved with the
three classication techniques applied: k-Nearest
Neighbors (k-NN), Support Vector Machines (SVM),
and Neural Networks (NN). In all cases, the features
extracted from the biosignals were used to classify
participants neutral, positive, negative, or mixed state
of emotion.
5.1 k-Nearest Neighbors (k-NN)
For our experiments, we have used MATLAB
1
and
k-NN implementation based on SOM Toolbox 2.0
(Vesanto et al., 2000). Besides the classication algo-
rithm described in Section 3.3, we have used a mod-
ied version, more suitable for calculating the recog-
nition rates. The output of the modied version is not
the resulting class, but a probability of classication
to each of the classes. This means that if there is a
single winning class; then, the output is 100% for the
winning class and 0% for all the other classes. If there
is a tie of two classes then the result is 50%for each of
them and 0% for the rest and so forth. All the recog-
nition rates of the k-NN classier in the current study
are obtained by this modied algorithm.
A correct metric is a crucial part of a k-NN classi-
er. A variety of metrics provided by the pdist func-
tion in MATLAB was applied. Different feature sub-
sets appeared to be optimal for different classes. Rani
et al. (2006) denoted the same issue in their empirical
review. If we use the feature subset optimized on the
objective classes; then, the recognition precisions in
other divisions lowers or improves only a little com-
pared to the improvement in the optimized class divi-
sion. The results of the best preprocessed input with
respect to the four emotion classes (i.e., neutral, posi-
tive, negative, and mixed) is 61.31%, with a cityblock
metric and k = 8.
Probability tables for the different classications
given a known emotion category are quite easy to ob-
tain. They can be estimated from confusion matrices
of the classiers by transforming the frequencies to
probabilities. Table 3 presents the confusion matrix
of the classiers used in this research.
1
http://www.mathworks.com/products/matlab/
5.2 Support Vector Machines (SVM)
We have used MATLAB environment and SVM-KM
Toolbox (Canu et al., 2005) for experimenting with
SVMs. We use input enhanced with the best prepro-
cessing described in the previous section. It was op-
timized for the k-NN classier; however, we expect it
to be a good input also for more complex classiers,
including SVM. This assumption was supported by
several tests with other normalizations. The inputs in
this section are normalized per signal per person. Af-
ter feature selection, the rst 5 principal components
from the PCA transformation were used.
The kernel function of SVM characterizes the
shapes of possible subsets of inputs classied into one
category. We applied both polynomial kernels, de-
ned as:
K
Poly
(x,y) = (x.y +1)
d
and Gaussian kernels, dened as:
K
Gaus
(x,y) = exp
_

|x y|
2
2
2
_
The correct kernel function is the most important part
of SVM.
A Gaussian kernel ( = 0.7) performed best with
60.71%correct classication. However, a polynomial
kernel with d = 1 has a similar classication perfor-
mance (58.93%). All the results are slightly worse
than with the k-NN classier.
5.3 Neural Networks (NN)
We have used a modied multi-layer perceptron
trained by back-propagation algorithm that is imple-
mented in the Neural Network Toolbox of MATLAB.
It uses gradient descent with moment and adaptive
training parameter. We have tried to recognize only
the inputs that performed best with the k-NN classi-
er.
In order to assess what topology of NN is most
suitable for the task, we ran a small test. In the ex-
periments with one hidden layer, we have tried 2 to
16 neurons and we run LOOVC for each network 100
times. The networks were trained with the xed num-
ber of 150 cycles and subsequently tested on the left-
out subject. The experiments with two hidden layers
were much slower; so, we made only 10 trials for each
combination of sizes of the layers. We have 12 12
different topologies, 10 trials for each of themand one
trial of LOOCV means to train and test 21 networks
(total: 30240 networks). Each network was trained
with 150 cycles.
Our experiments with different network topolo-
gies supported the claim from (Lawrence et al., 1996)
IS-21
Table 2: The recognition precision of the k-Nearest Neighbor (k-NN) classier, with and without ANalysis Of VAriance
(ANOVA) feature selection (FS) and with and without Principle Component Analysis (PCA) transform. # comp. denotes the
number of principal components used to reach the precision with FS.
Normalization no FS ANOVA FS # comp. ANOVA FS & PCA
no 45.54%
yes 54.07% 60.71% 5 60.80%
Table 3: Confusion matrix of the k-NN based classier of EDA and EMG signals for the best reported input preprocessing.
Real
Neutral Positive Mixed Negative
Neutral 71.43% 19.05% 9.52% 14.29%
Classied Positive 9.52% 57.14% 9.52% 21.43%
Mixed 4.76% 4.76% 64.29% 11.90%
Negative 14.29% 19.05% 16.67% 52.38%
that bigger NN does not always tend to over t the
data and the extra neurons are not used in the training
process. Bigger networks showed good generaliza-
tion capabilities. However, further enlargement of the
network did not lead to better results. For this reason,
we choose the topology with one hidden layer of 12
neurons.
An alternative method for stopping the adaptation
of the NN is using validation data. The data set is
split into three parts. In our case, we have used one
subject for testing, three subjects for validation and
seventeen subjects for training. The testing subject is
completely removed from the training process at the
beginning. After that, the network is trained using
seventeen randomly chosen training subjects. At the
end of each training iteration, the current network is
tested on the three validation subjects.
In order to evaluate the NN on different desired
outputs, we have performed the above described al-
gorithm for each subject as the testing subject and for
15 random triples of the remaining 23 subjects as the
validation data. The 15 networks trained on differ-
ent data create an ensemble and the nal result of the
classier is the most frequent output class. This way,
we can prot from the early stopping of the back-
propagation algorithm and still use all the training
samples for training of the whole classier. This pro-
cedure led to a 56.19% correct classication of the
four emotion classes.
6 DISCUSSION & CONCLUSIONS
Successful automatic classication of biosignals
could serve various purposes (Fairclough, 2009;
van den Broek et al., 2009). One of them is to ex-
tend consumer products, AI, and AmI with empathic
capabilities.
Throughout the last decade, various studies have
been presented with similar aims, reporting good re-
sults on the automatic classication of biosignals. For
example, Picard et al. (2001) reports 81% correct
classication on the emotions of one subject. More
recently, Kim and Andr e (2008) reported a recogni-
tion accuracy of 95% and 70% for subject-dependent
and subject-independent classication. Their study
included three subjects.
In comparison with Picard et al. (2001) and Kim
and Andr e (2008), this research incorporated data of
a large number (i.e., 24) of people, with the aim to de-
velop a generic processing framework. At rst glance,
with a recognition accuracy of 61.31%, its success is
questionable. However, taking in consideration the
generic processing pipeline, it should be judged as (at
least) reasonably good. Moreover, a broad range of
improvements are possible. One of them would be to
incorporate more biosignals in the processing frame-
work. Another directive could be to question the need
of identifying specic emotions, using biosignals for
MMI. Hence, the use of alternative, rather rough cate-
gorizations, as used in the current research, should be
further explored.
Also in this research, the differences among par-
ticipants became apparent. They can be denoted on
two levels: physiological and psychological. With
this we mean that people have different physiological
reactions on the same emotions and that people expe-
rience different emotions with the same stimuli (e.g.,
music or lms). Moreover, these two levels inter-
act (Fairclough, 2009; Marwitz and Stemmler, 1998).
Although our aim was to develop a generic model, it
seems to be questionable whether or not this can be
IS-22
realized. Various attempts have been made to deter-
mine peoples personal biosignals-prole; e.g., (Kim
and Andr e, 2008; Marwitz and Stemmler, 1998; Pi-
card et al., 2001; Rani et al., 2006). However, no gen-
erally accepted standard has been developed so far.
With respect to processing the biosignals, the cur-
rent research can be extended by a more detailed ex-
ploration of the time windows; e.g., with a span of
10 seconds (Fairclough, 2009; van den Broek et al.,
2009). Then, data from different time frames can be
combined and different normalizations can be better
applied to create some new features that could eas-
ier reveal emotions. For example, the information
concerning the behavior of the physiological signals
could be more informative than only the integral fea-
tures from the larger time window. Studying the short
time frames also provides a better understanding on
the relation between emotions and their physiological
correlates.
Preprocessing of the biosignals could also be im-
proved. First of all, we think that the feature selec-
tion based on an ANOVA is not sufcient for more
complex classiers such as Neural Networks. The
ANOVA tests gather the centers of random distribu-
tions that would generate the data of different cate-
gories; hereby assuming that their variances are the
same. However, a negative result of this test is not
enough to decide that the feature does not contain any
information. As an alternative for feature selection,
the k-NN classier can be extended by a metric that
would weigh the features, instead of omitting the con-
fusing or less informative features.
As can be derived from the discussion, various
hurdles have to be taken in the development of a
generic, self-calibrating, biosignal-driven classica-
tion framework for MMI purposes. The research and
the directives denoted in this article could help in tak-
ing the rst hurdles. When the remaining ones will
also be taken; then, in time, the common denomi-
nators of peoples biosignals can be determined and
their relation with experienced emotions can be fur-
ther specied. This would mark a new, biosignal-
driven, era of advanced MMI.
ACKNOWLEDGEMENTS
The authors thank Frans van der Sluis (University of
Twente, NL) and Joris H. Janssen (Eindhoven Uni-
versity of Technology, NL / Philips Research, NL) for
reviewing earlier drafts of this article.
REFERENCES
Aarts, E. (2004). Ambient intelligence: Vision of our future.
IEEE Multimedia, 11(1):1219.
Ader, R., Cohen, N., and Felten, D. (1995). Psy-
choneuroimmunology: interactions between the ner-
vous system and the immune system. The Lancet,
345(8942):99103.
Bimber, O. (2008). Brain-Computer Interfaces. IEEE Com-
puter, 41(10):[special issue].
Bishop, C. M. (2006). Pattern Recognition and Machine
Learning. New York, NY, USA: Springer.
Boucsein, W. (1992). Electrodermal activity. New York,
NY, USA: Plenum Press.
Burges, C. J. C. (1998). A tutorial on support vector
machines for pattern recognition. Data Mining and
Knowledge Discovery, 2(2):121167.
Canu, S., Grandvalet, Y., Guigue, V., and Rakotomamonjy,
A. (2005). SVM and kernel methods Matlab Toolbox.
Perception Syst` emes et Information, INSA de Rouen,
Rouen, France. URL: http://asi.insa-rouen.
fr/enseignants/

arakotom/toolbox/.
Carrera, P. and Oceja, L. (2007). Drawing mixed emotions:
Sequential or simultaneous experiences? Cognition &
Emotion, 21(2):422441.
Cover, T. M. and Campenhout, J. M. V. (1977). On the pos-
sible orderings in the measurement selection problem.
IEEE Transactions on Systems, Man, and Cybernet-
ics, SMC-7(9):657661.
Critchley, H. D., Elliott, R., Mathias, C. J., and Dolan, R. J.
(2000). Neural activity relating to generation and rep-
resentation of galvanic skin conductance responses:
A functional magnetic resonance imaging study. The
Journal of Neuroscience, 20(8):30333040.
Everitt, B. S. and Dunn, G. (2001). Applied Multivariate
Data Analysis, Second Edition. Arnold, London.
Fairclough, S. (2009). Fundamentals of physiological com-
puting. Interacting with Computers, [in press].
Frederickson, B. L., Manusco, R. A., Branigan, C., and Tu-
gade, M. M. (2000). The undoing effect of positive
emotions. Motivation and Emotion, 24(4):237257.
Grandjean, D. and Scherer, K. R. (2008). Unpacking the
cognitive architecture of emotion processes. Emotion,
8(3):341351.
Iglewicz, B. (1983). Robust scale estimators and con-
dence intervals for location, chapter 12, pages 404
432. New York, NY, USA: John Wiley & Sons, Inc.
James, W. (1893). Review: La pathologie des emotions by
Ch. F er e. The Philosophical Review, 2(3):333336.
Kim, J. and Andr e, E. (2008). Emotion recognition based
on physiological changes in music listening. IEEE
Transactions on Pattern Analysis Machine Intelli-
gence, 30(12):20672083.
King, B. M. and Minium, E. W. (2007). Statistical rea-
soning in psychology and education. New York, NY,
USA: John Wiley & Sons, Inc., 5th edition.
IS-23
Kreibig, S. D., Wilhelm, F. H., Roth, W. T., and Gross, J. J.
(2007). Cardiovascular, electrodermal, and respiratory
response patterns to fear- and sadness-inducing lms.
Psychophysiology, 44(5):787806.
Kring, A. M. and Gordon, A. H. (1998). Sex differ-
ences in emotion: Expression, experience, and physi-
ology. Journal of Personality and Social Psychology,
74(3):686703.
Lawrence, S., Giles, C. L., and Tsoi, A. (1996). What size
neural network gives optimal generalization? Conver-
gence properties of backpropagation. Technical Re-
port UMIACS-TR-96-22 and CS-TR-3617.
Leshno, M., Lin, V. Y., Pinkus, A., and Schocken, S. (1993).
Multilayer feedforward networks with a nonpolyno-
mial activation function can approximate any func-
tion. Neural Networks, 6(6):861867.
Marwitz, M. and Stemmler, G. (1998). On the status
of individual response specicity. Psychophysiology,
35(1):115.
Minsky, M. (2006). The Emotion Machine: Commonsense
Thinking, Articial Intelligence, and the Future of the
Human Mind. New York, NY, USA: Simon & Schus-
ter.
Picard, R. W. (1997). Affective Computing. Boston MA,
USA: MIT Press.
Picard, R. W., Vyzas, E., and Healey, J. (2001). Toward
machine emotional intelligence: Analysis of affec-
tive physiological state. IEEE Transactions on Pat-
tern Analysis and Machine Intelligence, 23(10):1175
1191.
Rani, P., Liu, C., Sarkar, N., and Vanman, E. (2006). An em-
pirical study of machine learning techniques for affect
recognition in human-robot interaction. Pattern Anal-
ysis & Applications, 9(1):5869.
Rottenberg, J., Ray, R. R., and Gross, J. J. (2007). Emotion
elicitation using lms, chapter 1, pages 928. New
York, NY, USA: Oxford University Press.
Russell, J. A. (1980). Acircumplex model of affect. Journal
of Personality and Social Psychology, 39(6):1161
1178.
Schuler, J. L. H. and OBrien, W. H. (1997). Cardiovascu-
lar recovery from stress and hypertension factors: A
meta-analytic view. Psychophysiology, 34:649659.
Solomon, G. F., Amkraut, A. A., and Kasper, P. (1974).
Immunity, emotions and stress with special reference
to the mechanisms of stress effects on the immune
system. Psychotherapy and Psychosomatics, 23(1
6):209217.
van den Broek, E. L., Janssen, J. H., Westerink, J. H. D. M.,
and Healey, J. A. (2009). Prerequisits for Affective
Signal Processing (ASP). In Proceedings of the In-
ternational Conference on Bio-inspired Systems and
Signal Processing, page [in press], Porto Portugal.
van den Broek, E. L., Schut, M. H., Westerink, J. H. D. M.,
van Herk, J., and Tuinenbreijer, K. (2006). Comput-
ing emotion awareness through facial electromyogra-
phy. Lecture Notes in Computer Science (Human-
Computer Interaction), 3979:5162.
van Tulder, M., Malmivaara, A., and Koes, B. (2007).
Repetitive strain injury. The Lancet, 369(9575):1815
1822.
Vapnik, V. N. (1999). An overview of statistical learn-
ing theory. IEEE Transactions on Neural Networks,
10(5):988999.
Vesanto, J., Himberg, J., Alhoniemi, E., and Parhankangas,
J. (2000). SOM toolbox for matlab. Technical Report
A57, Helsinki University of Technology. URL: http:
//www.cis.hut.fi/projects/somtoolbox/.
Westerink, J. H. D. M., van den Broek, E. L., Schut, M. H.,
van Herk, J., and Tuinenbreijer, K. (2008). Computing
emotion awareness through galvanic skin response
and facial electromyography, volume 8 of Philips Re-
search Book Series, chapter 14 (Part II: Probing in or-
der to Feed Back), pages 137150. Springer: Dor-
drecht, The Netherlands.
BRIEF BIOGRAPHY
Egon L. van den Broek obtained his MSc (2001) in
Articial Intelligence and his PhD (2005) in Content-
Based Image Retrieval (CBIR), both from the Rad-
boud University (RU), Nijmegen, The Netherlands
(NL). Previously, he has been junior lecturer (RU),
consultant, and assistant professor in Articial Intel-
ligence (AI) at the Vrije Universiteit (VU), Amster-
dam, NL. Currently, he is head of a group on Ad-
vanced Interface Design (Center for Telematics and
Information Technology, University of Twente, En-
schede, NL), coordinates a MSc track, is member of
the board of the post-doctoral professional study of
ergonomics (VU), is consultant for Philips Research,
and is visiting assistant professor in Articial Intel-
ligence (RU). He is involved in various national and
EU projects and is specialized in engineering cogni-
tion, affective signal processing, cognitive computer
vision, and perception. He has supervised 40+ BSc,
MSc, and PhD students and published 100+ articles
and book chapters, holds a patent, and developed the
online image retrieval system http://www.m4art.org.
IS-24
FULL PAPERS
LABEL FREE BIO SENSING METHOD USING RADIO
FREQUENCIES SPECTROSCOPY FOR CELL DETECTION AND
DISCRIMINATION
Claire Dalmay, Arnaud Pothier, Pierre Blondy
XLIM UMR 6172 Universit de Limoges/CNRS, France
[email protected], [email protected], [email protected]
Fabrice Lalloue, Marie-Odile J auberteau
Homostasie cellulaire et Pathologies, Universit de Limoges, France
[email protected]
Keywords: Bio sensor, Electrical bio-impedance, Microelectronics, RF planar devices.
Abstract: This paper presents an original label free bio sensing method allowing the study of electrical properties of
human cells and so potentially cell identification and discrimination. The proposed bio sensor is based on a
planar resonator operating at microwave frequencies, fabricated using a standard microelectronic process.
As the result its microscopic sensitive areas allow an improved detection at the cell scale which represents a
significant step in the study of many biological phenomenon. Thanks to a specific experimental protocol, we
present in this paper a simple method allowing electrical parameters measurement on a small number of
cells with a good accuracy.
1 INTRODUCTION
In recent years, biosensors known a great interest as
there is an important need for tools that can quickly
and accurately analyse biological elements like bio
molecules or cells. Current optical and chemical bio
detection techniques can effectively analyse
biological systems but present some drawbacks ;
especially their requirement of specific labels to
enhance the signal generation. These labelled
methods make the sample preparation more
complex, expensive and time consuming. In
addition, the sample can be largely chemically
altered prior analysis. In the other way, electronic
detection techniques are very interesting methods as
they allow the development of label free methods
(Kim et al. 2007). Thanks to microelectronic
technology a significant improvement of electrical
sensor detection performance can be expected since
resulting miniaturized biosensors are now able to
work at the cell scale.
In this paper, is presented an electric label free
method allowing to evaluate cell inside medium
permittivity and conductivity in the gigahertz
frequency domain. Actually, these two specific
parameters are influenced by the cell type and
morphology but also by their physiological state.
Culture chamber
Substrate (fused silica)
CPW access for
on wafer RF
measurements
Hydrophobic ring
Analysed cells

Figure 1: Schematic of the studied biosensor.
As example, tumorous cells are well known to
present a larger conductivity and permittivity than
normal cells (Blad and Baldetorp, 1996). Hence,
individual cell electrical properties measurement
represents a complementary tool allowing efficient
cell identification.
Developed biosensors are actually based on a
coplanar microwave resonator design (figure 1) able
to operate at radio frequencies with a significant
3

sensitivity to tiny concentrations of biological
medium that interacts with the sensor. Moreover its
planar configuration will make easier a coming
integration in microsystems with microfluidic flow-
through network enabling accurate cell sorting
applications as example.
2 BIO DETECTION METHOD
2.1 Biosensor Design
In this study a resonant structure has been favoured
because by nature much more sensitive to very small
cell concentration in comparison with wide band
device (Denef et al, 2004). But in the other way,
available analysis spectrum will be limited to a
narrow band around the sensor resonant frequency.
Wider band investigation will so require fabricating
several resonators with different resonant
frequencies.
Meandered inductor
Interdigital capacitor
Meandered inductor
Interdigital capacitor

Figure 2: RF Electromagnetic field distribution plot at
resonance frequency.
The developed micro biosensor has been
designed as a coplanar RLC resonator made with a
meandered inductor associated in parallel with an
inter-digital capacitor. In our case, used resonators
present a RF signal attenuation which becomes
maximum at the resonant frequency. As shown on
figure 2 at this frequency, the electromagnetic field
distribution is strongly concentrate in the capacitive
part of the device which represents the more
interesting interaction area for a capacitive detection.
Indeed, the introduction of any biological media
even in very small concentration will meaningfully
disturb the EM field distribution inducing a
detectable shift in the measured resonance frequency
of the sensor (figure 3). This frequency shift will be
all the more significant if cells are located close to
gaps between metallic lines where the
electromagnetic field is strong.
Frequency (GHz)
S
2
1
(
d
B
)
Unlooaded sensor
Sensor loaded by 4 cells
Sensor loaded by 6 cells
Sensor loaded by 8 cells
Cell number
increasing

Figure 3: Electromagnetic simulation of the cell number
influence on the sensor RF response:S
21
parameter relies
on the RF signal attenuation through the resonator.
Hence, the detection resonator performance
strongly relies on the biosensor sensitivity
capabilities where two parameters play a major role.
First, interaction between the EM field and cells to
be analysed must be maximized using an
appropriated sensor design with gaps between
metallic lines in the same order of magnitude of
analysed cell sizes: in the present case considered
gaps will be close to 10 m. Then, a sufficient
resonator unloaded quality factor (relative the
resonator intrinsic loss) has to be also considered; as
it controls how the resonant frequency pick will be
narrow and so the sensor frequency sensitivity to a
small frequency shift.
Once biological cells will be present on the
sensor surface, both their location and their number
will directly influence its response. As shown on
figure 3, following our approach the detection of a
low number of cells (at least less than ten) can be
expected.
In the end, a microfluidic network will certainly
be required but in order to demonstrate the sensor
capability, we have chosen to develop a specific
experimental protocol for instance, allowing an
easier test procedure with the cost of the difficulty to
work with real alive cells. This protocol will be
presented in the following paragraphs.
2.2 Biosensor Fabrication Process
Micro sensors are fabricated using standard
microelectronic process with biocompatible
materials (figure 4).
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
4


Figure 4: SEM photograph of the fabricated micro sensor.
A fused silica substrate has been preferred to
classical silicon one especially for its lower loss
properties in the RF frequency domain and also for
its transparency that makes easier the observation of
cells throw it. A classical photolithography allows to
define thin gold lines which are next electroplating
up to a thickness of 3 m. Then a SU8 photoresist
from Microchem is used to create a 20m thick well
localised culture micro-chamber on the sensor
surface.
2.3 Experimental Protocol
During characterizations, we have to ensure the
integrity of cells. Usually, a support biological
media, in which cells could be protected, is required.
As shown on figure 5, most of support medium
commonly used in biology are aqueous saline
solutions which present very strong losses for RF
signal especially when the sensor is fully cover with
it. Actually in this configuration, the biological
media alters in a too important manner the RF
performance of the device avoiding any accurate
detection.
5 10 15 20 25 30 35 0 40
-25
-20
-15
-10
-5
-30
0
1: Emptysensor
3: Distilledwater
2: PBS
4: Culture media
(GHz)
1
2
3
4
Biological
media
deposition
Dropof biological media
Frequency
S
2
1
(
d
B
)
5 10 15 20 25 30 35 0 40
-25
-20
-15
-10
-5
-30
0
1: Emptysensor
3: Distilledwater
2: PBS
4: Culture media
(GHz)
1
2
3
4
Biological
media
deposition
Dropof biological media
Frequency
S
2
1
(
d
B
)

Figure 5: Influence of different biological support media
20nl drops on the sensor RF performances.
A first solution could be to limit used biological
media to a very small volume, typically implying a
microfluidic flow-through system.
Another alternative will be to perform analysis in
a specific low loss support media, as done in
previous work (Dalmay et al., 2008) using ficoll (a
polymeric gel of sucrose). Hence, once dried, the
ficoll drop allows to protect cells to be analysed in a
low permittivity polymeric matrix with the cost of a
aleatory cell location on the sensor surface and in the
ficoll matrix that induces a significant error in their
electrical parameter extraction. Consequently, there
is a great interest in developing an experimental
protocol without any biological support media.
Actually with the proposed approach in this
paper, cells are directly grown on the sensor surface
submerged in a classical culture media. Few days are
required to allow a sufficient cell adhesion on the
sensor surface. Then sensor are washed in deionised
water following by paraformaldehyde 4 % bath
(PFA) in order to definitively fixe cells and so to
avoid cell degradation during the measurement
sequence. Since most of cell adhesion occurs
preferentially on the silica substrate than on gold
lines or on SU8 resist, most of fixed cells are located
only between gold lines in the culture micro-
chamber (figure 1). Number of cell on the sensor are
roughly controlled both with the cell concentration
initially dispersed in culture media and the culture
time.
After 5 days in
the culture
media
10 min
PFA4%with
deionised water
Sensor drying
Adherentcells
floatingcells
Biosensor
After 5 days in
the culture
media
10 min
PFA4%with
deionised water
Sensor drying
Adherentcells Adherentcells
floatingcells
Biosensor Biosensor

Figure 6: Experimental protocol process.
After PFA bath, cells are no longer alive, but
their original form, their intracellular content and
their electrical properties have been kept as in living
conditions. Then sensor are again washed in
deionised water and dried just before measurement.
LABEL FREE BIO SENSING METHOD USING RADIO FREQUENCIES SPECTROSCOPY FOR CELL DETECTION
AND DISCRIMINATION
5

3 EXPERIMENTAL RESULTS
All characterizations are performed using classical
microwave measurement techniques with on wafer
probing; as it allows a quick and a successive sensor
measurements.
First, for each sensor, the transmitted microwave
signal attenuation across the unloaded resonator is
recorded using a calibrated vector network analyser.
Then cell culture is performed; at the end, loaded
sensors with fixed cells are measured following the
same procedure than before cell growth. Hence, the
induced resonator frequency shift value, related to
the cells electrical properties can be extracted.
Frequency (GHz)
S
2
1
(
d
B
)
Unloadedsensor
EM simulation
Measurement
Sensorafter cell growth
370MHz

Figure 7: Biosensor measured response before and after
the cell growth and simulated one.
Figure 7 shows results of experimentations with
glial-cells-derived tumour glioblastoma coming
from human nervous system cells. Used biosensors
initially resonate at 16 GHz and shift down to 15.63
GHz when it is loaded with only 8 glial-cells (figure
8).

Figure 8: Photograph of the sensor after the cell growth.
Fullwave simulations, based on finite element
method (HFSS from ANSOFT), are then used to
extract individual cell electrical properties. Cells EM
modelling is done assuming that they are
homogenous, source-free and linear dielectric
volume. Hence, on a narrow frequency bandwidth
around the sensor resonant frequency, cell global
permittivity and conductivity can be extracted with a
good accuracy by fitting simulations data with
measured one, as shown on figure 7.
Hence in the case of analysed glial-cells, we
have obtained an effective permittivity value of 36
1 while global conductivity has been estimated
0.100 0.003 S/m at 16 GHz and 20C. These
results agree very well with previous analysis done
with ficoll media (Dalmay et al., 2008) and can be
compared to the effective permittivity of pure water
which is closed to 45 at 20C. Other
characterizations are currently done with other
cellular types, to demonstrate that it is possible with
this approach to discriminate between different cell
types.
4 CONCLUSIONS
An original label free bio-sensing approach for
cellular analysis at radio frequencies has been
demonstrated. Thanks to their sub millimetric size,
used sensors are able to work at the cell scale with a
very limited number of cells and can potentially be a
novel promising tool for cell discrimination. Further
work is ongoing to evaluate experimentally the
minimum number of cell analysis achievable and to
improve the sensor design and experimental process
for one single cell analysis.
REFERENCES
Young-Il Kim, Yunkwon Park, Hong Koo Baik, 2007.
Development of LC resonator for label-free
biomolecule detection, Sensors and Actuators A.
B. Blad, and B. Baldetorp, 1996. Impedance spectra of
tumour tissue in comparison with normal tissue; a
possible clinical application for electrical impedance
tomography, Physiol.Meas., vol. 17, pp. 105- 115.
T. W. Athey, M. A. Stuchly, S. S. Stuchly, 1982.
Measurement of radio frequency permittivity of
biological tissues with an open-ended coaxial line :
Part I, IEEE Trans. Microwave Theory Tech,. vol.
82, pp. 82-86.
N. Denef , L. Moreno-Hagelsieb, G. Laurent, R. Pampina,
B. Foultier, J . Remacle, D. Flandre, 2004. RF
detection of DNA based on CMOS inductive and
capacitive sensors, EUMW Conference Digest, pp.
669-672.
C. Dalmay, 2008. Label free biosensors for human cell
characterization using radio and microwave
frequencies, IEEE MTT-S International Microwave
Symposium Digest, IMS 2008.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
6
EVALUATION OF PSD COMPONENTS AND AAR PARAMETERS AS
INPUT FEATURES FOR A SVMCLASSIFIER APPLIED TO A
ROBOTIC WHEELCHAIR
Andr e Ferreira, Teodiano Freire Bastos-Filho, M ario Sarcinelli-Filho
Department of Electrical Engineering, Federal University of Espirito Santo, Vit oria, Brazil
{andrefer,tfbastos,mario.sarcinelli}@ele.ufes.br
Jos e Luis Martn S anchez, Juan Carlos Garca Garca, Manuel Mazo Quintas
Department of Electronics, University of Alcala (UAH), Alcal a de Henares, Spain
{jlmartin,jcarlos,mazo}@depeca.uah.es
Keywords:
Adaptive autoregressive parameters, Power spectral density components, Support-vector machines, Brain-
computer interfaces, Robotic wheelchair.
Abstract:
Two distinct signal features suitable to be used as input to a Support-Vector Machine (SVM) classier in
an application involving hands motor imagery and the correspondent EEG signal are evaluated in this paper.
Such features are the Power Spectral Density (PSD) components and the Adaptive Autoregressive (AAR)
parameters. Different classication times (CT) and time intervals are evaluated, for the AAR-based and the
PSD-based features, respectively. The best result (an accuracy of 97.1%) is obtained when using PSD com-
ponents, while the AAR parameters generated an accuracy of 94.3%. The results also demonstrate that it is
possible to use only two EEG channels (bipolar conguration around C
3
and C
4
), discarding the bipolar con-
guration around C
z
. The algorithms were tested with a proprietary EEG data set involving 4 individuals and
with a data set provided by the University of Graz (Austria) as well. The resulting classication system is now
being implemented in a Brain-Computer Interface (BCI) used to guide a robotic wheelchair.
1 INTRODUCTION
A Brain-Computer Interface (BCI) is a system that in-
cludes a way of acquiring the signals generated by the
brain activity, a method/algorithm for decoding such
signals and a subsystem that associates the decoded
pattern to a behavior or action (Sajda et al., 2008).
The BCI and its inherent challenges, involving areas
such as signal processing, machine learning and neu-
rosciences, have been the focus of several important
research groups. The results of this new technology
could be applied to improve the quality of life of many
people affected by neuromotor disfunctions caused by
diseases, like amyotrophic lateral sclerosis (ALS), or
injuries, like spinal cord injury.
A basic structure of a BCI, according to the previ-
ous denition, is presented in Figure 1. This paper is
related to the phases of feature extraction and feature
translation or classication, both indicated in the g-
ure. The objective here is to evaluate Power Spectral
Density (PSD) components and Adaptive Autoregres-
sive parameters as inputs for a Support-Vector Ma-
chine (SVM) classier. The SVM is supposed to be
able to distinguish two mental tasks related to hands
motor imagery, based on these two features extracted
from the EEG signal. Two data sets (a proprietary
one acquired in the University of Alcala and one pro-
vided by the University of Graz) are used to evaluate
the implemented algorithms. Congurations of three
EEG channels (bipolar aroundC
3
, C
z
and C
4
) and two
EEG channels (bipolar around C
3
and C
4
) are tested.
The preliminaries and the results of such evalua-
tion are hereinafter presented as follows: Section 2
contextualizes this work, introducing some previous
works involving a robotic wheelchair commanded by
a BCI; the methodology used to reach the objective
is explained in Section 3, where the feature extraction
and the classier are described in details. The results
obtained with the two data sets aforementioned and
some comments are presented in Section 4, which is
followed by Section 5, where the main conclusions of
this work are highlighted.
7
Figure 1: Brain-Computer Interface available at UFES.
2 BACKGROUND
A robotic wheelchair commanded through a BCI is
being developed at the Federal University of Espirito
Santo, Brazil. The users of such BCI can select move-
ments to be executed by the wheelchair from a set of
options presented in the screen of a PDA connected to
the BCI, as illustrated in Figure 1.
A drawback of this approach is the need of eye-
closing to generate the desired pattern, in this case an
ERS (Pons, 2008). An user who is not able to close
the eyes for a while to select an option of movement,
for example, will not get any prot using the current
version of the BCI implemented in the wheelchair. In
order to overcome such problem, other EEG informa-
tion should be used.
In such a context, hands motor imagery is being
tested here, in connection to a SVM-based classier,
to check the possibility of using this approach to im-
plement a BCI to be used to command the robotic
wheelchair aforementioned. The idea underlying this
study is to use imaginary hand movements, instead of
eye-closing, to generate recognizable EEG patterns.
3 METHOD
The focus of this paper is to evaluate the use of PSD
components and AAR parameters, associated to EEG
signals acquired in the region of the motor cortex of
the human brain, as inputs of a classier based on a
SVM. The system is supposed to classify two differ-
ent mental tasks related to hands motor imagery, aim-
ing at allowing to implement a BCI to be used to com-
mand a robotic wheelchair (Pons, 2008). In order to
perform such evaluation, the following methodology
was carried out:
1. evaluate two different approaches: PSD-SVM and
AAR/RLS
1
-SVM, according to the sketch of Fig-
ure 2;
2. evaluate different channel congurations:
[C
3
C
z
C
4
] and [C
3
C
4
]
2
;
3. PSD approach: evaluate for different time inter-
vals (3-5s, 4-6s, 5-7s, 6-8s and 7-9s);
4. AAR/RLS approach: evaluate for different Clas-
sication Times (CT) (Schl ogl et al., 1997). The
CTs used are 3s, 4s, 5s, 6s, 7s and 8s;
5. evaluate the algorithms using the proprietary
UAH dataset and search for the best conguration
(feature extractor and SVM classier);
6. apply such conguration to the Graz dataset and
evaluate the results.
Figure 2: A representation of the systems being evaluated.
3.1 Graz Dataset
The Graz dataset was provided by the Department
of Medical Informatics, University of Graz (Austria),
during the BCI Competition 2003. It is named Data
set III and is related to motor imagery. In this pa-
per, 140 trials of this dataset, and the respective la-
bels, were used, 70 related to left hand motor imagery
and 70 related to right hand motor imagery. Each trial
lasts 9 seconds, with a sampling rate of 128 Hz, re-
sulting in 1152 samples/channel/trial. The data was
obtained using a bipolar conguration around the po-
sitions C
3
, C
z
and C
4
, according to the 10-20 Interna-
tional System, as presented in Figure 3. In the same
1
Recursive Least Squares
2
Actually, the channels are bipolar, with electrodes
placed around these positions, as shown in Figure 3
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
8
gure (on the right side), an illustration of the proto-
col used during the experimental phase is presented.
After the 2 initial seconds, a beep sounds and a cross
is presented in the center of the screen, calling the
subjects attention to the beginning of the experiment.
One second later (t = 3 s), an arrow pointing left or
right is presented to the operator, suggesting which
mental task should be accomplished, and lasts for 6
seconds (until t = 9 s). The data was ltered, keep-
ing the spectrum ranging from 0.5 Hz to 30 Hz, and
visual feedback was used (more details can be found
in (Schl ogl, 2003)).
Figure 3: Electrodes Placement and the experimental proto-
col associated to the Graz dataset.
3.2 UAH Dataset
Experiments similar to those described in Sec-
tion 3.1 were accomplished at the University of Al-
cala (UAH), Spain. The mental tasks are the same of
the Graz dataset, also related to hands motor imagery.
The dataset was recorded from 4 normal subjects in
different sessions. Each session corresponds to 60 tri-
als (half to each one of the two mental tasks consid-
ered) and each trial was 9 s long, resulting in 9 min-
utes/session. Three subjects participated in 3 sessions
and one subject participates in 4 sessions, thus result-
ing in 780 trials. The bio-signal amplier g.BSamp
and the subsystem g.16sys compound the g.tec sys-
tem used to record the EEG data, the software being
implemented in Matlab. The data was also ltered to
keep only the spectrum from 0.5 Hz and 30 Hz, but
the volunteer had no visual feedback.
3.3 Feature Extraction: PSD
Due to the fact that EEG rhythms have been de-
ned mainly in the frequency domain, the Power
Spectrum Density (PSD) analysis of the signal is the
non-parametric technique used for feature extraction.
Other reasons that motivate this choice are the com-
putational efciency involved, the direct relation be-
tween PSD and power, power components can be in-
terpreted in terms of cerebral rhythms and the esti-
mations (via FFT) of spectral components are not bi-
ased as those estimated via AR models, as described
in (Mouri no, 2003).
The PSDis estimated here via he Welchs Method,
computed over sections of 1 s, averaging spectral es-
timates of 3 segments of 0.5 s each (64 samples, sam-
pling rate of 128 Hz) with 50% of overlap between
segments. The maximum size of each segment is im-
portant in order to consider the stationary behavior of
the EEG signal (Mouri no, 2003; McEwen and Ander-
son, 1975). A weighting Hanning window is applied
to the signal due to its considerable attenuation in the
side-lobes. The spectral components extracted from
the signal and used as features spans from 8 Hz to
30 Hz, with a frequency resolution of 2 Hz. Thus,
12 components are generated, in connection to each
channel. This feature extraction procedure is illus-
trated in Figure 4.
3.4 Feature Extraction: AAR/RLS
The other technique used for feature extraction is
based on Adaptive Autoregressive parameters (AAR),
estimated via Recursive Least Squares (RLS) algo-
rithm, as described in (Schl ogl et al., 1997; Haykin,
2001). This procedure is performed according to
E
t
=Y
t
a
T
t1
Y
t1
(1)
r
t
= (1 UC)
1
A
t1
Y
t1
(2)
k
t
= r
t
/(Y
T
t1
r
t
+1) (3)
a
t
= a
t1
+k
t
E
t
(4)
A
t
= (1 UC)
1
A
t1
k
t
r
T
t
, (5)
where
a
t
= [a
1,t
. . . a
p,t
]
T
(6)
Y
t1
= [Y
t1
. . .Y
tp
]
T
. (7)
The initial values adopted were A
0
= I, a
0
= 0
and UC = 0.007, and the model order was chosen as
p =6. Although the RLS algorithmhas a higher com-
putational complexity in comparison with the Least
Mean Squares (LMS), it has some advantages: the
faster convergence, the higher accuracy of the esti-
mate and the fact that no matrix inversion is neces-
sary. Figure 5 shows the temporal evolution of six
AAR parameters. In this case, the channel C
3
of the
rst trial included in the Graz dataset was considered.
3.5 Classier: SVM
Although the concept of Support-Vector Machines
(SVM) was introduced in COLT-92 (Fifth An-
nual Workshop on Computational Learning Theory)
(Boser et al., 1992), its evaluation in BCIs is quite re-
cent.
Briey speaking, the main idea of a SVM is to
nd an optimal separating hyperplane for a given fea-
ture set. Given a training set of instance-label pairs
EVALUATION OF PSD COMPONENTS AND AAR PARAMETERS AS INPUT FEATURES FOR A SVM
CLASSIFIER APPLIED TO A ROBOTIC WHEELCHAIR
9
Figure 4: Example of feature extraction using PSD components. Signals related to C
3
, C
z
and C
4
(bipolar) during hands motor
imagery. PSD is presented from 8 up to 30 Hz in dB/Hz.
Figure 5: Temporal evolution of six AAR parameters.
(x
i
, y
i
), i = 1, . . . , l, where x
i
R
n
and y {1, 1}
l
,
the SVM requires the solution of the optimization
problem
min
w,b,
1
2
w
T
w+C
l

i=1

i
, (8)
subject to
y
i
(w
T
(x
i
) +b) 1
i
(9)

i
0. (10)
Training vectors x
i
are mapped into a higher di-
mensional space (maybe innite) by the function .
The SVM nds a linear separating hyperplane with
the maximal margin in this higher dimensional space.
C > 0 is the penalty parameter of the error term. The
function K(x
i
, x
j
) (x
i
)
T
(x
j
) is called kernel. The
kernel function used in this paper is a Radial Basis
Function (RBF) dened as
K(x
i
, x
j
) = exp(x
i
x
j
)
2
), > 0. (11)
The choice of a SVM-based classier and a RBF
kernel function relies on previous works that consid-
ered this conguration (Shoker et al., 2005; Guler
and Ubeyli, 2007; Khachab et al., 2007). Further-
more, a SVM classier has improved the accuracy
in 13% when compared to LDA (Linear Discriminant
Analysis) and 16.3% when compared to NN (Neural
Networks), using the same features (Nicolaou et al.,
2008).
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
10
Table 1: PSD + SVM (UAH dataset).
2 channels (C
3
C
4
) 3 channels (C
3
C
z
C
4
)
Subject 3-5s 4-6s 5-7s 6-8s 7-9s 3-5s 4-6s 5-7s 6-8s 7-9s
S
01
71.1 71.1 73.3 73.3 66.7 73.3 80.0 66.7 75.6 64.4
S
02
68.7 80.0 73.3 73.3 71.1 75.6 75.6 73.3 73.3 66.7
S
03
68.3 76.7 66.7 66.7 66.7 68.3 70.0 66.7 65.0 70.0
S
04
75.6 91.1 82.2 84.4 84.4 73.3 86.7 82.2 84.4 75.6
Table 2: AAR/RLS + SVM (UAH dataset).
2 channels (C
3
C
4
) 3 channels (C
3
C
z
C
4
)
Subject 3s 4s 5s 6s 7s 8s 3s 4s 5s 6s 7s 8s
S
01
66.7 73.3 66.7 68.9 62.2 73.3 66.7 73.3 64.4 66.7 71.1 66.7
S
02
66.7 68.9 66.7 80.0 57.8 57.8 68.9 68.9 64.4 73.3 64.4 64.4
S
03
60.0 66.7 73.3 61.7 66.7 73.3 58.3 61.7 68.3 66.7 65.0 70.0
S
04
66.7 71.1 86.7 82.2 75.6 73.3 66.7 73.3 86.7 77.8 71.1 75.5
Table 3: PSD + SVM (Graz dataset).
2 channels (C
3
C
4
) 3 channels (C
3
C
z
C
4
)
Subject 3-5s 4-6s 5-7s 6-8s 7-9s 3-5s 4-6s 5-7s 6-8s 7-9s
S
Graz
88.6 97.1 85.7 74.3 77.1 85.7 94.3 85.7 80.0 71.4
Table 4: AAR/RLS + SVM (Graz dataset).
2 channels (C
3
C
4
) 3 channels (C
3
C
z
C
4
)
Subject 3s 4s 5s 6s 7s 8s 3s 4s 5s 6s 7s 8s
S
Graz
65.7 74.3 91.4 91.4 82.9 80.0 74.3 68.6 91.4 91.4 80.0 80.0
The scripts developed during this work are based
on the library libsvm (Chang and Lin, 2001).
4 RESULTS
Taking into account the data distribution, 75%of each
dataset was used for training and validation, while
the other 25% were used for test. After evaluating
two different techniques for feature extraction (based
on PSD components and AAR parameters), the re-
sults are presented in the following four tables. The
rst one (Table 1) shows the classication accuracy
obtained for each subject of the UAH dataset, when
PSD+SVM is used. The gray cells represents the
best classication accuracy found for each subject.
The higher values are related to the central period of
the experiment (4-6s) and, except by the subject S
01
,
these values are obtained with only two channels.
Table 2 contains the results for the other explored
conguration (AAR/RLS+SVM). Four subjects of the
UAH dataset are evaluated using different CTs and
channel conguration. Once more, the gray cells rep-
resents the best classication values for each subject
during the test. Equal values are all highlighted (gray
cells) to show in which situations they can appear. As
in PSD case, the higher classication rates are related
to the middle of the experiment (Table 1 and Table 2
(4-6s)). The best results can also be reached with only
2 channels, taking into account that all the high values
obtained with 3 channels appear on the left side of the
Table 2 (2 channels).
Thus, the best results with the UAH dataset can be
found using PSD+SVM, 2 channels (C
3
C
4
) and in the
middle of the experiment. A summary of the results
is presented in Table 5.
Table 5: Best Results (UAH dataset).
Subject Accuracy Conguration
S
01
80.0 PSD+SVM,C
3
C
z
C
4
,4-6s
S
02
80.0 PSD+SVM,C
3
C
4
,4-6s
S
03
76.7 PSD+SVM,C
3
C
4
,4-6s
S
04
91.1 PSD+SVM,C
3
C
4
,4-6s
As the next step of the proposed methodology, this
conguration was applied to the Graz dataset, in order
to evaluate it. The results obtained for this congura-
tion and the other are shown in Tables 3 e 4.
EVALUATION OF PSD COMPONENTS AND AAR PARAMETERS AS INPUT FEATURES FOR A SVM
CLASSIFIER APPLIED TO A ROBOTIC WHEELCHAIR
11
5 CONCLUSIONS
This paper evaluates the use of two set of features
(PSD components and AAR/RLS parameters of an
EEG signal) as inputs for a SVM classier, in order to
distinguish between two mental tasks related to hands
motor imagery.
The approach based on PSD (Welchs Method)
components and a SVM (RBF kernel) generated the
best results. The highest classication rates are re-
lated to the middle of the experiment, usually be-
tween seconds 4 and 6. It can be explained taking
into account that the subject needs some time to setup
him/herself (the cue, an arrow, is presented to the sub-
ject at instant t = 3 s) to the end of the trial (the trial
nishes at t = 9 s).
The best results can be accomplished using only
2 channels, four electrodes placed around positions
[C
3
C
4
] of the 10-20 International System.
After evaluating the system with the UAH dataset,
the algorithms were applied to the Graz dataset and
the best classication rate (accuracy) was 97.1%
(99.4% to mental task 1 and 93.5% to mental task 2).
The replacement of the method currently used to
select a symbol in a PDA, which requires a brief eye-
closing, by another based on motor imagery, such as
the one here discussed, is the next step of our research.
In other words, the idea is that motor imagery of a
hand (the one with higher accuracy) it is enough to
the user without eyes control to select desired symbols
in the PDA that will be translated into commands to
the robotic wheelchair or into some communication
outputs, also available in this system.
ACKNOWLEDGEMENTS
The authors thank CAPES, a foundation of the
Brazilian Ministry of Education (Project 150/07),
and FACITEC/PMV, a fund of the Vitoria City Hall
to support scientic and technological development
(Process 061/2007), for their nancial support to this
research.
REFERENCES
Boser, B. E., Guyon, I. M., and Vapnik, V. N. (1992). A
training algorithm for optimal margin classiers. In
COLT92: Proceedings of the Fifth Annual Workshop
on Computational Learning Theory, pages 144152,
New York, USA.
Chang, C.-C. and Lin, C.-J. (2001). LIBSVM: a library for
support vector machines. Software available at http:
//www.csie.ntu.edu.tw/

cjlin/libsvm.
Guler, I. and Ubeyli, E. (2007). Multiclass support vector
machines for EEG-signals classication. 11(2):117
126.
Haykin, S. (2001). Adaptive Filter Theory (4th Edition).
Prentice Hall.
Khachab, M., Kaakour, S., and Mokbel, C. (2007). Brain
imaging and support vector machines for brain com-
puter interface. In Proc. 4th IEEE International Sym-
posium on Biomedical Imaging: From Nano to Macro
ISBI 2007, pages 10321035.
McEwen, J. A. and Anderson, G. B. (1975). Modeling the
stationarity and gaussianity of spontaneous electroen-
cephalographic activity. (5):361369.
Mouri no, J. (2003). EEG-based Analysis for the Design
of Adaptive Brain Interfaces. PhD thesis, Universitat
Polit` ecnica de Catalunya, Barcelona, Spain.
Nicolaou, N., Georgeou, J., and Polycarpou, M. (2008).
Autoregressive features for a thought-to-speech con-
verter. In Proceedings of the International Conference
on Biomedical Electronics and Devices - BIODE-
VICES 2008, pages 1116, Funchal, Portugal.
Pons, J. L. (2008). Wearable Robots: Biomechatronic Ex-
oskeletons (1st Edition). Wiley, Madrid, Spain.
Sajda, P., Muller, K.-R., and Shenoy, K. (2008). Brain-
computer interfaces [from the guest editors]. Signal
Processing Magazine, IEEE, 25(1):1617.
Schl ogl, A. (2003). Data set: BCI-experiment.
http://ida.first.fraunhofer.de/projects/
bci/competition\_ii.
Schl ogl, A., Neuper, C., and Pfurtscheller, G. (1997). Sub-
ject specic EEG patterns during motor imaginary
[sic.: for imaginary read imagery]. In Neuper, C., edi-
tor, Proc. 19th Annual International Conference of the
IEEE Engineering in Medicine and Biology society,
volume 4, pages 15301532.
Shoker, L., Sanei, S., and Sumich, A. (2005). Distin-
guishing between left and right nger movement from
EEG using svm. In Sanei, S., editor, Proc. 27th
Annual International Conference of the Engineering
in Medicine and Biology Society IEEE-EMBS 2005,
pages 54205423.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
12
INVESTIGATION OF OPERATING PARAMETERS FOR A
SEMEN QUALITY ANALYSIS SYSTEM
S. Atherton, C. R. Evans, P. Roach, D. C. Hughes, G. McHale and M. I. Newton
School of Science and Technology
Nottingham Trent University, Clifton Lane, Nottingham
NG11 8NS, U.K.
[email protected]
Keywords: Sperm, Semen, Motility, Acoustic wave, QCM, Time of flight.
Abstract: To increase the success rate of Artificial Insemination (AI) in animals, it is important that the semen sample
is of a high quality. The quality is related to both the number and motility of sperm present. Numerous
methods of analysing semen samples exist, but these are generally expensive and/or laboratory based. A
useful alternative would be an inexpensive simple system that could be used in the field immediately prior
to insemination. We present a time of flight (ToF) technique using a quartz crystal microbalance (QCM). In
this system the sperm are introduced at one end of a liquid filled swim channel and self propel to a QCM
sensor at the other end. A chemical coating is applied to the QCM to bind the sperm and from the frequency
change the number of attached sperm and their ToF can be measured. We report the effect of temperature
and the introduction of small quantities of progesterone into the swim channel on the sperm ToF. Results
show the QCM can be used to detect the arrival of the sperm and that increasing temperature and the
presence of progesterone are both shown to decrease the ToF.
1 INTRODUCTION
Within the Artificial Insemination (AI) industry it is
important to be able to measure the concentration of
viable sperm in a sample. AI is a common procedure
in farm animals, more than 100 million inseminations
are performed globally every year. It is not only the
number of sperm in a semen sample that is crucial to
the success of the insemination process, but also the
motility of the sperm. Whilst having a large number
of motile sperm in a sample is not a guarantee of
fertility, it is an excellent indicator of semen quality.
Two optical methods of performing sperm
counts are the haemocytometer and counting
chambers. The drawback of these methods is that
multiple measurements are needed to achieve an
acceptable level of precision, resulting in a more time
consuming procedure. The reason for the relative
inaccuracy of these methods is due to the rapid
movement of the sperm under high magnification
and the tedious nature of the work for the human
operator. It is possible to perform a more objective
assessment of sperm motility using a computer
assisted semen analyzer (Mortimer, 2000), which is a
laboratory based instrument that can measure
different aspects of the sperm movement. To further
increase precision, a combination of fluorescent
staining and flow cytometry can be used to analyze
thousands of sperm in a sample (Christensen et al,
2005). The common drawback with all of the above
techniques is the price of the equipment itself and the
need for a skilled operator.
In some species the sample may be successfully
frozen and thawed before use however this is not
always the case. The sperm of other species cannot
be successfully frozen and so a fresh sample, with a
shelf life of only a few days, must be used.
Increasingly, AI is being used for equine applications
and recent changes in UK legislation (Artificial
Insemination of Mares Order 2004) mean that lay
(non-Veterinary) and farm workers are now being
trained to perform such inseminations. For sport
equine applications the cost of semen for a single
mare to be covered may exceed 1000. Semen
analysis is currently a specialist, subjective and
skilled process that is normally carried out under
laboratory conditions. Given these trends, a low cost,
simple to use and objective technique to assess the
quality of the semen, particularly under field
conditions just before insemination, would greatly
13
improve the quality and practice of artificial
insemination in animals. It would also provide an
easier and more cost effective method for monitoring
male animal fertility and breeding male welfare.
Acoustic wave sensors detect very small changes
in mass attached to their surface and often contain a
sensitizing layer that can recognize and bind the
species to be detected onto the mass sensitive
surface. The quartz crystal microbalance (QCM) is
the most widely used acoustic wave device for sensor
applications.
f =-2.26 x 10
-6
f
2
m/A (1)
The Sauerbrey equation (Sauerbrey, 1959) relates
the change of the crystals resonant frequency to the
change in rigid mass on the crystal surface; this is
shown for AT cut quartz in equation 1 where f (in
Hz) is the change in frequency that occurs for an
increase in mass m (in grams) on the surface of area
A (in cm
2
) with a crystal resonant frequency of f (in
Hz) and the constant comes from the crystal
materials properties. A well-designed oscillator
circuit can still resonate a crystal even under the high
damping caused by immersion in a liquid. The
change in mass rigidly attached to the surface still
causes a proportional change in frequency although
changes in other parameters such as the liquids
viscosity and density will also cause changes in
frequency. The acoustic wave will only sense mass
changes within a short distance into the liquid called
the penetration depth. (Kanazawa & Gordon, 1985)
Previous studies have shown up to 70% of the sperm
mass to be made up of water (Da Silva et al, 1992) so
it is not obvious how the attachment of a sperm will
change the QCM response. In a preliminary report
(Newton et al, 2007) we have fitted the resonance
curves of 5MHz QCM to the Butterworth van Dyke
model and this has shown that the sperm may be
treated a rigid mass and so a model based on the
Sauerbrey equation is appropriate when using an
effective mass of around 5pg. For other operating
frequencies or other species sperm this effective mass
would be different.
In this report we extend this preliminary work to
investigate the effect of environmental parameters on
the time of flight (ToF). For any practical
measurement technique it is essential the time the
measurement takes is sufficiently short to be usable.
For a portable field instrument then power
consumption may also be an issue therefore the first
parameter we investigate is operating temperature
and we consider a range from room temperature to
body temperature.
Progesterone is a steroid hormone involved in
female menstrual cycle, pregnancy and
embryogenesis of humans and other species. It is one
of a number of substances said to cause
hyperactivation of mammalian spermatozoa and its
presence may therefore affect the time of flight; the
effect of adding progesterone to the swim medium is
reported.

Figure 1: Magnified view of a boar sperm.
2 EXPERIMENTAL
Figure 2 shows a schematic diagram of the
experiment. This consists of an inlet port to a channel
filled with phosphate buffered saline (PBS) buffer.
At the other end of the channel is a quartz crystal
followed by a vent to air preventing pressure changes
being recorded in the QCM response when a semen
sample is added. Sperm are introduced at the inlet
port and are self propelled through the channel to the
QCM where they are detected. A volume of 20l of
the semen was used and added using a Gilson pipette.
The channel length was set to approximately 14.5 cm
and contained 4ml of PBS; note that for any practical
field instrument the swim channel length could be
considerably reduced to give an analysis time under 5
minutes. The sensing element in the experiments was
a 5MHz AT-cut quartz crystal (Testbourne 149211-
1). A Maxtek PLO-10 phase lock oscillator was used
to drive the crystal and the resonant frequency was
measured with an Agilent universal frequency
counter interfaced to a computer.
To sense the sperm it was necessary to get them
to adhere to the surface of the QCM. To achieve
sperm adhesion to the crystals they were coated in
either Poly-L-Lysine (Sigma-Aldrich) or cysteamine
(Sigma-Aldrich). Crystals were initially cleaning
with ethanol then ozone treated for 30 minutes. They
were then placed in either poly-L-lysine (as
supplied), or a cysteamine solution of 1mmol in
toluene and left overnight. The devices were then
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
14
washed in PBS buffer to remove any excess. The
cysteamine coating were found to be the most
reliable method of binding the sperm to the surface as
poly-L-lysine shows significant variability from
batch to batch.
Side view Top view
sample inlet port
buffer filled
channel
Quartz
crystal
inlet port
buffer filled
channel
vent
Spermdetection region

Figure 2: Schematic diagram of swim channel.
To provide temperature control to the
experiments the swim channel was housed inside an
Octagon 10 incubator. This allowed the temperature
to be controlled so as to investigate the effect this
would have on the sperm ToF. For these the
temperature was varied from 23-45
o
C. The incubator
also allowed the temperature to be kept constant
across all the progesterone experiments.
When looking at the effects of progesterone,
(Sigma-Aldrich) different concentration were added
to the PBS in the swim channel. Firstly 15.7mg of
progesterone was dissolved in 50ml of ethanol. This
mixture was diluted in PBS at concentration of 20-
90mol.
The porcine semen was supplied by a
commercial artificial insemination centre (J SR
Genetics, Driffield, UK). The semen was received
already mixed with a dilutant (Androhep), cooled to
a temperature of 17
o
C and packaged in plastic
bottles. The androhep allowed the semen to be
stored for up to 5 days at ambient temperature.
However, this does result in the concentration of
semen in the mixture being quite low. To get a more
concentrated sample a centrifuge was used to
separate the sperm from the androhep. To achieve
this, sealable microtubes were filled with 50l of the
androhep and semen mixture and these were
centrifuged for 40 seconds. This resulted in the
sperm being concentrated at the bottom of the tube.
The androhep was removed with the pipette and
50l of PBS was added to one of the tubes. The PBS
and sperm were mixed together, this mixture was
removed with the pipette and added to the next tube
and the process was repeated for all 15 tubes. What
was left was a more concentrated sperm sample
mixed with 50l of PBS.
3 RESULTS AND DISCUSSION
Figure 3 shows the QCM frequency response to
sperm binding on the surface. The arrow shows the
time at which the semen sample was introduce to the
inlet of the swim channel. The arrival of the sperm is
signified by a decrease in the frequency of the sensor
with the fastest ToF of approximately 20 minutes.
-45
-35
-25
-15
-5
0 50 100 150
time (minutes)
f
r
e
q
u
e
n
c
y

c
h
a
n
g
e

(
H
z
)

Figure 3: Graph showing the frequency decrease
indicating sperm arrival.
The frequency continues to drop as more sperm
make their way to the QCM and bind to the surface.
This continues for another 120 minutes until further
arrival of motile sperm finishes.
Using the previously determined sperm effective
mass and taking the rate of frequency change from
figure 1, the Sauerbrey equation can be used to
derive the rate of sperm arrival and this is shown in
figure 3. For use in a screening application, a simple
threshold number of detected sperm would be
required however this demonstrates that quantitative
analysis is also possible with this instrument.

Figure 4: Number of sperm arriving at the QCM over the
course of the experiment.
Figure 5 is a plot of the ToF of the sperm against the
temperature of the environment. The results show a
decrease in the ToF as temperature increases with
almost a 50% fall between room temperature and
body temperature. The scatter observed can be
attributed mainly to the experiments being
performed a differing lengths of time from the
INVESTIGATION OF OPERATING PARAMETERS FOR A SEMEN QUALITY ANALYSIS SYSTEM
15
receipt of the samples and the quality of the semen
degrades over time.
10
12
14
16
18
20
22
24
22 27 32 37 42
Temperature (Deg C)
T
o
F

(
m
i
n
)

Figure 5: The time of flight for fastest sperm arrival as a
function of temperature.
To further speed up the measurement,
progesterone was used to cause hyperactivation in an
attempt to decrease the sperm ToF; the results of this
are shown in figure 6. Comparing the non-
progesterone experiments with the progesterone
ones we see a significant decrease in the ToF of the
sperm however for the full range investigated there
was little effect from the progesterone concentration.
300
400
500
600
700
800
900
0.0 20.0 40.0 60.0 80.0 100.0
Progesterone concentration (mol)
T
i
m
e

o
f

f
l
i
g
h
t

(
s
)
Progesterone
No progesterone

Figure 6: Time of flight as a function of progesterone
concentration.
4 CONCLUSIONS
We have demonstrated that a time of flight
technique with an acoustic wave sensor provides a
viable method for determining the quality of a
semen sample both as a screening technique and as
an analytical tool. The cysteamine coating on the
QCM proved to be the more reliable method of
binding the sperm to the surface. Experiments
varying the temperature showed a general decrease
in ToF as temperature is increased suggesting that
body temperature would be the optimum value. The
presence of progesterone also reduces the ToF
however this was not concentration dependent over
the range investigated. Whilst the laboratory based
instrument reported here used commercial sensor
crystals, the cost of quartz crystals employed more
generally in electronic oscillator circuits are
inexpensive and are still offer a mass sensitive
surface. Using such crystals, pre-treated with
cysteamine, would reduce costs sufficiently to offer
the possibility of a disposable element. With a
modification to the swim channel length to bring
down the measurement time, this technique then
becomes a powerful tool for routine monitoring of
animal reproductive health and investigating factors
that affect the semen motility of animal.
REFERENCES
Mortimer, S. T., 2000, CASA - Practical Aspects, J .ournal
Andrology, 21, p.515-524
Christensen, P., Boelling, D., Pedersen, K. M., Korsgaard,
I. R., & J ensen, J ., 2005, J ournal of Andrology, 26
Sauerbrey, G., 1959, Zeitschrift fr Physik, p.155-206
Da Silva, L. B., Trebes, J . E., Balhorn, R., Mrowaka, S.,
Anderson, E., Attwood, D. T., Barbee, T. W., Brase, J .,
Corzett, M., Gray, J ., Koch, J . A., Lee, C., Kern, D.,
London, R. A., MacGowan, B. J ., Matthews, D. L., &
Stone, G., 1992, X-ray laser microscopy of rat sperm
nuclei, Science, 258, p.269-271
Kanazawa, K., & Gordon, J. G., 1985, Frequency of a
Quartz Crystal Microbalance in contact with a liquid,
J ournal of Analytical Chemistry, 57, p.1770-1771
Newton, M. I., Evans, C. R., Simons, J . J ., & Hughes, D.
C., 2007, Semen quality detection using time of flight
and acoustic wave sensors, Applied Physics Letters
90(15)
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
16
MULTIFOCAL ELECTRORETINOGRAPHY
Early Detection of Glaucoma based on Wavelets and Morphological Analysis
J . M. Miguel, S. Ortega, I. Artacho, L. Boquete, J . M. Rodrguez
Department of Electronics, University of Alcal, 28701 Alcal de Henares, Spain
[email protected], [email protected]
P. De La Villa
Department of Physiology, University of Alcal, 28701 Alcal de Henares, Spain
[email protected]
R. Blanco
Department of Surgery, University of Alcal, 28701 Alcal de Henares, Spain
[email protected]
Keywords: Wavelet transforms, Glaucoma, m-sequence, Multifocal electroretinogram, Morphological analysis.
Abstract: This article presents one of the alternative methods developed for the early detection of ocular glaucoma
based on the characterisation of mfERG (multifocal electroretinography) readings. The digital signal
processing technique is based on Wavelets, hitherto unused in this field, for detection of advanced-stage
glaucoma and the study of signal morphology by means of identity patterns for detection of glaucoma in
earlier stages. Future research possibilities are also mentioned, such as the study of orientation in the
development of the disease.
1 INTRODUCTION
Glaucoma is currently deemed to be a high-risk eye
disease since a large percentage of the population
suffer from its effects. The method proposed herein
has been developed for study and analysis of OAG
(open angle glaucoma), the commonest form in
todays society.
The sheer complexity of the disease and its
occultation make early and reliable detection
essential. The traditional techniques for clinical
analysis of the retina are based on indirect methods
(measurement of the intraocular pressure, visual
inspection of the eyeground, campimetric tests, etc).
Their main drawback is that they do not give
objective information on the functioning of the
retinal photoreceptors (Catal et al., 2005), essential
elements in the perception of light energy. A new
technique has recently been developed for obtaining
this retina-functioning information in a quick and
reproducible way; this technique is known as the
multifocal electroretinogram (mfERG). The mfERG
enables a functional exploration to be made of the
light sensitivity of the retinal cells and also the
spatial distribution of this sensitivity (J . M. Miguel
et al., 2007). The mfERG basically involves
recording the variations in retinal potential evoked
by a light stimulus and then mapping out the results
in a 2D or 3D diagram showing those regions that
respond to the visual stimuli (Sutter & Tran, 1992)
(Sutter EE., 2001).
The mfERG technique allows simultaneous
recording of local responses from many different
regions of the retina, building up a map of its
sensitivities. As in the conventional
electroretinogram (ERG), also called the full-field
electroretinogram, the potential is measured as the
sum of the electric activity of the retina cells. In the
full-field ERG, however, the signal recorded comes
from the whole retina surface, so it is hard to detect
smaller one-off defects that do not affect the whole
retina. The mfERG, by contrast, gives detailed
topographical information of each zone and can
therefore detect small-area local lesions in the retina
and even in its central region (fovea) (D. C. Hood et
al., 2003).
From the technical point of view, equipment is
needed for capturing the visually evoked potentials
17

at retina level (presented as a set of hexagons of
varying sizes and intensities). Due to the low
amplitude of the signals generated (down to
nanovolt level), the technique calls for suitable
hardware equipment (recording electrodes,
instrumentation amplifiers, digitalisation, etc) and
also signal processing algorithms (filtering,
averaging or smoothing procedures, rejection of
artefacts, etc) to ensure that the results are clinically
useful (M. F. Marmor et al., 2003).
This paper gives a description of the recording
and arrangement of the signals we have used in our
research, the signal analysis by the Wavelet
transform for recording possible advanced-stage
glaucoma markers, the detection of smaller lesions
by means of morphological analysis of the signal; it
also mentions possible future research lines.
2 METHODS
2.1 Obtaining the Signals
A total of 50 patients with diagnosis of advanced
open angle glaucoma (OAG) as well as an identical
number of healthy subjects were included in our
mfERG record database, used for obtaining markers
by means of the wavelet transform. Moreover, to
study the efficiency of our morphological analysis, a
second database was drawn up formed by 15
patients diagnosed with early-stage Glaucoma plus
an identical number of healthy controls.
The signal recording system was the VERIS 5.1
multifocal recording system (Electro-Diagnostic
Imaging, San Mateo, USA). The stimulus consisted
of an m-sequence applied to a group of 103
hexagons, as shown in figure 1, displayed on a 21-
inch monitor and covering a 45 arc of the retina.
The local luminance of each hexagon was 200 cd/m
2

in the on phase and less than 1.5 cd/m
2
in the off
phase, determined by the pseudorandom sequence.
The monitor frequency was 75 Hz and the m-
sequence was modified so that each step was
followed by 4 frames in the following order: flash-
dark-flash-dark, as shown in figure 2. In the flash
frames all the hexagons were illuminated with a
maximum luminance of 200 cd/m
2
, with a minimum
luminance of less than 1.5 cd/m
2
in the dark frames.
The background luminance of the rest of the monitor
surface surrounding the hexagons was held steady at
100 cd/m
2
. This stimulation protocol is especially
adapted for obtaining responses from the retinal
ganglion cells and their axons (Hagan R. P. et al.,
2006). It is based on the effect of the focal responses
(M) on the following global stimulus (F), which
amplifies the signals coming from the ganglion cells.

Figure 1: Geometry of the multifocal stimulus and
regrouping of the hexagons.
Basically, the protocol (M-F-O-F-O) consists of
five steps. In the first step (M) each hexagon follows
a luminous stimulation (200 cd/m
2
) determined by a
pseudorandom binary m-sequence. In the second
step the whole area is illuminated (200 cd/m
2
) (F),
followed by a dark sequence (O) (<1.5 cd/m
2
),
followed by another global flash (200 cd/m
2
) (F) and
then darkness again (O) (<1.5 cd/m
2
). This
stimulation will give us an acceptable signal-to-
noise ratio and also ensures a reasonably short
recording time (9 minutes).

Figure 2: Modification of the m-sequence.
The stimulus was displayed through
pharmacologically dilated pupils (minimum
diameter of 7 millimetres) using a Burian-Allen
bipolar contact lens (Hansen ophthalmics, Iowa
City, IA). Contact lens adaptation was facilitated by
a drop of topical anaesthetic (0.5% Proparacaine).
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
18

The residual spherical refractive error was corrected
by the VERIS autorefractor, mounted on the
stimulation monitor. The alignment of the patients
pupil with the monitor optic and the fixation stability
are controlled by an attached infrared camera. Each
monocular recording lasts about 9 minutes
(exponent of the stimulation m-sequence =13). To
make the process more comfortable for the patient,
the recording process was divided into eighteen 30-
second segments. Segments contaminated with
ocular movements were discarded and recorded
anew. The signals are amplified with a Grass
Neurodata Model 15ST amplification system (Grass
Telefactor, NH), with a 50,000 gain, filters with 10-
300 Hz bandwidth and a sampling interval of 0.83
milliseconds (1200 Hz).
Each participant was given a complete
ophthalmic exam, including general anamnesis, best-
corrected visual acuity, slit lamp biomicroscopy,
intraocular-pressure measurement using the
Goldmann applanation tonometer, gonioscopy,
dilated fundoscopic examination (90D lens), stereo
retinographs and a 24-2 SITA Humphrey automated
perimetry (Swedish Interactive Threshold
Algorithm. Carl Zeiss Meditec Inc.). A diagnosis of
open angle glaucoma was established where there
were at least two consecutive abnormal visual fields
in the Humphrey campimetry, (threshold test 24-2),
defined by: 1) a pattern standard deviation (PSD)
and/or corrected pattern standard deviation (CPSD)
below the 95% confidence interval; or 2) a
Glaucoma Hemifield Test outside the normal limits.
We define as abnormal an altitudinal hemifield in
the Humphrey visual field analysis giving three or
more contiguous sectors below the 95% confidence
interval, with at least one of them below the 99%
confidence interval. The visual field was dismissed
as unreliable if the rate of false positives, false
negatives or fixation losses was higher than 33%. A
control database was also established on the basis of
normal eye records established within the
longitudinal prospective study. All these normal eye
records had an intraocular pressure of 21 mmHg or
less (with no previous history of ocular
hypertension). An ophthalmic examination of the
optic papilla was also conducted to check that it fell
within the normal structural parameters.
The signals obtained from the 103 hexagons
were regrouped and averaged to build up a new 56-
sector map as shown in figure 1. The purpose of this
regrouping was to simplify the analysis and to
improve the signal-to-noise ratio. A 56-sector
topography was therefore chosen, similar to that
studied in automated campimetry, the clinical gold-
standard for evaluating the visual field. It should
also be noted here that sector 41 is the average of a
greater number of hexagons, since it is the area
containing the blind spot and, as such, more difficult
to analyse.
Two mfERG record databases were built up, one
containing healthy or control individuals and the
other glaucoma-affected individuals for study by
means of the Discrete Wavelet Transform (DWT).
Two other specific databases were also created to be
studied by means of an alternative technique,
Morphological Analysis, all made up by a complete
56-sector map as shown in figure 1.
Not all the sectors making up the map to be
analysed by the Wavelet Transform belonged to a
single patient; the map groups together 56 clearly
glaucoma-identified sectors from among the fifty
patients diagnosed with the same symptom.
Following a similar procedure, a sector map
comprising the control database was built up, this
time on the basis of healthy individuals.
As regards the databases used for the
morphological analysis, these were made up by two
15-record collections from the 56 sectors: the first
coming from 15 patients affected with early-stage
OAG and showing between 3 and 12 diseased
sectors, and the other built up from the 15 healthy
control subjects.
2.2 Study of Severe Lesions by Wavelet
Analysis
DWT was better than morphological analysis as a
mfERG-record analysis tool for detecting severe
retina lesions. Conversely, morphological analysis
was much more efficient for detecting early-stage
glaucoma by extracting certain markers present in
the records.
The great drawback of the Fourier transform-
based analysis is that the time information is
forfeited when the signal is transformed into the
frequency domain. The drawback is particularly
telling when the signal to be analysed is transitory in
nature or of finite duration, as in the case of mfERG
signals, whose frequency content changes over time.
The discrete wavelet transform (DWT) surmounts
this drawback by analysing the signal in different
frequencies with different resolutions, using regions
with windowing of different sizes and obtaining a
two-dimensional time-frequency function as a result.
Wavelet analysis uses finite-length, oscillating, zero-
mean wave forms, which tend to be irregular and
asymmetrical. These are the windowing functions
called mother wavelets. In principle there may be an
MULTIFOCAL ELECTRORETINOGRAPHY - Early Detection of Glaucoma based on Wavelets and Morphological
Analysis
19

infinite number of possible waves that are eligible
for use as wavelets, but in practice a more limited
number of wavelets are used, of well-known
characteristics, efficacy and implementation: Haar,
Daubechies, Coiflets, Mexican Hat, Symlets, Morlet,
Meyer, etc. In the study we are dealing with here a
great number of them were explored; it was with the
Bior3.1 wavelet that the best subjective results were
obtained for visual identification of certain markers
that help us to differentiate normal mfERG signals
from those belonging to subjects with advanced
glaucoma (J . M. Miguel et al., 2008).
The signal to be analysed is decomposed on the
basis of shifted and dilated versions of the mother
wavelet or analysing wavelet that we have decided
to use; this is all done by means of the correlation
between the signal to be decomposed and the
abovementioned versions of the mother wavelet.
Mathematically, the discrete wavelet transform
(DWT) is defined as:
/2
( , ) ( )2 (2 )
j j
n z
C j k f n n k

(1)
where the resulting C(j,k) is a series of coefficients
indicating the correlation between the function f(n)
to be decomposed and the wavelet a,b(t) dilated to
a scale a=2
j
and with a shifting b=k2
j
, with j,k Z.
The resulting C(j,k) includes time and frequency
information of the function f(n), according to the
values of j and k, respectively. In practice we obtain
two sets of time-function signals, one of them made
up by the signals A
1
to A
n
which represent
successive approximations of increasing smoothness
or declining frequency of the signal f(n), and the
other by D
1
to D
n
which represent the successive
details, also of falling frequency.
The signals were analysed by applying up to 5
levels of wavelet decomposition to each one of the
different sectors and for two different time windows:
one from 10 to 190 ms and another from 60 to 90
ms. The first contains the global response to the
multifocal stimulus used here and the second
contains the most important information on the
induced response generated by this type of stimulus.
Several superimposed records were obtained from
different sectors to obtain an overview of the
markers that might differentiate normal signals from
abnormal signals.
2.3 Study of Slight Lesions by
Morphological Analysis
The mfERG readings from patients with early-stage
glaucoma, with slight lesions or isolated sectors
developing the disease, do not show a uniform
pattern over the healthy or diseased retina sectors.
This makes the analysis thereof more critical. To
detect lesions of this type a morphological signal
study was conducted in the IC time interval (induced
component) falling between P1 and P2, as shown in
figure 3.

Figure 3: Morphology of the mfERG signal from one
sector.
Although the claim cannot be made across the
board for all cases, there is usually a series of
morphological characteristics held in common in the
records of healthy sectors, differentiating them from
the diseased ones. These are called identity patterns
(Brad et al., 2002). The identity pattern of the
healthy sectors shows little variation and contains a
quick signal response in and near the induced
component, thus building up more energy at mid
frequencies. This conduct reflects the behaviour of
the healthy retina cells, which tend to respond
quickly and efficiently to the mfERG stimulus. The
behaviour of a glaucomatous sector, on the other
hand, shows much more high frequency oscillatory
potentials in the IC interval, with a more blurred
definition of signal peaks and troughs and a long
drawn-out response. Given the signal characteristics
in said interval, our morphological analysis studies
the behaviour of the following signal parameters:
Localisation of points P1, N1 and P2.
Distance between P1 and P2.
Sample width at N1.
Slope in the interval N1 - P2.
Signal oscillations in the interval N1 - P2.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
20

The waveform of the mfERG reading changes
from one sector to another, depending on the retina
position of each one. To allow for this effect the
analyses have been carried out under different
performance parameters, depending on the sectors
position in the retina. Results show that the
individualised study of each sector zone gives our
method an enhanced spatial resolution.
3 RESULTS
In the DWT analysis, several superimposed records
were obtained from different sectors to obtain an
overview of the markers that might differentiate
normal signals from abnormal signals.

Figure 4: Detail D4 of the wavelet decomposition for 10
normal sectors (top) and 10 glaucomatous (bottom).
The top graph of figure 4 shows superimposed
the D4 details of the Wavelet decomposition
between 10 and 190 ms from ten different sectors
corresponding to various healthy individuals. The
bottom graph of the same figure shows a similar
representation for ten glaucomatous sectors and with
an identical topographical position to the former.
One of the most obvious features here is that the
signals corresponding to healthy individuals show
their greatest negative edge at about 70 ms, while
signals in the hexagons affected by glaucoma tend to
bottom out at about 45 ms. The efficiency of this
marker was quantified against a time window
running from 25 to 90 ms, looking for the greatest
negative edge. When this edge came in the first half
of the window the signal was classified as
glaucomatous, while if it came in the second half it
was classified as healthy.

Figure 5: A2 approximation of the wavelet decomposition
for 10 normal sectors (top) and 10 glaucomatous (bottom).
Figure 5 (top) shows superimposed the A2
approximations corresponding to the wavelet
decomposition between 60 and 90 ms of ten
different hexagons belonging to different healthy
individuals. The lower part of this figure shows a
similar representation for ten hexagons affected with
glaucoma and with the same topographical position
as those above. In this case a trough appears at about
73 ms for healthy signals, coming slightly later for
abnormal subjects. Since there might be more
troughs, the efficiency of this second marker is
quantified against a time window running from 65 to
MULTIFOCAL ELECTRORETINOGRAPHY - Early Detection of Glaucoma based on Wavelets and Morphological
Analysis
21

87 ms., seeking this trough. When the trough comes
in the first half of the window the signal was
classified as healthy, while if it came in the second
half it was classified as glaucomatous.
Table 1 shows the results, using both markers
separately, for true and false healthy and
glaucomatous out of a set of 56 sectors belonging to
different healthy individuals and 56 with glaucoma.
Table 1: Results obtained using DWT markers separately
(M=Marker, TH=True Healthy, FG=False Glaucomatous,
TG=True Glaucomatous, FH=False Healthy).
M TH FG TG FH
D4 55 1 48 8
A2 54 2 51 5
The morphological analysis of slight lesions
shows that the duration of the N1 interval is less in
healthy than in glaucomatous sectors, the time-lag of
P2 behind P1 is less in healthy than in glaucomatous
sectors, the amplitude of P2 has to be positive, the
glaucomatous signal shows greater sensitivity in P2
than in N1 and in P1 (accepting a 2% variation).
The disease also shows a change in the
deterioration of healthy sectors according to whether
the lesion is slight or severe (see Figure 6 from left
to right). This evolution can be seen in P2, changing
from a healthy sector morphology with a sharp P2
peak rising quickly from N1, to a flat morphology
with high frequency alterations in P2 (slight case)
and lastly to an even flatter P2 morphology (severe
case). The studys statistical results are shown in
table 2.
Table 2: Results of the morphological study TH=True
Healthy, FG=False Glaucomatous, TG= True
Glaucomatous, FH=False Healthy).
TH FG TG FH
80 % 20 % 90 % 10 %

Figure 6: P2 wave morphology trend.
4 CONCLUSIONS
The morphology of the signals recorded in each
hexagon varies according to the position that this
hexagon occupies in the retina and the type of
stimulus used. It is also known that the optic nerve
head component (ONHC) is the main cause of the
asymmetries in the records (Brad et al., 2002) (Wei
et al., 2007), whereby said component arrives in
each hexagon with a different time-lag depending on
the distance between the hexagon and the optic
nerve. This will enhance or cancel out some
components as a result of the different retina levels
below the hexagon under study. Loss of the ONHC
has already been mooted as an early indicator of
glaucoma (Nalini et al., 2006) (D. C. Hood, 2000),
so there is obviously a need for adjustment of the
various time windows and types of markers used in
this study, according to the position of the hexagon
in the retina map, to optimise and fine tune the
results obtained herein.
A more in-depth investigation needs to be
carried out to adjust the parameters obtained herein
by means of DWT analysis, to find out best values in
terms of the retinal quadrants and rings to which the
sector under study belongs, in view of the
abovementioned hexagon dependency.
The type of markers used herein and the tool
used to obtain them, i.e., the Wavelet transform,
make it impossible a priori to establish any
association with a specific physiological origin,
since there are no precedents to go on. It does not
fall within the remit of this study to establish a
physiological cause-effect relationship for the
marker but rather to search for technical tools to help
experts to diagnose glaucoma in humans in its early
stages of development.
It is obvious that a joint and complementary use
of all the techniques studied herein would be the best
way to improve OAG diagnosis. In this way the
sectors detected as healthy in the Wavelet study
would be introduced into the signal morphology
analysis to check whether there might be any slight
lesions that Wavelet analysis was incapable of
picking up.
ACKNOWLEDGEMENTS
This work was supported by grants from Comunidad
de Madrid-Universidad de Alcal (ref. n CCG06-
UAH/BIO-0711) and Ministerio de Educacin y
Ciencia (ref. n SAF2004-5870-C02-01) awarded to
Pedro de la Villa.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
22

REFERENCES
Brad Fortune, Marcus A. Bearse, J r, George A. Cioffi, and
Chris A. J ohnson, 2002. Selective Loss of an
Oscillatory Component from Temporal Retinal
Multifocal ERG Responses in Glaucoma. IOVS Vol.
43, No. 8, Association for Research in Vision and
Ophthalmology, pp. 2638-2647.
Catal Mora J ., Castany Aregall M., Berniell Trota J . A.,
Arias Barquet L., Roca Linares G., and J rgens Mestre
I., 2005. Electrorretinograma Multifocal y
Degeneracin Macular Asociada a la Edad. Archivos
de la Sociedad Espaola de Oftalmologa, Vol. 80 No.
7.
D. C. Hood, 2000. Assessing Retinal Function with the
Multifocal Technique. Progress in Retinal and Eye
Research. Vol. 19, No. 5, pp. 607-646.
D. C. Hood, J . G. Odel, C. S. Chen, and B. J . Winn, 2003.
The Multifocal Electroretinogram. J Neuro-
Ophthalmol, Vol. 23, No. 3, pp. 225-235.
Hagan R. P., Fisher A. C., and Brown M. C., 2006.
Examination of short binary sequences for mfERG
recording. Doc Ophthalmol, Vol. 113, pp. 21-27.
J . M. Miguel, R. Blanco, L. Boquete, J . M. Rodrguez, and
P. De la Villa, 2007. Electroretinography Multifocal.
Tcnicas y Aplicaciones. CISTI 2007, Porto, Portugal.
J . M. Miguel, R. Blanco, L. Boquete, J . M. Rodrguez, and
P. De la Villa, 2008. Multifocal Electroretinography.
Glaucoma Diagnosis by Means of TheWavelet
Transform, IEEE CCECE 2008. ISBN: 978-1-9244-
1643-1.
M. F. Marmor, D. C. Hood, D. Keating, M. Kondo, M. W.
Seeliger, and Y. Miyake, 2003. Guidelines for basic
multifocal electroretinography (mfERG). Documenta
Ophthalmologica, Kluwer Academic Publishers, Vol.
106, pp. 105-115.
Nalini V. Rangaswamy, Wei Zhou, Ronald S. Harwerth,
and Laura J . Frishman, 2006. Effect of Experimental
Glaucoma in Primates on Oscillatory Potentials of the
Slow-Sequence mfERG. IOVS Vol. 47, No. 2,
Association for Research in Vision and
Ophthalmology, pp. 753-767.
Sutter EE., 2001. Imaging visual function with the
multifocal m-sequence technique. Vision Res. Vol. 41,
pp. 1241-1255.
Sutter EE., and Tran D., 1992. The field topography of
ERG components in man--I. The photopic luminance
response. Vision Res. Vol. 32, pp. 433-446.
Wei Zhou, Nalini Rangaswamy, Periklis Ktonas, Laura J .
Frishman, 2007. Oscillatory potentials of the slow-
sequence multifocal ERG in primates extracted using
the Matching Pursuit method. Vision Research 47,
Elsevier, pp. 2021-2036.
MULTIFOCAL ELECTRORETINOGRAPHY - Early Detection of Glaucoma based on Wavelets and Morphological
Analysis
23
STUDY OF THE PROPERTIES OF BIOTIN-STREPTAVIDIN
SENSITIVE BIOFETS
Thomas Windbacher, Viktor Sverdlov, Siegfried Selberherr
Institute for Microelectronics, TU Wien, Guhausstrae 2729/E360, A-1040 Wien, Austria
[email protected], [email protected], [email protected]
Clemens Heitzinger, Norbert Mauser
Wolfgang Pauli Institute and Department of Mathematics, University of Vienna, Nordbergstrasse 15, A-1090 Wien, Austria
[email protected], [email protected]
Christian Ringhofer
Department of Mathematics, Arizona State University, Tempe, AZ 85287, U.S.A.
[email protected]
Keywords:
BioFET, Field-effect biosensor, Biotin-streptavidin, Simulation, Multi-scale problem, Interface conditions.
Abstract:
In this work the properties of a biotin-streptavidin BioFET have been studied numerically with homogenized
boundary interface conditions as the link between the oxide of the FET and the analyte which contains the bio-
sample. The biotin-streptavidin reaction pair is used in purication and detection of various biomolecules; the
strong streptavidin-biotin bond can also be used to attach biomolecules to one another or onto a solid support.
Thus this reaction pair in combination with a FET as the transducer is a powerful setup enabling the detection
of a wide variety of molecules with many advantages that stem from the FET, like no labeling, no need of
expensive read-out devices, the possibility to put the signal amplication and analysis on the same chip, and
outdoor usage without the necessity of a lab.
1 INTRODUCTION
Todays technology for detecting tumor markers,
antigen-antibody complexes, and pathogens is time-
consuming, complex, and expensive (Pirrung, 2002),
(Shinwari et al., 2006). For instance, a typical proce-
dure to detect a given DNA complex is to increase the
concentration by RT (reverse transcription) or PCR
(polymerase chain reaction), followed by a process
step that will add a label to the DNA enabling detec-
tion by light or radiation. After all these steps the
sample is applied to a microarray. The microarray
consists of an array of spots, and every single spot
is able to detect a different type of molecule. After
the reaction has taken place the array is read by an
expensive microarray reader.
Replacing the above sensing mechanism by an
electrical detection has several benets. First, the op-
tical microarray reader becomes superuous. Detec-
tion by FET (eld-effect transistor) makes the integra-
tion of amplifying and analyzing circuits on the same
chip possible, thus saving also equipment. The ad-
vanced development of semiconductor process tech-
nology allows mass production of such devices, de-
creasing the price dramatically. Various kinds of reac-
tion pairs are possible and have been studied, like de-
tection of DNA (Fritz et al., 2002), (Hahmand Lieber,
2004), (Gao et al., 2007), cancer markers (Zheng
et al., 2005), proteins, e.g. biotin-streptavidin (Im
et al., 2007), (Cui et al., 2001), (Gupta et al., 2008),
(Stern et al., 2007), albumin (Park et al., 2008), and
transferrin (Girard et al., 2006). In these papers dif-
ferent device types and materials were investigated
and provided different solutions for each problem. In
principle, every molecule that is charged in the solute
and that can be bound to the surface layer can be de-
tected by a BioFET. The eld of applications is very
wide and spans from DNA sequencing, point of care
applications, to controlling environmental pollution
and the spread of diseases. The BioFET can be easily
integrated into the chip environment. By putting a mi-
crouidic channel above the functionalized gate of the
24
BioFET the chip can be turned into a mini-laboratory
- the lab on chip. This enables better control of the
environmental parameters (e.g. local pH or detecting
the amount of a special protein) and gives the possi-
bility of local measurement (e.g. how a cell reacts to
a stimulus), thus providing a complete lab-on-a-chip.
However, there are still many problems to overcome
and a lot of research is needed. For instance, an in-
teresting way to avoid problems by poor isolation be-
tween device and solution has been shown by (Kim
et al., 2006).
2 METHOD
A BioFET consists of several parts: a semiconduc-
tor transducer, a dielectric layer, a biofunctionalized
surface, and the analyte (Figure 1). The semiconduc-
tor transducer is a conventional FET. The dielectric
layer is the gate oxide, and the biofunctionalized sur-
face contains immobilized biomolecule receptors at-
tached, so it is able to bind the desired molecule. The
analyte is in an aqueous solution. If a target molecule
binds to a receptor, the local charge density at the sur-
face changes and thus the potential in the semiconduc-
tor. The conductivity of the channel of the eld-effect
transducer is changed.
The binding of the target with the receptor hap-
pens at the Angstrom length scale, while the semi-
conductor device is in the micrometer length scale.
Thus a proper way of combining the semiconductor-
solution interface is crucial.
Analyte
p
n n
Oxide
Reference Electrode
Drain Source
Figure 1: Schematic diagram of a BioFET.
Transport in a FET with a gate length of
1m, is usually modeled via the drift-diffusion ap-
proach (Tang and Ieong, 1995), (Selberherr, 1984).
The aqueous solution is described by the Poisson-
Boltzmann equation.

0
(
Ana
(x, y)) =

S
q c

q
k
B
T
((x,y)

)
(1)
k
B
denotes Boltzmanns constant, T the temperature
in Kelvin, and S, where S contains the valences
of the ions in the electrolyte.
0
describes the permit-
tivity of vacuum, and q the elementary charge.

is
the chemical potential. c

is the ion concentration in

equilibrium, while
Ana
80 is the relative permittiv-
ity of water.
The sum describes the carrier densities arising
from the Boltzmann model. Assuming sodium-
chloride as salt, which is a 1 : 1 salt, the expression
given in (1) can be reduced to

0
(
Ana
(x, y)) =2q c

sinh(
q
k
B
T
((x, y)

)).
(2)
The charge on the surface due to chemical reaction
of the H
+
and OH

was modeled at pH = 7 with the

site-binding model (Shinwari et al., 2006):
Q
Ox
= q N
S
[H
+
]
b
K
a
e

q
k
B
T
(x,y)

K
b
[H
+
]
b
e
q
k
B
T
(x,y)
1 +
[H
+
]
b
K
a
e

q
k
B
T
(x,y)
+
K
b
[H
+
]
b
e
q
k
B
T
(x,y)
.
(3)
N
S
denotes the surface binding site density, while K
a
and K
b
are the equilibrium constants for charging the
surface positively and negatively, respectively. [H
+
]
b
describes the positive hydrogen ion concentration of
the bulk and is corrected to the activity of the hydro-
gen concentration by the e
q
k
B
T
(x,y)
terms.
The biomolecules are modeled in a physics-based
bottom-up approach. By calculating the charge and
dipole moment for a single molecule (see for ex-
ample Figure 2, (Poghossian et al., 2005)), a mean
charge density and a mean dipole moment density
of the boundary layer is obtained. This bridges
the gap between the Angstrom length scale of the
biomolecules and the micrometer dimensions of the
FET (Heitzinger et al., 2008a), (Heitzinger et al.,
2008b), (Ringhofer and Heitzinger, 2008), (Wind-
bacher et al., 2008).
The link between the gate oxide and the aque-
ous solution is realized by two interface conditions,
(Heitzinger et al., 2008a), (Heitzinger et al., 2008b),
(Ringhofer and Heitzinger, 2008), (Heitzinger and
Klimeck, 2007),

Oxid

y
(0, x)
0

Ana

y
(0+, x) =C(x), (4)
(0, x) (0+, x) =
D
y
(x)

Ana

0
. (5)
The x-axis is parallel oriented to the oxide sur-
face, while the y-axis points into the liquid. (0)
describes the potential in the oxide, while (0+) re-
lates to the potential in the solute. The rst equation
describes the jump in the eld, while the second in-
troduces a dipole moment which causes a shift of the
STUDY OF THE PROPERTIES OF BIOTIN-STREPTAVIDIN SENSITIVE BIOFETS
25
Figure 2: Biotin-streptavidin complex (http://www.pdb.org)
on the oxide surface. Two iso-surfaces for plus and minus
0.03
k
B
T
q

A
2
are shown.
-2e-07 -1e-07 0 1e-07
y [m]
-0.2
0
0.2
0.4
p
o
t
e
n
t
i
a
l

p
r
o
f
i
l
e

[
-
V
]
SiO2 Water and salt
SiO2 Water, salt, and biotin-streptavidin 10nm
Figure 3: Potential prole at the interface (from left to right:
semiconductor, oxide, solute).
potential taken into account by adjusting the potential
in the analyte (Figure 3).
3 SIMULATION
Three different types of dielectric were simulated.
SiO
2
as a reference, Al
2
O
3
, and Ta
2
O
5
as possi-
ble high-k materials, with relative permitivies of 3.9,
10, and 25 respectively. As solute 1mMol sodium-
chloride at pH = 7 was considered. The parameters
for the site-binding model can be found in Table 1
(Landheer et al., 2005). For each dielectric the un-
prepared state (just water and salt), the prepared state
(water, salt, and biotin), and the bound state when
the chemical reaction has taken place (water, salt,
and biotin-streptavidin) were calculated for two dif-
ferent mean distances between molecules ( = 10nm,
= 15nm). The data used for calculating charge and
dipole moment of biotin and streptavidin are obtained
from http://www.pdb.org (1SEW.pdb, Figures 2, 12).
The potential distribution across the device is shown
in Figure 4 and output curves were calculated for ev-
ery parameter combination mentioned above, assum-
ing a 100% binding efcency. The potential of the
reference electrode is set to 0.4V so that the FET will
be in moderate inversion as proposed by (Deen et al.,
2006).
Table 1: The parameters needed for the site-binding model
using different dielectric.
Oxide pK
a
pK
b
N
S
[cm
2
] Reference
SiO
2
2 6 5 10
14
(Bousse, 1982)
Al
2
O
3
6 10 8 10
14
(Bousse, 1982)
Ta
2
O
5
2 4 10 10
14
(Bousse et al., 1991)
Figure 4: Potential prole for Ta
2
O
5
water, salt, and biotin-
streptavidin at = 10nm average distance. Blue denotes
1V while red stands for 1V.
4 RESULTS
Figures 5, 6, and 7 show a decrease in the output cur-
rent for biotin attached to the surface in comparison
to the unprepared surface. This downward shift for
the bound state in comparison to the unbound state is
due to the increase of negative charges at the interface,
which is also conrmed by the difference between the
curves for = 10nm and = 15nm, since for 10nm
the molecules are more dense than by 15nm.
As can be seen in the Figures 5, 6, and 7 the bigger
the
r
of the dielectric the bigger is the output current.
Thus high-k materials deliver stronger output signals.
According to (Deen, 2007) however, higher
r
dielec-
tric constants may lead to higher trap densities and
thus to a decreased signal-to-noise ratio. Therefore a
trade-off between bigger output signal and signal-to-
noise ratio has to be met. Figure 8 shows the output
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
26
0 0,2 0,4 0,6 0,8
source to drain voltage [V]
0
10
20
30
40
d
r
a
i
n

c
u
r
r
e
n
t

[
A
/
m
]
SiO2 Water and salt
SiO2 Water, salt, and biotin 10nm
SiO2 Water, salt, and biotin 15nm
SiO2 Water, salt, and biotin-streptavidin 10nm
SiO2 Water, salt, and biotin-streptavidin 15nm
Figure 5: Output curve for SiO
2
for unprepared, prepared
but unbound, and bound state at = 10nm and = 15nm,
respectively.
0 0,2 0,4 0,6 0,8
source to drain voltage [V]
0
20
40
60
80
d
r
a
i
n

c
u
r
r
e
n
t

[
A
/
m
]
Al2O3 Water and salt
Al2O3 Water, salt, and biotin 10nm
Al2O3 Water, salt, and biotin 15nm
Al2O3 Water, salt, and biotin-streptavidin 10nm
Al2O3 Water, salt, and biotin-streptavidin 15nm
Figure 6: Output curve for Al
2
O
3
for unprepared, prepared
but unbound, and bound state at = 10nm and = 15nm,
respectively.
curves as a function of dielectric and molecule ori-
entation (0

means perpendicular to the surface and

90

means lying atly on the surface) leading to the

lowest output curves for 0

followed by 90

and the
curves without dipole moment for each group. Fig-
ures 10 and 11 show the small signal resistance as a
function of dielectric and molecule orientation, dis-
playing smaller values for higher relative permittivity

r
. A slightly larger differential resistance is observed
for perpendicular molecule orientation, in agreement
with the previous results shown in Figures 5, 6, and 7.
This is expected, because biomolecules are inhomo-
geneously charged. Therefore they possess a dipole
moment which enters into the boundary conditions (5)
and there should be a difference in the output curves
of the BioFET for different orientation angles in rela-
tion to the surface.
0 0,2 0,4 0,6 0,8
drain to source voltage [V]
0
50
100
150
d
r
a
i
n

c
u
r
r
e
n
t

[
A
/
m
]
Ta2O5 Water and salt
Ta2O5 Water, salt, and biotin 10nm
Ta2O5 Water, salt, and biotin 15nm
Ta2O5 Water, salt, and biotin-streptavidin 10nm
Ta2O5 Water, salt, and biotin-streptavidin 15nm
Figure 7: Output curve for Ta
2
O
5
for unprepared, prepared
but unbound, and bound state at = 10nm and = 15nm,
respectively.
0 0.2 0.4 0.6 0.8
drain to source voltage [V]
0
50
100
150
d
r
a
i
n

c
u
r
r
e
n
t

[
A
/
m
]
SiO2
Al2O3
Ta2O5
no dipole moment Angle 0 Angle 90
Figure 8: Output curves for SiO
2
, Al
2
O
3
, and Ta
2
O
5
for
calculation without dipole moment, angle 0

(perpendicular
to surface), and angle 90

(parallel to surface).
In the biochemical community there is an ongo-
ing discussion, if the orientation of the biomolecule
is relevant for sensing. Several papers have shown
contradictory results (Oh et al., 2005), (Wacker et al.,
2004), (Kusnezow et al., 2003), (Peluso et al., 2003),
(Turkova, 1999). All these papers are based on optical
detection. Although more study is needed, we men-
tion that for optical detection it is more important to
choose the linking molecule in a way that the reaction
is not hindered by steric effects (receptors block each
other) or the binding sites are blocked or even broken
by the crosslinker. In the case of a BioFET, however,
a eld-effect as working principle is used. Thus it is
important to have a linker that is as short as possible,
to be close to the surface. To increase the signal-to-
noise ratio, the linker should have as little charge as
possible. For example, in order to detect streptavidin,
STUDY OF THE PROPERTIES OF BIOTIN-STREPTAVIDIN SENSITIVE BIOFETS
27
-1e-07 -5e-08 0
y [m]
-0.4
-0.2
0
0.2
p
o
t
e
n
t
i
a
l

p
r
o
f
i
l
e

[
-
V
]
Al2O3 no dipole moment
Al2O3 Angle 0
Al2O3 Angle 90
Figure 9: Potential prole for biotin-streptavidin at =
10nm from left (semiconductor) to right (oxide).
0,2 0,4 0,6 0,8
source to drain voltage [V]
0
0,02
0,04
0,06
0,08
0,1
0,12
SiO2 biotin 10nm
SiO2 biotin 10nm Angle 0
SiO2 biotin 10nm Angle 90
Al2O3 biotin 10nm
Al2O3 biotin 10nm Angle 0
Al2O3 biotin 10nm Angle 90
Ta2O5 biotin 10nm
Ta2O5 biotin 10nm Angle 0
Ta2O5 biotin 10 Angle 90
s
m
a
l
l
s
i
g
n
a
l
r
e
s
i
s
t
a
n
c
e
[

m
]
Figure 10: Small signal resistance for SiO
2
, Al
2
O
3
, and
Ta
2
O
5
for calculation without dipole moment, angle 0

(perpendicular to surface), and angle 90

(parallel to sur-
face) at biotin only.
biotin is used as a binding agent. A biotin molecule is
attached to the surface with a neutral linker. Strepta-
vidin then binds to biotin thus forming a bound state.
The charge difference between the unbound state of a
biotin alone, which is negatively charged with a sin-
gle elementary charge and the bound state of biotin-
streptavidin, which is negatively charged with ve el-
ementary charges, is large enough for detection. We
also note that due to the tetrameric nature of strepta-
vidin it has four sites to bind biotin as shown in Figure
12. Therefore, the linker binding biotin to the surface
should be short enough in order to prevent binding
several biotin molecules to a single molecule of strep-
tavidin .
5 CONCLUSIONS
The model shows a strong dependence on surface
charges and indicates a detectable shift in the thresh-
old voltage depending on their orientation related to
the surface. The bound state (streptavidin-biotin) neg-
atively charged with ve elementary charges com-
pared to the unbound state (biotin) negatively charged
with one elementary charge leads to a reduced con-
ductivity, when hybridization has taken place. Also
the shift of the threshold voltage and output char-
acteristics due to different molecule orientations
(0

...perpendicular to surface, 90

...lying at on sur-
face) can be seen. This shows the usefulness of
the simulation method for the design of efcient
BioFETs.
0,2 0,4 0,6 0,8
source to drain voltage [V]
0
0,05
0,1
0,15
SiO2 biotin-streptavidin 10nm
SiO2 biotin-streptavidin 10nm Angle 0
SiO2 biotin-streptavidin 10nm Angle 90
Al2O3 biotin-streptavidin 10nm
Al2O3 biotin-streptavidin 10nm Angle 0
Al2O3 biotin-streptavidin 10nm Angle 90
Ta2O5 biotin-streptavidin 10nm
Ta2O5 biotin-streptavidin 10nm Angle 0
Ta2O5 biotin-streptavidin 10nm Angle 90
s
m
a
l
l
s
i
g
n
a
l
r
e
s
i
s
t
a
n
c
e
[

m
]
Figure 11: Small signal resistance for SiO
2
, Al
2
O
3
, and
Ta
2
O
5
for calculation without dipole moment, angle 0

(perpendicular to surface), and angle 90

(parallel to sur-
face) at bound state (biotin-streptavidin).
Figure 12: Scheme of the tetrameric protein streptavidin
and biotin.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
28
REFERENCES
Bousse, L. (1982). The Chemical Sensitivity of Elec-
trolyte/Insulator/Silicon Structures. PhD thesis.
Bousse, L., Mostarshed, S., Van Der Shoot, B., De Rooij,
N. F., Gimmel, P., and Gopel, W. (1991). Zeta po-
tential measurements of Ta
2
O
5
and SiO
2
thin lms.
Journal of Colloid and Interface Science, 147(1):22
32.
Cui, Y., Wei, Q., Park, H., and Lieber, C. M. (2001).
Nanowire nanosensors for highly sensitive and selec-
tive detection of biological and chemical species. Sci-
ence, 293(5533):12891292.
Deen, M. J. (2007). Highly sensitive, low-cost integrated
biosensors. In SBCCI 2007: 20th Symposium on Inte-
grated Circuits and System Design, page 1.
Deen, M. J., Shinwari, M. W., Ranu arez, J. C., and
Landheer, D. (2006). Noise considerations in eld-
effect biosensors. Journal of Applied Physics,
100(7):0747031 0747038.
Fritz, J., Cooper, E. B., Gaudet, S., Soger, P. K., and Man-
alis, S. R. (2002). Electronic detection of DNA by
its intrinsic molecular charge. In PNAS, volume 99,
pages 14121416.
Gao, Z., Agarwal, A., Trigg, A., Singh, N., Fang, C., Tung,
C., Fan, Y., Buddharaju, K., and Kong, J. (2007). Sil-
icon nanowire arrays for label-free detection of DNA.
Analytical Chemistry, 79(9):32913297.
Girard, A., Bendria, F., Sagazan, O. D., Harnois, M., Bihan,
F. L., Sala un, A., Mohammed-Brahim, T., Brissot, P.,
and Lor eal, O. (2006). Transferrin electronic detec-
tor for iron disease diagnostics. IEEE Sensors, pages
474477.
Gupta, S., Elias, M., Wen, X., Shapiro, J., and Brillson,
L. (2008). Detection of clinical relevant levels of
protein analyte under physiologic buffer using planar
eld effect transistors. Biosensors and Bioelectronics,
24:505511.
Hahm, J. and Lieber, C. M. (2004). Direct ultrasensitive
electrical detection of DNA and DNA sequence vari-
ations using nanowire nanosensors. Nano Letters,
4(1):5154.
Heitzinger, C., Kennell, R., Klimeck, G., Mauser, N.,
McLennan, M., and Ringhofer, C. (2008a). Modeling
and simulation of eld-effect biosensors (BioFETs)
and their deployment on the nanoHUB. J. Phys.:
Conf. Ser., 107:012004/112.
Heitzinger, C. and Klimeck, G. (2007). Computational as-
pects of the three-dimensional feature-scale simula-
tion of silicon-nanowire eld-effect sensors for DNA
detection. Journal of Computational Electronics,
6:387390.
Heitzinger, C., Mauser, N., and Ringhofer, C. (2008b).
Multi-scale modeling of planar and nanowire eld-
effect biosensors. submitted.
http://www.pdb.org.
Im, H., Huang, X. ., Gu, B., and Choi, Y. . (2007).
A dielectric-modulated eld-effect transistor for
biosensing. Nature Nanotechnology, 2(7):430434.
Kim, D., Park, J., Shin, J., Kim, P., Lim, G., and Shoji, S.
(2006). An extended gate FET-based biosensor inte-
grated with a Si microuidic channel for detection of
protein complexes. Sensors and Actuators, B: Chemi-
cal, 117:488494.
Kusnezow, W., Jacob, A., Walijew, A., Diehl, F., and Ho-
heisel,J.D. (2003). Antibody microarrays: An evalua-
tion of production parameters. Proteomics, 3(3):254
264.
Landheer, D., Aers, G., McKinnon, W., Deen, M., and
Ranu arez, J. (2005). Model for the eld effect from
layers of biological macromolecules on the gates of
meta-oxide-semiconductor transistors. journal of ap-
plied physics, 98(4):0447011 04470115.
Oh, S. W., Moon, J. D., Lim, H. J., Park, S. Y., Kim, T.,
Park, J., Han, M. H., Snyder, M., and Choi, E. Y.
(2005). Calixarene derivative as a tool for highly
sensitive detection and oriented immobilization of
proteins in a microarray format through noncovalent
molecular interaction. FASEB Journal, 19(10):1335
1337.
Park, K., Lee, S., Sohn, Y., and S.Y, C. (2008). BioFET
sensor for detection of albumin in urine. Electronic
Letters, 44(3).
Peluso, P., Wilson, D. S., Do, D., Tran, H., Venkatasub-
baiah, M., Quincy, D., Heidecker, B., Poindexter, K.,
Tolani, N., Phelan, M., Witte, K., Jung, L. S., Wagner,
P., and Nock, S. (2003). Optimizing antibody immo-
bilization strategies for the construction of protein mi-
croarrays. Analytical Biochemistry, 312(2):113124.
Pirrung, M. C. (2002). How to make a DNA chip. Angew.
Chem. Int. Ed., 41:12761289.
Poghossian, A., Cherstvy, A., Ingebrandt, S., Offenh ausser,
A., and Sch oning, M. J. (2005). Possibilities and lim-
itations of label-free detection of DNA hybridization
with eld-effect-based devices. Sensors and Actua-
tors, B: Chemical, 111-112(SUPPL.):470480.
Ringhofer, C. and Heitzinger, C. (2008). Multi-scale mod-
eling and simulation of eld-effect biosensors. ECS
Transactions, 14(1):1119.
Selberherr, S. (1984). Analysis and Simulation of Semi-
conductor Devices, volume Springer of ISBN: 3-211-
81800-6.
Shinwari, M. W., Deen, M. J., and Landheer, D. (2006).
Study of the elecrolyte-insulator-semiconductor eld-
effect transistor (EISFET) with applications in biosen-
sor design. Microelectronics Reliability.
Stern, E., Klemic, J., Routenberg, D., Wyrembak, P.,
Turner-Evans, D., Hamilton, A., LaVan, D., Fahmy,
T., and Reed, M. (2007). Lable-free immunodetection
with CMOS-compatible semiconducting nanowires.
Nature Letters, 445(1):519522.
Tang, T.-W. and Ieong, M.-K. (1995). Discretization of
ux densities in device simulations using optimum ar-
ticial diffusivity. IEEE Transactions on Computer-
Aided Design of Integrated Circuits and Systems,
14(11):13091315.
Turkova, J. (1999). Oriented immobilization of biologically
active proteins as a tool for revealing protein interac-
tions and function. Journal of Chromatography B:
STUDY OF THE PROPERTIES OF BIOTIN-STREPTAVIDIN SENSITIVE BIOFETS
29
Biomedical Sciences and Applications, 722(1-2):11
31.
Wacker, R., Schroder, H., and Niemeyer, C. M. (2004).
Performance of antibody microarrays fabricated by ei-
ther DNA-directed immobilization, direct spotting, or
streptavidin-biotin attachment: A comparative study.
Analytical Biochemistry, 330(2):281287.
Windbacher, T., Sverdlov, V., Selberherr, S., Heitzinger, C.,
Mauser, N., and Ringhofer, C. (2008). Simulation
of eld-effect biosensors (BioFETs). In Proc. Simu-
lation of Semiconductor Processes and Devices (SIS-
PAD 2008), pages P18/14, Hakone, Japan.
Zheng, G., Patolsky, F., Cui, Y., Wang, W. U., and Lieber,
C. M. (2005). Multiplexed electrical detection of
cancer markers with nanowire sensor arrays. Nature
Biotechnology, 23(10):12941301.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
30
AVALANCHE PHOTODIODES FOR HIGH-RESOLUTION PET
IMAGING SYSTEMS
R. Bugalho, B. Carrio, C. S. Ferreira, M. Ferreira, R. Moura, C. Ortigo
J . Pinheiro, P. Rodrigues, J. C. Silva, A. Trindade and J . Varela
1

LIP Lab. de Instrumentao e Fsica Experimental de Partculas, Avenida Elias Garcia 14, 1000-149 Lisboa, Portugal
[email protected], [email protected], [email protected], [email protected], [email protected], [email protected]
[email protected], [email protected], [email protected], [email protected], [email protected]
1
also at IST- Instituto Superior Tcnico, Av Rovisco Pais, 1049-001 Lisboa, Portugal
Keywords: Avalanche photodiode, Dark current, Gain, Positron emission mammography, Quality control.
Abstract: A high-resolution Positron Emission Tomography (PET) scanner prototype, named Clear-PEM, was
developed by the Portuguese PET Consortium in the framework of the Crystal Clear Collaboration (CCC).
This scanner is a PET prototype dedicated for breast cancer imaging mammography, based on a novel
readout scheme constituted by fine-pitch scintillator crystals, avalanche photodiodes (APD), low-noise high-
gain frontend amplifiers and a reconfigurable FPGA-based electronics readout system. The Clear-PEM
scanner is designed to exam both the breast and the auxiliary lymph node areas, aiming at the detection of
tumours down to 2 mm in diameter. The prototype has two planar detector heads, each composed of 96
detector modules. Each detector module is composed of a matrix of 32 identical 2x2x20 mm
3
LYSO:Ce
scintillator crystals, read at both ends by Hamamatsu S8550 APD arrays (4x8) for Depth-of-Interaction
(DOI) capability. The APD arrays were characterized through the measurement of gain and dark current as a
function of bias voltage, under controlled conditions. A set of 984 APD arrays followed a well defined
quality control (QC) protocol, aiming at the rejection of arrays not complying with the defined
specifications. From the total of 984, only 1 (0.1%) was rejected, reassuring the trust in these detectors for
prototype assembly and future applications.
1 INTRODUCTION
New methods for breast cancer diagnosis are object
of heavy research efforts. One such research line
relies on the use of Positron Emission based
technology applied to breast cancer detection. In
spite of initial very encouraging results in limited
clinical trials, whole body PET systems have
considerable operational costs, with a low patient
turnover incompatible with systematic screening and
a low spatial resolution which limits the minimum
lesion size that can be detected. This has led to the
development of dedicated PET scanners, targeting
breast cancer imaging applications (Thompson et al,
1995, Moses, 2004). The PEM units are designed to
explore localized regions of the body, usually the
breast, and adopt several design principles, like fine
pixelized crystals, to achieve a better spatial
resolution, as well as a large countrate capability.
This set of requirements has lead to the development
of a series of proofofprinciple and a few full
assembled Positron Emission Mammography (PEM)
scanner prototypes, several of which have been used
in preliminary clinical trials, based on high density
and highZ inorganic scintillator crystals. Of this
list, almost all lack the ability to measure the depth
ofinteraction and thus reconstructed images may
show significant aberrations due to the parallax
effect. An emerging technique to reduce this
aberration effect consists on the readout of the
scintillation light, produced in the crystal elements,
by two opposing photosensors and extracts the
coordinate of interaction along the longitudinal axis
from the asymmetry of light collection (Shao et al.,
2000). This is particular important since in a planar
detector with the active media located close to the
object under examination, the parallax effect can be
an important source of blurring in the spatial
resolution.
Several technical challenges need to be
addressed namely on how to readout the light from
the crystals without putting unacceptable amounts of
31

non-active media (like conventional
photomuiltplers) between the patient port and the
crystals, which would lead to a degradation on the
final image quality and lesion detection sensitivity.
To address these issues, the ClearPEM scanner
(Lecoq and Varela, 2002, Abreu et al, 2006), was
developed by the Portuguese PET Consortium,
composed by: Laboratrio de Instrumentao e
Fsica Experimental de Partculas (LIP), Instituto de
Biofsica e Engenharia Biomdica da Faculdade de
Cincias de Universidade de Lisboa (IBEB),
Instituto de Biomdico de Investigao de Luz e
Imagem (IBILI), Instituto de Novas Tecnologias
(INOV), Instituto de Engenharia de Sistemas e
Computadores Investigao e Desenvolvimento
(INESC-ID), Instituto de Engenharia Mecnica e
Gesto Industrial (INEGI), Hospital Garcia da Orta
(HGO), Instituto Portugus de Oncologia (IPO) in
Porto and with the participation of CERN
(Organisation Europene pour la Recherche
Nuclaire) through the international collaboration
Crystal Clear Collaboration.
The detector is based on pixelized LYSO:Ce
crystals optically coupled on both extremities to
avalanche photodiodes (APD) and readout by a fast,
low-noise electronic system. The choice of
avalanche photodiodes was dictated by its
compatibility with the implementation of a double
readout technique. The APDs also demonstrate good
energy resolution for a direct detection of X-rays
and for LYSO:Ce scintillator light readout, with low
multiplication noise and acceptable gain uniformity.
The Hamamatsu S8550 APD was selected for the
Clear-PEM detector, which offers a stable working
regime up to gain 200 (Abreu et al., 2007), and has a
dark current noise below 27 electrons ENC at room
temperature (Kapusta et al., 2003).
The total number of APD arrays needed for a
single scanner is 384, and the S8550 arrays were
manufactured by Hamamatsu Photonics Inc. J apan.
A total of 984 photodetectors were acquired for
assessment.
A quality control protocol and methodology were
developed in order to assess and characterize the
acquired APDs. In this paper the adopted quality
check procedures for the APD arrays is described
and the controlled parameters are presented (gain
dependence on bias voltage, gain variation at various
voltages, dark current dependence on the bias
voltage).

2 CLEAR-PEM IMAGING
SYSTEM
2.1 Detector Heads
The Clear-PEM imaging system (Figure 1) is based
on a two parallel detector heads design. The detector
consists in two compact and planar detector heads
with adequate dimensions for breast and axilla
imaging. A dedicated gantry was built in order to
allow the rotation of the detector heads in breast
exams as well as to permit exams of the axilla
region. Each detector head holds the scintillator
material matrices and the frontend readout circuitry
composed of multipixel APD photosensors,
frontend ASIC chips, freesampling ADCs and
Channel Link LVDS serializers. Auxiliary sub
systems are also assembled in the detector heads,
providing environmental monitoring, cooling, power
supply and system clock distribution. A dedicated
digital trigger and data acquisition system is used for
online selection of coincidence events with high
efficiency, large bandwidth and negligible dead-time
(Albuquerque et al 2006, Albuquerque et al 2008a).

Figure 1: The Clear-PEM imaging system (CAD image).
The structure of a detector head starts in the
detector module. The detector module is composed
by the LYSO:Ce crystal matrix, optically coupled
(through an optical glue) to a S8550 APD array, on
each end, for DOI measurements. The APD array is
mounted in a small PCB with a low footprint SMT
connector on the back side. The components of the
detector module are housed and sealed in a
dedicated plastic assembly. The assembly has two
empty slots in which two detector modules can be
plugin, defining a double module. 12 detector
modules are mechanically fixed and electrically
connected to a front and back Frontend Boards
(FEBs) forming a Supermodule (Figure 2)
(Albuquerque et al, 2008b).
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
32


Figure 2: Supermodule structure assembling 12x32
LYSO:Ce crystals and 24 APDs in double readout. Each
Frontend board has two 192 input channels ASICs.
The assembled modules underwent a quality
control measurements carried out at LIP (Amaral et
al., 2006). The electronics chips were mounted on
the external faces of the two FEBs. Connectors for
cables linking to the data acquisition system were
also mounted in the PCBs. In a detector head, eight
Supermodules, each with 12 modules, are mounted
together.
The final prototype has two planar parallel
detector heads with 192 detector modules, 6144
crystals and 384 APD arrays (12 288 APD pixels),
covering a 1715 cm
2
Field-of-View.
2.1.1 Crystals
The chosen scintillator for the Clear-PEM was a
inorganic crystal, LYSO:Ce. The LYSO:Ce is
cerium activated lutetiumyttriumorthosylicate
(Lu
2(1x)
Y
2x
SiO
5
:Ce) and has similar properties to
LSO:Ce (Lu
2
SiO
5
) (Melcher and Schweitzer, 1992).
The choice of this scintillator was dictated by its
high light output (27-30 photons/keV) compatible
with good energy measurements (9% at 662 keV
with a PMT readout), as well as a fast rise time and
decay (42 ns time constant) compatible with high
quality timing measurements and low dead time. Its
peak emission is shortblue/long ultraviolet (UV)
wavelength (420 nm).
For the Clear-PEM scanner, crystals with
dimensions of 2220 mm
3
were chosen. The 20
mm longitudinal size guarantees a large detection
efficiency for 511 keV photons and the
determination of the DOI coordinate with a
resolution of 2 mm (FWHM) (Abreu et al., 2006,
Amaral et al., 2006).
In total the scanner uses 6144 pixel LYSO:Ce
crystals (Figure 3a) divided in 192 matrices,
composed by a 48 LYSO:Ce crystal array (Figure
3b). In this configuration, the crystals are optically
isolated by 300 m BaSO
4
reflector walls. The
BaSO
4
provide the crystal support and provides the
diffuse reflecting surfaces needed to optimize the
light collection and the DOI measurements.

Figure 3: Photograph of (a) a sample of 2220 mm
3

LYSO:Ce crystals produced for the Clear-PEM scanner
and (b) an assembled BaSO
4
type matrix.
2.1.2 Photosensors
The S8550 APD array from Hamamatsu Photonics
with a 32 pixels in an 84 configuration was the
selected photodetector for the Clear-PEM scanner
(Figure 4a and b). The S8550 array is assembled
from two distinct monolithic silicon wafer parts, of
28 reverse type structure pixel elements, and
each 2x8 group is called an sub-array. The sub-array
that contains the APD pixels from A1 to H1, and
from A2 to H2, is called sub-array 1. The remaining
pixels are contained in the sub-array 2. Each sub-
array is biased independently. The pixels are
mounted on a 1 mm thick ceramic package with a
0.5 mm thick epoxy window (Kapusta et al., 2003).

Figure 4: (a) Schematic representation of the 32 pixel
Hamamatsu S8550 APD array (dimensions in millimetres)
and (b) photograph of a S8550 APD array.
Each Si pixel element has a 1.61.6 mm
2
,
compatible with an individual 1:1 readout of 22
mm
2
crosssection LYSO:Ce crystals. The element
pitch is 2.3 mm and all the pixels placed in the same
submatrix share the same common bias.
(a)
(b)
(a) (b)
AVALANCHE PHOTODIODES FOR HIGH-RESOLUTION PET IMAGING SYSTEMS
33

A compilation of the Hamamatsu S8550 main
electrical and optical characteristics are present in
Table 1. The effective APD gain (ratio of the total
number of secondary avalanche electrons produced
by the initial number of electronhole pairs due to
the scintillation light) is between 70 and 350 with a
specified interpixel gain variation less than 5%
r.m.s. and a dark current of 24 nA per pixel. The
terminal capacitance is 10 pF per pixel (Abreu et al.,
2007).
Table 1: Hamamatsu S8550 APD 32 channel electrical and
optical parameters (Abreu et al, 2007, Mosset, 2006).
Parameter Hamamatsu S8550
Pixel Size 1.6 x 1.6 mm
2
Pixel Pitch 2.3 mm
Window Type 0.5 mm epoxy resin
Peak Wavelength 600 nm
QE @ 420 nm 72-76%
Gain (M)
Gain Gradient @ M=70
Dark Current
Capacitance
50 350
3.6%/V
2.4 nA (pixel @ M=70)
10 pF (pixel @ M=70)
The S8550 APD operates at a bias voltage of
360-500 V, depending on the required gain. The
gain shows a temperature gradient of 2.4%/C at
gain 70. This is due to lattice vibrations in the silicon
structure of the APD which are enhanced as
temperature increases making more probable to
interact with avalanche secondary electrons.
Temperature drifts, if not controlled, may thus
originate gain drifts contributing to a detioration of
the energy resolution (Crespo et al., 2004,
Spanoudaki et al., 2005). The temperature gradient
is also function of the bias voltage, which means that
at higher gains the APDs show an increased
susceptibility to temperature drifts. All this imply
that the system has to operate under stable thermal
conditions. The gain gradient as function of the
polarizing bias is observed to vary as function of the
gain. For the highest gain the bias polarization
supplies must be very stable with controlled amounts
of tension ripple (Abreu et al., 2007).
2.2 Electronic Systems
The frontend and data acquisition electronics
systems are key components in the developed PET
system, enabling a high detection sensitivity and low
random background noise, without compromising
the spatial resolution, allowed by fine segmented
crystals. The data acquisition and trigger electronics
of the ClearPEM scanner is composed by three
main blocks (Albuquerque et al., 2006):
The Frontend Electronics System, performs
signal amplification, channel selection and
analog multiplexing, analog to digital
conversion and paralleltoserial translation;
The Trigger and Data Acquisition System,
which implements the temporary data storage,
firstlevel trigger (L1) computation and data
transfer to the acquisition computer (DAE
server);
The Software Trigger (L2), implemented in the
DAE server, in which the received data from
the DAE crate is unpacked, extraction
algorithms (time, energy, DOI) applied and
the trigger revalidated;
The ondetector electronics includes the
amplifier and analog multiplexing integrated circuits
(frontend ASIC). The front-end system is based in a
data-driven synchronous design that identifies and
multiplexes the analog signals of channels above
threshold, reducing the number of channels by a
factor of 96. To transmit the signals from the
detector heads, digital serializers are used to
minimize the number of lines connecting to the
trigger and data acquisition system (Rodrigues,
2007).
The offdetector system receives the serialized
digitized data streams, applies a coincidence trigger
based on the computation of the pulses amplitude
and timing, and pushes the data into the data
acquisition computer. The trigger and data
acquisition logic is implemented in large FPGAs
with 4 and 3 million gates respectively, with the
trigger algorithm decomposed in a sequence of
elementary operations executed in pipeline mode.
The system was designed to operate at a maximum
input clock frequency of 100 MHz and be able to
sustain a data acquisition rate of 1 MHz with
efficiency above 9095%, under a maximum total
single photon background rate in the detector of 10
MHz. The communication between the readout
system and the data acquisition computer is
established through a serial highspeed link
allowing the offdetector crate to be located several
meters away from the DAE server data acquisition
computer.
2.3 Mechanical Systems
The Clear-PEM mechanical system was designed to
allow the exam of both the breast and the auxiliary
lymph nodes. The system is used in conjunction with
a shielded examination table that enables exams to
be performed with the patient kept in prone position.
Configurable openings in the examination table
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
34

allow the exam of both breasts with the detector
heads positioned in each side of the breast (Abreu et
al., 2006). During the exam the detector heads can
rotate around the detector main axis in order to
collect data at several angular orientations as
required for tomographic reconstruction.
The examinations of the breast region close to
the chest and of the axilla region are performed in a
front-back configuration. In these exams one
detector head is facing the breast (complementary
exam), or the shoulder (axilla exam), under the
scanner table and the other is positioned against the
patient back.
3 QUALITY CONTROL OF
AVALANCHE PHOTODIODES
The scanner construction starts with the quality
control (QC) and gain characterization of all APD
arrays acquired by the PET consortium, in order to
assess which ones are useable in the final prototype.
The project commercially acquired 984 APD arrays
which needed a fast evaluation. An automatic
measuring setup was developed, based on the direct
measurement of the APD-array common cathode
current when illuminated by a blue LED light (to
simulate the LYSO:Ce scintillation). This automatic
measuring setup was named APD Tester. The blue
light (420 nm) from the LED is homogeneously
distributed by a light guide through all APD pixels.
The characterization consists in the measurement
of the bias voltage needed to operate the APD arrays
at gains 70, 150 and 350, the variation of gain per
volt (dM/dV) at each of these gain regions, and also
the Dark Current (Id), for the same gain regions.
Each individual APD sub-array is characterized
independently (1968 sub-arrays, 9 characterization
parameters each, gives a total of 17712 measurement
values).
For the QC part, the chosen acceptance values at
gain 70 were:
Bias Voltage 500V;
Gain Variation 4.0%/V;
Dark Current 160 nA (10 nA per pixel);
The parameters were selected accordingly to the
Hamamatsu specifications (and tolerances) and are
applied to each sub-array. If one of the parameters is
not approved, the APD array is marked as bad and
sent back to the manufacturer.

3.1 APD Tester
The APD Tester is a dedicated electronic setup for
the Hamamatsu APD array S8550 gain and dark
Current quality control and characterization (Figure
5). One of the main features of this dedicated
electronic system is the automatization of the
measurement procedure, in order to save time and
manpower. It measures 16 APD arrays (32 sub-
arrays) in a single run (or several), in about 4 hours.
It also controls the blue LED, used to simulate the
scintillation light.

Figure 5: APD Tester, an electronic setup designed to the
automatic characterization of the S8550 APD Array.
The electronic was completely designed and
developed at LIP. The Tester was assembled inside a
metal box (Faraday cage) due to the need of a
protection from any outside signal interference, and
to keep the APD arrays away from the outside light.
The LED is located inside the box and is coupled to
a light guide to provide an equal light distribution to
all APD sub-arrays (and pixels). The tester setup is
composed by two major parts: HV Control and
Digital Control.
The digital control is composed by HV relays to
select the APD sub-array to be measured, from the
possible 32. The relays are controlled through a
FPGA Cyclone II (ALTERA DE2 Board) via serial
port. The LED is also controlled by the digital part.
The HV Control receives the HV from a Keithley
6487 Picommeter/Voltage Source, controlled via
serial port, and delivers it to the enabled APD sub-
array. The automatic setup, and the Keithley, are
controlled via serial ports by a LabView program,
which selects the measurement type and collects the
data into an excel file.



AVALANCHE PHOTODIODES FOR HIGH-RESOLUTION PET IMAGING SYSTEMS
35

3.2 Experimental Setup
The experimental setup is composed by the APD
Tester (in the Faraday cage), a Keithley 6487
Picoammeter / Voltage Source and a PC. The gain
calculation assumes that when the APD sub-array is
biased at 30V (with the LED on), the gain is 1. From
that value the M=70, 150 and 350 are calculated. For
the dark current measurements the LED is turned off
and the residual current is measured. The schematic
of the experimental setup can be found in Figure 6.

Figure 6: Schematic representation of the experimental
setup for gain and dark current measurements.
3.3 Measurement Procedure
Each measurement run is accompanied by a stability
check procedure (performance on a reference APD
array) in order to evaluate the setup in terms of
significant variations on the studied parameters. The
setup operates at a stable temperature (through AC
control) of approximately 24C, in order to minimize
temperature dependence deviations. The complete
measurement took about 30 working days (2 run sets
per day approximately). The stability check consists
in the measurement of a predefined APD array, the
control APD. Each measurement run characterized
15 new APD plus the control APD.
3.3.1 Bias Voltage Quality Control and
Characterization
The 984 APD matrices (1968 sub-arrays) were
characterized in bias voltage terms and Table 2
summarizes the results.
All APD passed the HV QC at M=70 (the bias
voltage values were all below 500V) -
Figure 7. The HV values oscillated between 360
and 459 V - Figure 8.
Table 2: Bias Voltage average values for gains M=70, 150
and 350 from the 984 APD arrays (also the maximum and
minimum HV).
Bias Voltage (V) Sub-array 1 Sub-array 2
M=70
Average (r.m.s) 419 22 418 21
Minimum 360 364
Maximum 458 459
M=150
Average (r.m.s) 434 22 434 21
Minimum 375 379
Maximum 473 474
M=350
Average (r.m.s) 443 22 442 21
Minimum 384 389
Maximum 482 481

Figure 7: High Voltage per APD sub-array distribution at
M=70.

Figure 8: High Voltage histogram for M=70 values.
3.3.2 Gain Variation Quality Control and
Characterization
The gain variation of the 984 APD matrices was
characterized and Table 3 summarizes the results.
All but one APD sub-array had a gain variation
of less than 4%/V, passing the Gain variation quality
control at M=70 Figure 9.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
36

The APD that didnt pass the QC procedure had
a dM/dV of about 22.7 % on the second sub-array
(the other had a normal value, 3.2 %/V).
Table 3: Gain Variation average values for gains M=70,
150 and 350 from the 984 APD arrays (also the maximum
and minimum dM/dV).

Gain Variation
(%/V)
Sub-array 1 Sub-array 2
M=70
Average (r.m.s) 3.6 0.1 3.6 0.6
Minimum 3.2 3.3
Maximum 4.0 22.7
M=150
Average (r.m.s) 6.3 1.8 6.0 0.8
Minimum 5.4 5.4
Maximum 26.5 19.4
M=350
Average (r.m.s) 14.0 4.7 13.5 4.6
Minimum 9.2 6.0
Maximum 30.7 31.6

Figure 9: Gain Variation per APD sub-array distribution at
M=70.
3.3.3 Dark Current Quality Control and
Characterization
The Dark Current characterization of the 984 APD
matrices was completed and Table 4 summarizes the
results.
All APD sub-array passed the Dark Current QC
at M=70 (all the values were below the stipulated
maximum limit of 160nA, less than 10 nA per APD
pixel according to Hamamatsu Photonics) Figure
10.
4 CONCLUSIONS
The quality control and characterization procedure
for the acquired APD arrays for the Clear-PEM
prototype was defined and implemented. A
dedicated automatic APD characterization
electronics was developed and built. The 984
Hamamatsu S8550 APD arrays were submitted to
the QC and characterization procedure at gains 70,
150 and 350, in terms of bias voltage, dark current
and gain variation per volt.
Table 4: Dark Current average values for gains M=70, 150
and 350 from the 984 APD arrays (also the maximum and
minimum Id).

Dark Current
(nA)
Sub-array 1 Sub-array 2
M=70
Average (r.m.s) 27.1 13.3 24.2 10.1
Minimum 7.8 8.2
Maximum 88.9 67.6
M=150
Average (r.m.s) 43.1 21.1 40.0 19.2
Minimum 10.1 12.1
Maximum 138.4 300.0
M=350
Average (r.m.s) 107.1 67.0 103.7 65.9
Minimum 15.9 17.6
Maximum 334.2 700.0

Figure 10: Dark Current per APD sub-array distribution at
M=70.
The QC (M=70) had very good results: average
bias voltage of 419 and 418V for sub-array1 and 2
respectively, average dark current of 27.1 and
24.2nA, for sub-array1 and 2, and average gain
variation of 3.6%/V for both sub-arrays.
From a total of 984 tested APD arrays only 1
(0.1%) didnt pass the QC procedure (due to a
22.7%/V gain variation in the second sub-array).
The small variance on the different electrical
characterization parameters points out that current
available APDs are suitable for high-integration PET
prototypes which could not be implemented using a
classical photomultiplier readout.


AVALANCHE PHOTODIODES FOR HIGH-RESOLUTION PET IMAGING SYSTEMS
37

ACKNOWLEDGEMENTS
The authors would like to thank colleagues from the
Portuguese PET Consortium and the Crystal Clear
Collaboration for their suggestions and contribution.
This project is financed by AdI (Agncia de
Inovao) and POSI (Operational Program for
Information Society), Portugal. The work of P.
Rodrigues and A. Trindade was supported by FCT
under grants SFRH/BPD/37233/2007 and
SFRH/BPD/37226/2007. The work of R. Bugalho,
B. Carrio, C. S. Ferreira, R. Moura, C. Ortigo and
J . F. Pinheiro was supported by AdI.
REFERENCES
Abreu, M. C., Aguiar, J . D., Almeida, F. G., Almeida, P.,
Bento, P., Carrio, B., Ferreira, M., Ferreira, N. C.,
Gonalves, F., Leong, C., Lopes, F., Lous, P.,
Martins, M. V., Matela, N., Mendes, P. R., Moura, R.,
Nobre, J ., Oliveira, N., Ortigo, C., Peralta, L.,
Pereira, R., Rego, J ., Ribeiro, R., Rodrigues, P.,
Sampaio, J ., Santos, A. I., Silva, L., Silva, J . C., Sousa,
P., Teixeira, I. C., Teixeira, J . P., Trindade, A.,and
Varela, J . (2006). Design and evaluation of the Clear-
PEM scanner for positron emission mammography.
IEEE Trans. Nuc. Sci., 53:7177.
Abreu, M. C., Amaral, P., Carrio, B., Ferreira, M.,
Moura, R., Ortigoo, C., Rato, P., and Varela, J .
(2007). Characterization and quality control of
avalanche PhotoDiode arrays for the ClearPEM
detector modules. Nucl. Instr. and Method.,
576(1):1922.
Albuquerque, A., V. Bexiga, Bugalho, R., Carrio, B.,
Ferreira, C. S., Ferreira, M., Godinho, J . Gonalves,
F., Leong, C., Lous, P., Machado, P., Moura, R.,
Neves, P., Ortigo, C., Piedade, F., Pinheiro, J . F,
Rivetti, A., Rodrigues, P., Silva, J . C., Silva, M. M.,
Teixeira, I. C., Teixeira, J . P., Trindade, A., Varela, J .
(2008) Experimental characterization of the 192
channel ClearPEM frontend ASIC for multipixel
APD readout. Sub. Nucl. Instr. And Method.
Albuquerque, E., Almeida, F. G., Almeida, P., Augusto,
S., Bexiga, V., Bugalho, R. Carmona, S. , Carrio, B.,
Ferreira, C. S., Ferreira, N. C., Ferreira, M., Godinho,
J ., Gonalves, F., Guerreiro, C., Leong, C., Lous, P.,
Machado, P., Martins, M. V., Matela, N., Moura, R.,
Neves, P., Oliveira, N., Ortigo, C., Piedade, F.,
Pinheiro, J . F., Rego, J ., Relvas, P., Rivetti, A.,
Rodrigues, P., S, D. N., Sampaio, J., Santos, A. I.,
Silva, M. M., Teixeira, I. C., Teixeira, J. P., Silva, J.
C., Trindade, A., Varela, J . Performance evaluation of
a highly integrated APD/ASIC double-readout
supermodule with 768 channels for Clear-PEM, In
2008 IEEE Nuclear Science Symposium Conference
Record.
Amaral, P., Carrio, B., Ferreira, M., Moura, R., Ortigo,
C., Rodrigues, P., Silva, J . C., Trindade, A., and
Varela, J . (2006). Performance and quality control of
ClearPEM detector modules. Nucl. Instr. and
Method. in press.
Albuquerque, E., Bento, P., Leong, C., Gonalves, F.,
Nobre, J ., Rego, J ., Relvas, P., Lous, P., Rodrigues,
P., Teixeira, I. C., Teixeira, J . P., Silva, L., Silva, M.
M., Trindade, A., and Varela, J . (2006). The Clear-
PEM electronics system. IEEE Trans. Nuc. Sci.,
53(5):27042711.
Crespo, P., Kapusta, M., Pawelke, J ., Moszyski, M., and
Enghardt, W. (2004). First inbeam PET imaging with
LSO/APD array detectors. IEEE Trans. Nuc. Sci.,
51(5):26542661.
Kapusta, M., Crespo, P., Wolski, D., Moszyski, M., and
Enghardt, W. (2003). Hamamatsu S8550 arrays for
highresolution scintillator matrices readout. Nucl.
Instr. and Method., A504:139142.
P. Lecoq and J . Varela, Clear-PEM, a dedicated PET
camera for mammography. Nucl. Instrum. Meth. vol
A 486, pp.1-6, 2002.
Melcher, C. L. and Schweitzer, J . S. (1992). Cerium
doped lutetium oxyorthosilicate: a fast, efficient new
scintillator. IEEE Trans. Nuc. Sci., 39(4):502505.
Moses, W. W. (2004). Positron emission mammography
imaging. Nucl. Instr. and Method., A525(12).
Mosset, J . B. (2006). Developpement dun module de
detection phoswich LSO%LuYAP pour le prototype
de camera `a positrons ClearPET. PhD thesis, Faculte
des Sciences de base de lEcole Polytechnique
Federale de Lausanne.
Rodrigues, P.,. (2007). Study and Development of the
Clear-PEM Trigger and Data Acquisition System
ClearPET. PhD thesis, Instituto Superior Tcnico,
Universidade Tcnica de Lisboa.
Spanoudaki, V., McElroy, D. P., Zell, K., and Ziegler, S. I.
(2005). Effect of temperature on the stability and
performance of an LSOAPD PET scanner. In 2005
IEEE Nuclear Science Symposium Conference
Record, pages 27852789.
Shao, Y., Silverman, R. W., Farrell, R., Cirignamo, L.,
Grazioso, R., Shah, K. S., Visser, G., Clajus, M.,
Tumer, T. O., and Cherry, S. R. (2000). Design
studies of a high resolution PET detector using APD
arrays. IEEE Trans. Nuc. Sci., 47(3):10511057.
Thompson, C. J . , Murthy, K., Picard, Y. , Weinberg, I. N.
, Mako F. M. (1995) Positron Emission
Mammography (PEM): a promising technique to
detect breast cancer. IEEE Trans. Nucl. Sci. 42 1012
1017.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
38
2.4 GHZ WIRELESS ELECTROMYOGRAPH SYSTEM
WITH STATISTICALLY OPTIMAL AUTOMATIC GAIN CONTROL
Design and Performance Analysis
Andrea Morici, Giorgio Biagetti and Claudio Turchetti
DIBET Dipartimento di Ingegneria Biomedica, Elettronica e Telecomunicazioni
Universit` a Politecnica delle Marche, I-60131 Ancona, Italy
{a.morici,g.biagetti,turchetti}@deit.univpm.it
Keywords:
Wireless sensors, Surface EMG, IEEE 802.15.4, LR-WPAN, AGC, Laplacian distribution.
Abstract:
In this paper a wireless system for non-invasive surface electromyography (SEMG) is presented. The use of a
wireless technology, that substitutes cabled electrodes with a wireless link, allows the number of sensors on the
body to be increased without affecting the patients freedom of movement. Problems in this setup, that extend
from energy consumption minimization, to satisfaction of wireless link operational bandwidth and distance
requirements, and from the necessity of embedding hardware in an appreciably small device, to making it not
too expensive to nal customers, have been deeply analyzed and solved. In this context, low rate wireless
personal area networks (LR-WPANs) proved to be a good choice for the realization of low-cost embedded
wireless electrodes for electromyography. Following these considerations, a low-cost electromyographical
wireless device, based on an off-the-shelf IEEE 802.15.4-compatible RF transceiver, have been designed and
realized, and optimized signal processing algorithms developed to enhance the system accuracy. In particular,
due to the wide range of possible amplitudes for the SEMG signal, an optimal automatic gain control, based on
a detailed statistical signal analysis, have been developed to reduce the distorsion at the output of the quantizer.
1 INTRODUCTION
Electromyography (EMG) is a useful diagnosis tech-
nique in the eld of neurophysiology, used for evalu-
ating and recording physiologic properties of muscles
both at rest and while contracting. EMG can either
use needle electrodes (intramuscular EMG) or sur-
face electrodes (surface EMG) (Berzuini et al., 1985).
Naturally, only the latter is a non-invasive technique,
and is the one with which we deal in this paper.
The prediction of muscle force from EMG (Stau-
denmann et al., 2006) may aid with the diagnosis of
some medical conditions in which the electrical ac-
tivity of the muscles or nerves is not normal, such as
nerve compression or injury, nerve root injury, and
with other possibly muscle-related problems such as
deambulation difculties. This technique is not only
useful in hospital environments, but also in rehabili-
tation and motion analysis laboratories, as well as in
tness centers, as it provides an assessment of the
electrical activity generated by contracting muscles
during movements (Jansen et al., 2003). Today the
non-invasive Surface Electromyography (SEMG) has
become very popular owing to the great variety of
applications it can be used in. Recently, to mention
just few examples, EMG has proven to be useful as a
sensor for measuring everyday playing behaviour of
children (Kawakami et al., 2007), as an interface for
inputting characters to a computer (Miyazawa et al.,
2006) and for studies of gait dynamics in free-running
insects (Lemmerhirt et al., 2006).
A typical SEMG analysis may exercise multi-
ple sensors positioned on the patients body, each of
which may require one or more data channels. A
wireless recorder system is thus demanded so that the
patients freedom of mobility does not decrease with
the number of sensors applied. Some issues that an
SEMG recorder system must face are related to its
input sensivity, as it is well established that the am-
plitude of the EMG signal ranges from 0 to 10 mV
(peak-to-peak) or 0 to 1.5 mV (rms) and that is con-
taminated by several sources of noise, thus a high-
gain differential amplier with a CMRR of at least
80 dB is required. Other problems are related to the
number of channels to be acquired: successful anal-
ysis of motion activity involving a group of muscles
39
needs a multiple channel recording, able to evaluate
more than one EMG channel simultaneously. In this
paper a multi-channel SEMG recorder system, con-
sisting of a wireless wearable sensor interfaced to a
standard PC for back-end data analysis, has been de-
veloped to satisfy the requirements of portability, high
sensivity and low-cost.
The research objectives of this article is to pro-
pose a new design of an embedded system made with
off-the-shelf components and suitable for the above
purposes. A statistical analysis of the EMG signal
is provided and exploited to optimize the algorithms,
used to enhance the performance of the low-cost com-
ponents employed in the system, to levels adequate
for the application. This is done by means of statisti-
cally optimized gain adjustment system that is able to
achieve a better SNR respect to xed-gain implemen-
tations.
2 ELECTROMYOGRAPHIC
SIGNALS
A brief introduction on the EMG signals is neces-
sary before presenting the hardware structure of the
system. The source of EMG signal is the electrical
potential generated by muscle cells when they con-
tract. Using a surface electrode, only the general pic-
ture of muscle activation is monitored, whereas the
activity of just a few bers can be observed using nee-
dle electrodes. The amplitude of the resulting sig-
nal can range from less than 50 uV to about 5 mV.
As the usable energy of the signal is limited to the
0 to 500 Hz frequency range, with the dominant en-
ergy being in the 50 to 150 Hz range, the sampling
frequencies needed for its acquisition are in the or-
der of kilohertzs, generally from 1 kHz upwards. In
order to get EMG voltage signals, a differential pair
of Ag/AgCl electrodes is commonly used. These will
pick up the voltage difference and through two very
short shielded conductors it will be fed to an instru-
mentation amplier on the acquisition device, that ad-
justs the amplitude of the EMG signal to an appropri-
ate range for the analog to digital converter (ADC) as
will be shown next.
An example of a typical EMG signal is shown in
Figure 1, which reports its time progress as acquired
by our system at a sample rate of 2 kHz.
The periods where the signal amplitude is low and
at correspond to the muscle being at rest, and the
recorded signal is dominated by electrical noise from
both the environment and the acquisition system. The
parts in which the amplitude is increased above such
noise oor correspond to muscle contractions. The
0 5 10 15
4.5
3
1.5
0
1.5
3
4.5
time [s]
a
m
p
l
i
t
u
d
e

[
m
V
]
Figure 1: EMG signal recorded on a biceps muscle with the
presented system.
subject of the acquisition is a biceps humerus muscle
of a young adult male who was asked to perform re-
peated liftings of his arm following a randomised se-
quence. Some contractions have been kept for several
seconds. The exercise was executed in a non stressed
condition.
2.1 Statistical Analysis for Optimal
Automatic Gain Control
A detailed knowledge of the statistical EMG signal
amplitude is necessary to devise optimum strategies
to properly amplify the EMG signals before feeding
them to the quantizer used for the analog to digital
conversion, since for proper evaluation of muscle ac-
tivity it is of utmost importance that the possibility of
saturation is kept to a minimum, while not degrading
the signal too much due to quantization.
To analyze the statistical properties of an EMG
signal it is necessary to separate the two main parts
previoulsy mentioned: rest condition and activation.
The former is primarily dominated by the noise gener-
ated by the electrical devices in the acquisition board,
and by their susceptibility to irradiated emissions. In-
deed, it can easily be veried that it is white noise
with a Gaussian probability density function (PDF).
The latter part consists in the recording of the over-
all effect of the activation potentials during the mus-
cle contraction. Figure 2 shows an estimate of the
PDF of the recorded signal cleared of the noise-only
portions. Conrming already known ndings (Clancy
and Hogan, 1999), the EMG signal, let us call it x(t),
displays a Laplacian PDF, dened as
f
x
(x) =
1
2b
exp

|x |
b

(1)
where is the mean value, supposedly zero, and b is
the mean absolute value (MAV), that is, b = E{|x|},
and is usually one of the most important features to
be extracted from the EMG signal.
For reference, Figure 3 reports the same
experimentally-obtained distribution plotted against
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
40
3 2 1 0 1 2 3
0
0.2
0.4
0.6
0.8
1
1.2
amplitude [mV]
p
r
o
b
a
b
i
l
i
t
y

d
e
n
s
i
t
y

[
1
/
m
V
]
pdf estimate from data
interpolated Laplacian
Figure 2: Probability density function of the EMG signal of
Fig. 1 compared to its best tting Laplacian distribution.
3 2 1 0 1 2 3
0
0.2
0.4
0.6
0.8
1
1.2
amplitude [mV]
p
r
o
b
a
b
i
l
i
t
y

d
e
n
s
i
t
y

[
1
/
m
V
]
pdf estimate from data
interpolated Gaussian
Figure 3: Probability density function of the EMG signal of
Fig. 1 compared to its best tting Gaussian distribution.
the best-tting Gaussian distribution. There is a clear
mismatch between the two, conrming that the Lapla-
cian better describes the EMG signal statistics.
From a statistical point of view, as the non-
linearity affects the signal amplitude, an optimal au-
tomatic gain control (AGC) must be able to minimize
the distorsion at the output of the quantizer.
Let us then suppose that the quantizer has a satura-
tion level L, and that in between it performs the ideal
staircase quantization described by
y =

x +1/2 |x| < L

L sign(x) |x| L
(2)
then, the expected MAV at the quantizers output is
E{|y|} =
1
b

L+1/2
0
x +1/2e
x/b
dx +
1
b

+
L+1/2
Le
x/b
dx
(3)
where the second term at the right hand side is the
contribution due to saturation effects. The rst inter-
gral can easily be evaluated as follows

L+1/2
0
x +1/2e
x/b
dx =
L

n=1

n+1/2
n1/2
ne
x/b
dx (4)
0 0.1 0.2 0.3 0.4 0.5
14
12
10
8
6
4
2
0
relative input MAV
M
A
V

e
s
t
i
m
a
t
e

r
e
l
a
t
i
v
e

e
r
r
o
r

[
%
]
1024levels quantizer
512levels quantizer
256levels quantizer
Figure 4: Error of MAV estimate after quantization and sat-
uration as a function of input MAV (relative to full scale)
for different quantizer resolutions.
which, after a few simple manipulations, yields
E{|y|} =
1 e
L/b
e
1/(2b)
e
1/(2b)
(5)
Ideally, we would like to operate the quantizer
with an input signal whose amplitude b is such that its
post-quantization estimate E{|y|} is as close as pos-
sible to b. A measure of the error introduced by the
quantizer is the relative MAV error (E{|y|} b)/b,
shown in Figure 4 for different quantizer resolutions
as a function of the relative input signal amplitude
b/L. As can be seen from the graphs, if the input
level is kept close to about 0.1L with a proper AGC
mechanism, which will be described in a following
section, even low resolution quantizers can give ex-
tremely good results.
3 WIRELESS SENSOR NODE
The wireless node we propose comprise all the elec-
tronics needed to measure the biological parameters
of interest, processing them both prior and after their
conversion to the digital domain, and to transfer them
through a wireless link to a nearby observation sta-
tion. All wireless board components were chosen
keeping in mind energy saving, low-cost, high inte-
gration and good electrical property, for the analog
parts, consistently to EMG signal requirements.
Figure 5 shows the block diagram of our wireless
node, while Figure 6 shows a photo of a prototype of
the wireless embedded board for electromyography.
The sensor, that was designed to be of peel-and-
stick type, detects at the skin surface a differential
voltage signal that is amplied by a low-noise dif-
ferential amplier and low-pass ltered in the band
of interest. This is a solution known as active elec-
trode, for the differential amplier is placed as close
as possible to the detection surface of the electrodes,
2.4GHZ WIRELESS ELECTROMYOGRAPH SYSTEM WITH STATISTICALLY OPTIMAL AUTOMATIC GAIN
CONTROL - Design and Performance Analysis
41
uC
VOLTAGE
REFERENCE
A1
A2
DIGITAL
POTENTIOMETER
LPF
LPF
G1
G2
IIC
ADC
PWM ADC
GPIO
SPI
IEEE 802.15.4
PHY
ACCELEROMETER
BIASING
SEMG
Figure 5: Block scheme of the wireless electromyograph
node.
so as to improve the immunity to the noise induced by
external radiated interferences. The signal, so ampli-
ed, is fed through an RC anti-imaging low-pass lter
that limits the upper bound of the frequency spectrum
to be acquired. Its cut-off frequency was chosen to
be approximately 530 Hz.
The active sensor prototype used a two chip so-
lution for the microcontroller part and for the wire-
less physical (PHY) transceiver, to facilitate labo-
ratory testing and debugging, but in future realiza-
tions it will more convenient to switch to Platform-in-
Package (PiP) solutions, where the above mentioned
two chips are integrated into one package, in order to
save cost, space, and PCB design difculties.
The board also comprise the off-chip RF part, in-
cluding PCB traces designed for impedance match-
ing and discrete components chosen to achieve good
impedance matching between the PHY transceiver
and the chip antenna. This kind of antenna was cho-
sen because of its small size and characteristics opti-
mized for our operating wireless band.
The node also includes a three-axis accelerometer
that can provide useful data for motion analysis when
combined to EMG measurements. Moreover, it pro-
vides three additional data channels to demonstrate
the bandwidth capacity in streaming four simultane-
ous real-time channels from the sensor node.
Acceleration data can also be an efcient way to
give commands to the board, for instance to enable
low power operation modes by putting the electronic
devices of the board in standby during inactivity peri-
ods.
Figure 6: The wireless electromyograph embedded board.
Actual size is 50 mm 30 mm.
Figure 7: Details of the instrumentation amplier congu-
ration used.
3.1 Signal Acquisition and Conditioning
The analog signal chain for the EMG signal requires
special attention. Because of the wide range in which
EMG signals can vary, a cascade of two variable
gain intrumentation ampliers was used. A detailed
schematic diagram of their connection is shown in
Figure 7.
These ampliers are also capable of convertingthe
differential EMG signal to single ended, with a high
CMRR, and although the circuit was optimized for
operation with a gain close to 1000 (60dB), by vary-
ing both gains it is possible to change the overall gain
by 40 dB to accomodate very different and possibly
extreme application conditions.
Indeed, the algorithm in the sensor can choose
gains appropriate to different condition of use: for ex-
ample, if you are walking, sensor board will have to
set sensivity parameters to adequate values to sense
low level excitation potentials. On the contrary,
during sport trainings muscles are probably more
stressed, and so the sensor node can adjust gain pa-
rameters to achive best measuring performance.
A dual digital potentiometer serves to make the
gain variable, changing between a set of 256 different
values of resistance for each stage. This level of pre-
cision is very useful when the node is moved between
different muscle masses that need different gain lev-
els, and allows for the ne-grained AGC that ensures
optimal quantization.
The analog signal so amplied is subsequently
converted to a digital signal by the ADC included
in the microcontroller, operating at a 2 kHz sampling
rate and with a maximum resolution of 10 bits.
To provide a voltage bias to the output of the am-
pliers, an integrated low-cost, band-gap reference
was used. The resistance seen between the output ter-
minal of the band-gap reference and ground is only
a fraction of an ohm, ensuring that the amplier in-
trinsic insensibility to common-mode voltages is not
reduced because of it. With a high-CMRR input stage
and a small battery-powered sensor node mounted
close to the patients skin, so as to minimize stray ca-
pacitance to ground, the use of a reference electrode
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
42
to set the patients body potential was deemed unnec-
essary. Proper biasing of the input stage is provided
directly to the main electrodes by means of a resistive
network connected to the internal reference potential.
Despite all these precautions, it is still possible
that the detected signal presents a DC component, or
an input bias, high enough to cause saturation of some
amplier. To avert this possibility the circuit employs
an active bias control, fed between the two amplifying
stages by means of a PWM signal, that is able to can-
cel both amplier bias and (small) DC components
that can arise when the patient moves other parts of
their body.
This active offset compensation technique was
deemed to be superior to simple AC coupling, for it
can use more complex lters. Moreover, since the
communication channel between the node and the
base station is bidirectional, these lters can be run
on the PC side and thus there are potentially no lim-
its to their complexity. For instance, we made use of
fourth-order elliptical digital lters running on the PC
side, with their group delay properly compensated in
the reported plots. Such a solution proved to be very
good at removing the spikes, usually due to patients
motion, commonly found in EMG traces, but would
have been much more costly to implement in hard-
ware.
3.2 AGC Algorithm
AGC poses a different problem than offset and bias
compensation, since the latter are usually slowly
time-varying phenomena for which the round-trip de-
lay, due to the communication with the base station
system, does not cause harmful degradation of the
control loop stability and performance. On the con-
trary, the EMG signal can have quite abrupt transis-
tions. As a consequence, a simple but yet effective
AGC algorithm was devised, so that it could be run
on the wireless node to offer the quickest possible re-
sponse time.
As previously stated, the purpose of the AGC is
to keep the input MAV level to the quantizer as close
as possible to 0.1L, where L is the ADC saturation
level. To this end, an estimation

b(t) of the MAV is
computed with a rst-order recursive digital lter,

b(t) = (1 )

b(t 1) +|z(t)/g(t)| (6)

with z(t) being the ADC output, g(t) the amplier
gain, and controls the lter bandwidth. Good re-
sults have been obtained with 1/64, which corre-
sponds to a cut-off frequency of about 5 Hz.
The optimum gain is then calculated as
g(t +1) =
0.1L

b(t)
(7)
from which the actual gain g(t +1) to be used next
is chosen among the available gains, in steps of ap-
proximately 2 dB, with the help of a 22-entry look-up
table.
3.3 Wireless Data Transmission
LR-WPAN are emerging technologies for medium
distance low data rate communications. A protocol
to manage this kind of networks has been dened by
IEEE 802.15.4, which describes both a MAC layer
and a PHY layer. The operating frequencies of the
wireless link can be 868MHz, 915MHz or 2.4GHz,
for an available data rate respectively of 20 kbps,
40 kbps and 250kbps. Our active sensor operates
at 2.45 GHz in the ISM band to achieve maximum
throughput. In this band there are 16 channels, each
5 MHz wide. Typical distances covered by this tech-
nology ranges from 30 m to 70 m in open spaces. It
can be easily extended by the use of an RF power am-
plier joined to an LNA. Typically they are the same
as for other ISM wireless technologies such as Blue-
tooth and Wi-Fi. In customized applications IEEE
802.15.4 could imply difculties in respecting timing
constraints posed by real-time streaming of data, such
as the one we need to perform in this context. We
hence decided to only use the capabilities of the PHY
layer of IEEE 802.15.4, customizing the MAC layer
to our purposes. A number of active SEMG sensors,
depending on how many data channels each uses, can
comunicate to the base station (BS) in a star topology
on the same RF channel, using a custom beaconed
time-division multiple access (TDMA) MACscheme.
The BS is itself composed by an IEEE 802.15.4 com-
pliant transceiver and its task is to make data available
to the PC by the use of an USB link. For stream-
ing data from multiple sensors and for achieving full-
duplex operation it is necessary to assign time slots
to each sensor and to transmit/receive transactions.
Transmission and reception has to be scheduled by
devising an adequate timing of the active sensor con-
sidering the strict requirements of the ADC sampling
time. The adopted transceivers have particular tim-
ings regarding the transmission over the air of a data
packet. There is a warmup period t
warmup
= 144s be-
fore the effective bitstream can be relayed, followed
by a t
cooldown
= 10 s cooldown period. Timings are
then coherent with those reported in Figures 8 and 10:
t
pkt
(B) =t
warmup
+t
header
+Bt
byte
+t
trailer
+t
cooldown
(8)
t
tx
t
pkt
(B)

B=B
tx
(9)
with t
byte
= 32 s as per IEEE 802.15.4 specications,
and where the payload length B
tx
is comprised of the
EMG data bytes B
EMG
and of the acceleration data
2.4GHZ WIRELESS ELECTROMYOGRAPH SYSTEM WITH STATISTICALLY OPTIMAL AUTOMATIC GAIN
CONTROL - Design and Performance Analysis
43
Preamble SFD FLI CRC
144 us 192 us 64 us
4 bytes
1
byte
2 bytes
Payload
1
byte
Figure 8: PHY packet transmission timings.
Header
3.968 ms
4 bytes 96 bytes 2 bytes
EMG Z / BATT
18 bytes
ACC Gains
4 bytes
Figure 9: Packet payload content.
bytes B
ACC
, apart from side channels (e.g. gain lev-
els) and node status information, as shown in Fig. 9.
Packet length must be chosen as a compromise be-
tween latency and channel bandwitdth utilisation, as
shown below.
Let us call t
adc
the time needed by the ADC to ll
B
i
bytes of the payload, for the EMG and each of the
three accelerometer channels:
t
adc
=
B
i
N
s
n
i

1
F
s
(i)
(10)
where N
s
is the number of bytes per sample, n
i
the
number of channels per sensor type in each node,
sampled at a sample rate F
s
(i) (usually n
EMG
=1 for the
single EMG channel sampled at F
s
(EMG)
= 2 kHz, and
n
ACC
=3 for the three accelerometer channels sam-
pled at a much lower sample rate F
s
(ACC)
=125 Hz).
Then we have the constraint
t
rx
> t
pkt
(B)

B=B
rx
, t
rx
=t
adc
mt
tx
(11)
where B
rx
is the number of bytes needed for the con-
trol channel, and m is the maximum number of wire-
less nodes simultaneously active.
Sensor
Boards
Base
Station
Timeline
Wireless
Link
TX
RX
TX
RX
RX
TX
RX
TX
Personal
Computer
~4.4 ms
x m
~1 ms
~4.4 ms
x m
~1 ms
Cabled
Link
#1
#2
#m
Figure 10: Communication ow scheduling of the various
system components.
Table 1: Relative mean square error of the post-quantizer
MAV estimate, for different quantizer resolutions, and for
either a xed gain of 60 dB and for AGC-determined gains.
gain policy 8 bit 9 bit 10 bit
xed gain 36.3dB 47.5dB 55.1dB
AGC 57.1dB 63.8dB 67.1dB
The samples of the myo-electric and acceleromet-
ric signals are packed for a better efciency of trans-
mission. Each packet has a header dened by means
of a protocol that has been developed for this specic
application, in order to performthe temporal synchro-
nization and the recovering of data ow in a sensor
network. With the payload structure shown in Fig. 9,
dened for ADC resolutions of up to 12 bits, and with
the previously mentioned sample rates, the theoret-
ical maximum number of nodes m operating on the
same RF channel is 6, with an extra time slot available
for retransmissions in case of errors. The actual limit
depends on the particular RF environment and back-
ground noise, which affects the need and frequency of
retransmissions, and is currently being investigated.
4 EXPERIMENTAL RESULTS
The prototype board was tested and used to acquire
a sample EMG signal, reported in Fig. 1. Although
proper shielding of the prototype was not employed
due to the presence of auxiliary debugging and de-
velopment connections, a quite good signal-to-noise
ratio was achieved. Power consumption resulted in
about 10mA, with many power-saving optimizations
that can still be implemented in the control software,
thus making it possible to achive a battery life of 5
to 10 hours of continuous operation out of a standard
coin-size rechargeable Li-ion cell.
In order to show the effectiveness of the proposed
AGC for EMG signal acquisition, some other experi-
ments were also made. Figure 11 shows some of the
results. It displays an EMG signal used as a refer-
ence, and the amplier gains that were chosen by the
AGC algorithm for each portion of the signal. The re-
sult was then quantized, and the quantized signal used
to compute an estimate of the MAV. The MAV was
smoothed with a fourth-order elliptical lter and the
result shown in the same gure, and compared with
that obtained from the unquantized version.
The results of the comparisons are shown in Ta-
ble 1, which compares the errors made with different
gain policies and different quantizer resolutions. As
can easily be seen, the adoption of AGC permitted
us to obtain an increase in the accuracy of the MAV
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
44
0 10 20 30 40 50 60 70 80
1.0
0.5
0.0
0.5
1.0
time [s]
a
m
p
l
i
t
u
d
e

[
m
V
]
0 10 20 30 40 50 60 70 80
60
80
100
time [s]
g
a
i
n

[
d
B
]
0 10 20 30 40 50 60 70 80
0.0
0.1
0.2
0.3
time [s]
M
A
V

[
m
V
]
Figure 11: Operation of the AGC and resulting MAV es-
timate. From top to bottom: original EMG signal, optimal
gain as selected by the AGC algorithm, post-quantizer MAV
estimate.
estimation equivalent to more than two extra bits of
resolution in the ADC, thus making it possible to use
cheaper components without any serious degradation
of the performance.
5 CONCLUSIONS
A complete wireless electromyographic system was
developed, comprising a wearable sensor board, in-
teracting with software running on a common PC for
elaboration of the acquired data, and remote control
of the acquisition circuitry for optimal system perfor-
mance.
Moreover, the employed technology offers the ca-
pability of conguring reasonably large sensor net-
works for the analysis of several concurrent muscle
ber activities, also with base stations distributed in
large area laboratories, and for example makes the
analysis of running patients easier. The wireless EMG
sensor makes the patients free of cumbersome wires
and heavy transmitters and, as their movement are
more natural, the resulting analysis is more adherent
to reality. In comparison with other wireless technolo-
gies, these devices have a lower power consumption, a
longer battery life, and the networks they realize can
have a greater number of nodes and cover a longer
distance.
In future works the microcontroller will be substi-
tuted by a low-cost DSP, embedding all the peripher-
als on it and augmenting the on-board signal process-
ing capabilities.
REFERENCES
Berzuini, C., Figini, M. M., and Bernardinelli, L. (1985).
Evaluation of the effectiveness of EMG parameters in
the study of neurogenic diseases-a statistical approach
using clinical and simulated data. IEEE Transactions
on Biomedical Engineering, BME-32(1):1527.
Clancy, E. A. and Hogan, N. (1999). Probability density
of the surface electromyogram and its relation to am-
plitude detectors. IEEE Transactions on Biomedical
Engineering, 46(6).
Jansen, B. H., Miller, V. H., Mavrofrides, D. C., and Jansen,
C. W. S. (2003). Multidimensional EMG-based as-
sessment of walking dynamics. IEEE Transactions
on Neural Systems and Rehabilitation Engineering,
11(3).
Kawakami, G., Nishidaa, Y., and Mizoguchi, H. (2007).
In situ measurement of playing children by wireless
wearable electromyography. In Proc. IEEE Sensors
Conference 2007, Atlanta, GA, USA.
Lemmerhirt, D. F., Staudacher, E. M., and Wise, K. D.
(2006). A multitransducer microsystem for in-
sect monitoring and control. IEEE Transactions on
Biomedical Engineering, 53(10).
Miyazawa, K., Ueno, A., Mori, H., Hoshino, H., and
Noshiro, M. (2006). Development and evaluation of a
wireless interface for inputting characters using lapla-
cian EMG. In Proc. 28th IEEE EMBS Annual Inter-
national Conference, New York City, NY, USA.
Staudenmann, D., Kingma, I., Daffertshofer, A., Stegeman,
D. F., and van Die en, J. H. (2006). Improving EMG-
based muscle force estimation by using a high-density
EMG grid and principal component analysis. IEEE
Transactions on Biomedical Engineering, 53(4).
2.4GHZ WIRELESS ELECTROMYOGRAPH SYSTEM WITH STATISTICALLY OPTIMAL AUTOMATIC GAIN
CONTROL - Design and Performance Analysis
45
AN ASYNCHRONOUS PROGRAMMABLE PARALLEL
2-D IMAGE FILTER CMOS IC BASED ON THE
GILBERT VECTOR MULTIPLIER
Rafa Dugosz
Swiss Federal Institute of Technology in Lausanne, Institute of Microtechnology
Rue A.-L. Breguet 2, CH-2000, Neuchtel, Switzerland
[email protected]
Vincent Gaudet
University of Alberta, Department of Electrical and Computer Engineering
ECERF Building, Edmonton Alberta, T6G 2V4, Canada
[email protected]
Keywords: Parallel and asynchronous 2-D analog filter, Gilbert vector multiplier, Ultra-low power dissipation.
Abstract: A novel analogue power-efficient 2-D programmable finite impulse response image filter is proposed. This
solution is based on the current-mode Gilbert-vector-multiplier operating in the weak inversion region,
which allows for ultra low power operation. The main advantage is in the asynchronous and parallel
calculation of all pixel values without using any clock generator. The filter is a programmable structure that
allows programmability of different filter masks both low-pass and high-pass. An experimental filter
integrated circuit with a resolution of 6x1 pixels dissipates in measurements a power of 30 W at a data rate
of 30 kframes/s in a 180 nm CMOS technology. One of the intended applications of our proposed image
filter is in data compression in wireless endoscopic capsules.
1 INTRODUCTION
The advent of inexpensive CMOS image sensors has
recently led to the design of wireless capsules for
endoscopic diagnosis (Meng et al., 2004) (Xie et al.,
2006). Such capsules usually contain the CMOS
image sensor, signal processing circuitry, and a
communication block between the capsule and a
transceiver located outside the patients body that
records collected data. Such a system, together with
light-emitting diodes and a battery, is placed in a
small pill that is hermetically packaged to be robust
against enzymes and acid in the human digestive
tract. Such capsules must last 8 to 24 hours as they
travel throughout the digestive tract while capturing
about 50,000 photos that are then used for diagnostic
purposes. Thus the capsule must collect and transmit
an extremely large amount of information (Miaou,
2005). This problem is usually addressed by
different compression techniques e.g. a discrete
wavelet transform (DWT) (Miaou, 2005) and
recently reported near-lossless compression
algorithm for the images with a Bayer color filter
array (CFA) (Xie et al. 2006).
In these systems, analog data is typically
captured by a CMOS image sensor and then
processed in a specialized digital circuit (Xie et al.
2006). In this paper we propose a programmable
analog, asynchronous, parallel 2-D finite impulse
response (FIR) filter that can be used as a pre-
processing filter prior to the analog-to-digital
conversion (ADC) as well as in the filter banks in
the DWT-based compression algorithms. Our
proposed filter uses the current-mode Gilbert vector
multiplier circuit operating in the weak inversion
regime, leading to ultra-low power dissipation and
energy savings (Winstead, 2004), which is one of
main criterions in endoscopic capsule applications.
One of the advantages of this approach is the
speedup that results from the parallel and
asynchronous calculation of all pixel values without
using a clock generator.
Research has been conducted into analog image
filters for many years. An example image filter with
an array of 16x16 pixels that dissipates 165mW at
46

the I/O data rate of 5x10
7
events per second has been
described by (Serrano-Gotarredona et al., 2008).
This solution allows for connecting of many chips
into bigger systems with higher resolutions.
Image filters are also implemented using cellular
neural networks (CNN). An example filter of this
type, designed in a 0.5 m CMOS technology, with
a resolution of 64x64 pixels at an I/O analog data
rate of 1 MSps, dissipates 1.5 W (Linan et al., 1999).
The experimental prototype filter described in
this paper has been designed for a resolution of 6x1
pixels, but as the circuit features a modular structure,
therefore it can be redesigned for higher resolutions.
Based on the simulation and the measurement
results, we project a power dissipation of 22 mW for
a 64x64 pixel filter operating at a throughput of 10
Mpixels/s (measurements) or even 4 Gpixels/s
(simulations) that means a significant improvement
in comparison to the solutions described above.
The paper is organized as follows. In next
section we present the circuits that we use for FIR
filtering. An experimental filter realized in TSMC
180nm CMOS processis described in the subsequent
section together with selected measurement results.
Finally the last section concludes this paper.
2 GENERAL IDEA OF THE
PROPOSED IMAGE FILTER
A basic 2-D FIR filtering scheme is shown in Fig. 1.
Pixels from a 2-D input signal A are first multiplied
by filter coefficients from a mask H. The products of
these operations are summed, thus producing pixels
for an output image B, according to the following
equation (for e.g. 3x3 mask H):

= =
+ + =
3
1
3
1
) , ( ) 2 , 2 ( ) , (
n m
m n h m y n x A y x B (1)
When the image filtering is implemented in DSP
systems the mask is moved over the input image A
and the output pixels are calculated sequentially,
which is a time and power consuming process.

A- an input 2-D signal - an output 2-D signal B

1
2
2
2
3
1
4
7
9
4
2
2
2
5
1
3
8
3
6
3
4
6
8
4
1
1
2
2
1
8
4
3
1
6
1
1
7
5
3
9
6
3
1
2
1
2
2
2
1
1
1
maskH
y
y
x
x
3
9
1
7
8
7
2
2
1
317
7
2
2
4 2
1
7
1
1
1
2
2 2 2
2

Figure 1: Two-dimensional FIR image filtration.
On the other hand, the solution proposed in this
paper allows for a parallel calculation of all pixels
B
xy
. As a result time that is required to calculate a
single image B is not dependent on the number of
pixels in this signal. The proposed filter additionally
is a programmable structure that can be quickly
reprogrammed to perform both the low-pass and the
high-pass filtering, which is an important feature e.g.
in realizing 2-D DWT filter banks.
The basic circuit used in the proposed filter is the
current-mode Gilbert vector-by-scalar multiplier
(GVSM) shown in Fig. 2 (Winstead, 2004). A vector
of the output currents P
xy
={I
p1
, I
p2
, , I
pN
} is
calculated according to the following formula:
T
0
T
h I
h
h I
P
xy
N
i
i
xy
xy
= =

=

(2)
in which h={h
1
, h
2
, , h
N
} is a vector of
currents, which in the proposed application are
proportional to the filter coefficients, while I
xy
is the
current proportional to an input pixel A
xy
. The
currents h
i
are adjusted off-line and then kept
constant during filtering, although they can be
reprogrammed very quickly if necessary.
xy
I
p1
I
h0
I
p0
I
hN
I
pN
I
h1
I
p
i
I
S
=

I

Figure 2: Gilbert scalar-by-vector multiplier (GSVM).
p
412
p
413
p
141
p
142
p
143
p
441
p
443
p
111
P
xy
h
2
h
1
xy
I
p
442
output
pixel
xy
I
p
121
p
131
p
141
p
211
p
221
p
pixel
p
input
411

412
p
411
p
441
p
442
p
111
p
141
p
142
p
112
p

Figure 3: Calculation scheme in our proposed filter: two
possible views.
The calculation scheme in the proposed filter is
illustrated in Fig. 3. Each column in this 3-D block
contains a single GSVM circuit that provides the
AN ASYNCHRONOUS PROGRAMMABLE PARALLEL 2-D IMAGE FILTER CMOS IC BASED ON THE GILBERT
VECTOR MULTIPLIER
47

vector P
xy
. Notice that lengths of this vector as well
as of vector h are equal to the number of the filter
coefficients that have unique values. For example,
the low-pass filter mask, shown in Fig. 1, contains
only 2 different values, namely: {1, 2}, so the length
of the vector h is in this case equal to 2.
S
xy
p
3
xy
p
3 xy
p
3 xy
p
3 xy
p
3 xy
p
3 xy
p
3
p
xy2
p
xy2
p
xy2
p
xy2
p
xy2
p
xy2
p
xy2
p
xy1
p
xy1
p
xy1
p
xy1
p
xy1
p
xy1
p
xy1
p
xy2
p
xy2
xy
p
3 xy
p
3 c c b b a a i i
c c b b a a i i
c c b b a a i i
p
xy1
p
xy1
d
1
1
3
1
d
1
1
3
2
d
1
2
3
1
d
1
2
3
2
d
1
3
3
1
d
1
3
3
2
d
3
3
3
1
d
3
3
3
2
d
3
3
2
2
d
3
3
2
1
d
1
3
2
1
d
1
3
2
2
d
1
2
2
1
d
1
2
2
2
d
1
1
2
2
d
1
1
2
1
d
1
1
1
1
d
1
1
1
2
d
1
2
1
1
d
1
2
1
2
d
1
3
1
1
d
1
3
1
2
d
3
3
1
2
d
3
3
1
1
I
xy
Ih2
Ih3
Ih1
p
xy
p
xy
p
xy
p
xy
p
xy
p
xy c c b b a a i i
p
xy
p
xy
) ( (+) ) ( (+) ) ( (+) ) ( (+)
I

Figure 4: Realization of a pixel block with a program-
mable connection map.
(a)

1 2 1
2 2 2
1 2 1

k I
k I
h
h
=
=
1
2
2
1

00
n
m
10 10
10 10
10
10 10
10
10
00
00
00
00 00
00 00
00

(b)

0 1 0
1 0 1
0 1 0

k I
k I
h
h
=
=
1
1
2
1

+
00
00
00
00
00 10
00
sign
00 00
00 00 00
00
00
00
00
00
m
n
01

(c)

1 2 1
2 4 2
1 2 1

k I
k I
h
h
=
=
1
3
2
1

+1
n
m
10
10 10
10
10 10
10
10
01
10
01
01 01
00 00
00 00
10
+3
+1
+3
-1

Figure 5: Example connection maps for: (a) simple low-
pass filter from Fig. 1, (b) simple high pass filter, (c) low-
pass filter with combined coefficients. In the last case
value 2 is realized as 3-1, while value 4 as 3+1.
The GSVM circuit calculates currents I
p
that are
products of the input pixels A
xy
and particular
coefficients h
i
. As a result, the structure shown in
Fig. 3 provides all necessary partial products I
p
(p
for short) in parallel, and these can be then used to
calculate any output pixel B
xy
. To realize it, a proper
connection map is required between the signals p
xyz

and the output signals B
xy
. Notice that particular
products p
xyz
can be used in several output samples,
and therefore an equivalent number of copies of
each of these signals must be available. A single
pixel block, which provides all necessary copies of
p
xyz
, is shown in Fig. 4. Particular branches
(connections) in this circuit are controlled using
configuration bits d
n,m,z,l
that are programmed off-
line. These bits establish only the connection map,
but values of the filter coefficients finally depend on
values of the currents I
h
.
)
I
DC
p
1
p
2
p
(+)
1
p
(-)
1
d
12
d
11
d
22
d
21
d
N2
d
N1
p
(+)
d
...1
p
(-)
d
...2
I
DC
p
N
i i
= B
xy
( )
i i
(

Figure 6: Calculation scheme of the output signal B
xy
.
It is convenient to collect the configuration bits d
into a single 4-dimensional matrix D, in which the
first two dimensions (n, m) are determined by
dimensions of the filter mask H. The 3
rd
dimension
(z) is determined by the length of the vector P
xy
, i.e.
by number of unique filter coefficients, while the 4
th

dimension is always equal to 2. Theoretically
products from a given vector P
xy
can be used in
calculation of nm output samples. In this situation
bits that are under position n, m in the matrix D
determine how the products from this vector will be
used to calculate a given signal B
xy
. The 3
rd

dimension determines which product(s) from a given
vector P
xy
will be added to a given signal B
xy
, while
the last dimension determines if this product(s) will
be added with a positive or the negative value.
Taking this principle into account, an example
matrix D in case of the mask shown in Fig. 1 would
have dimensions 3, 3, 2, 2, so the total number of the
configuration bits d is in this case equal to only 36.
Several example connection maps for different
filter masks both the high-pass and the low-pass,
together with required currents representing
particular filter coefficients h
i
, are shown in Fig. 5.
In presented examples 2 unique filter coefficients are
assumed for simplicity.
It is worth noting that when two different filters
have an equal connection map and differ only by
value of selected coefficients then switching
between such filters requires only adjusting the
currents I
h
. It is also worth noting that although the
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
48

number of coefficients in the vector h is usually low,
these coefficients can be combined together to
provide an effectively larger number of coefficients,
as in the example map shown in Fig. 5 (c).
Products p that are added as negative values are
summed in a separate junction, then inverted in a
single NMOS-type current mirror, and finally added
to the output signal, as shown in Fig 6. The potential
problem in this approach occurs when sum of the
negative values is larger than sum of the positive
values, which may turn off the output current mirror.
This problem has been solved by introducing an
additional constant current I
DC
, which can be
optionally added to the output signal in such a case.
3 CMOS IMPLEMENTATION OF
THE PARALLEL FILTER AND
EXPERIMENTAL RESULTS
An experimental image processing block has been
realized by authors in a 180 nm CMOS process. In
this prototype the image resolution is equal to 6 x 1
pixels, while the mask has dimensions 3 x 1. Such
parameters are sufficient to verify the concept, but as
the filter has a modular structure, higher resolutions
can be easily realized. For example, the filter with
8x8 pixels with 3x3 masks has been successfully
verified in the HSPICE simulations.

Figure 7: Experimental programmable image filter
realized in three versions with different transistor
dimensions.
The internal structure of the chip is shown in Fig.
7. The common problem in the Gilbert multipliers
used in the proposed filter, in which transistors
operate in weak inversion is influence of the
transistor mismatch on the filter accuracy. To verify
influence of this effect the experimental filter has
been designed in three versions that differ only in
the transistor dimensions (filter 1, filter 2, filter 3).
The measurement results show that mismatch has a
limited influence only on a DC value at particular
filter outputs, which can be easily corrected at the
digital side, after analog-to-digital conversion.
0.3
0.35
0.4
0.45
0.5
0.55
0.0008 0.0009 0.001 0.0011 0.0012 0.0013 0.0014 0.0015

0.105
0.11
0.115
0.12
0.125
0.13
0.0008 0.0009 0.001 0.0011 0.0012 0.0013 0.0014 0.0015 Time [s]
Vout [V]
B1 B6
B2 B5
B3
B4

Figure 8: Measurements for small signals for the low pass
filter (top) input signals A3 and A4 (bottom) the filters
outputs B1-B6. Results for are for the filter 2.
2.9
3.1
3.3
3.5
3.7
3.9
4.1
4.3
4.5
4.7
7 8 9 10 11 12 13 Ti me [us]
I out [uA]
B3, B4
B2
B5
B1
B6
T_ f <1 us

0.3
0.32
0.34
0.36
0.38
0.4
0.42
0.44
0.46
0.48
0.0001 0.0006 0.0011 0.0016 0.0021 Ti me [s]
Vout [V]
T_f =0.1ms
B5
B6
B1
B2
B3, B4

Figure 9: Low pass filter (top) HSPICE postlayout
simulations and (bottom) the measurement results.
Selected measurement results and post-layout
simulations have been presented to demonstrate the
filter performance. The first and the second
experiments, shown in Figs. 8 and 9, are for lowpass
filter mask, H =[1, 1, 1] for small and for large input
signals respectively. Fig. 8 illustrates additionally
the input waveform applied to the A3 and A4 inputs.
The other inputs are constant and equal to the
bottom value of this pulse signal.

AN ASYNCHRONOUS PROGRAMMABLE PARALLEL 2-D IMAGE FILTER CMOS IC BASED ON THE GILBERT
VECTOR MULTIPLIER
49

2.9
3.1
3.3
3.5
3.7
3.9
4.1
4.3
4.5
4.7
7 8 9 10 11 12 13 Ti me [us]
I out [uA]
B3, B4
B2
B5
B1
B6
T_f <1 us

0.3
0.32
0.34
0.36
0.38
0.4
0.42
0.44
0.46
0.48
0.0001 0.0006 0.0011 0.0016 0.0021 Ti me [s]
Vout [V]
T_f =0.1ms
B5
B6
B1
B2
B3, B4

Figure 10: High pass filter (top) HSPICE postlayout
simulations and (bottom) the measurement results.
Table 1: Example results for the low-pass filter [1,1,1]
(see Fig. 8).
Input [A] A
1
A
2
A
3
A
4
A
5
A
6

2.5 2.5 5.098 5.098 2.5 2.5
Out [A] B
1
B
2
B
3
B
4
B
5
B
6

calculated 3.880 4.461 4.757 4.757 4.176 3.595
simulated 3.595 4.475 4.761 4.761 3.885 3.002
measured 3.562 4.456 4.768 4.754 3.851 2.940
er_S/C % 15.68 0.77 0.22 0.22 16.03 32.60
er_M/C %
17.50 0.27 -0.60 0.20 17.88 36.03
er_M/S % -1.81 -1.04 0.39 -0.41 -1.86 3.43
Table 2: Example results for the high-pass filter [1,-1,1]
(see Fig. 9).
Input [A] A
1
A
2
A
3
A
4
A
5
A
6

2.5 2.5 5.098 5.098 2.5 2.5
Out [A] B
1
B
2
B
3
B
4
B
5
B
6

calculated
2.195 2.420 2.186 2.420 2.654 2.645
simulated
1.885 2.407 2.186 2.405 2.686 3.001
measured
1.969 2.416 2.184 2.402 2.643 2.913
er_S/C %
32.83 1.38 -0.01 1.63 -3.40 -37.7
er_M/C %
23.94 0.40 0.23 1.86 1.12 -28.4
er_M/S %
8.89 0.97 -0.24 -0.23 -4.52 -9.26
A detailed comparison for the experiment with
large input signals is provided in Table 1. To
investigate the dynamic parameters of the filter, the
input image, A, was varied during the test. The A
1
,
A
2
, A
5
and A
6
input samples were constant during the
test and equal to 2.5 A, while the others (A
3
, A
4
)
were oscillating in the range between c.a. 2.19 and
5.1 A. Theoretical (calculated), simulated and
measured values of the output pixels are given in the
4
th
, 5
th
and 6
th
rows respectively.
The results have been compared in terms of the
errors given in last three rows, in which S/C means
simulated vs. calculated results etc. All presented
results are for supply voltage 0.8 V.
The results for an example highpass filter with a
mask H =[-1 1 0] are shown in Fig. 10 and
compared in details in Table 2.
Notice that error values in signals B
2
-B
4
are
below 1-2 %. This is regularly observed for different
input images and filter masks. On the other hand,
errors concerning the B
1
, B
6
and for the low-pass
filter additionally B
5
outputs (due toA
6
being zeroed)
are larger, which is due to the border effect where
samples are calculated using only two inputs A
i
.
The noise at the output is at a level of 1 mV the
currents are measured as voltages on 100 k
resistors. The measured dynamic range depends on
the average value of the input pixels as well as on
type of the filter mask. The example measured
values are as follows:

a) for large signals (about 3 A),
low pass filter SNR =36dB (6 bits); see Fig. 9
b) for large signals (about 3 A),
high pass filter SNR =30dB (5 bits); see Fig. 10
c) for small signals (about 0.6 A),
low pass filter SNR =24dB (4 bits); see Fig. 8
d) for small signals (about 0.6 A),
high pass filter SNR =21dB (3 bits)

The maximum data rate depends on values of the
input signals and the transistors dimensions used in
GSVM circuits. The example data rates are as
follows:

a) filter 1 (W/L=15/5 m)
is equal to 15.15 [kframes/s],
b) filter 2 (W/L=8/2 m)
is equal to 21.74 [kframes/s],
c) filter 3 (W/L=6/1 m)
is equal to 33.33 [kframes/s]
In our experimental filter each frame consists of
6 pixels. Power dissipation depends on values of the
input currents I
xy
. For small input values c. 0.6 A
power dissipation is equal to 5 W, while for larger
values c. 3 A it equals to 30 W.
It is worth noting that filter accuracy, which has
an influence on the gray scale depth, does not differ
significantly between the simulations and the
measurements, although the environmental noise
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
50

limits the possible SNR. The noise is kept at a level
of 1 mV, which limits the filter accuracy by c. 6 dB
(1 bit). The accuracy is lower in case of high-pass
filters, which is due to additional signal inverting.
On the other hand the data rate significantly
differs between simulations and measurements. The
lower data rate attained in measurements is due to
large capacitances of pads as well as the setup
environment. The conclusion is that when the filter
is part of a bigger system, in which the output analog
signals are further processed in the chip, the data
rate can be higher and closer to the simulation
results. In this case energy dissipated per calculation
of a single pixel will be much smaller.
4 CONCLUSIONS
A 2-D analog, current-mode FIR filter has been
proposed in this paper. The main building block in
our filter is the Gilbert scalar-by-vector multiplier
that allows for an ultra-low power dissipation due to
transistors operating in weak inversion.
The proposed filter is a programmable solution,
with a very simple logic block and small number of
programming bits that allow for realization of
different filter masks both the high-pass and the low-
pass. One of the main advantages is a parallel and
asynchronous calculation of all output pixels without
using clock generators that typically are source of
feedthrough noise.
To verify the proposed idea, three experimental
image filters with different transistor dimensions
have been designed in a 180 nm CMOS technology.
Attenuation observed in post-layout HSPICE
simulations reaches a level of about 55 dB.
Theoretical analysis and measurement results
concerning the influence of the transistor-threshold-
voltage mismatch on the filter properties shows that
even in the worst case scenario, attenuation higher
than 36-dB (6 bits) can be achieved, for the
mismatch that is at the level of 2-3 %, which is
sufficient for many practical applications e.g. in
endoscopic capsules. The attained lower attenuation
is caused by an environmental noise that is present
in the input signal, as shown in Fig. 8 (top).
The filter performance is summarized in Table 3.
The data rate is given in image frames/s as in our
proposed filter this parameter does not depend on
number of pixels in a single frame. The data rate
attained in measurement significantly exceeds those
usually required is endoscopic capsules i.e. several
frames/s (Xie et al., 2006). This creates the
possibility to switch off the circuit for most of the
time and to save energy, which is one of the key
criteria in wireless endoscopic capsules.
Table 3: Summary of the image filters performance.
Parameter Small signals Large signals
Voltage supply 0.8 V 0.8 V
effective data rate(measur.) 15 kframes/s 15 kframes/s
effective data rate(simulate.) 350 kframes/s 1 Mframes/s
SNR (dB) 21-24 (3-4 bits) 30-36 (5-6 bits)
Power dissipation (6 pixels) 5 W 30 W
Energy/pixel (measured) 55 pJ 250 pJ
Energy/pixel (simulated) 2.3 pJ 5 pJ
Process TSMC CMOS 0.18 m
Die area - one filter: 6 pixels 350 x 150 m(0.052 mm
2
)
Image/mask resolution 6 x 1 / 3 x 1
ACKNOWLEDGEMENTS
The work is supported by EU Marie Curie Outgoing
International Fellowship No. 021926
REFERENCES
Meng, M.Q.-H. Tao Mei, J iexin Pu, Chao Hu, Xiaona
Wang, Yawen Chan, Wireless robotic capsule
endoscopy: state-of-the-art and challenges, 5th World
Congress on Intelligent Control and Automation
(WCICA), 2004, Vol. 6, J une 2004, pp.5561 - 5555
X. Xie, G. Li, X. Chen, X. Li, Z. Wang, A Low-Power
Digital IC Design Inside the Wireless Endoscopic
Capsule, IEEE J ournal of Solid-State Circuits, Vol.
41, Issue 11, Nov. 2006, pp. 2390 - 2400
Shaou-Gang Miaou, Shih-Tse Chen, Fu-Sheng Ke,
Capsule endoscopy image coding using wavelet-
based adaptive vector quantization without codebook
training, 3rd International Conference on Information
Technology and Applications (ICITA), Vol. 1, July
2005, pp. 634 - 637
C. Winstead, Analog Iterative Error Control Decoders,
Ph.D disserta-tion, University of Alberta, ECE
Department, Edmonton, ABa, 2004.
G. Linan, P. Foldesy, S. Espejo, R. Dominguez-Castro, A.
Rodriguez-Vazquez, A 0.5m CMOS 106 transistors
analog programmable array processor for realtime
image processing, 25th European Solid-State Circuits
Conference (ESSCIRC), 1999, pp. 358-361.
R. Serrano-Gotarredona, T. Serrano-Gotarredona, A.
Acosta-J imenez, C. Serrano-Gotarredona, J . A. Perez-
Carrasco, A. Linares-Barranco, G. J imenez-Moreno,
A. Civit-Ballcels, and B. Linares-Barranco, "On Real-
Time AER 2D Convolutions Hardware for
Neuromorphic Spike Based Cortical Processing,"
IEEE Trans. on Neural Networks, vol.19, No.7, pp.
1196-1219, J uly 2008
AN ASYNCHRONOUS PROGRAMMABLE PARALLEL 2-D IMAGE FILTER CMOS IC BASED ON THE GILBERT
VECTOR MULTIPLIER
51
SYNCHRONIZING AN X-RAY AND
ANESTHESIA MACHINE VENTILATOR
A Medical Device Interoperability Case Study

David Arney

, Julian M. Goldman

, Susan F. Whitehead

and Insup Lee

University of Pennsylvania, Philadelphia, PA, U.S.A.

[email protected], [email protected]

MD PnP Program, CIMIT, Cambridge, MA, U.S.A.

[email protected], [email protected]
Keywords:
MDPnP interoperability, Plug-and-play, Interoperable interconnected medical devices, X-ray ventilator, For-
mal methods, Verication model, Checking apnea, Patient safety.
Abstract:
When a x-ray image is needed during surgery, clinicians may stop the anesthesia machine ventilator while the
exposure is made. If the ventilator is not restarted promptly, the patient may experience severe complications.
This paper explores the interconnection of a ventilator and simulated x-ray into a prototype plug-and-play
medical device system. This work assists ongoing interoperability framework development standards efforts
to develop functional and non-functional requirements and illustrates the potential patient safety benets of
interoperable medical device systems by implementing a solution to a clinical use case requiring interoper-
ability.
1 INTRODUCTION
Medical devices are a key element in the modern
health care environment. They assist medical staff by
automatically measuring physiologic parameters such
as blood pressure, oxygen level, and heart rate, or ac-
tively inuence these parameters by means of infu-
sion pumps for analgesia and insulin or ventilators for
breathing support. Almost all modern medical care
rely on electronic medical devices.
Despite the pervasive use of medical devices
throughout modern health care, most devices work
on their own and in isolation. In contrast, interop-
erable devices would allow connections for sharing
patient data, device status, and enabling external con-
trol, even between devices from different manufactur-
ers. Such interoperability would lead to clear bene-
ts for the care provider and the patient such as more
accurate assessment of the patients health and error-
resilient systems through safety interlocks, closed-
loop control, and automatic hot swappable backups.
To realize these benets, the MD PnP program
at the Center for Integration of Medicine & Innova-

This research was supported in part by NSF CNS-

0509327, NSF-CNS-0610297, NSF CNS-0720703, and
NSF CNS-0834524.
tive Technology at the Massachusetts General Hospi-
tal (CIMIT.org) has been developing techniques and
standards to facilitate medical device interoperability
via MD PnP (Medical Device Plug-and-Play), similar
to the plug-and-play of PC devices.
This paper describes a prototype MD PnP case
study that was conducted for two purposes: (1) for
the MD PnP program to extrapolate functional and
non-functional requirements for the interoperability
standards in progress, and more importantly, (2) to
develop a demo interoperable medical device system
which would illustrate the benets of the work by im-
plementing a solution to a clinical use case requiring
interoperability.
The rest of the paper is organized as follows. Sec-
tion 2 describes the clinical use case which motivated
this case study. Our problemstatement and challenges
are in Section 3. Section 4 describes the details of our
system implementation, and Section 5 tells how we
modeled and veried the system and generated code
from the model. Finally, our conclusions are in Sec-
tion 6.
52
2 CLINICAL USE CASE
This project was driven by a specic clinical use case.
This use case was documented by the Anesthesia Pa-
tient Safety Foundation to illustrate a potential safety
problem with the way x-ray images are usually taken
during surgery.
A 32-year-old woman had a laparoscopic
cholecystectomy [gall bladder removal] per-
formed under general anesthesia. At the sur-
geons request, a plane lm x-ray was shot dur-
ing a cholangiogram [bile duct image]. The
anesthesiologist stopped the ventilator for the
lm. The x-ray technician was unable to re-
move the lm because of its position beneath
the table. The anesthesiologist attempted to
help her, but found it difcult because the
gears on the table had jammed. Finally, the
x-ray was removed, and the surgical proce-
dure recommenced. At some point, the anes-
thesiologist glanced at the EKG and noticed
severe bradycardia. He realized he had never
restarted the ventilator. This patient ultimately
expired. (Lofsky, 2004)
It is common practice to stop the anesthesia ma-
chine ventilator for a short time during surgery when
this type of x-ray is performed. This ensures that the
patients chest and abdomen are not moving when the
exposure is made, thus providing a sharper image.
This does not harm the patient provided that the ven-
tilator is restarted promptly. Difculties arise only if
the ventilator is not restarted for some reason. This
kind of problem can be mitigated by using intercon-
nected devices. If the anesthesia machine ventilator
can synchronize with the x-ray, then it is no longer
necessary to manually stop the ventilator to make the
exposure.
Synchronization between a camera and external
devices like a ash is not new. Typically, the cam-
era sends a trigger signal to the ash at the right time.
Similarly, the ventilator could synchronize with the
x-ray machine. Since ventilators are not built to send
synchronized signals to x-ray machines, we designed
our system to have a third device which sits between
the ventilator and x-ray, reads status messages from
the ventilator, and makes the decision about when to
trigger the x-ray. This third component is called the
supervisor and is described in detail in Section 4. Sys-
tems which synchronize x-rays and ventilators have
been built in the past, see for instance (Langevin et al.,
1999), but these systems must be built one at a time
for specic devices and are limited to experimental
use. Ventilators and x-ray machines are manufac-
tured by many companies. Cross-manufacturer inter-
operability would allow synchronized systems to be
built from any combination of devices that support
the functionality. The aim of the MD PnP program
is to develop techniques and standards that facilitate
medical device interoperability in order to allow such
systems to be easily assembled and used clinically.
3 PROBLEMSTATEMENT AND
CHALLENGES
Our goal was to explore the safety and engineering
issues involved in building a system that would al-
low the x-ray machine to take a clear image of the
patient without the need to turn off the ventilator. Fur-
thermore, we wanted to build a system which would
illustrate the benets of interoperability in the med-
ical domain. Interoperable medical devices are de-
vices which are capable of connecting to each other
to share data or to allow external control. Such de-
vices must have an external interface, and the design
of these interfaces is the subject of several ongoing
standards processes such as ISO/IEC 11073, Health
Level 7 (HL7), and others. The use case we addressed
specically requires interoperability supporting exter-
nal control. The implementation we developed is not
intended to be used clinically. This project is essen-
tially a research platform for understanding the core
issues with interfacing these devices in this particular
use case.
Most medical devices currently manufactured are
not designed to be interoperable. The challenges we
faced in building this system are generally faced by
anyone trying to connect medical devices and are a
major reason such interconnection is not more com-
mon. Medical devices generally have proprietary in-
terfaces which are only documented in technical man-
uals or other material not openly available. We were
fortunate to have the cooperation of Dr ager, the man-
ufacturer of the ventilator we used. The interface of
the ventilator was designed to be used for diagnosis
of machine faults and to send data to the electronic
medical record, not as a source of real-time status in-
formation. Thus, it runs at a relatively slow rate, and
the lowmaximumsample rate (5 - 10 samples per sec-
ond) was the limiting factor in designing our control
algorithm.
A further challenge in interconnecting medical
systems is proving the safety of the resultant system.
Safety is dened as freedom from unnecessary risk,
where risks are unmitigated hazards. FDA provides
guidance on risk minimization for medical devices.
(U.S. Department of Health and Human Services,
Food and Drug Administration, Center for Drug Eval-
SYNCHRONIZING AN X-RAY AND ANESTHESIA MACHINE VENTILATOR - A Medical Device Interoperability
Case Study
53
uation and Research (CDER), Center for Biologics
Evaluation and Research (CBER), 2005) The risk as-
sessment process starts with a hazards analysis or fail-
ure modes and effects analysis (FMEA). These docu-
ments gather potential hazards and their mitigations,
that are used in writing requirements and safety prop-
erties. The risk analysis process and how we used
hazards to derive safety properties with which to ver-
ify the system is described in Section 5.
Our development process started with informal
system requirements which were used to build a state
machine model of the desired system behavior. We
checked this model for safety properties using model
checking software and then generated code from the
model to produce the supervisor.
4 SYSTEMDESCRIPTION
Figure 1 shows the overall architecture of our ap-
proach. This architecture follows closely that of the
draft standard for integrating the clinical environment
(hereafter referred to as ICE) (ASTM F29 WK19878,
2008). The major components are a set of medical
devices, a network controller, a supervisor, the pa-
tient, and a caregiver. Medical devices connect to
each other and the supervisor through the network
controller. The devices connections to the network
controller may go through physical adapters and data
format converters if their connectors and formats are
not directly compatible. The network controller may
also connect to an external network such as a hospital
information system. The supervisor runs the control
software for the system. Supervisor software for our
system is the subject of Section 4.2. The supervisor
hosts the user interface for the caregiver and may also
contain a data logger, which records network activity
and information from the devices.
MD PnP requires three phases of operations each
with its own safety, security, and functional require-
ments. The rst phase is device discovery and con-
nection establishment, when devices are rst con-
nected to a MD PnP network. When a new device
is connected to the network, the devices capabilities
need to be communicated to the rest of the system.
The second phase is normal operation of the plug-
and-play system. During this phase, the devices trans-
mit data they produce and receive commands or data
from other parts of the system. The nal phase is
disconnection. When devices are removed from the
system, the supervisor must decide how to respond.
If the device was necessary for continued operation
of the supervisor program, then the supervisor might
notify the user and shut down. If the device was not
Figure 1: Conceptual Architecture Overview.
necessary, the supervisor might be able to continue
operating in a limited manner.
Our system implementation, which follows the
conceptual architecture, is shown in Figure 2. The
devices we used were a Dr ager anesthesia machine
ventilator and a simulated x-ray machine. The role of
network controller and adapter is lled by the Live-
Data RTI software program. LiveData Inc. is a com-
pany which produces software to integrate medical
devices for common display data. We worked with
LiveData to connect the ventilator and simulated x-
ray. Their software translates the proprietary medical
device formats and makes the data available through
a single interface.
The supervisor program runs on the same com-
puter as the LiveData RTI software. Finally, the pa-
tient was represented with a physical lung simulator
consisting of a bellows and spring. While a simple
lung simulator does not capture all the nuances of a
real patient, it is sufcient for this application. Lung
movement is the factor we can control in taking a
clear x-ray, and a supervisor which can synchronize
with a simulated lung can be expected to do the same
with a real patient.
Our demo is not a full MD PnP system. It is
an interconnected medical device system rather than
an interoperable system. An interconnected system
is one in which devices are functionally connected
through an interface. It differs from an interoperable
system in that the devices are hard-coded. The sys-
tem is built around specic devices and will not oper-
ate with other, similar devices. It also does not fully
implement the three phases described above. The
demo is designed to illustrate the possibilities of in-
terconnected systems and show the kinds of systems
which interoperability would permit. It is not pos-
sible at this time to build a fully interoperable MD
PnP system, since the standards are still under de-
velopment. We believe that limited systems such as
this demo still have value in identifying functional
and non-functional requirements for the standards in
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
54
Supervisor
Server
ModBus /
Ethernet
Logitech IF /
USB
SOAP /
Ethernet
MediBus /
Serial
PLC
Light Button
Webcam
Simulated XRay
Ventilator
LiveData
Figure 2: Overview of the System.
progress and illustrating the benets of the interoper-
ability work.
4.1 Hardware
The system consists of three major hardware compo-
nents. These are an anesthesia machine ventilator, an
x-ray machine, and a supervisor computer. The venti-
lator breathes for the patient by pumping gas into their
lungs and allowing them to exhale on their own. The
x-ray machine takes radiographs, and the supervisor
coordinates the actions of the other components. Each
of these devices has its own physical interface and
communication protocol, all of which are different.
Medical devices are not generally developed with the
intention of interconnection. Any external interfaces
which are present are usually used for logging sta-
tus information or debugging. There is presently lit-
tle incentive for manufacturers to follow standards in
building these interfaces, and few standards for them
to follow. Thus, a wide range of interfaces are found
on various devices.
A fourth component is the LiveData server, which
translates formats between the other devices. The
LiveData server communicates with the anesthesia
machine ventilator using Dr agers MediBus protocol
over a 9600 baud serial line, with the x-ray machine
using the Modbus protocol over ethernet, and with the
supervisor using SOAP on HTTP on TCP/IP over eth-
ernet.
The x-ray machine is simulated using a PLC con-
troller, a webcam, a small red light, and a pushbut-
ton. The PLC allows the light to be turned on and
off and the pushbuttons status to be read over ether-
net. The webcam is a standard USB webcam which is
controlled using proprietary software.
The supervisor computer, LiveData server, and
PLC are connected with a standard ethernet switch.
In our demo, the supervisor software and LiveData
server were usually run on the same computer.
4.2 Software
The systems software is divided between the supervi-
sor and the LiveData server. The supervisor controls
the other devices in the system - it correlates infor-
mation from the other devices and makes the deci-
sion when to trigger the x-ray. Supervisors in gen-
eral are responsible for implementing the parts of the
system which are specic to a particular clinical sce-
nario. The supervisor checks to see whether the re-
quired devices are present, collects data from the var-
ious connected devices, and sends commands to the
devices according to the particular scenario. For this
demo, the supervisor gathers data from the ventilator
and sends the signal to trigger an x-ray exposure.
The LiveData server receives SOAP requests from
the supervisor and translates them into requests to
individual devices in their proprietary formats, then
takes the replies from the devices and formats them
as SOAP responses.
The development of the supervisor software is de-
scribed in more detail in Section 5.
4.3 SOAP
The SOAP interface is used for communication be-
tween the supervisor and the LiveData server. A typi-
cal SOAP transaction goes as follows:
1. the user requests the list of variables from the
LiveData server
2. they use the generic get or set methods to do op-
erations on those variables
3. the server receives the command, translates it to
the Dr ager MediBus protocol used for the ventila-
tor, and sends the command along
4. the ventilator sends its response to the server,
which translates it and passes the response to the
user
5. the response is returned in an XML wrapper
which contains typing information for the re-
turned values
SOAP is a standard protocol for web services. We
used it here primarily because it was supported by the
LiveData program. It worked well enough for this
application, which had relatively slow data rates and
small amounts of data being passed, but it does have
appreciable overhead. All queries and responses are
SYNCHRONIZING AN X-RAY AND ANESTHESIA MACHINE VENTILATOR - A Medical Device Interoperability
Case Study
55
passed as XML messages which need to be generated
on the sending side and parsed when received.
Another issue was the latency of the SOAP server.
Some of the latency is inherent in encoding and de-
coding XML messages and in passing the messages
through a translator instead of directly sending them
from one device to another. Additional latency in our
demo system resulted from a commonly used conges-
tion control method. When many small packets are
sent over a TCP/IP network, the data being sent is a
small portion of the transmitted packet - most of the
packet is taken up with headers. This can lead to the
network becoming congested when small packets are
sent quickly.
Nagles algorithm (Nagle, 1984) is used to con-
catenate these small packets together to reduce over-
all network overhead at the expense of delayed mes-
sage delivery. The data in many small packets can
be bundled into one large packet, reducing the over-
head but increasing the transmission time of the early
packets. In this context, we had plenty of bandwidth
and were much more concerned about the timely de-
livery of messages, so turning off this feature greatly
reduced the latency of the SOAP server.
4.4 Synchronization Algorithms
The supervisor uses information fromthe ventilator to
decide when to trigger the x-ray. The synchronization
algorithm denes exactly how this decision is made.
Figure 3 shows the respiratory cycle graphed as pres-
sure over time. The pressure increases until the end
of inspiration (at time Tinsp after start of breath), at
which point it drops off quickly through expiration.
There is usually a pause between the end of exhalation
and the start of the next breath. For this case study,
we want to support taking an x-ray when the lung
was not moving signicantly. This occurs when the
patient is relatively still at the peak of inspiration or
between the end of expiration and the start of the next
breath. An exposure is possible if the time the patient
is still exceeds the time needed for the exposure plus
the latency between triggering the x-ray and the actual
exposure.
Pressure
Start of Breath Start of Breath
Tinsp
1 / Frequency
Time
Figure 3: Respiratory Cycle.
4.4.1 Synchronization Method 1: Dead
Reckoning
The rst method used to determine when to trigger
the x-ray is simple dead reckoning using the time of
last breath, time of inspiration, and frequency. The
variables used for this method are shown in Figure 4.
All times are in seconds.
name description
T
now
current time
T
lb
time of last breath
T
nb
time of next breath
T

a small offset time to accommodate jitter

T
trigger
time to send trigger signal to the X-ray
T
exp
time of X-ray exposure
f req frequency, breaths / minute
f low instantaneous ow rate
Figure 4: Variables for dead reckoning.
If we know the time of the start of the last breath
and the frequency of breathing, then it is trivial to cal-
culate the time of the start of the next breath.
T
nb
=T
lb
+60/ f req (1)
There is probably time to trigger the x-ray just be-
fore start of the next breath, as long as the patient has
nished exhaling before the start of the next inhala-
tion.
T
trig
=T
nb
T
exp
T
delta
(2)
We can check whether the patient has actually n-
ished exhaling by sampling the instantaneous ow
rate just before the start of the next breath. If it is
close to zero, then the patient is not inhaling or exhal-
ing and is still enough to allow taking the x-ray.
1. Get values of the variables T
now
, T
lb
, f req
2. Calculate T
trig
3. Sleep for T
trig
T
now
seconds
4. Wake up and sample f low
5. If f low =0, trigger X-ray
else, start over
This method of synchronization makes many as-
sumptions. The most critical assumption is that the
respiratory frequency is not going to change between
the last breath and the next one. If it does, or if the
system setup changes in other ways, this method of
synchronization will not work. The check of instanta-
neous ow rate should prevent the system from trig-
gering the x-ray when the patient is moving, but the
system may not be able to take an image in situations
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
56
where a different synchronization method would al-
low an exposure.
4.4.2 Synchronization Method 2: Dynamic
Another way to calculate the trigger time is to sam-
ple the real-time ow rate rapidly enough to build a
picture of the ow graph. We experimented with two
techniques for doing this. The variables used in the
following descriptions are listed in Table 5.
name description
f low instantaneous ow rate
T
f low
time of last f low sample
S
current
value of current ow sample
T
current
time of current ow sample
S
last
value of last ow sample
T
last
time of last ow sample
slope calculated slope value
Threshold slope threshold
Figure 5: Variables for dynamic synchronization.
We originally envisioned sampling at a high
enough rate to be able to integrate the total ow vol-
ume by multiplying the sampled ow rate by the time
interval of the samples. This would allow the supervi-
sor to trigger the x-ray at the right time no matter what
changes were made to the ventilators programming
or how the patient reacted. However, the ventilator
was not able to provide samples at a high enough rate
to enable this method to be used. The SOAP server
and interface introduced additional latency and jitter
into the samples, which further reduced their useful-
ness for this purpose.
Our second idea was to use the slope of the ow
signal to nd when inspiration is about to end. This
meant taking two or more samples, calculating the
rate of change of the ow rate between them, and
triggering when this rate of change was low enough.
The problem we ran into here is that the ow graph
tails off very rapidly, making it unlikely that we would
get even a pair of samples in the short time when the
breath is about to end. The low sample rate made this
problem worse.
1. prime S
last
, T
last
, S
current
, T
current
with two consec-
utive samples
2. S
last
= S
current
3. T
last
= T
current
4. S
current
= f low
5. T
current
=T
f low
6. slope =S
current
S
last
/T
current
T
last
7. if slope < Threshold and f low is near 0, trigger
x-ray
else loop back to 2.
In the end, we found that dynamic synchroniza-
tion is possible only at relatively low respiratory rates
- under 8 to 10 breaths per minute. The dead reck-
oning method functions at much higher rates, up to
approximately 25 to 30 bpm depending on the other
ventilator settings. The supervisor program for our
demo checks the respiratory rate and chooses whether
to use the dynamic or dead reckoning method accord-
ingly.
4.5 Alarms
The system should not trigger the x-ray if the venti-
lator has active alarms. The ventilator will take care
of displaying the alarm condition to the caregiver and
sounding alarms, so the supervisor just has to detect
that the ventilator has active alarms and not trigger the
x-ray on that respiratory cycle. It does this by getting
a summary of all active alarms and warnings from the
ventilator. If the list of active alarms is not empty,
then the supervisor will not trigger the x-ray. This
technique is easy to implement and covers the most
common situation where the alarm sounds sometime
before the supervisor decides to trigger the x-ray. This
is sufcient for the demo, but an implementation with
a real x-ray machine and a real patient would have
to take into account factors such as the alarm being
raised after the supervisor checks the alarm status but
before the exposure is made.
In the case where this happens, many conditions
which would cause a ventilator alarm will not affect
the synchronization algorithm. These include alarms
like low gas levels, overpressure, some sensor fail-
ures, etc. Any alarm that does not indicate an unex-
pected change in ventilator settings will not stop the
supervisor from being able to synchronize. Alarms
for major mechanical malfunctions are very rare, but
would indicate conditions where we would not want
an exposure to be made - though any failure which
stopped the ventilator from operating would mean
that the patients chest was not moving. The prob-
lem with taking an exposure during an alarm is not
that the image would be blurred, but that the safety of
any caregivers responding to the alarm could be com-
promised. Caregivers are also protected by the use of
a dead man switch that the x-ray technician holds
during the exposure. If the switch is released, the x-
ray will not be taken. The time interval where there
was an active alarm and the exposure was being made
would be a fraction of a second, but this should be
taken into account in the risk management process.
SYNCHRONIZING AN X-RAY AND ANESTHESIA MACHINE VENTILATOR - A Medical Device Interoperability
Case Study
57
Any system using a real x-ray machine would also
need to take into account alarms from the x-ray, and
any system using medical devices which are capable
of pushing alarms rather than having them polled (as
we did with this ventilator) would also need to con-
sider possible race conditions between the alarm han-
dling and synchronization parts of the supervisor.
5 MODELING, VERIFICATION
AND CODE GENERATION
The software for the supervisor is the key element of
the system. The supervisor is the new piece which
facilitates communication between the other devices.
As was described in Section 4.2, the supervisors role
in this demo is to gather data from the ventilator, de-
cide when to trigger the x-ray, and send the signal to
the x-ray machine at the correct time. The supervi-
sor interacts with the caregiver to get input such as
whether to make the exposure during inspiration or
expiration and to provide the caregiver with status in-
formation and, ultimately, with the x-ray image.
The functioning of the supervisor program is crit-
ical to the safety of the system, so we devoted a sig-
nicant amount of time and effort to ensuring its cor-
rectness.
The supervisor software development process
started with gathering informal requirements. These
requirements were collected during discussions with
caregivers and biomedical engineers and included
functional requirements such as when the exposure
is made, the red light on the x-ray box should light
up and safety requirements like the caregivers x-
ray trigger button must be held down for the x-ray ex-
posure to be made. These requirements were rened
and expanded upon throughout the development pro-
cess. For instance, when we started development we
did not know that we would need a dead-reckoning
synchronization algorithm in addition to the dynamic
method and thus did not include any requirements
about when the supervisor should use one or the other
of these techniques.
A state machine model of the supervisor was built
and then veried to meet essential safety properties.
We used the model to generate Java code which then
ran the demo. This development process is described
in more detail in the following sections.
We began by modeling the supervisor program as
an extended nite state machine (EFSM). This format
was chosen because it is expressive enough to capture
the behavior of the program and tools are available
to automatically translate the EFSM specication into
the input languages of a number of tools.
Verication. Once the system was modeled as a
state machine, we used a tool to translate it into the
input format for the model checker UPPAAL. The
model checker was used to simulate the system, to
test the system for general properties like deadlock,
and to test more specic properties. These activities
suggested changes to the EFSM specication, and the
process went though several iterations. Eventually,
we produced an EFSM specication which satised
all the safety requirements.
The safety requirements for the system were gath-
ered by talking with clinicians and working though
an informal hazard analysis process. For a device in-
tended for use with patients, this process would be
much more thorough.
The primary hazard introduced by this system is
triggering the x-ray at the wrong time. This could po-
tentially endanger the x-ray technician or other clini-
cians. Triggering the x-ray when the patient is mov-
ing will result in a blurred x-ray and the need to take
another exposure, meaning additional radiation expo-
sure for the patient. Another hazard is that an image
might not be taken even though it is possible. This is
less signicant, since the system will inform the clin-
ician that the exposure was not possible and try again
on the next breath. The exposure is delayed slightly,
but this is a small cost compared to that of a failed ex-
posure. The EFSM model of the system was checked
for structural properties like deadlock (that the sys-
tem cant get stuck) and for specic safety proper-
ties. These focused on when the x-ray is triggered,
since this is the single safety-critical action the sys-
tem takes. We checked that the trigger signal was sent
only at the correct time (as described in the algorithms
in 4.4.1 and 4.4.2) and that the system would not trig-
ger unless the ow rate reported by the ventilator was
near zero.
AG xray = exposing implies T
now
=T
nb
TexpT

(3)
Formula 3 is used for checking the system when
it is being used to make an exposure at the peak
of expiration (the lung is empty) in dead reckoning
mode. This specication is in linear temporal logic
(LTL) and it says that whenever the x-ray machine
is in a state where it is exposing (AG xray = expos-
ing) the current time must be the time of the next
breath minus the exposure time minus a small offset
(T
now
= T
nb
Texp T

). This means that if there

is any possible way that the EFSM could have the x-
ray in the state exposing when it is not that time, the
model checker will showit as a counterexample. Sim-
ilar formulas are used for checking exposure times for
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
58
inspiration.
AG xray=exposing implies f low<= f low threshold
(4)
Formula 4 states that when the x-ray is exposing,
the instantaneous ow rate must be less than the ow
threshold. This threshold is dened to be low enough
that the lung will not be moving enough to blur the
image, but also high enough to allow an exposure
when there are very small movements.
Implementation Generation. The nal EFSM
specication was used to automatically generate Java
code which was used in the demo implementation.
The demo includes a handwritten GUI frontend which
is the user interface and the supervisor application,
which is largely generated code. The generated code
interacts with some handwritten functions which per-
form low-level actions. For instance, the model sim-
ply uses values like f low, while the generated code
replaces references to such variables with calls to
handwritten library functions which actually provide
the values.
Demo. The demo starts with a screen describing the
clinical use case. This is followed by giving the user
a choice of taking an image at the peak of inspira-
tion (when the lungs are full) or the peak of expira-
tion (when the lungs are empty). The user is asked to
conrm their choice and taken to a screen describing
the image-taking process. The user is asked to play
the role of an x-ray technician and to pick up a physi-
cal button which they will hold while the exposure is
made. In a non-synchronized x-ray, this button would
trigger the x-ray directly. In our system, the button is
held down to give the system permission to make the
exposure. The clinician holds the button for several
seconds while the system waits for the lung to reach
the proper phase of respiration and the system checks
to make sure the button is held before taking an im-
age. If the clinician decides that it is not safe to make
an exposure (e.g., if someone walks into the room),
they can simply release the button and no exposure
will occur. This allows us to keep a human in the loop
as an additional safety precaution. Assuming the but-
ton is held down, when the lung reaches the proper
phase the exposure is made and the webcam image is
displayed on the screen.
The system consisted of many components from a
variety of sources, written in several languages. The
main difculty in implementing the demo was inte-
grating these diverse components into a single, func-
tional system. As was described in Section 4, the
system was tied together using LiveData and SOAP.
While there were signicant disadvantages to this
approach (especially in terms of latency), we were
successful in making a working demo. This demo
was shown at the CIMIT Innovation Congress and
as a Scientic Exhibit at the American Society of
Anesthesiologists annual meeting and presented at the
High Condence Medical Devices, Software and Sys-
tems and Medical Device Plug-and-Play Interoper-
ability workshop (Arney et al., 2007).
6 CONCLUSIONS
We successfully built a system which was able to syn-
chronize the ventilator with a simulated x-ray ma-
chine, demonstrating that the approach is feasible. In
the process, we learned lessons for building more gen-
eral systems. These include the importance of recog-
nizing the limitations of device interfaces in the su-
pervisor algorithm design and the need to have super-
visors which can respond to the changing settings of
the devices. We had two synchronization algorithms,
one which was more accurate but only usable at low
breath rates and a less accurate but faster algorithm
for high breath rates. We used formal methods in the
development of the supervisor and have presented a
methodology for ensuring that the integrated device
systems meet their specied safety properties.
This work started with an unfortunate use case,
resulting from the lack of a respiratory pause feature
on the ventilator and the ventilators inability to syn-
chronize with the x-ray machine. The exposure that
our demos brought to this problem has led to a pro-
posed change to the international anesthesia worksta-
tion standard. Hopefully in the future such changes
and the introduction of safe, inter-connected systems
will help to improve patient safety.
ACKNOWLEDGEMENTS
We would like to thank the following people who
were involved in creating the x-ray ventilator syn-
chronization demo. Without their contributions, this
work would not have been possible.
Steve Boutrus, Tufts Medical School
Philippe Cortes, Compiegne Univ. of Technology,
France
Jennifer Jackson, BWH Biomedical Engineering
Shankar Krishnan, MGH Biomedical Engineering
Ersel Llukacej, LiveData, Inc.
Heidi Perry, Draper Laboratory
Tracy Rausch, DocBox, Inc.
SYNCHRONIZING AN X-RAY AND ANESTHESIA MACHINE VENTILATOR - A Medical Device Interoperability
Case Study
59
Jeff Robbins, LiveData Inc.
Rick Schrenker, Biomedical Engineering
Dan Traviglia, Draper Laboratory
Sandy Weininger, U.S. Food and Drug Administra-
tion
REFERENCES
Arney, D., Goldman, J., Lee, I., Llukacej, E., and White-
head, S. (2007). Use Case Demonstration: X-Ray /
Ventilator. In High Condence Medical Devices, Soft-
ware, and Systems and Medical Device Plug-and-Play
Interoperability, 2007, page 160.
ASTM F29 WK19878 (2008). New Specication for
Equipment in the Integrated Clinical Environment -
Part I: General Requirements for Integration.
Langevin, P. B., Hellein, V., Harms, S. M., Tharp, W. K.,
Cheung-Seekit, C., and Lampotang, S. (1999). Syn-
chronization of Radiograph Film Exposure with the
Inspiratory Pause. Am. J. Respir. Crit. Care Med.,
160(6):20672071.
Lofsky, A. S. (2004). Turn Your Alarms On! APSF
Newsletter, 19(4):4160.
Nagle, J. (1984). Request for Comments: 896, Congestion
Control in IP/TCP Internetworks. Technical report,
Ford Aerospace and Communications Corporation.
U.S. Department of Health and Human Services, Food and
Drug Administration, Center for Drug Evaluation and
Research (CDER), Center for Biologics Evaluation
and Research (CBER) (2005). Guidance for Indus-
try Development and Use of Risk Minimization Ac-
tion Plans. Technical report, Ofce of Training and
Communication, Division of Drug Information.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
60
AUTOFLUORESCENCE SPECTROSCOPY OF A HUMAN
GASTROINTESTINAL CARCINOMA CELL LINE
Design of Optical Sensors for the Detection of Early Stage Cancer
D. S. Ferreira
1
, M. Henriques
2
, R. Oliveira
2
, J . H. Correia
1
and G. Minas
1
1
Algoritmi Centre, Universidade do Minho, Campus Azurm, 4800-058 Guimares, Portugal
2
Department of Biological Engineering, Universidade do Minho, Campus Gualtar, 4710-057 Braga, Portugal
[email protected], [email protected], [email protected], [email protected]
[email protected]
Keywords: Autofluorescence, Fluorophores, Cancer, Optical sensors.
Abstract: Human tissues show autofluorescence (AF) emission spectra when excited by ultraviolet or short-
wavelength visible light. The intensity and shape of these spectra are dependent on the tissues pathological
state and, therefore, its measurement gives information about the degree of malignant transformations that
could lead to cancer. In this article, it is characterized the AF spectra of one human gastrointestinal
carcinoma cell line (CACO-2). The obtained results showed significant AF signal for the presence of amino
acids. The spectral information obtained can be used for the design of fluorescence optical sensors that will
be incorporated on an endoscopic capsule, for measuring the AF emission spectra of normal and cancer
cells. This integrated optical system will innovate on the diagnosis of early stage cancer.
1 INTRODUCTION
Cancer of the gastrointestinal (GI) tract is the second
most common cause of cancer death in the United
States and industrialized countries. Patients with a
family history of colon cancer, familial polyposis,
long-standing inflammatory bowel disease, and
Barretts esophagus are at high risk for developing
this kind of cancer. The new innovations in
endoscopy, including capsule endoscopy, made the
GI tract one of the most frequently and completely
examined system. Despite the greater access, early
detection of neoplasia is difficult during routine
endoscopy due to the absence of typical
morphological structures and, as a result, is still
limited to blind random biopsies in patients at high
risk of developing cancer, in a both inefficient and
costly process. Therefore, more sensitive endoscopic
screening tools, which enable differentiation
between premalignant and benign lesions during
endoscopy, are of scientific and clinical interest
(Prosst, 2002; Banerjee, 2004; Georgakoudi, 2006).
When human tissues are illuminated with
ultraviolet or short wavelength visible light, they
emit fluorescence light of a longer wavelength
(Figure 1). This tissue autofluorescence (AF) arises
from endogenous molecules within the tissue, called
fluorophores (Haringsma, 1999; DaCosta, 2002).

The GI tract tissues are composed by a complex
combination of several fluorophores that occur in
different concentrations and at different depths. The
mucosa, submucosa and muscularis propria have
distinct fluorophore compositions, so that even
though each fluorophore has a distinct fluorescence
spectrum, the total fluorescence measured comprises
contributions from the fluorophores in the various
layers. Hence, excitation and fluorescence emission
wavelength bands are often broad, relatively
featureless and overlap with one another, so that
identifying individual fluorophores in a given tissue
spectrum is difficult (Haringsma, 1999;
DaCosta, 2002).

Some research groups have explicitly studied the
AF emission of established cell lines (DaCosta,
2005). These studies, by the use of single living
cells, allow the isolation of the contribution of
intracellular fluorescence in the tissular emission,
through the elimination of the extracellular matrix
emission and the influence of light absorption and
scattering. Therefore, such studies provide a better
understanding of the biochemical changes associated
with malignant transformation. Also, cellular
61
reference spectra are needed for the modeling of the
tissue fluorescence, in view of spectral
quantification which should be very useful for
diagnostic purposes. To quantify the relatively weak
AF created by one single fluorophore, sophisticated
algorithms have been developed (Stepp, 1998;
Prosst, 2002; Villette, 2006).
Mucosa
Submucosa
Muscularis propria
Fluorophore
Chromophore
Specular
Reflectance
Incident
Light
Diffuse
Reflectance
Scattering
Absorption
Fluorescence
Emission

Figure 1: Lighttissue interactions include reflectance,
scattering, absorption and fluorescence. Tissue
fluorescence is originated by the absorption of light by
fluorophores.
Among the significant fluorophores in the GI
tract are: tryptophan and tyrosine (aromatic amino
acids present in cells); NADH and NADPH (cellular
metabolism related coenzymes), which assist
oxidation and reduction processes and are found
mainly in mitochondria; and flavins and flavin
nucleotides, which are mostly bound to enzymes and
concentrated in the mitochondria (DaCosta, 2005).

Each of these fluorophores has its characteristic
excitation and emission spectrum (Table 1). Thus,
different excitation wavelengths result in the
activation of different fluorophores.
Table 1: Excitation and emission maxima of some
endogenous fluorophores (DaCosta, 2002).

Endogenous
fluorophores
Excitation
maxima (nm)
Emission
maxima (nm)
Amino acids
Tryptophan 280 350
Tyrosine 275 300
Phenylalanine 260 280
Metabolic cofactors
FAD, Flavins 450 515
NADH 350 450
NADPH 336 464
Over the last decade a number of optical
techniques have been developed to try to detect early
GI neoplastic lesions. These optically-based
methods have the potential to detect the very earliest
mucosal changes at the micro structural and
molecular levels (Haringsma, 1999). Light induced
autofluorescence spectroscopy, for example, is based
on the analysis of the fluorescence emission of
endogenous fluorophores, providing information that
can be used to characterize changes that take place
as tissues become diseased. It involves delivery of
the excitation light via placement of a fiber-optic
probe in direct contact with tissue. The fluorescence
is collected by the same probe and delivered to the
input of a spectrometer. Achieving discrimination
between diseased and healthy tissue relies upon the
identification of differences either in emission
intensity, spectral distribution, fluorescence lifetime,
or crucially, a combination of these (Stringer, 2004;
Georgakoudi, 2006).
The capsule endoscopy is a relatively recent
method for GI tract evaluation that uses an
endoscopic capsule (EC). Since, presently, some of
the most promising methods for detecting cancer at
an early stage are based on tissues fluorescence, the
EC can play an important role in this kind of
diagnosis, once it can be equipped with miniaturized
fluorescence CMOS optical sensors. These sensors
should be incorporated in the EC for measuring the
AF emission spectra of normal and cancer cells, and
should have the higher quantum efficiency in a
defined spectral range (Delvaux, 2006; Dias, 2007).
This integrated system will certainly innovate in the
diagnosis of GI cancer.

It is important to notice that
current endoscopic capsules do not have added
diagnostic functions, such as spectroscopic analysis,
which are commonly available in conventional
endoscopes. Therefore these developments will be a
major step towards creating a new platform for
diagnosis.
In this article, it is characterized the AF spectra
of one human GI carcinoma cell line (CACO-2). The
results obtained showed significant AF signal for the
presence of amino acids, namely tryptophan. The
resulting spectral information of the CACO-2 cell
line is a good initial approach for the design of
fluorescence optical sensors, to be incorporated on
an EC. This study was performed using
spectrofluorimetry on cell monolayers, with several
excitation wavelengths, resulting in several
fluorophores emission spectra. Some of the reported
results appeared promising, being in accordance
with previously published results.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
62
2 EXPERIMENTAL
2.1 Human Carcinoma Cell Line
Culture
The experimental studies were performed using one
human GI carcinoma cell line - CACO-2 - that was
purchased from the European Collection of Cell
Cultures (ECACC). These cells were grown in 88%
Minimum Essential Medium Eagles (EMEM),
supplemented with 10% Fetal Bovine Serum (FBS),
1% Non Essential Amino Acids (NEAA), and 1%
Glutamine (2mM) in a humidified atmosphere, at
37C and 5% CO
2
. Cells were grown in 25 cm
2

flasks to 70 80% confluence and their morphology
was routinely inspected.
2.2 Preparation of CACO-2 Cell Line
for Spectrofluorimetry
The flasks of cultured cells were examined under the
inverted microscope to observe cell confluency.
After that, the old medium was removed and
discarded and cells were washed with warm (37C)
phosphate-buffered solution (PBS). A small volume
of trypsin (~1 mL) solution was added to the flasks
and the flasks were placed back in an incubator at
37C. Five to ten minutes after, cells were examined
under the inverted microscope to confirm that most
of the cells had rounded up and were detaching from
the flask surface. Finally, it was added 4 mL of fresh
growth medium to the cell suspensions. The number
of viable and dead cells/mL was determined using a
Neubauer haemocytometer and the trypan blue dye
exclusion method.
A total of eight experiments were performed for
the CACO-2 cell line. To support the cells for these
experiments it was used glass cover slips placed
inside 6-well plates; after that the final cell
suspensions were divided and placed within these 6-
well plates. The cells used were grown as
monolayers over the cover slips. After 24, 48 or 72
hours (depending on the experiment), cover slips
were collected from the 6-well plates and packed in
Petri dishes, with 3 mL of PBS solution, for the
experiments. Spectral analysis was then performed
within less than half an hour after cells were placed
in Petri dishes.



2.3 Spectrofluorimetry of CACO-2 Cell
Line
Fluorescence spectra were performed on human
carcinoma cell monolayers. The samples were
measured using a SPEX

FluoroLog

- a high
sensitivity spectrofluorometer with a SPEX1680
(0,22m) Double Spectrometer. The excitation light
was provided by a Xenon lamp of 450W with a DC
450RAM power supply, from EUROSEP
Instruments. The equipment was connected to a
computer to control and collect the data. The slit
widths of both spectrometers were adjusted in order
to provide a good spectral resolution in the
excitation and emission path.
2.4 Fluorescence Spectra of CACO-2
Cell Line
Emission spectra were recorded from 270690 nm,
for an excitation wavelength ranging from 260 to
350 nm. These excitation wavelengths were chosen
in order to allow a comparison with data obtained
from literature. Reference spectra were obtained
with pure PBS solutions. All the spectra were
normalized subtracting the reference spectra from
the fluorescence spectra. The fluorescence emission
spectra were also corrected for the spectral response
of the spectrofluorometer. The recordings were
performed for several cell concentrations, depending
on the experiment (24, 48 or 72 hours). These cells
were maintained in PBS during the measurement,
which could in some way put in risk cell viability.
All measurements were performed under the same
experimental conditions.
3 RESULTS AND DISCUSSION
The analysis of the two epithelial cell lines AF was
performed using spectrofluorimetry on cell
monolayers. The measurements were performed on
cell monolayers to use epithelial cells as close as
possible to their in vivo tissular physiological
conditions, as cells grown as monolayers undergo
less stress than cells in suspension. This was
confirmed on previous studies on normal cells from
primary cultures, which easily grow as adherent
monolayers looking like an epithelium structure
while showing a poor viability in suspension.
Furthermore, cell monolayers can be considered as
an optically thin medium, while the acquired spectra
from cell suspensions are strongly affected by
AUTOFLUORESCENCE SPECTROSCOPY OF A HUMAN GASTROINTESTINAL CARCINOMA CELL LINE -
Design of Optical Sensors for the Detection of Early Stage Cancer
63
scattering, inducing spectral distortion (Villette,
2006).
The experiments started measuring the AF signal
of the CACO-2 cell line, with 24, 48 and 72 hours of
growth, placed inside a 6-well plate and slightly
immersed in PBS. The fluorescence measurements
were made for different excitation wavelengths (
ex
)
- 260, 270, 280, 335, and 350 nm - and the emission
was collected in the range 270-690 nm. The average
number of viable and dead cells was determined in
each experiment (Table 2).
Table 2: Number of viable and dead CACO-2 cells, after
24, 48 and 72 hours in culture.
Time in culture
(hours)
Live Dead
Number of
cells/mL
24 16,5x10
4
1,25x10
4

48 22,0x10
4
1,40x10
4

72 63,5x10
4
2,50x10
4


In all experiments, for
ex
of 260, 275 and
280 nm the emission spectra are very similar in
shape (Figures 2, 3 and 4) - all of them exhibit a
broad emission band from 310 to 370 nm, with a
peak around 340 nm, which is primarily caused by
amino acids, namely tryptophan.
However, the AF intensity diverges among all
the acquired spectra, being related with culture time.
The maximum AF intensity is achieved for a 48
hours cell culture, decreasing then as time elapses.
This can be explained by the number of viable and
dead cells. From 24 to 48 hours there is a
considerable increase in the number of viable cells,
which is translated by AF signal intensification.
From 48 to 72 hours, despite the increase in the
number of viable cells, there is a reduction in AF
signal intensity. This intensity drop (for the
minimum value) can be justified not only by the
raise in the number of dead cells, but also by an
increase in cell fragility, as cells are in culture for a
longer time, supplied by the same growth medium
(its important to notice that the culture medium was
not discarded and replaced by fresh growth medium
in the period of 72 hours).
For other spectra acquired using
ex
of 350 and
335 nm, for the detection of NADH and NADPH,
respectively, the resulting fluorescence intensity was
approximately zero, when compared to the amino
acids AF intensity (Figures 5 and 6). This may be
justified by the fact that the AF emission from these
fluorophores, although existing, is very weak and
doesnt superimpose over the PBS signal (reference
signal). Probably these molecules are not present in
significant amounts to contribute to a high quality
and strong signal. It may also be possible that some
other molecules partly affect the measurements by
absorbing the fluorescence or by emitting
fluorescence in the same bandwidth. These results
were very similar for all the experiments performed
(24, 48 and 72 hours growth).
0.00E+00
2.00E+10
4.00E+10
6.00E+10
8.00E+10
1.00E+11
270 310 350 390 430 470 510 550
Emissionwavelength(nm)
A
u
t
o
f
l
u
o
r
e
s
c
e
n
c
e

i
n
t
e
n
s
i
t
y

(
a
.

u
.
)
260nm
275nm
280nm

Figure 2: Autofluorescence emission spectra of a CACO-2
cell line, with 24 hours growth, obtained for different
excitation wavelengths: 260, 275 and 280 nm.
0.00E+00
5.00E+10
1.00E+11
1.50E+11
2.00E+11
2.50E+11
270 310 350 390 430 470 510 550
Emissionwavelength(nm)
A
u
t
o
f
l
u
o
r
e
s
c
e
n
c
e

i
n
t
e
n
s
i
t
y

(
a
.

u
.
)
260nm
275nm
280nm

Figure 3: Autofluorescence emission spectra of a CACO-2
cell line, with 48 hours growth, obtained for different
excitation wavelengths: 260, 275 and 280 nm.
0.00E+00
1.00E+10
2.00E+10
3.00E+10
4.00E+10
5.00E+10
270 310 350 390 430 470 510 550
Emissionwavelength(nm)
A
u
t
o
f
l
u
o
r
e
s
c
e
n
c
e

i
n
t
e
n
s
i
t
y

(
a
.

u
.
)
260nm
275nm
280nm

Figure 4: Autofluorescence emission spectra of a CACO-2
cell line, with 72 hours growth, obtained for different
excitation wavelengths: 260, 275 and 280 nm.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
64
Recent studies have investigated the spectra of
human intestinal mucosa using a spectrofluorometer.
The measurements were performed using short
wavelengths for excitation, and revealed utility in
detecting neoplasia. The AF intensity increased with
neoplasia and had the spectral line shape of
tryptophan, indicating that AF emission from
tryptophan might represent a viable approach to the
detection of malignancy within colonic tissue
(Banerjee, 2000, 2002). A different study had
confirmed that the amount of amino acid related
fluorescence (emission between 300 and 380 nm) is
greater in adenomatous polyps than in the normal
tissues (DaCosta, 2003). Therefore, it can be said
that the results obtained, namely for amino acids AF
emission, are a good initial approach for the design
of optical sensors for the detection of GI early stage
cancer.
2.50E+09
1.50E+09
5.00E+08
5.00E+08
1.50E+09
2.50E+09
360 400 440 480 520 560 600 640
Emissionwavelength(nm)
A
u
t
o
f
l
u
o
r
e
s
c
e
n
c
e

i
n
t
e
n
s
i
t
y

(
a
.

u
.
)
350nm

Figure 5: Autofluorescence emission spectra of a CACO-2
cell line, with 48 hours growth, obtained for an excitation
wavelength of 350 nm.
2.50E+09
1.50E+09
5.00E+08
5.00E+08
1.50E+09
2.50E+09
345 385 425 465 505 545 585 625
Emissionwavelength(nm)
A
u
t
o
f
l
u
o
r
e
s
c
e
n
c
e

i
n
t
e
n
s
i
t
y

(
a
.

u
.
)
335nm

Figure 6: Autofluorescence emission spectra of a CACO-2
cell line, with 48 hours growth, obtained for an excitation
wavelength of 335 nm.

4 CONCLUSIONS
There is a great interest in developing
autofluorescence-based spectroscopic systems for
the detection of early stage cancers. Understanding
the autofluorescent tissue components and how these
components change in concentration and distribution
with disease is essential in terms of optimizing
diagnostic techniques. Several endogenous
fluorophores have been identified and alterations in
concentrations of these are suggested for
discrimination of normal and malignant tissues.
In this article, it is characterized the AF spectra
of one human GI carcinoma cell line (CACO-2),
using several different excitation wavelengths,
which targeted different molecular species. The
obtained results showed significant AF signal for the
presence of amino acids, namely tryptophan. The
resulting spectral information of the CACO-2 cell
line is consistent with previously published results of
malignant colonic tissues, and is a good initial
approach for the design of fluorescence CMOS
optical sensors, to be incorporated on an EC, for the
differentiation of normal and malignant tissues. It is
important to notice, however, that when comparing
fluorescence spectra of cell monolayers with those
from in vivo or ex vivo tissues is has to be considered
the possible changes due to the physicochemical
microenvironment.
REFERENCES
Banerjee, B., Agarwal, S., Miedema, B., Perez R.,
Chandrasekhar, H. (2000). Use of a shorter
wavelength autofluorescent band to separate
adenomatous from hyperplastic polyps of the colon.
Gastrointestinal Endoscopy, 51, AB149.
Banerjee, B., Chandrasekhar, H. (2002). Use of a single
autofluorescence emission ratio for the detection of
colonic neoplasia. Gastrointestinal Endoscopy, 55,
AB129.
Banerjee, B., Henderson, J ., Chaney, T., Davidson, N.
(2004). Detection of Murine Intestinal Adenomas
Using Targeted Molecular Autofluorescence.
Digestive Diseases and Sciences, 49, 54-59.
DaCosta, R. S., Wilson, B. C., Marcon, N. E. (2002). New
optical technologies for earlier endoscopic diagnosis
of premalignant gastrointestinal lesions. Journal of
Gastroenterology and Hepatology, 17, S85-S104.
DaCosta, R. S., Andersson, H., Wilson, B. C. (2003).
Molecular Fluorescence ExcitationEmission Matrices
Relevant to Tissue Spectroscopy. Photochemistry and
Photobiology, 78, 384-392.
DaCosta, R. S., Andersson, H., Cirocco, M., Marcon, N.
E., Wilson, B. C. (2005). Autofluorescence
AUTOFLUORESCENCE SPECTROSCOPY OF A HUMAN GASTROINTESTINAL CARCINOMA CELL LINE -
Design of Optical Sensors for the Detection of Early Stage Cancer
65
characterization of isolated whole crypts and primary
cultured human epithelial cells from normal,
hyperplastic, and adenomatous colonic mucosa.
Journal of Clinical Pathology, 58, 766-774.
Delvaux, M., Gay, G. (2006). Capsule endoscopy in 2005:
facts and perspectives. Best Practice & Research
Clinical Gastroenterology, 20, 23-39.
Dias, R. A., Correia, J . H., Minas, G. (2007). CMOS
Optical Sensors for being incorporated in Endoscopic
Capsule for Cancer Cells Detection. Proceedings of
ISIE 2007, 2747-2751.
Georgakoudi, I. (2006). The color of cancer. Journal of
Luminescence, 119, 75-83.
Haringsma, J ., Tytgat, G. (1999). Fluorescence and
Autofluorescence. Baillires Clinical
Gastroenterology, 13, 1-10.
Prosst, R. L., Gahlen, J . (2002). Fluorescence diagnosis of
colorectal neoplasms: a review of clinical applications.
International Journal of Colorectal Disease, 17, 1-10.
Stepp, H., Sroka, R., Baumgartner, R. (1998).
Fluorescence endoscopy of gastrointestinal diseases:
basic principles, techniques, and clinical experience.
Endoscopy, 30, 379-386.
Stringer, M., Moghissi, K. (2004). Photodiagnosis and
fluorescence imaging in clinical practice.
Photodiagnosis and Photodynamic Therapy, 1, 9-12.
Villette, S., Pigaglio-Deshayes, S., Vever-Bizet, C.,
Validire, P., Bourg-Heckly, G. (2006). Ultraviolet-
induced autofluorescence characterization of normal
and tumoral esophageal epithelium cells with
quantitation of NAD(P)H. Photochemical and
Photobiological Sciences, 5, 483-492.



BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
66
A LOW-POWER INTEGRATED CIRCUIT FOR ANALOG SPIKE
DETECTION AND SORTING IN NEURAL PROSTHESIS SYSTEMS
A. Bonfanti

, T. Borghi

, R. Gusmeroli

, G. Zambra

, A. S. Spinelli

, A. Oliynyk

L. Fadiga

and G. Baranauskas

Department of Robotics, Brain and Cognitive Sciences, The Italian Institute of Technology - Genova, Italy

Dipartimento di Elettronica e Informazione, Politecnico di Milano - IU.NET, Milano, Italy

Dipartimento di Scienze Biomediche e Terapie Avanzate, Sezione di Fisiologia umana, Universit a di Ferrara - Ferrara, Italy
[email protected], [email protected], [email protected], [email protected],
[email protected], [email protected], [email protected], [email protected]
Keywords:
Prosthetic device, Action potential (AP), Multichannel recording system, Spike detection, Spike sorting.
Abstract:
Since the proof that prosthetic devices directly controlled by neurons are viable, there is a huge increase in
the interest in integrated multichannel recording systems that register neural signals with implanted chronic
electrodes. One of the bottlenecks in such compact systems is the limited rate of data transmission in the wire-
less link, requiring some sort of data compression/reduction. To solve such a problem, we propose an analog
low power integrated system for action potential (AP) detection and sorting. In this system, AP detection is
performed by a double threshold method that reduces the probability of false detections while AP sorting is
based on the measurement of peak and trough amplitudes and spike width. The circuit has been implemented
in 0.35 m CMOS technology with power consumption of 70 W per channel including the pre-amplier.
The system was tested with real traces. Compared to standard AP sorting techniques, the proposed simple AP
sorter was able to correctly assign to single units over 90% of detected APs. Thus, our system preserves most
information encoded by APs and we estimate that for a typical trace the required bandwidth per channel will
be less than 4 kbps or 400 kbps for 100 channel.
1 INTRODUCTION
In recent years, in a number of laboratories it has
been shown that signals obtained with multichan-
nel extracellular recordings are sufcient to con-
trol the movements of a prosthetic device in real
time (Hochberg et al., 2006). However, we are still
far away from any device to be clinically tested be-
cause of enormous technological and scientic chal-
lenges (Stieglitz, 2007). According to current think-
ing, a future neural prosthetic device should con-
tain a set of neural ampliers connected to a signal-
processing unit and a wireless link to transmit the
amplied signals down to an actuator (e.g., a robotic
arm, a remote controller, a mouse, and so on). Unfor-
tunately, the low power state-of-the-art wireless sys-
tems cannot transmit data at the high rate required for
this kind of application where power budget is lim-
ited to 800 W/mm
2
(Harrison and Charles, 2003).
For example, 100 channels sampled at 30 kHz with
10 bits resolution would require to transmit 30 Mbps
while low power wireless systems can handle only
< 3 Mbps (Olsson et al., 2005). Thus, to enable
data transmissions from multichannel neural ampli-
ers, it is necessary to compress or reduce raw neural
data. Since in a typical neural trace only action po-
tentials (APs) contain any useful information, a very
efcient way of neural signal reduction is to trans-
mit a single bit for each AP. Researchers in (Harri-
son et al., 2007) used simple threshold detectors to
identify APs in the amplier signal and then trans-
mitted zeros when no AP were detected and ones
for each threshold crossing event. However, each
recording site usually collects the activity from sev-
eral nearby neurons and the aforementioned method
does not provide information on which neuron red
at a given time. This may be a serious drawback in
motor prostheses applications where it is important
to isolate the activity of each neuron to better predict
the intended movement (Hochberg et al., 2006). This
paper presents a low-power circuit that permits neu-
ron discrimination while drastically reducing the re-
quired data bandwidth per channel. This circuit ex-
tracts three parameters for a single spike, namely the
67
peak and trough amplitudes and the time between the
peak and the trough. It has been shown that the use
of these AP features is sufcient to adequately sorts
APs and is actually superior to spike spike separa-
tion using more parameters (Vibert and Costa, 1979).
According to (Olsson et al., 2005), for sampling rate
of 20 kHz at 5 bit resolution, a similar AP feature
extraction method yields data compression close to
90%. Although a similar in function analog device
has been described (Horiuchi et al., 2004) a digital ex-
traction of these three parameters has been presented
also (Olsson et al., 2005), in this work we introduce
signicant innovations: a) the implementation of a
new spike-detection algorithm that decreases several-
fold the rate of false detections; b) an improved peak
and trough detectors with very fast recovery time per-
mitting the detection of closely spaced APs, and c)
the use of an additional feature, namely the time dif-
ference between peak and trough occurrence. More-
over, the amplication of neural signals is performed
with an amplication topology that provides the best
known trade-off between noise and power consump-
tion. To conrm the efciency of our device, we
tested it employing articial and real signals and we
conclude that, in the majority of cases, with few in-
dependent units present in a trace, the extracted three
parameters can be sufcient to achieve adequate sin-
gle unit separation.
2 SYSTEM ARCHITECTURE
The system architecture is shown in Fig. 1 and com-
prises a low-noise amplier (LNA), a spike detec-
tor and a spike sorter. The spike detector splits the
amplied signal in two paths and exploits a double-
threshold strategy, which will be discussed in detail
in Section 2.2. The signal coming from the LNA
is also processed by the spike sorter, which extracts
from each detected spike three features, namely the
peak and trough amplitudes and the time interval be-
tween peak and trough. The detailed circuit and its
performance descriptions follows below.
2.1 Low-power and Low-noise Neural
Amplier
Pre-amplication is obtained through a double-stage
active lter, shown in Fig. 2. The rst stage is
an ac-coupled inverting high-pass lter, employing
two MOS-bipolar pseudoresistors as feedback ele-
ments (Harrison and Charles, 2003). This enables
the synthesis of high-value resistance without large
area consumption. Mid-band gain is set to G
1
=
VH
C
GM
LNA
WI NDOW
GENERATOR
CK
RN
FF
D Q SPI KE DET
PEAK
DETECTOR
TROUGH
DETECTOR
TROUGH PEAK
WIDTH
DETECTOR
VI N
POS SPI KE
NEG SPI KE
VDD
TI ME WI NDOW
VL
WI NDOW
CONTROL SPI KE
DETECTOR
SPI KE
SORTER
WI DTH
T
IM
E
W
IN
D
O
W
P
O
S
S
P
IK
E
Figure 1: Block scheme of the detection and sorting system.
+V
DD
-V
DD
C
1
TELE
OPAMP
V
I N
V
REF
+V
DD
2 STAGES
OPAMP
-V
DD
C
1
C
2
C
3
C
4
C
3
C
4
C
2
V
OUT
Figure 2: Overall schematic of the pre-amplier (LNA).
C
1
/C
2
= 66, with C
1
= 10 pF and C
2
= 150 fF.
The higher cut-off frequency is determined by the
gain-bandwidth product of the rst operational am-
plier (GBWP
1
) and is set to about GBWP
1
/G
1
=
20 kHz, while the high-pass pole frequency was de-
signed to be about 10 Hz as determined by C2 and
MOS pseudoresistors . The second stage was de-
signed to reject the rst stage offset while provid-
ing further signal amplication and ltering. This
stage employs the same pseudoresistor elements in
the feedback path, with a gain G
2
= 50 (achieved
with C
3
= 7.5 pF and C
4
= 150 fF). The operational
amplier in the rst stage was designed to have an
input-referred noise in the pre-amplier band lower
than 5 V
rms
. Furthermore, since this circuit is
thought to be used in an implantable system with a
large number of electrodes, the current consumption
was minimized by designing a telescopic operational
amplier (see Fig. 3); in fact, this well-known con-
guration gives excellent noise performance thanks
to the limited number of transistors and can achieve
the best trade-off between power consumption and
noise. The input PMOS and current mirror NMOS
transistors are biased in weak-inversion and strong in-
version region (Harrison and Charles, 2003), respec-
tively, since (W/L)
p
(W/L)
n
(see Table 1, where
IC is the inversion coefcient which is << 1 in the
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
68
weak-inversion region). This implies that M
P
transis-
tors have a transconductance much higher than M
N
ones and hence the equivalent input-referred noise is
mainly determined by the the input transistors of the
differential pair. Considering both the icker noise
and the thermal noise, the input referred power noise
density results:
E
2
n
eq
=
8KT
g
mp
+
2K
(1/ f )
p
C

ox
W
p
L
p
1
f
, (1)
with equal to 1/(2) in weak-inversion region (
0.7). In this working region the transconductance may
be approximated by:
g
m

=
I
D
U
T
, (2)
and so the input-referred thermal noise spectral den-
sity results:
E
2
n
th
=
8kTU
T

2
I
bias
, (3)
where I
bias
is the total current consumption of the tele-
scopic cascode amplier. Setting an upper limit for
the thermal noise contribution in the 20-kHz ampli-
er band of 3 V
rms
, a bias current of 4 A has to be
chosen. The icker noise contribution was carefully
minimized choosing a large area for the input PMOS
transistors, W
p
L
p
= 400 1 m
2
, that implies a
noise corner frequency lower than 100 Hz assuming a
icker noise coefcient K
(1/ f )
p
2 10
26
V
2
F. A
gure of merit proposed in (Harrison and Charles,
2003) and widely adopted to compare different design
is the Noise Efciency Factor (Harrison and Charles,
2003), dened as:
NEF =V
in,rms

2I
tot
U
T
4kT BW
, (4)
where V
in,rms
is the input-referred rms noise, I
tot
is the
total supply current and BW is the bandwidth of the
amplier. For the aforementioned sizing strategy, one
can nd from Eqs. (4) and (3) that the minimum NEF
value is

2/ = 2.02, which is the minimum value
achieved with a conventional differential pair as input
stage (Wattanapanitch et al., 2007).
Table 1: Transistor Operating Points.
(W/L) m V
ov
[mV] IC g
m
[A/V]
M
P
400/1 103 0.17 70
M
N
50/1 535 61 8
M
CAS
5/40 17 0.56 50
V
OUT
V
DD
=1 . 5V
M
p
(400/1 ) V
P V
N
M
p
(400/1 )
(50 /1 )
(5/40)
(5/40)
M
n
M
n
C
L
M
bi as
(50/1 )
V
bi as
=-0. 25V
(200/3)
-V
DD
=-1 . 5V
I
bi as
=4mA
M
cas
M
cas
50nA
Figure 3: Schematic of the telescopic operationaL amplier
used in the pre-amplier rst stage.
Figure 4: Performance of different spike detection algo-
rithms. a) False detection rate as a function of the threshold
amplitude normalized by ; b) Comparison of receiver op-
erating characteristics (ROC) in a low-SNR condition. P
det
and P
f alse
represent the probabilities to correctly and falsely
detect a spike.
2.2 Spike-detection Algorithm and
Circuit Implementation
The amplied signal is then sent to a spike-detection
circuit. The detection algorithm is based on simul-
taneous detection of the peak and trough of a spike
using two different thresholds, with a low-pass l-
ter placed on the trough detection channel. A spike
is detected only if a trough is found within a time
interval after the peak. For simplicity, we will call
the proposed algorithm that includes low-pass lter-
ing for trough detection the two-threshold with lter-
ing method, while the same algorithm without lter-
ing will be simply called the two-threshold method.
MATLAB
R
simulations were rst run to assess the
performance of the algorithm. For tests, the real-spike
waveforms obtained from in-vivo and in-vitro record-
ings were used to build the simulated traces by su-
perimposing a series of non-overlapping APs evenly
spaced at 5.5 ms onto a realistic noise (taken from
a recording without any spikes). The overall perfor-
mance of the algorithm was evaluated both by deter-
mining for different threshold values the probability
A LOW-POWER INTEGRATED CIRCUIT FOR ANALOG SPIKE DETECTION AND SORTING IN NEURAL
PROSTHESIS SYSTEMS
69
of false detections alone and by the receiver operat-
ing characteristic (ROC). Fig. 4a shows that, for low
thresholds, the two-threshold with ltering method
can reduce two- to threefold the rate of false detec-
tions compared to a standard single-threshold method,
while the improvement is less than 20% if no lter is
used. Such a difference is explained by the fact that
the trough is usually slower and smaller than the peak,
and, without a proper ltering, the trough detector is
much more likely to respond to the background noise.
Meanwhile, ROC analysis Fig. 4b shows that the two-
threshold with ltering method performs much better
than the other two methods, especially for low signal-
to-noise ratio (SNR). For instance, for a SNR (de-
ned as the ratio between the peak-to-trough ampli-
tude and six times the standard deviation () of the
noise) of 0.7, this method still detects APs reason-
ably well while the other two methods performno bet-
ter than detection by chance. Based on these results,
we designed an analog circuit to implement the two-
threshold with ltering algorithm. As shown in Fig. 1,
the conditioned signal is split in two paths. The rst
signal path has a comparator with a selectable posi-
tive threshold V
H
. Whenever the threshold is crossed,
the comparator triggers a time-window generator, de-
picted in Fig. 5. The output of the set-reset ip-
op is normally in a low logic state (Q = 0), keep-
ing the capacitor charged. When a positive peak of
AP is detected, the NMOS swiftly discharges the ca-
pacitor and the resulting drop of the transistor drain
voltage turns off the same NMOS, thus enabling the
capacitor re-charging. When the recharging capaci-
tor voltage crosses the trigger threshold, the cycle is
repeated. This block acts then as a clock generator,
feeding the counter that generates the output, namely
the time window. When the counter completes its cy-
cle, it resets the ip-op, bringing everything to the
initial state. The current I
CHARGE
can be externally
tuned in order to produce time windows of different
lengths. For example, by setting I
CHARGE
= 20 nA
we obtain a time window of 1 ms with a 10 pF ca-
pacitor. The second signal path has a low-pass lter
(implemented with a G
M
C cell with tunable band-
width) followed by a second comparator with nega-
tive threshold V
L
. Finally, outputs of the two channels
feed a D-type ip-op acting as an AND port: its out-
put is true only when a negative crossing happens dur-
ing the assertion of the time window. The following
parameters of the circuit were used for the tests here:
5 kHz for the low-pass lter, 0.6 for the ratio of the
two threshold amplitudes and about 2 ms for the time
window duration.
S
R
Q
Q
V
DD
-V
DD
COUNTER
EOC CK
I
CHARGE
C
POS SPI KE
WI NDOW
Figure 5: Simplied schematic of the time-window genera-
tor.
2.3 Spike-sorter Circuit
The amplied signal is also processed by a spike-
sorter block (see Fig. 1) that measures the peak and
the trough amplitudes and the width of the detected
spike. The peak detector circuit is shown in Fig 6(a).
If the ENABLE signal is low, the circuit acts as a tradi-
tional voltage follower, while when ENABLE is high
the source follower M
6
is biased with the small leak-
age current of transistor M
7
and it is able to follow
only the rising edge of the input signal. In the imple-
mented circuit, the ENABLE signal comes from the
time-window generator output: when the amplied
signal crosses the positive threshold, the time-window
signal is set high, forcing the detector to track the
peak. This control signal allows detecting peaks with
very different amplitudes and very close to each other,
that was not possible with the implementation in (Ho-
riuchi et al., 2004) because of the slow discharge after
a peak detection. Two aspects were taken into con-
sideration in the design of the peak detector. First,
the M
1
M
4
amplier gain was maximized (to about
50 dB) in order to reduce the detector offset (which
could cause a systematic error in the peak amplitude
evaluation). Second, the gate-source capacitance of
the follower transistor M
6
was minimized in order to
keep low the voltage drop caused by the input voltage
that drops below the peak voltage.
The trough detector is the PMOS equivalent of the
Fig 6(a) circuit. In this case, the enable signal for the
trough detector is the spike-detector output itself.
The width detector circuit, shown in Fig. 6(b), is a
time-to-amplitude converter, based on the charging of
a capacitor with a constant current, with two control
signals, ENABLE and RESET. The latter is the time-
window signal (TIME WINDOW), while ENABLE is
the logic exclusive OR between TIME WINDOW and
SPIKE DET. The crossing of the positive threshold
activates the current generator, M
5
, and switches off
M
6
. The capacitor C = 10 pF is charged by a 30 nA
constant current until the crossing of the negative
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
70
V
DD
=+1 . 5V
1 mA
V
I N
C
C
PEAK
ENABLE
C
PEAK
1 mA
3pF
5pF
M
1
M
2
M
3
M
4
M
5
M
6
M
7
M
8
M
9
-V
DD
=-1 . 5V
(a) Peak detector
RESET
WI DTH
ENABLE
1 0pF
M
1
M
2
M
3
M
4
M
5
M
6
30nA
V
DD
=+1 . 5V
-V
DD
=-1 . 5V
(b) Width detector
Figure 6: Simplied schematic of the peak detector (a) and
width detector (b).
threshold. Then, the capacitor voltage is kept constant
to a value proportional to the time between the posi-
tive and the negative threshold crossing, until the time
window signal goes down discharging the capacitor.
The gain of the width detector is about 3 V/ms. Note
that all the three sorter output signals (peak, trough
and width signals in Fig. 1) are meaningful only when
the spike detector output is high. If a spike is de-
tected, both the falling edge of the TIME WINDOW
signal and of the SPIKE DET signal reset the three
spike sorter outputs meaning that the SPIKE DET sig-
nal may be used as a trigger for read-out of the analog
sorter outputs in order to convert these signals into a
digital format.
3 CIRCUIT TEST
A test structure with a pre-amplier, a spike detec-
tion circuit and an analog sorter was implemented
in 0.35-m AMS CMOS process occupying a total
area of 0.24 mm
2
(see Fig. 7). The LNA was char-
acterized using an HP35665A Dynamic Signal Ana-
lyzer. The measured transfer function is presented in
Fig. 8, showing a mid-band gain of 71.5 dB and a
29 Hz22 kHz pass-band. The same gure shows the
transfer function of the LNA followed by the G
M
C
lter present in the trough path of the spike detector.
The upper cut-off frequency is set to about 5 kHz
by tuning the bias current of the G
M
C low-pass
cell. The measured input-referred noise power spec-
trum is shown in Fig. 9. It was obtained by divid-
ing the output noise power spectral density by the
square of the overall transfer function. The input-
referred thermal noise is about 25 nV/

Hz, as ex-
pected for the set bias current, while the noise cor-
ner frequency is about 800 Hz, higher than the esti-
mated value. This result is likely to be due the incor-
rect model of the icker noise for a transistor working
in the weak-inversion/subthreshold region, that is the
case for the input PMOS transistors. Integrating the
input-referred noise spectrum over the pre-amplier
band, we get a 4 V
rms
input noise, while the mea-
sured current consumption is 4.7 A for the rst stage
LNA and 0.5 A for the operational amplier in the
second stage. The Noise Equivalent Factor (NEF) of
the LNA is 2.45 which is the best result reported to
date (Fig. 10). However, this NEF is higher than the
theoretical limit obtained in the previous section, due
to the presence of the unaccounted icker noise and
of a second amplication/ltering stage. The spike
detection block was tested with simulated APs sig-
nals immersed in a realistic background noise fed into
the integrated circuit using a TTI TGA12104 arbitrary
waveform generator. Fig. 11 shows the amplied ar-
ticial neural activity, the control signals that trigger
the spike detector (output of positive- and negative-
threshold comparators) and the spike detector out-
put itself. Note the false-positive threshold crossing
shown in gure: a simple threshold algorithm would
interpret it as a true spike ring event, while our algo-
rithm rejects it correctly, thus increasing the detection
reliability. The current consumption of the spike de-
tector is largely determined by the comparators (each
one draws 2 A), and by the time window generator.
The latter dissipates only when the positive threshold
is crossed and in the case of 100 Hz ring rate, the
average current consumption is 10 A. The G
M
C
lter has a negligible current consumption since its
bias current is about 0.1 A. In addition, to verify
the correct measurement of the three features (peak,
trough and width amplitudes) for detected spikes, the
spike sorter circuit was tested with simulated wave-
forms. The amplied trace was recorded as well as
peak, trough and width outputs of the spike sorter. A
snapshot of the measured signals is shown in Fig. 12.
The three features measured by the spike sorter were
compared with the same features extracted from the
acquired LNA output using MATLAB
R
. For ampli-
ed spikes of 400 mV peak amplitude the error was
never larger than 30 mV. This value corresponds to
an error of 8 V for a 106 V peak amplitude input
A LOW-POWER INTEGRATED CIRCUIT FOR ANALOG SPIKE DETECTION AND SORTING IN NEURAL
PROSTHESIS SYSTEMS
71
Figure 7: Die photo of the proposed circuit.
1 10 100 1k 10k 100k
20
30
40
50
60
70
80
only LNA
-40dB/dec
+40dB/dec
LNA + G
M
-C filter
T
r
a
n
s
f
e
r

F
u
n
c
t
i
o
n

[
d
B
]

Frequency [Hz]
Figure 8: Measured transfer functions for the LNA ampli-
er (22-kHz bandwidth) and for the LNA + G
M
C lter
(5-kHz bandwidth).
spike. In the same way we obtained a maximum error
of 50 s for a spike width of 500 s. The power con-
sumption of the spike sorter block was kept very low:
the peak and the trough detector draw2 Aeach while
the width detector has a negligible consumption. The
power dissipation of the whole system (output buffers
excluded) is about 70 W fora 3-V power supply. The
main characteristics of the whole circuit are summa-
rized in Table 2.
4 TEST WITH SIGNALS
To test the efciency of the spike sorter, we used a
variety of traces obtained during in-vivo recordings in
monkeys employing tungsten electrodes (impedance
in the range of 0.5 2 M). The signals were down-
loaded to the waveform generator and then fed into
the IC through an attenuation lter to simulate the
real amplitude of the neural signal and the input re-
sistance of typical electrodes. For AP detection and
sorting, the peak threshold was set to about 4 times
the standard deviation of the background noise while
10 100 1k 10k 100k
-160
-150
-140
-130
-120
1/f
2
25nV/ Hz
1/f
I
n
p
u
t

R
e
f
e
r
r
e
d

N
o
i
s
e

[
V

2
/
H
z
]

Frequency [Hz]
Figure 9: Measured input-referred noise spectrum of the
LNA.
1 10 100
0.1
1
10
100
[Harrison, 2003]
[Perelman, 2007]
[this work]
[Olsson, 2005]
[Mohoseni, 2004]
[Harrison, 2007]
[Wattanapanitch, 2007]
N
E
F

-

t
h
e
o
r
e
t
i
c
a
l

l
i
m
i
t

Current Consumption [uA]

Figure 10: Difference between measured NEF and mini-
mum theoretical limit (2.02) of published neural differen-
tial ampliers as a function of the supply current.
Figure 11: Snapshot of measured signals. From top to bot-
tom: LNA output signal, outputs of the comparator with
negative and positive threshold and spike detector output.
the trough detector threshold was 0.6 times the peak
threshold. All these tests yielded similar results, so
that here we only describe a typical case obtained with
an in vivo recording trace (see Fig. 13). We compared
the chip sorted waveforms with the results obtained
employing the Principal Components (PCs) analysis
of AP waveforms and identifying clusters of events
in the space of the three largest PCs (Lewicki, 1998).
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
72
Table 2: System electrical characteristics.
Technology CMOS 0.35m AMS
Amplier + spike-detector + spike sorting area 0.24 mm
2
Power supply 3.0 V
Pre-amplier midband gain 71.5 dB
Pre-amplier bandwidth 29 Hz22 kHz
Pre-amplier input noise (29 Hz22 kHz) 4 V
rms
Overall pre-amplier NEF 2.45
Current consumption (output buffers excluded)
Pre-amplier (1
st
stage) 4.7 A
Pre-amplier (2
nd
stage) 0.5 A
Spike detector (for a 100 Hz ring rate) 14.1 A
Spike sorter 4.1 A
Total 23.4 A
Figure 12: Snapshot of the measured signals of the ana-
log sorter: amplied neural signal (gold line), peak detec-
tor output (red line), trough detector output (blue line) and
width detector output (green line).
In this case, the AP waveforms were obtained by de-
tecting the events of threshold crossing and collect-
ing about 0.3 ms of trace before and 1.3 ms after
the event. Since we used real data, there is no as-
surance that AP shapes are sorted correctly. There-
fore, to asses the quality of unit separation, we used
AP occurrence auto- and cross- correlograms (Har-
ris et al., 2000). The method is based on the fact that
each neuron has a certain refractory period of 23 ms
after ring an AP during which no spike can be gen-
erated. Thus, if all AP shapes are derived from a sin-
gle neuron, there should be no APs occurring at in-
tervals less than the refractory period of the neuron.
This absence of APs at brief intervals will correspond
to no events around the midpoint (0 ms interval) in
the auto-correlogram (Fig. 13). In the case of a pure
noise signal, the probability to detect events around
the midpoint (i.e.: events separated by small time in-
tervals) would be equal to the probability of detect-
ing events separated by large time intervals. Thus,
by comparing the frequency of events at very short
and very long intervals it is possible to estimate the
quality of the achieved unit separation. In the ex-
ample shown in Fig. 13, both methods were able to
detect two units and the remaining events turned out
to be noise. While in the PCA analysis both units
have clean refractory periods with no events for inter-
vals of < 3 ms, the chip separated units have a small
background noise (Fig. 13). However, compared to
the large interspike intervals, the frequency of these
short interspike interval events was very low (< 10
fold) suggesting that less than 10% of identied units
were misclassied. Similarly, the number of events
detected by our spike sorter was > 90% of the ones
detected by PCA analysis conrming that our simple
sorter is able to achieve 90% efciency of the PCA
analysis method. Similar results were obtained with
the simulated test signals downloaded from the public
database (http://www.vis.caltech.edu/

rodri/
Wave_clus/Waveclus_home.htm, data not shown)
thus conrming the conclusions reached with the real
signals.
5 CONCLUSIONS
The described analog integrated circuit for on-line AP
waveform separation was designed with intention to
be a part of a compact integrated multichannel sys-
tem where low power consumption, small size and
data compression/reduction are key factors to be con-
sidered during the systems design. The bandwidth re-
quired to transmit the three AP features depends on
the neurons ring rate. For a typical electrode signal
containing 2 3 neurons ring rate at < 50 Hz, and
with 10 bits of resolution, the estimated bandwidth to
transmit one channel data processed by our device or
less than 4 5 kbps. Thus, information contained in
100 channels can be wireless transmitted with a de-
vice capable of 500 kbps data rates that is well within
the range of current state of the art systems. The three
features selected to perform the discrimination of dif-
ferent spikes were proved to work with real signals.
Our implementation overcomes some of the problems
A LOW-POWER INTEGRATED CIRCUIT FOR ANALOG SPIKE DETECTION AND SORTING IN NEURAL
PROSTHESIS SYSTEMS
73
Figure 13: Comparison of PCA- and chip-based spike sorting. a) Unit 1 and 2 and noise plotted in the PCA space.
b)Waveforms of single units. c) Cross- and auto-correlograms of waveforms selected with PC analysis. d) Single wave-
forms plotted in the chip-extracted three dimension space of features. Isolated clusters are clearly visible. e) Single units
waveforms assigned by employing the chip-based sorting. f) Cross- and auto-correlograms of the single units identied with
the chip-based sorting method.
of similar systems, such as a slow recovery of peak
detectors after sensing an AP (Horiuchi et al., 2004),
while remaining an extremely low-power circuit with
overall power density of (<300 W/mm
2
) that makes
this circuit suitable for multichannel implantable sys-
tems.
ACKNOWLEDGMENTS
The authors would like to thank M. L. Grossi for her
help in the circuit design. This work was partially
supported by IIT (Italian Institute of Technology) and
by EC funds to LF (RobotCub IP).
REFERENCES
Harris, K., Henze, D., Csicsvari, J., Hirase, H., and Buzsaki,
G. (2000). Accuracy of tetrode spike separation as
determined by simultaneous intracellular and extra-
cellular measurements. Journal of Neurophysiology,
84:401414.
Harrison, R. and Charles, C. (2003). A low-power low-
noise cmos amplier for neural recording applica-
tions. IEEE J. Solid-State Circuit, (6):958965.
Harrison, R. R., Watkins, P. T., Kier, R., Lovejoy, R., Black,
D. J., Greger, B., and Solzbacher, F. (2007). A low-
power integrated circuit for a wireless 100-electrode
neural recording system. IEEE J. Solid-State Circuit,
(1):123133.
Hochberg, L., Serruya, M., Friehs, G., Mukand, J., Saleh,
M., Caplan, A., Branner, A., Chen, D., Penn, R., and
Donoghue, J. (2006). Neuronal ensemble control of
prosthetic devices by a human with tetraplegia. Na-
ture, (442):164171.
Horiuchi, T., Swindell, T., Sander, D., and Abshier, P.
(2004). A low-power CMOS neural amplier with
amplitude measurements for spike sorting. In Pro-
ceedings of ISCAS, pages 29322395.
Lewicki, M. (1998). A review of methods for spike sort-
ing: the detection and classication of neural action
potentials. Network: Computation in Neural Systems,
9(4):5378.
Olsson, R. and Wise, K. (2005). A three-dimensional
neural recording microsystem with implantable data
compression circuitry. IEEE J. Solid-State Circuit,
(40):27962804.
Olsson, R. H., Buhl, D. L., Sirota, A. M., Buzsaki, G.,
and Wise, K. D. (2005). Band-tunable and multi-
plexed integrated circuits for simultaneous recording
and stimulation with microelectrode arrays. IEEE
Trans. Biomed. Eng., 52:13031311.
Stieglitz, T. (2007). Neural prostheses in clinical practice:
biomedical microsystems in neurological rehabilita-
tion. Acta Neurochir Suppl, 97:411418.
Vibert, J. and Costa, J. (1979). Spike separation in multiu-
nit records: a multivariate analysis of spike descrip-
tive parameters. Electroenceophal. Clin. Neurophys.,
(47):172182.
Wattanapanitch, W., Fee, M., and Sarpeshkar, R. (2007).
An energy-efcient micropower neural recording am-
plier. IEEE Trans. Biomed. Eng., 1(2):136147.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
74
A BIOLOGICAL MONITORING MODULE BASED ON A
CERAMIC MICROFLUIDIC PLATFORM
Walter Smetana
1
, Bruno Balluch
1,2
, Ibrahim Atassi
1,2
, Khatuna Elizbarowna Gvichiya
2
Erwin Gaubitzer
3,2
, Michael Edetsberger
3,2
and Gottfried Khler
3

1
Institute of Sensor and Actuator Systems, Vienna University of Technology, Gusshausstrasse 27-29, Vienna, Austria
2
OnkoTec GmbH, Vestentting 1, Waidhofen/Thaya, Austria
3
Max. F. Perutz Laboratories, Department of Biomolecular Structural Chemistry, University of Vienna, Vienna, Austria
[email protected], [email protected], [email protected], [email protected]
[email protected], [email protected] and [email protected]
Keywords: LTCC-technology, Microfluidic, FEA.
Abstract: A 3-dimensional mesofluidic biological monitoring module has been successfully designed and fabricated
using a low-temperature co-fired ceramic (LTCC) technology. This mesofluidic device consists of a
network of micro-channels, a spherical mixing cavity and measuring ports. A selection of appropriate
commercially available ceramic tapes has been chosen with regard to their biocompatibility performance.
Specific processing procedures required for the realization of such a complex structure are demonstrated.
Three dimensional numerical flow simulations have been conducted to characterize the concentration
profiles of liquids at a specific measuring port and verified by experiment.
1 INTRODUCTION
Microfluidic and mesofluidic analytical systems are
becoming increasingly popular in chemical and
biomedical applications due to the need of small
volume reagents, small wastes and short reaction
times. Microfluidic devices may be classified into
functionally limited labs-on-chip (LOC) and micro
total analysis systems (TAS). Most LOCs are
single function and single layer devices such as
mixers, separation channels, etc. Micro total analysis
systems, on the other hand, are more complicated
and are capable of performing many functions such
as mixing, reaction, separation, etc. on a single
module. Generally these devices handle nanoliters of
reaction volumes and often accomplish their specific
tasks in milliseconds of reaction times. Most of these
systems are based on silicon, glass, poly-
methylmethacrylate (PMMA) or polydimethyl-
siloxane (PDMS) substrates (Anderson, 2000). A
range of rapid-prototyping methods using laser-
techniques for the fabrication of microfluidic
devices are reported in literature. Micro-stereo-
lithography, selective laser sintering, laser writing
method and microcladding techniques are applied
for generating complex 3-dimensional (3D)
microparts of polymer, metal and metal-matrix
composite (Kathuria, 2001, Yu, 2006).
Micropatterned ceramic components may be
fabricated in a rapid prototying process combining
stereolithography for the supply of master models
with low pressure moulding (Knitter, 2003). But true
3D microfluidic structures cannot be easily imple-
mented using these techniques due to process
limitations or material properties. The use of LTCC-
technology for microfluidic devices enables the
realization of multiple 3D microchannels, a feature
not easily attainable in other MEMS technologies
(Gongora-Rubio, 2001; Golonka, 2006). Therefore,
based on this experience LTCC technology has been
considered as an adequate approach to realize a
compact temperature controlled monitoring module
for biological reactions with low sample consump-
tion which may be considered as a TAS device.
The process technique for making a 3D-
architecture with LTCCs is rather simple and
standardized. Device production in LTCC
technology covers the machining, punching or laser
drilling of vias and channels on individual layers.
The individual layers are stacked and laminated in a
heated platen press or in a heated isostatic press.
Subsequently the laminated stack is exposed to the
firing cycle which is a somewhat critical process
where heating rate, dwell time at burnout
75

temperature and total firing cycle time have to be
matched to the thickness of the ceramic stack.
The field of nonlinear chemical kinetics has been
investigated about half a century. Still only a few
complex chemical reactions have been described by
means of an experimentally backed system of
coupled chemical equations. Moreover, in biological
relevant nonlinear systems most of the reactants, e.g.
proteins etc., are expensive to prepare and generally
only available in limited quantities. Therefore, it is
essential to use tiny reaction volumes for continuous
flow experiments, as realized in this reaction cell.
This hereby presented contribution outlines the
procedures of fabrication and characterization of a
reaction module designed to follow quantitatively
complex biochemical regulatory reaction networks
in vitro as a basis for advanced mathematical fitting.
This device consists of a mixing module which
provides fast mixing of reactants and four sensor
ports for simultaneous measurements. The
functionality was proved using the well known
oscillatory behaviour of the chlorite-iodide reaction
(Kgler, 2008).
2 DESIGN
Figure 1 shows the schematic of the monitoring
module also used as model for establishing finite
element analyses (FEA).
The module comprises a spherical reactor cell
where continuous mixing of the reagent fluids is
provided. Besides this mixing chamber the module
is equipped with pH-, oxygen-, temperature- and
iodide sensitive sensors for reaction monitoring as
well as with SMA-connectors for glass fibres
required for absorption or fluorescence
spectroscopic analyses (figure 2). Pumping a thermal
fluid through embedded ducts provides temperature
control. A network of micro-channels with cross
section dimensions varying from 200 m x 200 m
up to 2 mm x 2 mm are connecting the different
measuring sections within the ceramic module.
Special attention has to be spent on the selection of
appropriate ceramic tape material with regard to
biocompatibility and suitability to build up a
complex 3-dimensional structure which contains a
large number of cavities and channels.


Figure 1: Scheme of the monitoring module with reactor
cell and channel system.

Figure 2: Biological monitoring module (completely
assembled with sensor-, inlet-, outlet- ports and sockets for
optical fibers).
3 DEVICE FABRICATION
3.1 Biocompatibility Testing
Three different lead-free tapes have been considered
for this application like the ESL 41020, Ferro A6
and CeramTec GC-tape. The tapes should be
biocompatible but nevertheless cells should not
adhere and proliferate to a large extent on their
surfaces. The latter requirements are critical with
regard to clogging of channels by cell
agglomerations. In order to compare the
biocompatibility, proliferation, viability and
adherence of HeLa (human, cervix epithelial) cells
grown on sintered LTCC-tapes have been evaluated
using standard test procedures.



BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
76

3.1.1 Proliferation Test
As a first biocompatibility testing the ability of
HeLa cells to proliferate on different LTCC tapes in
contrast to glass and standard plastic surfaces was
observed. The influence on cell proliferation was
tested with the Bromodeoxyuridine (BrdU) assay
(Gire, 1998) (Calbiochem, USA). BrdU
incorporation is detected immunochemically with
unlabelled primary antibodies and HRPO labeled
secondary antibodies.
HeLa cells were exposed in a 96 well plate (1 x
10
5
cells/well) with the different test disks (4 mm x
4 mm) for 23 hours in DMEM (Dulbecco's mod.
Eagle-Medium) containing 4.500 mg Glucose, 4.5
mM L-Glutamine, 44 mM Na-bicarbonate, 100
units/ml Penicillin, 100 g/ml Streptomycin, 0.9
mM Na-pyruvate and 10 % fetal calf serum in a 6
volume % CO
2
humified atmosphere at 37 C
(standard conditions). During the final 3 hours of
incubation BrdU is added. The BrdU concentration
is measured using a multi-plate reader (BIORAD) at

abs
= 455nm and
reference
=655 nm.

Figure 3: Biocompatibility testing for HeLa cells on fired
LTCC- samples (CeramTec GC, ESL 41020, FERRO A6),
glass-cover slides (Assistent, Germany), Copper (Cu) and
standard plastic dishes (control-sample); Colour marking:
black: proliferation rates, red: percentage of viable cells,
blue: percentage of total cells compared to control group.
The results of the proliferation test are shown in
figure 3 (black bars). With exception of the
CeramTec GC-tape and the Cu- disks, the detected
proliferation-rates of all other test samples are equal
or even higher (glass support) compared to the
growing rate of the control group on plastic support
(100 8.7 %). Also the deviation is comparable to
the control group. In contrast CeramTec tapes show
a 30-40 % reduction in proliferation. Nevertheless
this reduced proliferation is low in comparison with
cells incubated with Cu-disks, which under the same
conditions show a proliferation rate of 2-3 %
compared to control group.
3.1.2 Viability Testing
Next to the determination of proliferation rates it is
necessary to analyze the ability of cells to adhere to
surfaces and to determine the percentage of viable
cells. For this test series HeLa cells were incubated
in standard culture dishes with a diameter of 10 cm
(about 5 x 10
5
cells/dish) for 23 hours in DMEM
under standard conditions. The fired LTCC tapes to
be tested, glass cover slides and Cu-plates were
used with lateral dimensions of about 2.5 cm x 4 cm.
Cells were incubated in the presence of these
materials and on standard culture surfaces as control.
The number of cells and their viability was
evaluated with propidium-iodide (Zarnai, 2001)
using a Nucleocounter (Chemometec, Denmark).
The results of the viability tests are presented in
figure 3.
No significant reduction in number of total cells
(blue bars) has been detected for all LTTC samples
and glass as their percentage of total cells, in relation
to the control group, are comparable within the error
limits caused by inaccuracies of seeding the cells.
Only samples grown on Cu show about 40 % less
cells than the other samples.
Additionally not any significant reduction in
percentage of viable cells (red bars) was shown for
cells grown on LTCC tapes or glass support. Again,
only samples grown on Cu-plates show a diminution
of about 40 %.
3.1.3 Adhesion Testing
The side walls of channels and cavities are formed
by the laser machined edges of tapes. To observe
cell adhesion, which may rely on tape material but
also on surface finishing a test method has been
designed which enables to examine the potential
adhesion of cells on the laser machined edges of the
tapes. To evaluate the influence of the ceramic tapes
on the adhesion of cells on glass surfaces, CeramTec
and ESL ceramic tapes with laser-micro machined
tapering channels with decreasing channel width
(see figure 4-A) were mounted on standard glass
cover slides (Assistant, Germany) with a super
adhesive (Loctite). HeLa cells were seeded on the
cover slides (10
5
cells) and incubated over night in
DMEM medium under standard conditions. The
growth behaviour of the cells was evaluated using a
light transmission microscope.
A BIOLOGICAL MONITORING MODULE BASED ON A CERAMIC MICROFLUIDIC PLATFORM
77


Figure 4: A: Test sample of a ceramic tape coupon
carrying a continuous row of channel segments with
decreasing width (central circular opening is 10 mm in
diameter, the channel width is 2.122, 1.175, 0.672, 0.394,
0.228, 0.168 mm and 2.208, 1.224, 0.653, 0.396, 0.102,
0.054 mm for channels 1 to 6 of the ESL (D, F) and
CeramTec (C, E) tape, respectively. C, E: HeLa cells
grown on glass cover slides with a mounted fired
CeramTec-tape at channels 0-1 (C) and 2-3 (E). D-F:
HeLa cells grown on glass cover slides with a mounted
fired ESL-tape at channels 0-1 (D) and 3-4 (F). B: HeLa
cells grown on glass cover slides surrounded by mounting
glue (arrow). Micrographs were performed with a Zeiss
Axiovert S100TV at 10x magnification and a Digital
Camera (Nikon DMX1200).
Exemplary the results for the CeramTec and the
ESL tapes are shown in figure 4. It could be
demonstrated that the initial circular opening and
channel 1 machined in the CeramTec GC tape and
the ESL tape show only a poor influence on the
adherence of HeLa cells (figure 4-C, D; cell density
in circular opening: 250 300 cells/mm
2
, channel 1:
200 220 cells/mm
2
) but already at channels 2 and 3
(figure 4-E) with a width of about 0.6 mm a
significant decrease in adherence can be observed
for CeramTec tapes (cell density in channel 2: 45
cells/mm
2
, channel 3: 19 cells/mm
2
) as no significant
number of cells is observed at this area. In contrast
the fired ESL-tape shows quite a different cell
adherence performance. Also in channels 3-4 (figure
4-F) a rather dense cell population of cells (cell
density in channel 3 and 4: ca. 100 cells/mm
2
) and
even aggregation near to the edges are observed.
Only for smaller channels with width of about 0.2
mm a decrease of adherence can be observed (data
not shown). The mounting glue has not any effect on
the adherence behavior of HeLa cells to the glass
surface (figure 4-B).
3.1.4 Summary of Biocompatibility Testing
It can be concluded that all three LTCC materials
tested do not affect the viability of the cells, as the
total number of cells counted after the same period
was found identical within an error interval of 10
% for all LTCC-materials, standard plastic as well
as glass. An increased cell death compared to the
control group could not be found for any of the
standard materials. Only copper showed a significant
influence on viability, cell adherence and
proliferation. If only the proliferation characteristic
is considered, a distinction in tape performance can
be made. Whereas the ESL-tape shows a similar
behaviour with respect to the control assay of a
standard plastic support, FERRO-tape reacts similar
to a glass support and shows an increased
proliferation. Only the fired CeramTec-tape exhibits
a significant decrease in proliferation (figure 3).
These results are also validated by the adherence
test. When cell growth on a glass support was
restricted by sidewalls of a narrow channel
configuration, the cells reacted differently to ESL-
and CeramTec-tapes: Whereas on ESL tapes the
cells grow even in channels of very small width, it
was not the case for channels machined into
CeramTec-tapes. For these samples the area near to
the glass ceramic interface was widely free of cells
and an efficient growth of cells was only detectable
in rather wide channels. The reasons of the
difference in cell adhesion on the channel walls are
still under investigation. It might be related to the
difference in surface energy as result of laser
machining. The quality of the cutting edges and the
surface of channel walls are defined by the
absorption of laser energy in the tape material which
may vary in dependence on the composition of the
selected LTCC- material.
The results show, that on the one hand side fired
ceramic tapes are generally biocompatible and do
not restrict the viability of cells, and on the other
hand side cells adherence can be specifically
influenced by the material selected. CeramTec GC
tape should be preferred, when adherence to the
walls of the chamber or channels should be
restricted as e.g. in microscopic observation
chambers where single cells are preferentially
observed in the center of an optical window and a
cell agglomeration along the side walls or growth
into micro fluidic flow channels should be avoided.



BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
78

3.2 Processing Procedures
The CeramTec GC tape has been proved as a
suitable candidate for the realization of the
monitoring module not only with regard to the
results of the biocompatibility tests but also due to
its performance characteristic during processing.
Since the module comprises a complex network of
channels and cavities a tape material is required
which does not tend to sag during firing. Beyond the
green GC-tape with a thickness of 325 m provides
an adequate ruggedness for handling.
The three dimensional LTCC-structure is
realized by forming a stack of adequately laser
machined (diode pumped Nd:YAG-laser (Rofin
Sinar, Germany) equipped with an acoustic optical
switch, operating power: 12 W at TEM00-mode)
single layers of tapes which are laminated and
finally exposed to a firing process. For the
realization of the complex module a single tape
shows a rather delicate structure since it is
penetrated by large-area laser machined
perforations. A standard LTCC-processing
technology cannot be easily applied for the
realization of the module since it comprises 133
ceramic layers, which deviates from the number of
tapes usually applied for conventional applications.

Figure 5: Inside view of reactor cell with various channels
and ducts (collated stack of tapes before lamination).
Special attention has to be paid on the lamination
of the ceramic tapes since the module contains a
large number of cavities and channels (e.g. the
reactor cell has a cavity volume of 1 cm
3
). The
finished sheets are collated in a mould (figure 5) and
aligned by registration pins providing that
successive layers are rotated by 90 to compensate
for the texture (preferential orientation) induced by
the fabrication of green tape. The lamination of the
stack of tapes has been carried out in a heated platen
press (Wabash).
A range of experimental work has been
conducted (Wang, 2008) in order to optimize
lamination parameters which should contribute to
provide the shape integrity of channel and cavity
structures. Sagging and delamination are typical for
relatively wide channels (width equal to 500 m or
more) whilst contraction is characteristic for narrow
channels (width equal to 200 m and less). The
application of sacrificial material for filling cavities
and channel structures is a valuable approach to
avoid sagging as well as the risk of delamination.
The selection of an appropriate sacrificial material is
absolutely essential for this complex microstructure
device. It has been found out that the main task of
the sacrificial material is to provide a uniform
pressure distribution within the LTCC-tape stack
during lamination. An additional supporting function
of cavity structures during firing is not required
since the considered LTCC material shows in all the
phases of the firing process an adequate strength and
stability. This substance should evaporate during the
burnout phase of the sintering process without
damaging the structure of the module. So the
approach was to find a material, which is
accommodated to the shrinkage performance of the
tape. An obstruction of materials shrinkage should
be avoided. Some authors recommend carbon black
as sacrificial volume material (SVM) which may be
applied as tape or paste (Birol, 2005). This material
decomposes and exhausts at rather high temperature
when sintering of tape already starts. Different
polymer materials have been tested as potential
candidates acting as SVM since they decompose and
burn out at a temperature <400 C (before tape-
shrinkage is starting).
Best results have been attained with PMMA
chosen as SVM. It is also part of the organic
constituents of the considered tape and exhausts free
of residues at the burnout temperature of tape. The
organic sacrificial material has been used in powder
form. Channels and cavities are filled by a vacuum
sucking technique. It has been found out that a
uniform pressure of only 30 bar (sample
temperature: 70 C, lamination time: 3 minutes) has
to be applied onto the ceramic layer stack which
enables to maintain the rectangular cross-section of
the buried channels while still avoiding delamination
of the ceramic batch. The applied pressure was
reasonably lower than recommended for
conventional applications.
Thermal gravimetric analyses have been
conducted to optimize the sintering profile
(especially to provide a complete and slow burn out
of the organics before sintering) in order to avoid
A BIOLOGICAL MONITORING MODULE BASED ON A CERAMIC MICROFLUIDIC PLATFORM
79

crack formation or delamination. The binder
decomposition process was studied by means of
Thermogravimetric Analysis (TGA) coupled with a
Mass-Spectrometer (MS) to detect simultaneously
the evolved gases (Balluch, 2008). Additionally, the
binder burnout was analyzed by Dynamic Scanning
Calorimetry (DSC). TGA revealed multistage
decomposition behaviour of the binder system. The
binder degradation is predominately governed by
exothermic reactions in the temperature range up to
400 C and the course of the heat flow may be
correlated with the evolution of carbon dioxide. The
heat flow of GC-tape shows two exothermic peaks:
one at 250 C and a major one at 364 C. For the
GC-tape the first exothermic peak corresponds to the
maximum degradation rate in the TGA as can be
seen in the insert of figure 6.

Figure 6: Comparison of the heat flow signal with the ion
current for carbon dioxide for GC-tape.
The results of this study yield a useful approach
to establish the firing profile of the considered
LTCC with regard to heating rate and intermediate
dwell time for the preheat phase of firing process.
Based on the results of TGA it becomes evident that
for the burnout stage of the GC-tape firing process
an appropriate dwell time depending on the mass of
LTCC-module has to be established. It has to be
provided that the PMMA starts to pyrolyze slowly
which enables to exhaust the gaseous decomposition
products via the channels. In contrast a rapid
decomposition of the organic filler induces a sudden
intensified production of gaseous burnout products
which cannot escape adequately via the channels and
finally results in a destruction of the LTCC module.
The burnout phase of the firing schedule has to be
adapted to these requirements whereas an adequate
dwell time (depending on the mass of the LTCC
module) at the critical degradation temperature of
the filler is of great importance. Practical
experiments have shown that a dwell time at 250 C
is ignorable if an adequate slow heat rate is selected.
A sufficient long dwell time at 350 C for the total
binder burnout is obligatory to be provided.
Firing of the samples has been conducted in a
box furnace of Heraeus (heating rate for temperature
range 25 C 350 C: 0.8 C/min, holding time at
350 C: 6 h, heating rate for temperature range 350
C 500 C: 0.8 C, holding time at 500 C: 4 h,
heating rate for temperature range 500 C 920 C:
1.8 C/min, holding time at peak temperature of 920
C: 1 h, cooling rate: 2 C/min). Another parameter,
which has to be considered during the firing cycle,
was the temperature distribution within the sample.
A temperature gradient >6 C within the ceramic
module during the total firing cycle has to be
avoided with regard to potential crack formation and
fracture of the module (Kluge, 2006).
4 DEVICE CHARACTERIZATION
The influence of arrangement of inlet-channels
along the perimeter of the reactor cell and the sink
on the bottom of the cell on the flow performance of
liquids within the reactor cell has been studied for
different flow conditions by means of FEA using the
FLUENT program package. The characterization of
the device is based on time - dependent flow
simulations. The numerical model describes the flow
of dyed and pure water entering through different
inlets at varying mass flow rates and passing into the
spherical cavity of the reactor cell and the channel
system. The local distribution of fluid at the
spectroscopic port is predicted and contrasted with
the results attained by spectroscopic analyses of light
absorption at the considered port.
Exemplarily the flow characteristic for a
representative assumption has been derived by
numerical simulation and validated by experiment.
The flow condition described starts with filling the
cavity by injection of pure water at the radial inlets 1
and 2 (figure 1 and 7). The mixing initiated as a step
inflow of dyed water (0.5 % trypan blue) is imposed
on the radial inlet 2, which means that a stream of
constant mass flowrate and dyeconcentration is
applied at the respective inlet (figure 7). In the
meantime the average residence time of dye
concentration is measured at the optical port (figure
1).
The propagation of the streamlines depends
strongly on the position of the inlet along the
meridian of the spherical cavity as can be seen in
figure 7 for the corresponding inlet arrangement.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
80

The inlet 1 (In 1) is placed 0.45 mm higher than
inlet 2 (In 2). Within a period of 2.5 s the
streamlines of the fluid entering inlet 1 are already
diffusing into the interconnecting channels while
those of the fluid starting from inlet 2 are still ending
in the mixing chamber.

Figure 7: Propagation of streamlines of fluids entering
inlet In 1 (blue) and inlet In 2 (red) with flow rates of
2 ml/min within a period of 2.5 s.
The model of the complete module was
constructed and meshed with the program package
GAMBIT (Fluent Inc.) Mesh elements were mainly
hexahedral and only few tetrahedral elements were
required for completely meshing the module. In all,
the model (including the recirculation tube) has a
volume of 3 cm and includes a total number of
292,194 mesh elements, of which 161,374 are
hexahedral, 128,737 are tetrahedral and 2,083 are
pyramidal.
The curves in figure 8 describe the increase of
the dye volume in the optical port from initially pure
water to the mixed state. For the supply of inlets
with liquids two Flodos Stepdos 03 pumps were
used and for providing the flow circulation a Flodos
NF 5 diaphragm pump was applied. The flow rates
considered in simulation and experiment are 2
ml/min at both inlets. At the optical port the
absorbance was measured using a uv-vis array-
spectrometer (EPP2000-50 m Slit, StellarNet Inc.).
The absorbancecharacteristic in a scaled
presentation is shown in figure 8. The steady state
mixed flow condition at the optical port corresponds
to a dye concentration of 47 %. It can be noticed that
the ideal and the computed curves are almost
identical. However, the experimentally derived
characteristic shows a slightly less rate of rise as
well as oscillating peaks. The oscillating peaks may
be attributed to the pulsating characteristic of the
pumps in use. Furthermore it must be noted that the
equilibrated state concentration condition is attained
also within the same period if the injection flow rate
at the inlet ports has been varied synchronously
while keeping the total sum of mass flow rates
constant. If all three inlet ports are used
simultaneously for the supply of the module
different mass flow rates have to be selected to
provide the desired mean concentration profile at the
optical port.

Figure 8: Dye concentration vs. time characteristic at the
optical port.
Besides the performed dye distribution tests, the
functionality of the microfluidic reaction module has
been demonstrated by observing the time evolution
of the iodine concentration due to the iodide-chlorite
reaction, which involves only 2 inorganic ions and is
characterized by an established set of reaction
mechanisms according to the LLKE-model, which is
named after its authors Lengyel, Li, Kustin and
Epstein (Kgler, 2008). This complex reaction
model involves an autocatalytic step and the
coupling of different reactions can cause complex
reaction patterns like oscillatory behaviour when
measured under certain conditions in continuous
flow. An intermediate of the chemical reaction leads
also to production of the starting compound, which
accelerates the reaction until the respective reaction
partner is used up, but when the reaction is
performed under continuous flow conditions
oscillatory behaviour results which can be observed
continuously. Figure 9 shows the oscillations
observed for the iodine absorption measured around
470 nm (i.e. in a range between 456 to 484 nm).
This means, that the optical density of the product is
measured as a mean value over 60 pixels of the
CCD-array (full black curve). The red curve shows
the final result when the time dependence is
additionally averaged over 20 seconds. The figure
demonstrates that such an autocatalytic reaction can
be measured in the microfluidic reaction module,
although the total reaction volume and also the
optical path of the optical detection port is very
A BIOLOGICAL MONITORING MODULE BASED ON A CERAMIC MICROFLUIDIC PLATFORM
81

small. Nevertheless oscillatory behaviour can also
found under such conditions. The device is sensitive
enough to record small concentration changes giving
rise to changes in the optical density below 0.001. In
future experiments the microfluidic module will be
used for measuring complex kinetics in biochemical
systems. In that case, limited amounts of reactive
compounds, e.g. of enzymes, are available and the
use of small reaction volumes is an essential
prerequisite to allow quantitative measurements.

Figure 9: Oscillations of the I
2
absorption observed in the
module produced by the reaction of I
-
and ClO
2-
in
continuous flow.
5 CONCLUSIONS
The influence of different ceramic materials on the
viability and proliferation of HeLa cells has been
tested as a first basic characterization of the
biocompatibility. In respect to their use in micro
fluidic devices the growth of cells in channels of
different width has been visualized. The LTCC-
technology has been proved as a very versatile
method to build up complex three-dimensional
multilevel channel structures including even large-
volume cavities, which are suitable for biological
and diagnostic applications.
ACKNOWLEDGEMENTS
The financial support of this work by the WWTF,
project MA05 "Inverse methods in biology and
chemistry as well as by the FFG, project N209
"Nanoflu" is greatly acknowledged by the authors.
This project is integrated within EU 4M Project
(Contract Number NMP2-CT-2004-500274). The
maintenance with ceramic tapes and many useful
advices during the course of tape processing by Dr.
C. P. Kluge (CeramTec AG) are especially
appreciated.
REFERENCES
Kathuria, Y. P. (2001) An overview of 3D structuring in
microdomain, J. Indian Inst. Sci. vol. 81, pp. 659-
664.
Anderson, J . R., Chiu, D. T., J ackman, J ., Cherniavskaya,
O., McDonald, J . C., Wu, H., Whitesides, S. H. and
Whitesides, G. M. (2000) Fabrication of
topologically complex three-dimensional microfluidic
systems in PDMS by rapid prototyping, Anal. Chem.,
vol. 72, no. 14, pp. 3158-3164.
Yu, H., Balogun, O., Li, B., Murray, T. W. and Zhang, X.
(2006) Fabrication of three dimensional
microstructures based on single-layered SU-8 for lab-
on-chip applications, Sensors Actuat. A Phys.,
vol.127, pp. 228-234.
Knitter, R., Bauer, W. and Ghring D. (2003)
Microfabrication of ceramics by rapid prototyping
proces chains, J. Mechan. Eng. Sci. vol. 217, pp. 41-
51.
Gongora-Rubio, M. R., Espinoza-Vallejos, P., Sola-
Laguna, L. and Santiago-Aviles, J . J . (2001)
Overview of low temperature co-fired ceramics tape
technology for meso-system technology (MsST),
Sensors Actuat. A Phys., vol.89, pp. 222-224.
Golonka, L. J ., Zawada, T., Radojewski, J ., Roguszczak,
H., and Stefanow, M. (2006) LTCC Microfluidic
System, Int. J. Appl. Ceram. Technol., vol. 3, no. 2,
pp. 150-156.
Kgler, Ph., Gaubitzer, E. and Mller, St. (2008)
Parameter identification for chemical reaction
systems using the adjoint technique and sparsity
enforcing stabilization - a case study for the Chloride
Iodide reaction, submitted for publication to J. Phys.
Chem.
Gire, V. and Wynfford-Thomas, D.W. (1998)
Reinitiation of DNA Synthesis and Cell Division in
Senescent Human Fibroblasts by Microinjection of
Anti-p53 Antibodies, Mol. Cell. Biol., vol.18, no. 3,
pp.1611-1621.
Zamai, L., Canonico, B., Luchetti,F., Ferri, P., Melloni, E.,
Guidotti, L., Cappellini, A., Cutroneo, G., Vitale, M.
and Papa, S. (2001) Supravial Exposure to Propidium
Iodide Indentifies Apoptosies on Adherent Cells,
Cytometry vol. 44, pp.57-64.
Kluge, C. P.(2006) private communication.
Birol, H., Maeder, T., J acq, C., Straessler, S. and Ryser, P.
(2005) Fabrication of Low-Temperature Co-fired
Ceramics micro-fluidic devices using sacrificial
carbon layers, Int. J. Appl. Ceram. Technol., vol. 2,
pp. 364-373.
Balluch, B., Lftl, S., Seidler, S., Smetana, W. (2008)
Binder Burnout Studies of LTCC Tapes by Means of
Thermal Analyses, Proc. 32 nd Int. Conf. of IMAPS,
Pultusk, Poland: in print.
Wang, X. S., Balluch, B., Smetana, W., Stangl, G. (2008)
"Optimization of LTCC-channel Fabrication for
Micro-Systems", Proc. of 30st ISSE 2008, pp.420-425.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
82
MINIATURIZED ELECTROCHEMICAL SENSING SYSTEMS
FOR IN VITRO AND IN VIVO BIOMEDICAL APPLICATIONS
V. I. Ogurtsov, K. Twomey, N. V. Bakunine, C. Mc Caffrey, J . Doyle, V. Beni and D. W. M. Arrigan
Tyndall National Institute, University College Cork, Ireland
[email protected], [email protected], [email protected], [email protected]
[email protected], [email protected], [email protected]
Keywords: Miniaturized, Electrochemical, Sensor, Sensing system, In vitro, In vivo, Potentiostat.
Abstract: Development of miniaturized electrochemical sensing systems for in vitro and in vivo biomedical
applications is discussed. The systems are based on high sensitivity potentiostatic instrumentation, which is
suitable for chemical and biochemical sensors. The in vitro application is an 8 channel hand-held PC-
controlled system with user-friendly interface. This is capable of implementing different electrochemical
potentiodynamic techniques. The in vivo applications are realized using two approaches: a small sized PCB
with commercially available ICs, and a specially developed on-chip system. The performance of the
systems is validated through electrochemical characterization of a microarray sensor.
1 INTRODUCTION
Electrochemical sensors play an important role in
biomedical applications, including clinical disease
diagnostics (Peng et al., 2008), (Seo et al., 2008),
(Ndamanisha and Guo, 2008), (Wang and Ha, 2007),
(McLaughlin et al., 2002), drug testing (Abbaspour
and Mirzajani, 2007), microbiological pollutant
determination (Morales et al., 2007), and detecting
and quantifying DNA and proteins (Shiddiky et al.,
2008). These sensors give rapid and sensitive
measurements, and can detect solid, liquid or
gaseous analytes. If we consider the disease
diagnostic applications, this diversity can be
illustrated with the following examples. Nitric oxide
sensors (Peng et al., 2008) have found application in
the monitoring of neural conditions such as stroke,
heart attack and epilepsy, through the link between
elevated nitric oxide levels and these diseases.
Various glucose sensors are available for in vitro or
continuous in vivo monitoring of blood, urine and
saliva to diagnose diabetes (Seo et al., 2008). There
are oxygen sensors (McLaughlin et al., 2002) for
monitoring blood oxygen levels in pathways to
different organs. Recently sensors for uric acid have
been developed which indicate the presence of gout
and Lesch-Nyhan diseases (Ndamanisha and Guo,
2008). Sensors, which measure pH levels within the
gastrointestinal tract, can detect the presence of
intestinal diseases like gastroesophageal reflux
disease (GERD) (Wang and Ha, 2007). Also,
electrochemical sensors for in vivo drug monitoring
are being explored to monitor drugs administered to
treat different diseases e.g. anti-inflammatory drugs
for the treatment of intestinal diseases (Abbaspour
and Mirzajani, 2007).
In the area of clinical diagnosis, there is a
growing emphasis on earlydisease detection.
Through this capability, serious diseases such as
heart disease and cancer can be more successfully
treated. To achieve this challenging goal,
developments in three key areas are essential. Firstly
more sensitive and more diverse sensors should be
created. Sensors with a lower limit of detection, for
example, would be crucial for early diagnostics
(before full-scale disease symptoms develop), and an
expansion of the number of available sensors for
analytes or biomarkers linked to the disease is
important to improve accuracy of disease diagnosis.
Secondly these sensors need to move from their
current use in highly specialized clinical laboratories
to areas of point-of-care (POC) such as at the
patients bedside. To enable this move,
developmental work in the area of instrumentation is
necessary, which will provide easy access to these
methodologies for the patient and physician, e.g.
user-friendly portable systems or in vivo systems
such as implantable devices that will allow
continuous supervision of the patient with access
close to the source of disease. This ability would
allow for more accurate monitoring of the disease
and provide early disease diagnosis.
83

There have been notable advances in suitable
sensors for early stage disease detection (Peng et al.,
2008), (Seo et al., 2008), (Ndamanisha and Guo,
2008), (Wang and Ha, 2007), (McLaughlin et al.,
2002), (Abbaspour and Mirzajani, 2007), (Morales
et al., 2007), (Shiddiky et al., 2008), However, there
remains a need to address the developments in
appropriate systems to allow progression of these
sensors outside of the lab to POC and to in vivo
applications. Currently, sensing instrumentation is
mostly bench-top based equipment, which is large
and not adapted to work outside of the lab
environment. Therefore, the development of
miniaturized sensing instrumentation is essential.
In this study, we discuss approaches to develop
miniaturized sensing systems for in vitro and in vivo
applications based on a potentiostatic solution,
which is commonly used with different
electrochemical sensors, and different sensing
methods (amperometry, voltammetry,
impedometry). Although, fundamentally the systems
for in vitro and in vivo applications are built on
similar circuitry, they vary significantly due to
different requirements in size, power, sensitivity,
biocompatibility and functionality. For the in vitro
applications, we have developed a portable, hand-
held, miniaturized, multichannel potentiostatic
system, which is optimized for operation with a
sensor array on chip. The system has the appropriate
accuracy and sensitivity required by biomedical
applications, and would be suited for use in a
hospital or at a GPs office. For the in vivo
applications, we have developed two variants of the
system that are both miniaturized and can operate
over extended periods on low power, again with the
required accuracy and sensitivity. The first solution
is based on commercially available low-power, low-
noise, micro-sized small outline integrated circuit
(SOIC) components and chips. The second device
presents a specialized on-chip system that is
developed in-house. The in vivo systems have
similar performance specification but the PCB-based
system costs less. Both of these systems are
optimized for low-power operation, making them
applicable for implantable devices, where an
appliance is implanted and remains long term in the
body, operating on a continued basis.
2 SENSING SYSTEM
STRUCTURE
In general, for both system solutions, the
electrochemical potentiostatic sensing system
consists of an electrochemical cell incorporating the
sensor or sensor array, and two main analog
electronic units, the potentiostat and a
transimpedance amplifier, and a microcontroller
unit. These analog units connect to the
microcontroller unit that controls the measurement
process, and provides data acquisition and
connectivity to the personal computer (PC) as shown
in figure 1.
Microcontroller
Multichannel
potentiostat
Transimpedance
amplifier
Microcell
PC

Figure 1: Block diagram of electrochemical sensing
system for in vitro and in vivo applications.
The electrochemical cell is usually a three
electrode structure comprising a counter electrode
(C), reference electrode (R) and working electrode
(W), which is immersed into or covered by sample
solutions to be analyzed. Electrochemical or
biochemical reactions in the cell are detected with
electronics, in particular with a transimpedance
amplifier. The reactions are dependent on the W
potential, therefore, a stable potential at the W
should be provided. The W potential stability is
secured by the R, which supplies a reference
potential independent of the environment, and the
potentiostatic unit, which maintains this W potential,
with respect to the R, to be equal to a stimulation
signal generated by the microcontroller unit. The
shape of the stimulation signal depends on the
sensing methodologies (i.e. DC for amperometry,
staircase for voltametry and so on). A simplified
schematic of the electrochemical potentiostatic
sensing system (without microcontroller unit) is
shown in figure 2

Figure 2: Simplified schematic of the potentiostatic
sensing system.
Implementation of the electrochemical sensing
systems for in vitro and in vivo biomedical
applications is each governed by a different set of
requirements. For in vitro real- time systems, these
2
3
1
Op3
R
W
Vout
2
3
1
Op2
2
3
1
Op1 C
Vin
Current pass
CELL
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
84

requirements are high sensitivity, rapid rate of
measurement and corresponding signal processing,
accurate data interpretation, a multi-functional
instrument and simplicity of its operation by the end
user. For in vivo systems the main challenging
requirement besides sensitivity is the size of device,
which must adhere in some cases to millimeter size
scale, and continuous operation over extended
periods of time (i.e. up to one year, or longer for
some implantable devices). Both of these restrictions
impose a serious limitation on the power and size of
the electronic components which can be used for the
system design.
2.1 In vitro System
In keeping with these requirements for the in vitro
applications, a multi-channel potentiostatic system,
optimized for operation with an 8 working electrode
on-chip sensing system, has been developed. The
instrument is based on low-noise, high precision,
pico-ampere input current operational amplifiers
AD8602 (leakage current of 200 fA) and Analog
Devices microconverter, ADuC812, which enables
generation of the stimulation signal (via the in-built
DAC), feeds sensing data into the system (via the in-
built 8 channel ADC), and provides preliminary
signal processing and communication with the PC
(via an RS232 communications port). Running on
the PC is dedicated software with user-friendly
graphical interface, which provides final signal
processing and data interpretation. The prototype of
the system, with overall size 170x110x40 mm and
weight 250 g, is shown in figure 3. The associated
user-friendly interface software (developed in
LabVIEW, figure 4) allows the user to specify the
type of voltammetric technique and its setting (start
and finish potential, scan rate for each channel), and
to visualize obtained experimental data.

Figure 3: The multi-channel potentiostatic system suitable
for in vitro applications.
In order to validate and characterize the system,
a microelectrode array was fabricated using CMOS
techniques. It consisted of an array of gold working
electrodes surrounded by a counter electrode. The
final packaged die connects to a PCB dipstick which
allows for easy connection and testing of the sensor
chip. The solutions with analyte can be placed
directly on the sensor, or the dipstick can be
integrated into an associated fluidic system, or it can
be simply immersed into a container with the sample
to be analysed. The performance of the developed
system can be estimated by example of cyclic
voltammetry applied to the 8 Ws in ferrocene
carboxylic acid solution in simultaneous mode. The
measurements that were carried out over different
scan rates are shown in figure 5a, and the
measurements that were carried out over different
start and finish potentials are shown in figure 5b.
These experiments demonstrate that the systems
current sensitivity is greater than 1nA, with a noise
level less than 50pA p-p.

Figure 4: PC Interface for the multi-channel potentiostatic
system.

a.

b.

Figure 5: Cyclic voltammograms recorded simultaneously
with the instrument for oxidation of Ferrocene carboxylic
acid at an eight microelectrode array: (a) with different
scan rates set for each channel (from 50mV/s to 400mV/s),
(b) with different starting and finishing potentials set for
each channel.


MINIATURIZED ELECTROCHEMICAL SENSING SYSTEMS FOR IN VITRO AND IN VIVO BIOMEDICAL
APPLICATIONS
85

2.2 In vivo System
In keeping with requirements for the in vivo
applications, a potentiostatic system based on a PIC
microcontroller, has been developed. This
microcontroller family contains a number of
microcontroller models including those which are
available in a smaller size, consumes less power and
has more RAM memory than the ADuC, making
them more suitable for this application. The
PIC18F2520 has been chosen which features
6x6mm, 1256 bytes RAM and current consumption
of 2.6mA during normal operation. This is in
contrast to the ADuC812 features of 8x8mm, 256
bytes RAM and 6.2mA during normal operation. In
the programming of the microcontroller, close
attention has been given to reduce the power
consumed during device operation by using
appropriate electronic components with power
switch on/off mode and inbuilt microcontroller
possibilities. There are two realizations of this
system. The first one is a PCB-based device with
separate commercially-available low power and
small size ICs forming a one-channel potentiostatic
system in accordance with figure 1. The size of the
system prototype is 12x24 mm, which is suitable for
a variety of in vivo applications. To validate the
performance of the system, cyclic voltammetry was
carried out in 0.5M sulfuric acid solution at a gold
macroelectrode, figure 6.

Figure 6: Cyclic voltammogram for in vivo PCB-based
system at the macro gold electrode in 0.5M sulphuric acid
solution
The second in vivo system is a specially
developed on-chip system, which is a realization of
the PCB in vivo device in an integrated single die.
The implementation of the developed system on a
silicon substrate was made by using a C5P0 CMOS
5 micron silicon breadboard. There is a digital core
of 1248 gates and an analog oriented section
containing 32 operational amplifier blocks, 1500
units of poly-poly capacitor, approximately 2M of
resistance in 40 sticks of 6 segments each, and a
range of more specialized components. The
schematic of the breadboard operational amplifier
unit and final chip appearance, are shown in figures
7 and 8 correspondingly. The performance of the on-
chip system can be seen from the example of cyclic
voltammetry of a macro gold W in 0.5M sulphuric
acid solution, figure 9, which substantiates its
operability. At present both of the in vivo systems
have similar performance specification but due to
the initial small batch size, the PCB-based system
costs less. The potential advantages of the on-chip
system over PCB based device (a smaller size, a
reduction in noise and power, a decrease of leakage
current, which leads to an improvement of the
system performance) will be realized when complete
integration of sensor and electronics on one chip is
developed.


Figure 7: P-input CMOS OPAMP. Figure 8: Fabricated
on-chip potentiostat
die, 8x10mm.

Figure 9: Cyclic voltammogram for in vivo on-chip system
at the macro gold electrode in 0.5M sulphuric acid
solution
3 CONCLUSIONS
Miniaturized sensing systems for in vitro and in vivo
biomedical applications suitable for point-of-care
applications have been presented that will facilitate
early-stage disease diagnosis. The systems are based
on potentiostatic instrumentation, which is
commonly employed with different electrochemical
sensors. For in vitro requirements, an 8 channel PC-
controlled system has been developed, capable of
carrying out a number of sensing methods. For the in
vivo applications, two systems have been realized,
both based on the PIC microntroller. The first is a
PCB with separate commercially-available low
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
86

power and low size ICs forming a one-channel
potentiostatic system. The second is a specially
developed integrated on-chip system. Future work
will include optimization of the developed systems
for specific biomedical application according to end-
user requirements. Additionally for the on-chip
system the chip will be redesigned to improve its
characteristics and reduce the fabrication cost.
ACKNOWLEDGEMENTS
Financial support of this work by CFTD/04/112,
CFTD /05 /122 Enterprise Ireland and National
Access Programme NAP68 projects is gratefully
acknowledged.
REFERENCES
Peng, Y., Hu, C., Zheng, D., & Hu, S. (2008). A sensitive
nitric oxide microsensor based on PBPB composite
film-modified carbon fiber microelectrode. Sensors
and Actuators B-Chemical, in press.
Seo, H., Park, D., Park, J . (2008). Fabrication and
characterization of platinum black and mesoporous
platinum electrodes for in-vivo and continuously
monitoring electrochemical sensor applications. Thin
Solid Films, 516(16), 5227-5230.
Ndamanisha, J . C., & Guo, L. P. (2008). Electrochemical
determination of uric acid at ordered mesoporous
carbon functionalized with ferrocenecarboxylic acid-
modified electrode. Biosensors & Bioelectronics,
23(11), 1680-1685.
Wang, M., & Ha, Y. (2007). An electrochemical approach
to monitor pH change in agar media during plant
tissue culture. Biosensors & Bioelectronics, 22(11),
2718-2723.
McLaughlin, G. W., Braden, K., Franc, B., & Kovacs, G.
T. A. (2002). Microfabricated solid-state dissolved
oxygen sensor. Sensors and Actuators B-Chemical,
83(1-3), 138-148.
Abbaspour, A., & Mirzajani, R. (2007). Electrochemical
monitoring of piroxicam in different pharmaceutical
forms with multi-walled carbon nanotubes paste
electrode. Journal of Pharmaceutical and Biomedical
Analysis, 44(1), 41-48.
Morales, M. D., Serra, B., de Prada, A. G. V., Reviejo, A.
J ., & Pingarron, J . M. (2007). An electrochemical
method for simultaneous detection and identification
of Escherichia coli, Staphylococcus aureus and
Salmonella choleraesuis using a glucose oxidase-
peroxidase composite biosensor. Analyst, 132(6), 572-
578.
Shiddiky, M.J.A., Rahman, M.A., Choel, C.S., Shim, Y.
(2008). Fabrication of disposable sensors for
biomolecule detection using hydrazine electrocatalyst.
Analytical Biochemistry, 379(2), 170-175.
MINIATURIZED ELECTROCHEMICAL SENSING SYSTEMS FOR IN VITRO AND IN VIVO BIOMEDICAL
APPLICATIONS
87
IMPLEMENTATION OF AN AUTOMATED ECG-BASED DIAGNOSIS
ALGORITHMFOR A WIRELESS BODY SENSOR PLATFORM
Francisco J. Rinc on
1
, Laura Guti errez
1
, M onica Jim enez
1
, Vctor Daz
2
, Nadia Khaled
3
David Atienza
1,4
, Marcos S anchez-

Elez
1
, Joaqun Recas
1
and Giovanni De Micheli
3
1
DACYA/UCM, Madrid, Spain
{francisco.rincon, lagutier, monicajimenez}@fdi.ucm.es, {jrecas, marcos}@s.ucm.es
2
Depto. Salud y Rendimiento Humano/UPM, Madrid, Spain
[email protected]
3
LSI/EPFL, Lausanne, Switzerland
{nadia.khaled, giovanni.demicheli}@ep.ch
4
ESL/EPFL, Lausanne, Switzerland
[email protected]

Keywords:
Wireless Body Sensor Networks, Biomedical signal processing, Electrocardiogram.
Abstract:
Wireless Body Sensor Networks (WBSN) are poised to become a key enabling technology of personal systems
for pervasive healthcare. Recent results have however shown that the conventional approach to their design,
which consists in continuous wireless streaming of the sensed data to a central data collector, is unsustainable
in terms of network lifetime and autonomy. Furthermore, it was established that wireless data communica-
tion is responsible for most of the energy consumption. To address the energy inefciency of conventional
WBSNs, we advocate an advanced WBSN concept where sensor nodes exploit their available, yet limited
processing and storage resources to deploy advanced embedded intelligence and processing, to reduce the
amount of wireless data communication and consequently energy consumption. More specically, this paper
addresses the design and optimization of an automated real-time electrocardiogram (ECG) signal analysis and
cardiovascular arrhythmia diagnosis application for a prototype sensor node called Wireless 25 EEG/ECG sys-
tem. The satifactory accuracy of this on-line automated ECG-based analysis and diagnosis system is assessed
and compared to the salient off-line automated ECG analysis algorithms. More importantly, our results show
an energy consumption reduction of 80% to 100% with respect to conventional WBSNs, when our analysis
and diagnosis algorithm is used to process the sensed ECG data to extract its relevant features, which are
then wirelessly reported to the WBSN central data collector, after the node can automatically determine the
potential cardiovascular pathology without human monitoring.
1 INTRODUCTION
AND RELATED WORK
Wearable personal health systems for pervasive mon-
itoring and healthcare are widely recognized to
be a key enabling integration technology for next-
generation advanced citizen-centric eHealth delivery
solutions (Lo and Yang, 2005). Through enabling
continuous biomedical monitoring and care, they hold
the promise of improved personalization and qual-
ity of care, increased ability of prevention and early

This work is partially supported by the Spanish Gov-

ernment Research Grant TIC 2005-5619 and the Integrated
Systems Center of EPFL under the project Performance
monitoring for professional and recreation sports using
Wireless Sensor Networks (2007-2008).
diagnosis, and enhanced patient autonomy. Further-
more, wearable personal health systems can help the
eHealth sector realize rapid sustained market growth,
reduction of healthcare and public costs.
To provide the necessary accurate, integrated and
long-term assessment and feedback, these wearable
personal health systems must sense, acquire, moni-
tor and analyze a large number of physiological and
metabolic parameters, both during physical activity
and rest. The number as well as the nature of the pa-
rameters of interest depends on the actual biomedical
application/scenario and target population. Neverthe-
less, it is largely accepted that Wireless Body Sen-
sor Networks (WBSN) will be the underlying com-
mon architecture and technology of these personal
health systems. More specically, the WBSN will
88
consist of a number of sensor nodes attached to the
patient body, each sensor node potentially compris-
ing of 5 components (Lazzer et al., 2002; Culler
et al., 2004): sensors, actuators, a microprocessor,
a wireless transceiver and an energy source. Each
WBSN node ensures the accurate sensing and capture
of its target physiological data, its (pre-) processing
and wireless communication to the other nodes and
the wearable Personal Digital Assistant (PDA). This
PDA will be responsible for the storage, organization,
complementary analysis and fusion of the collected
information, its user-friendly representation, and its
dissemination to the relevant medical staff or cen-
tral monitoring service through private and/or public
wireless access networks (Lo and Yang, 2005).
State-of-the-art commercial products and experi-
mental prototypes of personal health monitoring sys-
tems merely apply on-board analog ltering to the
sampled sensed data, before it is either logged on a
bulky patient unit for off-line analysis, or wirelessly
transmitted to a remote monitoring system (Jovanov
and et al., 2005; LifeShirt, ; SmartShirt, ). The ob-
trusiveness and off-line nature of the analysis of the
rst approach compromises its acceptance and appli-
cability to pervasive healthcare, whereas the second
approach is not sustainable in free-living conditions in
terms of autonomy. Indeed, current results in WBSN,
e.g., prosthesis processing (Kemere and et al., 2004)
or Electroencephalogram (EEG) / Electrocardiogram
(ECG) monitoring (L ofgren and et al., 2007)) indi-
cate that an unaffordable amount of energy would be
spent in the wireless communication, if no local sig-
nal processing is present and most of the acquired
data is wirelessly streamed to the PDA. Moreover,
similar conclusions can be derived from the Code-
Blue project (Project, ), which is a WBSN that targets
biomedical monitoring by including a set of devices
to collect ECG and oxigen saturation data, which can
be transmitted through a wireless network to a wide
range of receiving devices that can display the data
in real time. The conclusions of this project outline
that the largest proportion of energy is consumed in
the wireless data transmission, and requires monitor-
ing of the received data by a doctor or biomedical
specialist; thus, the WBSN nodes are not able to re-
port any physical anomaly. Therefore, we advocate
in this paper an advanced WBSN concept where sen-
sor nodes exploit their available processing and stor-
age resources to deploy advanced embedded intelli-
gence and processing, which will be optimized for
enhanced functionality and autonomy. More partic-
ularly, in this paper, we investigate the feasibility and
benets of such an advanced WBSN for an automated
electrocardiogram (ECG) signal analysis and cardio-
vascular arrhythmia recognition application, using a
prototype sensor node called Wireless 25 EEG/ECG
system (Penders et al., 2007).
A signicant amount of research effort has been
devoted to the automated analysis of ECG signals.
Some of the proposed methods are able to classify
a set of Arrhythmias depending on special correlated
characteristics of the ECG signal, for instance, using
Multicategory Support Vector Machines (Khadtare
and Sahambi, 2004). Other methods are based on
the underlying detection of the major ECG charac-
teristic waves, namely the QRS complex, P and T
waves (S ornmo and Laguna, 2005). As a matter of
fact, the performance of an automated ECG analy-
sis system using the second approach critically de-
pends on the reliable detection of these ducial waves.
The most salient methods proposed for the auto-
mated detection of the ECG ducial waves belong to
three categories: ltering or adaptive thresholding,
wavelet transform-based and (nonlinear) multiscale
transform-based (S ornmo and Laguna, 2005). The
latter approach was evidenced to have less noise sen-
sitivity than adaptive thresholding, and to avoid the
problem of position deviation exhibited by wavelet-
based techniques. Therefore, in this paper, we con-
sider a multiscale morphological derivative (MMD)
transform-based algorithm to realize automated ECG
characteristic wave detection.
While the retained MMD transform-based algo-
rithm was validated by simulation (Sun et al., 2005),
its translation into a robust, efcient and reliable au-
tomated diagnosis capability embedded in our wear-
able sensor node calls for the porting and (non-
straightforward) optimization of this algorithm to
adapt it to the sensor nodes limited processing re-
sources. In general, this porting and optimization
effort is key to translate the recent biomedical sig-
nal processing advances into autodiagnosis tools, and
hence to enable pervasive healthcare. As a result, the
main contributions of the paper are:
The design of a real-time ECG-based diagnosis al-
gorithm, including a new run-time ECG signal re-
construction module, based on the off-line MMD
algorithm, and a diagnosis module able to identify
various anomalies in the cardiovascular function.
The porting and optimization of the new real-time
ECG-based diagnosis algorithm on the Wireless
25-channel EEG/ECG sensor node platform.
The application of the new diagnosis algorithm
for autodiagnosis on-board the sensor node to sig-
nicantly reduce the amount of data to be wire-
less transmitted, and consequently, dramatically
reduce the sensor nodes energy consumption and
extend its battery life.
IMPLEMENTATION OF AN AUTOMATED ECG-BASED DIAGNOSIS ALGORITHM FOR A WIRELESS BODY
SENSOR PLATAFORM
89
The rest of the paper is organized as follows. In
Section 2, we overview the software and hardware
architecture of the WBSN node used in this work.
Then, in Section 3, we present the ECG-based diag-
nosis algorithm ported to our WBSN node. Next, in
Section 4, we describe the performance and energy
consumption trade-offs between node processing and
the communication tasks. Finally, in Section 5, we
summarize the main conclusions of this work.
2 WIRELESS BODY SENSOR
NODE ARCHITECTURE
As previously mentioned, this work uses as target
platform a Wireless 25-channel EEG/ECG sensor
node (Penders et al., 2007). This small sensor node
can monitor up to 25 different bio-potential signals
and transmit in real-time wirelessly the sensed infor-
mation to a data collector device. Furthermore, it
can be combined with other instances of the same
node to dene complex WBSN topologies. The node
has a very elaborated hardware/software architecture,
which enables porting different types of biomedical
signal processing applications, use different hardware
components, and customize the communication stack
for various wireless communication protocols. In this
section, we summarize the main features of the har-
ware/software architecture of this node.
2.1 Hardware Architecture
The hardware architecture of the Wireless 25-channel
EEG/ECG sensor node is partitioned in three main
blocks, which relate to the three main tasks in WSNs,
namely, sensing, processing and wireless transmis-
sion of the acquired information to a base station (e.g.,
a PC, PDA or a mobile phone).
The sensing task is performed by a 25-channel
ultra-low-power Application-Specic Integrated Cir-
cuit (ASIC) that is able to extract the bio-potential sig-
nals. The core of the processing part is a TI MSP430
ultra-low-power microcontroller (Lutz Bierl, 2000).
This small 16-bit microcontroller has a low active
power (0.6 nJ/instruction), as well as a a very low
stand-by power (2W). In addition, it features a fast
wakeup from stand-by to active mode (6s) and an
on-chip 12-bit analog-to-digital converter, suitable for
biomedical signal processing. Finally, it also includes
60kB of ash program memory and 2kB of RAM to
upload applications and store processed data from the
different sensors.
Then, the wireless communication task is carried
out using a Nordic nRF2401 transceiver (Semicon-
Radio
nRF2401
C
MSP430
25-ch
ASIC
Radio
driver
C
driver
ASIC
driver
MAC
TinyOSkernel
Application
Figure 1: OS-Based EEG/ECG 25-ch sensor node architec-
ture overview.
ductor, 2000), which is an ultra-low-power 2.4 GHz
communication chip. This component has a very low
power consumption, with only 10.5mA at an output
power of -5dBm and 18mA in receive mode. Also,
it includes built-in power-down modes that allow to
switch-off the radio when not used, and special trans-
mission modes to reduce the processing needed by the
microcontroller for every new packet sent or received.
Overall, this radio is a low-power solution for WBSN,
which has relatively low duty cycle rates.
2.2 Software Architecture
The software architecture of the platform follows a
layered modular approach in which each hardware
component (sensors, microcontroller and radio) is a
separate software module. This modular multi-layer
structure, depicted in Figure 1, is supported thanks
to the inclusion of TinyOS (Culler, 2006), a light
event-based operating system specially designed for
WSNs. It has been written using the nesC (Gay et al.,
2003) programming language, which is an extension
of the C language optimized for the memory restric-
tions of sensor networks. Thus, it is possible to eas-
ily port new biomedical signal processing applica-
tions using the nesC language, and to use the under-
lying hardware driver support provided by TinyOS
to access different hardware blocks in the architec-
ture. In addition, the abstraction from the hardware
blocks makes possible to modify/replace the hardware
blocks included in each instance of the node, with-
out modifying the remaining blocks or the upper lay-
ers (communication protocols, applications, etc.). In
particular, in this research work, while a porting for
the TI MSP430 microcontroller was already available
within the TinyOS hardware support, we had to de-
velop the new drivers to support the Nordic nRF2401
radio module and the 25-channel EEG/ECG ASIC in
the operating system, to be able to use them from the
ECG-based diagnosis application layer.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
90
3 ECG-BASED DIAGNOSIS
ALGORITHM
The design of an ECG-based diagnosis algorithm to
detect possible cardiovascular diseases is a very com-
plex process that can be divided in two main phases.
First, a real-time detection of the characteristic waves
of ECG signals (Sun et al., 2005; Sun et al., 2002)
(e.g., complex QRS, P and T wave, etc.) is an es-
sential module for the quality of the detection of the
cardiovascular diseases; thus, we have developed a
run-time ECG wave identication module, which is
described in Section 3.1. Second, once the funda-
mental features of the input ECG signal are identi-
ed, they can be used to detect if the heart is suffer-
ing an anomalous behavior of the cardiovascular func-
tion and diagnose corresponding pathologies. This
last phase of the proposed ECG-based diagnosis al-
gorithm is described in Section 3.2.
3.1 Run-Time ECG Signal
Reconstruction
Our run-time ECG signal reconstruction phase is
based on a multiscale morphological derivative
(MMD) transform-based singularity detector (Sun
et al., 2005). This initial detector is an off-line al-
gorithm able to reliably identify the Q wave, R peak,
S wave and the onsets and offsets of the P wave and
T wave, which are fundamental blocks of any ECG
signal, as shown in Figure 2.
Although the ECG signal reconstruction using the
MMD algorithm is quite accurate (Sun et al., 2005),
the fundamental issue to directly apply for WBSN is
the fact that MMD requires as input data the com-
plete wave read from a patient during a large amount
of time, which is subsequently processed to tune the
thresholds to apply for each concrete sample. Unfor-
tunately, the memory resources of a WBSN sensor
node are very limited (only few kilobytes) to min-
imize its size and power consumption; thus, taking
into account the high sampling frequency needed for
a reliable ECG signal processing, only a few seconds
of information can be stored, which is insufcient to
accurately tune the thresholds for MMD. Moreover, if
all the data is streamed from the sensor node to a de-
vice with larger capacity (e.g., a base station) to pro-
cess them, the sensor node would run out of battery
in a short period of time, because of the elevated data
transmission rate that is needed and the high energy
consumption of the radio. As a result, WBSN require
an on-line approach for ECG signal reconstruction, in
which the complete ECG signal is processed directly
in the sensor nodes and only the relevant results are
(a)
A
B
C
D
E
F
G
H
I
J
K
(b)
(c)
Figure 2: Results obtained using the MMD detector. (a)
Original ECG input signal (b) Multiscale morphological
transform of the original signal (c) Points and waves de-
tected by the algorithm.
A=onset of the P wave, B=P peak, C=offset of the P wave, D=onset
of the Q wave, E=onset of the R wave, F=R peak, G=offset of the
R wave, H=offset of the S peak, I=onset of the T wave, J=T peak,
K=offset of the T wave.
sent to the data collector device or base station.
In the following we propose a run-time process-
ing of the ECG signal, based on a MMD transform-
based singularity detector, which can be directly ex-
ecuted in a WBSN sensor node. In the rst phase of
our run-time ECG ducial wave detection algoritm,
the received input signal is ltered for noise reduction
and baseline wander correction (S ornmo and Laguna,
2005). To this end, we have implemented a morpho-
logical lter on the input signal (Sun et al., 2002),
where at all instances of time during the execution,
a window of the last 300 samples are stored in a cir-
cular buffer. Hence, since the sampling frequency of
the 25-channel EEG/ECG node is 200Hz, the current
implementation of the algorithm uses the last 1.5s of
the input signal as history of processed signal.
Once each sample is ltered, the following step in
an MMD detector requires the application of a mul-
tiscale morphological transform on the ltered data
to detect the ducial points in the original ECG sig-
nal. However, due to the scarce memory resources
in the node, the transformed signal cannot be stored
in the node; thus, the morphological transform is dy-
namically recalculated when needed during the fol-
lowing steps of the run-time algorithm. Then, the lo-
cal maxima and minima of the transformed signal can
be selected, based on some characteristics of the mor-
IMPLEMENTATION OF AN AUTOMATED ECG-BASED DIAGNOSIS ALGORITHM FOR A WIRELESS BODY
SENSOR PLATAFORM
91
Table 1: Steps followed by the proposed run-time ECG sig-
nal reconstruction algorithm. All the operations are per-
formed over the transformed data.
1. The thresholds Th
r
and Th
f
are selected based
on an adaptive thresholding from the histogram
of the transformed signal
2. r
peak
= local minima with absolute amplitude
larger than Th
r
If r
peak
was not detected, go to 1, otherwise
continue
3. r
on
=rst local maxima on the left of r
peak
with
absolute amplitude larger than Th
f
r
o f f
= rst local maxima on the right of r
peak
with absolute amplitude larger than Th
f
4. q
wave
=rst local minima on the left of r
on
with
absolute amplitude larger than Th
f
5. s
wave
= rst local minima on the right of r
o f f
with absolute amplitude larger than Th
f
6. p
o f f
= rst local maxima on the left of q
wave
with absolute amplitude larger than Th
f
p
on
=rst local maxima on the left of p
o f f
with
absolute amplitude larger than Th
f
7. t
on
= rst local maxima on the right of s
wave
with absolute amplitude larger than Th
f
t
o f f
=rst local maxima on the right of t
on
with
absolute amplitude larger than Th
f
phological trasform and a dynamic thesholding, up-
dated at run-time using a window of the last received
samples. These detected local maxima and minima
are then subsequently used to locate the characteristic
points of the wave. In this regard, since the sampling
frequency is 200Hz, the microcontroller only has 5ms
between two consecutive samples to perform all the
calculations on the data to nd the local maxima and
minima. As this time interval is too short to execute
the complete detection of maxima and minima points,
we have divided the algorithm into 7 phases, which
are listed in Table 1. Each of these phases can be ex-
ecuted in the interval between two consecutive sam-
ples. Thus, when a new data is sampled, ltered and
included in the circular buffer, one of the 7 phases is
executed. For example, at the beginning of the pro-
cess, when no R peak is yet present in the considered
window, steps 1 and 2 are repeated continuously be-
tween data samples. Then, when a R peak is detected
in step 2, steps 3 to 7 are applied in sequence, such
that, each of them is used in the time slot between
two consecutive samples. After the complete identi-
cation of the current input ECG wave is nished, the
process restarts from step 1 for the next wave.
For illustration purposes, one of the results ob-
tained using the proposed run-time ECG signal re-
construction algorithm in an input wave is depicted
in Figure 2. This gure shows three different waves,
the rst one (or a) is the original ECG signal given
as an input to the algorithm. The second wave (or
b) is the original signal after applying the multiscale
morphological transform, all the points detected by
the algorithm are marked. The last one (or c) shows
the detected waves and peaks over the original signal.
This last wave illustrates where the Q wave, R peak,
S wave and the onsets and offsets of the P and T wave
are found using our run-time algorithm.
3.2 Arrhythmia Detection Phase
After processing the original ECG signal and obtain-
ing the ducial points of the wave (Figure 2), our
diagnosis module is applied to check if the signal
presents any anomalous behavior, which may hint that
the patient is suffering a cardiovascular pathology.
The diagnosis module checks iteratively ve dif-
ferent conditions based in the points detected in the
ECG signal, according to the valid ranges reported
in (Schamroth, 1971; P erez-G omez, 1985), namely:
1. The time from D to H must be less or equal to
0.10s.
2. The time interval fromAto D must be in the range
from 0.12s to 0.20s.
3. The amplitude of J must always be positive.
4. The time from D to F must not be longer than
0.03s.
5. The QT interval rule, which establishes a relation
between the interval from D to K. This rule indi-
cates the valid interval between the heart beat rate,
and the last RR interval (i.e., the interval from the
last R peak to the current one). To nd the valid
QT interval, Bazets formula (1) is used to calcu-
late the QT coefcient for an input signal (QTc)
as shown in Equation 1. Then, the valid values of
the QTc are reported in Table 2.
QTc =
time interval f rom D to K

previous RR interval
(1)
As a result, in the current version of our diagnosis
algorithm, according to the results of checking these
previous ve conditions, the sensor node reports to
the base station that the heart of the monitored patient
can be suffering one of the pathologies of Table 3.
4 EXPERIMENTAL RESULTS
The complete proposed ECG-based diagnosis algo-
rithm has been ported and optimized to be executed
in the 25-channel EEG/ECG node. The statistics
and energy values shown in the rest of the paper
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
92
Table 2: Normal values of the QTc. HBR=Heart Beat Rate
(per minute), RR=RR interval. Values in last two columns
are in seconds.
HBR RR QTc and normal limits
40 1.5 0.46 (0.41 - 0.51)
50 1.2 0.42 (0.38 - 0.46)
60 1 0.39 (0.35 - 0.43)
70 0.86 0.37 (0.33 - 0.41)
80 0.75 0.35 (0.32 - 0.39)
90 0.67 0.33 (0.30 - 0.36)
100 0.60 0.31 (0.28 - 0.34)
120 0.50 0.29 (0.26 - 0.32)
150 0.40 0.25 (0.23 - 0.28)
180 0.33 0.23 (0.21 - 0.25)
200 0.30 0.22 (0.20 - 0.24)
were obtained using the model of the 25-channel
EEG/ECG systemfor PowerTOSSIM, which was val-
idated in (Rinc on et al., 2008), showing variations of
less than 4% between the simulation framework and
the measurements of the nal platform.
4.1 Validation of the ECG-based
Diagnosis Algorithm
The accuracy of the algorithm running on the 25-
channel EEG/ECG system was tested with several in-
put signals taken fromthe QT database (Laguna et al.,
1997). This database was specially created for eval-
uation of algorithms that detect waveform boundaries
in ECG signals. It consist of 105 fteen-minute ex-
cerpts of two-channel ECG recordings. It also con-
tains a subset of beats (at least 30 in each recording)
that have been manually annotated by cardiovascular
analysis experts.
In our rst set of experiments, we have com-
pared the performance of the proposed algorithm
with MMD and other state-of-the-art ECG detec-
tion algorithms from the literature for detection of
waveform boundaries in ECG signals: the original
MMD off-line algorithm (Sun et al., 2005), an adap-
tive thresholding-based detector (TD) (Daskalov and
Christov, 1999) and a wavelet transform-based detec-
tor (WD) (Li et al., 1995).
We report three parameters in these comparisons,
namely, their mean error (m), standard deviation ()
and Sensitivity (Se). We have used these parameters
because they give a complete overview of the overall
behavior of each algorithm, as m shows how close the
detection results using an algorithm are to the manu-
ally annotated results, represents the stability of the
detection and Se (dened in 2) measures the detection
sensitivity (where TP is the number of true detections
and FN is the number of manual annotations that are
not detected by the algorithm).
Table 3: Pathologies detected by diagnosis module.
Problem Possible pathology
Time from D to H
longer than 0.10s
(it could even reach
0.12s).
Block of the His bundle.
Supraventricular rhythm with
aberrant conduction.
Abnormal conduction over acce-
sory pathways.
Ventricular rhythm or pacemaker
rhythm.
The time interval
from A to D is
longer than 0.20s
(long PR interval).
Disorder in the conduction be-
tween atriums and ventricles at
the atrioventricular node level,
His bundle (or its branches) or
Purkinje system.
The time interval
from A to D is
shorter than 0.12s
(short PR interval).
Presence of an anomalous ac-
cesory pathway that produces a
faster conduction or the presence
of a rhythm with origin in the
atrioventricular union, in the left
atrium or in the lower part of
the right atrium. Generally, this
anomaly is due to a ventricular
preexcitation.
The amplitude of J
is negative (T wave
is negative).
Primary alterations of the repo-
larization phase (due to ischemia
or myocardial infarction, suba-
cute pericarditis or myocarditis).
Secondary alterations of the re-
polarization phase (due to alter-
ations of the ventricular repolar-
ization).
The time from D
to F is longer than
0.03s.
Delay in the ventricular activa-
tion time.
QT interval longer
than the values
specied in Table
2.
The ventricular repolarization
has slowed down, which can
be due to acquired or congen-
ital causes. It is related to the
appearance of arrhythmias.
QT interval shorter
than the values
specied in Table
2.
This problem is usually related to
the use of some medicines, hy-
percalcemia or hyperpotassemia.
Se =
TP
TP+FN
100 (2)
Table 4 shows the results obtained from these
comparisons and the accepted standard deviation tol-
erances given by the Common Standards for Electro-
cardiography (CSE) committee. As this table illus-
trates, the values of the mean error are quite low and
can be compared to the values obtained with MMD,
TD and WD. Only in the case of T
on
, the mean er-
ror is very high, but this is not a problem, since T
on
is not relevant for the cardiovascular pathologies tar-
getted by the diagnosis module. Then, the compari-
son between the proposed run-time ECGsignal recon-
struction algorithmand the MMD detector in terms of
IMPLEMENTATION OF AN AUTOMATED ECG-BASED DIAGNOSIS ALGORITHM FOR A WIRELESS BODY
SENSOR PLATAFORM
93
Table 4: Comparative results of our run-time ECG charac-
teristic wave detection and other state-of-the-art ECG de-
tection algorithms.
Technique Parameter Pon P
of f
QRSon QRS
of f
Ton T
of f
Ours Se(%) 92.4 92.4 100 100 96.6 91.7
m(ms) 1.4 14.9 -7.8 8.2 53.6 12.8
(ms) 15.6 13.3 22.6 16.8 21.6 20.9
MMD Se(%) 97.2 94.8 100 100 99.8 99.6
m(ms) 9.0 12.8 3.5 2.4 7.9 8.3
(ms) 9.4 13.2 6.1 10.3 15.8 12.4
TD Se(%) 96.2 97.0 99.9 99.9 98.8 98.9
m(ms) 10.3 -5.7 -7.3 -3.6 23.3 18.7
(ms) 14.1 13.6 10.9 10.7 28.3 29.8
WD Se(%) 89.9 89.9 100 100 99.1 99.2
m(ms) 13.0 5.4 4.5 0.8 -4.8 -8.9
(ms) 12.7 11.9 7.7 8.7 13.5 18.8
CSE (ms) 10.2 12.7 6.5 11.6 - 30.6
is very important, since the proposed algorithm is
based on the original off-line MMD detector. As ex-
pected, the values of the standard deviation achieved
by the MMD method with the complete signal anal-
ysis instead of the limited run-time history are better.
Nonetheless, the degradation of standard deviation in
the case of the new run-time approach is very limited.
In particular, the reasons for this limited degradation
are the following:
The proposed run-time algorithm is optimized to
work in real-time in a sensor node platform with a
simple microcontroller. Floating point operations
are very costly in terms of time for such a sim-
ple microcontroller, and they had to be converted
into integer operations, with the resultant loss of
precision in the ltering and the multiscale mor-
phological transform operations.
Dynamic thresholding based on the histogram of
the signal in the new approach is more difcult
than with the off-line approach, since only a small
windowof data (the last 1.5s of signal) is available
to choose the correct thresholds. Sometimes this
is also a source of errors in the detections.
However, it is very important the fact that results
using the proposed run-time approach are not far from
the original algorithm, especially considering that, in
exchange of a bit of accuracy, it is possible to process
the input data in real-time and the patient can be in-
formed of a possible cardiac pathology in real-time.
Therefore, the time to react and attend the patient in
case a problemoccurs is minimal, and then the conse-
quences of the problem and the probability of the ap-
pearance of a more serious pathology derivated froma
late identication of the problem are greatly reduced.
4.2 Energy Features of Diagnosis
Sensor Node Platform in WBSN
One of the main problems in WBSNs is energy con-
sumption, mainly due to the high sampling frequency
required by these applications. Sensors generate lots
Table 5: Typical ranges of heart beat rate per minute at rest
and practising aerobic and anaerobic exercise for different
kind of adult people.
Sedentary Fit Sportperson
At rest 70 - 90 60 - 80 40 - 60
Aerobic 110 - 130 120 - 140 140 - 160
Anaerobic 130 - 150 140 - 160 160 - 200
Table 6: Energy savings in the radio due to transmission
using the run-time ECG detector instead of data streaming.
Sedentary Fit Sportperson
At rest 91 - 93% 92 - 94% 94 - 96%
Aerobic 87 - 89% 86 - 88% 84 - 86%
Anaerobic 85 - 87% 84 - 86% 80 - 84%
of data that need to be transmitted to the base station,
for that reason, the time that sensors can be in stand-
by mode is shorter than in other kind of applications.
In these networks, the nodes are attached to the body,
therefore batteries must be very small, to allow the
person to do a normal life. These small batteries have
a short lifetime and after some time they need to be
replaced. If the sensors are implanted in the body, the
replacement of the batteries is not feasible. A possible
alternative could be energy scavengers, but in general,
energy scavengers are not able to supply the node con-
tinuously, since the average power generated by one
of these devices is usually much lower than the av-
erage power consumed by the nodes. Normally, they
are coupled to batteries and recharge them, extending
their lifetime.
Analyzing the energy consumption of the different
hardware blocks of a node using typical applications
for WBSNs, we can extract that the main cause of en-
ergy waste is the transmission and reception of data
via radio. The radio is responsible for between 70%
and 90% of the total amount of energy consumed by
the whole sensor node. Then, one of the main chal-
lenges in WBSNs is trying to reduce as much as possi-
ble the energy consumption of the node, therefore we
need to minimize the energy consumed by the radio,
since the radio is the main source of energy waste.
In this section, we consider and analyze three dif-
ferent approaches: data streaming, the use of the pro-
posed run-time ECG signal reconstruction algorithm
and the complete embedded diagnosis system, cou-
pling the diagnosis module introduced in 3.2 to the
basic run-time ECG algorithm.
First, we analyze the most basic application, that
simply performs data streaming, which means that all
the data read by the sensors is transmitted to the base
station without any intermediatte operation. All the
information will be processed off-line in the base sta-
tion. In order to perform a reliable ECG streaming, a
sampling frequency of about 200Hz is required. Tak-
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
94
ing into account that the packet payload is 18 bytes
and the size of each sample is 12 bits, we can con-
clude that 1000 packets are sent in 60s.
Using the run-time ECG signal reconstruction al-
gorithm previously described, the sampled data are
preprocessed before being sent. In this way, a packet
is sent only when a R peak is detected. The sent
packet contains when the R peak has been detected,
the position of the rest of detected points respect the
R peak and the amplitude of P peak, R peak and T
peak. The number of packets sent in 60s using this
algorithm would be equal to the heart beat rate per
minute. Typical values of heart beat rate per minute
are shown in Table 5. Table 6 shows the energy sav-
ings in the radio due to transmission of packets ob-
tained with the use of the proposed algorithm. In
order to verify these theoretical results, some exper-
iments were run in the simulator. The result of one
of these simulations is presented in Figure 3. This
plot shows that the energy consumption due to packet
transmission in the case of the ECG streaming appli-
cation is 50.30mJ in a simulation of 60 real seconds.
If the run-time ECG algorithm is used, the energy
consumption drops to 3.73mJ for an input signal of
76 heart beats per minute, which means a reduction
of 92.6%, as it was predicted in Table 6.
As previously mentioned, a diagnosis module has
been implemented and successfully tested. This mod-
ule evaluates, taking as input the points detected by
the run-time ECG signal reconstruction algorithm, if
the ECG signal of the patient is normal or it presents
any anomalous behavior. When a problem is de-
tected, a packet is sent, containint the code of the di-
agnosed pathology and the points of the wave where
the pathology was detected. In the worst case, when
a problem is detected in every heart beat, the energy
savings with regard to the data streaming case are the
same as if we do not use the diagnosis module. In the
best case, when no pathology is detected because the
ECG signal is normal, no transmission is produced
and therefore the reduction in the number of trans-
missions reaches 100% improvement.
5 CONCLUSIONS
Energy consumption minimization is one of the ma-
jor design challenges to enable WBSN-based solu-
tions for personal healthcare systems. In this paper,
we have demonstrated the feasibility of exploiting
the limited processing and storage resources to de-
ploy a novel real-time automated ECG analysis and
arrhythmia diagnosis algorithm on an actual sensor
node platform. This goal was achieved through judi-
0,00
10,00
20,00
30,00
40,00
50,00
60,00
ECG streaming Run-time ECG detector
E

(
m
J
)
Figure 3: Energy consumption due to packet transmission
for the ECG streaming application and the run-time ECG
signal reconstruction algorithm during a period of 60s.
cious application-level optimizations, taking into ac-
count the underlying hardware architecture and the
real-time constraints of the automated diagnosis ap-
plication. The satisfactory accuracy of both the auto-
mated ECG analysis (ECG characteristics wave de-
tection) and ECG-based arrhythmia diagnosis algo-
rithms was assessed. Finally, we also demonstrated
that the deployment of these advanced algorithms for
ECG analysis and autodiagnosis enables a dramatic
reduction of the energy consumption (between 80%
and 100%, depending on the sensed signal), com-
pared to the conventional WBSN approach that sim-
ply wirelessly streams the sensed data to the central
node. These excellent results show the potential of
embedding advanced processing and intelligence in
sensor nodes for extending the lifetime of WBSNs,
and motivates further research in characterizing the
optimal trade-off between embedded signal process-
ing and wireless communications for these energy-
constrained wireless networks.
REFERENCES
Culler, D. (2006). Tinyos: Operating system design for
wireless sensor networks. Sensors, pages 4149.
Culler, D., Estin, D., and Srivastava, M. (2004). Overview
of sensor networks. Computer, pages 4149.
Daskalov, I. K. and Christov, I. I. (1999). Automatic detec-
tion of the electrocardiogram T-wave end. Med Biol
Eng Comput, 37(3):348353.
Gay, D., Levis, P., von Behren, R., Welsh, M., Brewer,
E., and Culler, D. (2003). The nesC language: A
holistic approach to networked embedded systems. In
PLDI03:Programming Language Design and Imple-
mentation.
Jovanov, E. and et al. (2005). A WBAN system for ambula-
tory monitoring of physical activity and health status:
applications and challenges. In International Confer-
ence of the IEEE Engineering in Medicine and Biol-
ogy Society.
Kemere, C. and et al. (2004). Model-based decoding for
reaching movement for prosthetic systems. In In-
IMPLEMENTATION OF AN AUTOMATED ECG-BASED DIAGNOSIS ALGORITHM FOR A WIRELESS BODY
SENSOR PLATAFORM
95
ternational Conference of the IEEE Engineering in
Medicine and Biology Society.
Khadtare, M. S. and Sahambi, J. (2004). ECG arrhythmia
analysis by multicategory support vector machine. In
AACC, pages 100107.
Laguna, P., Mark, R. G., Goldberger, A. L., and Moody,
G. B. (1997). A database for evaluation of algorithms
for measurement of QT and other waveform intervals
in the ECG. Computers in Cardiology, pages 673
676.
Lazzer, S., Feng, J., Koushanfar, F., and Potkonjak, M.
(2002). System-architectures for sensor networks is-
sues. In IEEE International Conference on Computer
Design (ICCD).
Li, C., Zheng, C., and Tai, C. F. (1995). Detection of ECG
characteristic points using wavelet transforms. IEEE
Transactions on Biomedical Engineering, 42:2129.
LifeShirt. http://www.vivometrics.com.
Lo, B. and Yang, G. (2005). Key technical challenges and
current implementations of body sensor networks. In
Second International Workshop on Wearable and Im-
plantable Body Sensor Networks.
L ofgren, N. and et al. (2007). EEG entropy estimation using
a Markov model of the EEG for sleep stage separation
in human neonates. In International Conference of the
IEEE Engineering in Medicine and Biology Society.
Lutz Bierl, T. I. (2000). MSP430 family mixed-signal mi-
crocontroller application reports. Technical Report TI-
06-2000.
Penders, J., Gyselinckx, B., de Vicq, N., and Torfs, T.
(2007). Body area networks for multi-modal biomed-
ical monitoring. In pHealth.
P erez-G omez, F. (1985). Cardiac Pacing. Editorial Grouz.
Project, C. B. http://www.eecs.harvard.edu/ mdw/proj/ code-
blue/.
Rinc on, F., Paselli, M., Recas, J., Zhao, Q., S anchez-

Elez,
M., Atienza, D., Penders, J., and Micheli, G. D.
(2008). OS-Based Sensor Node Platform and En-
ergy Estimation Model for Health-Care Wireless Sen-
sor Networks. In Design, Automation and Test in Eu-
rope (DATE 08), number ISSN: 1530-1591/05.
Schamroth, L. (1971). The disorders od cardiac rhythm.
Blackwell Scientic Publications.
Semiconductor, N. (2000). nRF2401 transceiver data
sheets. http://www.nordicsemi.com/.
SmartShirt. http://www.sensatex.com.
S ornmo, L. and Laguna, P. (2005). Bioelectrical Signal Pro-
cessing in Cardiac and Neurological Applications. El-
sevier Academic Press.
Sun, Y., Chan, K. L., and Krishnan, S. M. (2002). Ecg signal
conditioning by morphological ltering. Computers in
Biology and Medicine, 32(6):465479.
Sun, Y., Chan, K. L., and Krishnan, S. M. (2005). Charac-
teristic wave detection in ecg signal using morpholog-
ical transform. BMC Cardiovascular Disorders.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
96
DEVELOPMENT OF A MYOELECTRIC CONTROLLER
BASED ON KNEE ANGLE ESTIMATION
Alberto Lpez Delis, J oo Luiz Azevedo de Carvalho, Adson Ferreira da Rocha
Francisco Assis de Oliveira Nascimento and Geovany Arajo Borges
Department of Electrical Engineering, Universidade de Brasilia, Brasilia, DF, Brazil
[email protected], [email protected], [email protected], [email protected], [email protected]
Keywords: Electromyographic signal, Prosthesis control, Microcontrolled bioinstrumentation, Feature extraction,
Dimensionality reduction, Neural network.
Abstract: This paper presents the development of a bioinstrumentation system for the acquisition and pre-processing
of surface electromyographic (SEMG) signals, as well as the proposal of a myoelectric controller for leg
prostheses, using algorithms for feature extraction and classification of myoelectric patterns. The
implemented microcontrolled bioinstrumentation system is capable of recording up to four SEMG channels,
and one electrogoniometer channel. The proposed neural myoelectric controller is capable of predicting the
intended knee joint angle from the measured SEMG singals. The controller is designed in three stages:
feature extraction, using auto-regressive model and amplitude histogram; feature projection, using self
organizing maps; and pattern classification, using a Levenberg-Marquadt neural network. The use of SEMG
signals and additional mechanical information such as that provided by the electrogoniometer may improve
precision in the control of leg prostheses. Preliminary results are presented.
1 INTRODUCTION
The use of microprocessors in myoelectric control
has grown notably, benefitting from the functionality
and low cost of these devices. Microprocessors
provide the ability to employ advanced signal
processing and artificial intelligence (AI) methods as
part of a control system, while easily conforming to
control options, and adjusting to the input
characteristics. They also provide the ability to
implement pattern-recognition-based control
schemes, which increases the variety of control
functions, and improves robustness.
Surface electromyographic (SEMG) signals
provide a non-invasive tool for investigating the
properties of skeletal muscles (Sommerich et al,
2000). The bandwidth of the recorded potentials are
relatively narrow (50-500 Hz), and their amplitude is
low (50 V - 5 mV) (De Luca, 2006). These signals
have been used not only for monitoring muscle
behavior during rehabilitation programs (Monseni-
Bendpei et al, 2000), but also for the mechanical
control of prostheses. In this context, it is important
to be able to correctly predict which movement is
intended by the user. The SEMG signal is very
convenient for such application, because it is non-
invasive, simple to use, and intrinsically related to
the users intention. However, there are other useful
variables, especially those related to proprioception,
for example: the angle of a joint, the position of the
limb, and the force being exerted.
This project is supported under the development
of an active leg prosthesis prototype (Figure 1). The
prosthesis has three degrees of freedom: one for the
knee (sagittal plane), and two movements for the
foot (sagittal and frontal plane). The three degrees of
freedom are associated to the angles
1
,
2
and
3
,
controlled by DC reduction motors.
The prototype will be fixed to the patients upper
leg through a fixing capsule, where the SEMG
sensors will be located. The prosthesis will receive
control commands through digital signal processing,
feature extraction, and pattern classification.
Specifically, for the development of an active leg
prosthesis that also possesses ankle and foot axes, it
is necessary to use other sources of information
besides SEMG (Ferreira et al, 2005). Thus, the use
of myoelectric signals combined with other variables
related to proprioception may improve the reliability
in closed-loop control systems. In addition, the
bioinstrumentation system should be as immune to
noise and interference as possible. This can be
97

achieved by proper board and shielding design, as
well as the use of filters whenever they are
necessary.

Figure 1: Mechanical structure of the prosthesis prototype.
Figure 2 presents the typical main components of
a general myoelectric controller based on pattern
recognition. The SEMG signals are acquired by
surface electrodes placed on the skin over muscle(s)
of the user. The signals originating from the
electrodes are pre-amplified to differentiate the
small signals of interest, and then are amplified,
filtered and digitized. Finally, the information is
transferred to a myoelectric controller (Asghari and
Hu, 2007).

Figure 2: Typical main components of a general
myoelectric controller based on pattern recognition.
In the design and implementation of a myoelectric
controller, the systems precision is essential for a
realistic accomplishment of the users intention. The
precision is an important factor on the development
of multi-sensory controllers, and can be improved by
extracting more information from the muscles state,
and using a classifier that is capable of improving
this information. The controller should be capable of
learning the muscular activation patterns that are
used in natural form for typical movements. It also
needs robustness against the condition variations
during the operation. The response time cannot
create delays that are noticeable to the user.
This article presents a micro-controlled
bioinstrumentation prototype system as part of the
development of an active leg prosthesis structure
that allows the acquisition and processing of
electromyographic signals and other data related to
the articulate movement, specifically the angle of the
knee. The information obtained is processed in order
to obtain appropriate myoelectric patterns for
prosthesis control. Preliminary results on the design
of algorithms for the estimation of the knee angle,
based on patterns recognitions techniques, are
presented.
2 METHODS
The front end stage of the designed
bioinstrumentation system acquires up to four
SEMG channels. The SEMG are measured on a pair
of agonist and antagonist muscles of the leg (Fig. 3).
An electrogoniometer is used to measure the flexion
and extension angles of the knee joint (Fig. 3c).


(a) (b) (c)
Figure 3: Experimental setup. Surface electrodes are
placed over a pair of agonist and antagonist muscle groups
of the leg: (a) vastus intermedius, (b) semitendinosus. An
electrogoniometer is used to measure the flexion and
extension angles of the knee joint (c).
Differential amplifiers, used in the bipolar
configuration, significantly reduce the common
mode interference signals (CMRR > 110 dB). A
band-pass filter between 10 Hz and 500 Hz
frequency range is used. It is composed by a low-
pass filter and high-pass filter with a programmable
gain stage from digital potentiometers, controlled by
the microcontroller. These elements allow the setting
of the SEMG amplitude levels based on the
measurements from the patient. To minimize power
consumption and increase noise immunity,
operational amplifiers with J FET inputs were used.
To obtain adequate myoelectric amplitude, an
overall gain of up to 20000 can be programmed at
the front end (De Luca, 2006).
A second block, micro-controlled and optically
isolated from the front end (Figure 4), centralizes all
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
98

the functions associated with the analog/digital
conversion process, implementing the digital gain
control for the front end amplifiers and synchronized
sampling of SEMG signals. The microcontrollers
belong to the ARM SAM7S64 ATMEL family of
high performance processors, based on 32-bit RISC
architecture with an integrated group of peripherals
that minimize the number of external components.


Figure 4: Block diagram of the bioinstrumentation system.
A 13-bit A/D converter with Serial Peripheral
Interface (SPI) is used for signal sampling, and
allows discriminating small amplitude levels. The
electrogoniometer channel is coupled to the system,
and generates an electric signal corresponding to the
angular position ranging from 30 to 240 degrees.
The sampling frequency of each channel is 1744.25
Hz. Figure 5 presents example data acquired during
an experimental measurement.


Figure 5: Recorded SEMG signals (rectus femoris and
opposite muscles) and angle of the knee joint during a
10-second experiment.
The microcontroller is linked through RS-485
protocol to the central processor of the prosthesis,
which is responsible for coordinating the tasks in the
control process. Besides the RS485 protocol, which
provides the interaction of the block with the central
processor, RS-232C and USB interfaces are
available for the communication with a PC when the
system is configured in stand alone mode (Figure 6).
In this mode, the system allows the visualization of
the state of the experiments during their realization
using a LCD display. The instrumentation system is
designed using low power consumption components,
which increases the systems portability.


Figure 6: Bioinstrumentation module (with accessories)
configured in stand alone mode.
2.1 Adaptive Filter Implementation
The power line interference usually has its first
harmonics (60 Hz, 120 Hz, 180 Hz, and 240 Hz) in a
portion of the spectrum with major SEMG energy
concentration. The use of an analog notch filter may
distort the signal; therefore it should only be used
when really necessary. Generally, the best option is
to use an adaptive notch filter. An embedded
subroutine in the ARM-SAM7S64s core
implements an adaptive notch filter in real time. This
filter maintains a running estimate of the 60 Hz
interference, and the current noise at time t can be
estimated from the previous two noise estimates
(Hamilton, 1996), as shown in equations (1) and (2),

) 2 ( ) ( ) ( nT t e nT t e N t e = (1)

where T is the sample period and N=2cos(2.60.T).
In the filter, the output is generated by subtracting
the estimated noise, e(t), from the input signal, x(t).
The expression presented in equation (2) is used to
implement the filter.
)] ( ) ( [ )] ( ) ( [ ) ( nT t e nT t x t e t x t f = (2)
If f(t)>0, then the estimate was too low, so we
adjust the estimate upward by incrementing d:
d T nT e T nT e + + = + ) ( ) ( (3)
If f(t) <0, the estimate was too high, so the
estimate is decremented:
d T nT e T nT e + = + ) ( ) (
(4)
DEVELOPMENT OF A MYOELECTRIC CONTROLLER BASED ON KNEE ANGLE ESTIMATION
99

As d increases, the filter adapts more rapidly, and
exhibits a broad bandwidth. Similarly, as d
decreases, the filter adapts more slowly, and has a
narrower bandwidth. The selection of the d factor is
empiric, based on test realizations, and its value is
small compared to the dynamic range of the A/D
converter (Hamilton, 1996). Figure 7 shows the
adaptive filtering of a SEMG signal measured on the
rectus femoris muscle.








Figure 7: Adaptive filtering performed on a SEMG signal
contaminated with power-line interference.
2.2 Myoelectric Controller
Presenting the myoelectric signal directly to a
classifier is impractical, because of the
dimensionality and the random characteristics of the
signal. Its necessary that the signal is represented as
a vector with reduced dimensionality, i.e., a feature
vector. The myoelectric controller algorithm
proposes the use of three stages for feature
extraction and pattern classification. The first stage
consists in the mixture of feature vectors from time
domain and spectral analysis. A second stage will
perform the reduction of the feature space, allowing
the increase in the number of SEMG input sensors
without affecting the performance of the control
process. The last stage has the goal of estimating the
knee angle.
2.2.1 Feature Vector Extraction
Given the stochastic nature of the myoelectric
signal, it can be considered as a time series, and
modeled as a linear combination of their past and
present values. The autoregressive model is a
convenient structure for model identification,
especially when the computations of velocity and
response time are important, as in the recognition
and classification of myoelectric patterns. The
autoregressive coefficients provide information
about the muscular contraction. The estimate of the
coefficients is performed using a recursive least
squares (RLS) technique, with a forgetting factor.
This method gives more weight to the most recent
samples at the moment of the iteration cycle. The
parameters are calculated recursively (Ljung, 1987),
as presented in equations (5), (6) and (7):

^
1
^
1
^
] [

+ =
k
T
k k k k k
y L
(5)
k k k
T
k k
k
T
k k k
k k
P
P P
P P

1
1
1 1

(6)
k k
T
k k
k k
k
P
P
L


1
=

(7)

where
k
^

are the vector coefficients that are

estimated at discrete time k;
k
are the regressive
vectors,
k
P is the inverse correlation matrix and
k
L is the gain vector of the filter. The forgetting
factor
k
controls the system response time. The
coefficient estimated at instant k can be interpreted
as a characteristic of the SEMG signal within the
time interval specified by the forgetting factor, and it
is a way of determining the angular displacement
that the patient wants to impose to the prosthesis
(Ferreira et al, 2005). The coefficients form a feature
vector for the pattern classification processes.
Recent research (Hargrove et al, 2008) has
demonstrated that a functional and efficient
configuration consists of a mixture of feature vectors
on the time domain with autoregressive coefficients.
This configuration provides good classification
precision, and is computationally efficient, which
facilitates its implementation in embedded systems.
It is also more robust to the displacement of the
surface electrodes (Hargrove et al, 2008).
This work uses a mixture of the autoregressive
vector with the EMG Histogram method. The EMG
Histogram is an extension of the Zero Crossing and
the Willison amplitude (Zardoshti-Kermai et al,
1995). Myoelectric signals reach relatively higher
levels during the contraction process, compared to
the base line amplitudes. Thus, vectors obtained
from the histogram provide a measure of the
frequency in which the signal reaches each level of
amplitude, associated with different histogram bins.

For the implementation of the histogram, a
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
100

symmetrical interval with respect to base line over
the SEMG register is established and the same is
subdivided into 9 bins. These bins represent
intervals of amplitude in which the SEMG signal is
grouped.
The resultant feature vectors (autoregressive and
histogram) are concatenated, and then used as the
input vector of the feature projection stage.
2.2.2 Feature Projection
A feature projection stage is used to reduce the
dimension of the feature space of the input vectors,
before pattern classification process, using
supervised neural networks. This reduction is
performed using an unsupervised Kohonen
self-organizing map (SOM) neural network. The
groups of vector coefficients obtained from each
SEMG channel using the RLS and histogram
methods are transformed into two-dimensional
vectors. With the reduction in input dimension, the
SOM is able to reduce noise and absorb the large
variations that appear in the original features. In
addition, the SOM can shorten the training time of
the supervised neural network. The unsupervised
SOM can find the winning neuron on a 2-D map to
represent the original pattern. To find the output
neuron (winning node), the following steps are used,
according to the learning rule of the Kohonen
Feature map (Haykin, 1999).
Step 1: Choose random values for the initial
weight vectors W
j
(0).
Step 2: Find the winning neuron y
c
at time step t
(Similarity Matching), by using the minimum-
distance Euclidean criterion:
t j t W t x y
j c
,...., 2 , 1 , ) ( ) ( min arg = =
(8)

Step 3: Update the synaptic weight vectors of all
neurons by using the following update rule:

)] ( ) ( [ ) ( ) ( ) ( ) 1 (
,
t W t x t h t t W t W
j y j j j
c
+ = + (9)

where ) (t is the learning rate, and ) (
,
t h
c
y j
is the
neighbors function centred around the winner.
) (t and ) (
,
t h
c
y j
are varied dynamically during the
learning stage, in order to obtain optimal results.
Step 4: Go back to step 2 until no changes in the
feature map are observed.
The inputs of the Kohonens SOM are features
from each channel, and the output is the 2-D
coordinate (on the x and y axes) on the 2-D
topological net. A 2-D coordinate is a condensed
feature for each channel (Figure 8).
2.2.3 Myoelectric Classification
Multi-layer neural networks have been successfully
applied to many difficult and nonlinear problems in
diverse domains and there is considerable research
on methods to accelerate the convergence time of
the multi-layer feedforward neural network
algorithm (Battiti, 1992 and Charalambous, 1992).
The method used in this paper is the Levenberg-
Marquadt (LM) algorithm (Hagan and Menhaj,
1994), which consists in the use of the nonlinear
least squares algorithm to the batch training of
multi-layer perceptrons. The LM algorithm can be
considered a modification of the Gauss-Newton
method. The key step in the LM algorithm is the
computation of the J acobian matrix. The LM
algorithm is very efficient when training networks
which have up to a few hundred weights. Although
the computational requirements of the LM algorithm
become much higher after each iteration, this is fully
compensated by its higher efficiency. This is
especially true when high precision is required
(Hagan and Menhaj, 1994). Figure 8 presents the
complete block diagram of the myoelectric
controller.

Figure 8: Block diagram of the proposed myoelectric
controller algorithm.
DEVELOPMENT OF A MYOELECTRIC CONTROLLER BASED ON KNEE ANGLE ESTIMATION
101

^
) (i x
3 RESULTS
As a prototype implementation, the training and
validation processes were performed in off-line
mode, and the algorithms described above were
implemented in Matlab. At a later stage, the full
validation of the controller will be the executed from
an embedded system running on a Linux platform.
For this demonstration, SEMG measurements
were captured from a healthy subject using 10 mm
Ag/AgCL surface electrodes placed on a pair of
antagonistic muscles, associated with the flexing and
extension movements of the knee (Figure 3). The
electrodes were arranged in bipolar configuration,
and gel was used to reduce the resistance between
electrodes and skin. The distances between the
centers of the electrodes was 3-5 mm, and the
reference electrode was placed over the lateral
condyle bone. A total of 10 measurements were
performed, divided in two groups of signals -
training and validation - acquired during walks with
different speeds, with duration of 10 seconds.
For training purposes, it is essential to know
information about the input and output, comparing
the dimensional vectors obtained from the SOM
network to the displacement angle measured by the
electrogoniometer sensor. Figure 9 shows the
estimated angle compared to the measured angle
from the electrogoniometer. Although the estimated
angle follows the measurement satisfactorily, the
output of the LM network presents the impulsive
noise (9a), which is canceled using a moving
average recursive filter with a 50-sample window
(9b). This filter keeps the changes levels or slopes
that are present in the angle estimated and present a

Figure 9: Comparison of the estimated knee angle (blue)
to the measured angle from the electrogoniometer (red):
(a) before filtering; (b) after filtering.
delay of (M - 1) / 2 samples, where M is the number
of samples in the average (Smith, 1999). The results
obtained with 50 samples of average were
satisfactory, decreasing the variance and conserving
the changes levels (Figure 9b).
A preliminary comparison was performed
between the proposed algorithm and the methods
proposed by Ferreira et al. (2005). The proposed
algorithm it is an alternative to the latter approach,
which consists in using the AR model for feature
extraction, and a LM multi-layer perceptron neural
network for pattern classification. The evaluation
was based on the classification error, which was
calculated using the following equation:
(%) 100
) (
) ( ) (
1
2
1
2
^

=
=
N
i
N
i
i x
i x i x
error



(10)

where ) (i x and represent the angular values
from the electrogoniometer sensor and the angle
estimated vectors respectively, N is the dimension of
the vectors. Table 1 presents the averages error rate
of classification measured in the group of validation
signals.
The results in Table 1 show that the proposed
algorithm achieved better classification then the
method proposed by Ferreira et al. (2005). This
means that the proposed method is more accurate in
estimating the knee joint angle from the myoelectric
signals.
Table 1: Comparison based the average rate classification.
Ferreira et al. (2005) 8.02% 4.2
Proposal Algorithm 5.86% 1.6
4 CONCLUSIONS
This paper presents the current state of development
of a bioinstrumentation system for active control of
leg prostheses. Features of the system and of the
signal processing algorithm used in the myoelectric
controller were presented. The system allows the
acquisition of SEMG signals with a maximum
amount of signal information and a minimum
amount of contamination from electrical noise. The
results show that the system has great potential for
future developments in leg prostheses. Another
feature of the system is that it works not only as a
part of the prosthesis control, but also in stand alone
mode. Preliminary analysis showed that the
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
102

computational complexity of the proposed algorithm
increases for each iteration during execution of the
LM network. Future work will aim optimize the
code for its execution in real time.
ACKNOWLEDGEMENTS
This work was partially supported by CAPES and
CNPq.
REFERENCES
Asghari, M. O. and Hu, H. 2007. Myoelectric Control
System - A survey, Biomedical Signal Processing and
Control, 2 275-294.
Battiti, R. 1992. First and second order methods for
learning: Between steepest decent and Newtons
method, Neural Computation, 4 141-166.
Charalambous, C. 1992. Conjugate gradient algorithm for
efficient training of artificial neural networks, Circuit,
Devices and Systems, IEE Proceedings, 301-310.
De Luca, 2006. Encyclopedia of Medical Devices and
Instrumentation, J ohn G. Webster Ed., 98-109.
Ferreira, R. U., da Rocha, A. F., Caso, J r C. A., Borges,
G. A., Nascimento, F. A. O., Veneziano, W. H. , 2005.
Reconhecimento de Padres de Sinais de EMG para
Controle de Prtese de Perna, Proc. XI Congresso
Brasileiro de Biomecnica, Brasil.
Hagan, M. T. and Menhaj, M. B. 1994. Training
feedforward networks with the Marquadth Algorithm,
IEEE Trans. Neural Networks 5 989-993.
Hamilton, P. S. 1996. A Comparison of Adaptive and
Nonadaptive Filters for Reduction of Power Line
Interference in the ECG, IEEE Trans. Biomed. Eng.,
43 105-109.
Hargrove, L., Englehart, K. and Hudgins, B. 2008. A
training strategy to reduce classification degradation
due to electrode displacements in pattern recognition
based myoelectric control, Biomedical Signal
Processing and Control, 3 175-180.
Haykin, S. 1999. Neural Networks: A Comprehensive
Foundation, New J ersey: Prentice Hall.
Ljung, L. 1987. Linear System Identification, Prentice-
Hall, Inc: Englewood Cliffs.
Monseni-Bendpei, M.A., Watson, M.J . and Richardson, B.
2000. Application of surface electromyography in the
assessment of low back pain: a literature review.
Phys. Ther. Rev., 2 93-105.
Sommerich, C.M., J oines, S.M., Hermans, V. and Moon,
S.D. 2000. Use of surface electromyography to
estimate neck muscle activity, J. Electromyography
Kinesiology, 6 377-398.
Smith, S. W. 1999. The Scientist and Engineers Guide to
Digital Signal processing, 2(ed.), San Diego:
California Technical Publishing.
Zardoshti-Kermani, M., Wheeler, B. C., Badie, K. and
Hashemi, R. M. 1995. EMG Feature Evaluation for
Movement Control of Upper Extremity Prostheses,
IEEE Trans. on Rehabilitation Eng., vol 3 324-333.
DEVELOPMENT OF A MYOELECTRIC CONTROLLER BASED ON KNEE ANGLE ESTIMATION
103
CATECHOL DETETION USING AN OPTICAL MEMS SENSOR
Peter H. Dykstra, Stephan T. Koev, Reza Ghodssi
MEMS Sensors and Actuators Lab, Department of Electrical and Computer Engineering
Institute for Systems Research (ISR), University of Maryland, College Park, MD 20742, U.S.A.
[email protected], [email protected], [email protected]
Gregory F. Payne
University of Maryland Biotechnology Institute (UMBI), Rockville, MD 20850, U.S.A.
[email protected]
Liangli Yu
Department of Nutrition and Food Science, University of Maryland, College Park, MD 20742, U.S.A.
[email protected]
Keywords: Catechol, Chitosan, Waveguides, Absorbance.
Abstract: We report the successful fabrication and testing of an optical MEMS sensor for the detection of the toxic
phenol, catechol. Catechols presence in food and drinking water posses a health concern due to its harmful
effects on cell respiration. By-products of catechol oxidation have demonstrated increased absorbance
changes in a chitosan film in the UV and near UV range. Our reported sensor takes advantage of this unique
absorbance property to detect catechol by measuring the change in light intensity at 472 nm, thus
eliminating non-specific responses that occur from other oxidized chemicals which do not cause the
absorbance change. Concentrations as low as 1 mM catechol are detected while control experiments
including ascorbic acid display no measurable response.
1 INTRODUCTION
Monitoring the safety of our water supply by using
portable, efficient and inexpensive devices has
become an area of growing interest over the past
decade. The growth of industry has contributed to
the contamination of ground waters with various
potentially dangerous organic chemicals. Catechol is
a synthetic phenol commonly generated in factory
processes and it has been proven to have detrimental
health effects (Starek, 2003). One monitoring
solution is through the use of
microelectromechanical systems (MEMS) to create
on-chip sensors which can give field personnel
accurate and sensitive data on-site regarding the
testing of the water supply.
Recently reported sensors for catechol detection
either use optical or electrochemical measurement
schemes. Optical sensors making use of absorbance
(Abdullah et al., 2006) and fluorescence (Wu et al.,
2004) detection have been reported but the necessity
for bulky, external measuring equipment has
hindered the fabrication of such devices on chip.
Electrochemical sensors typically employ a standard
three electrode system with the working electrode
covered by an immobilization matrix such as
calcium carbonate to entrap oxidizing enzymes
(Shan et al., 2007). Although these devices result in
high sensitivity, the enzyme activity degrades over
time and can be directly affected by certain
conditions such as the pH of the solution making
these sensors difficult to calibrate.
Our reported device utilizes an optical
absorbance technique in a sealed microfluidic
channel for the detection of catechol. In order to
amplify the effect of the detected absorbance by
catechol oxidation, the aminopolysaccharide
chitosan is deposited in the microfluidic channel,
thus intersecting the pathway of the light through the
device. The proposed detection scheme does not
require the use of enzymes yet still remains selective
to phenolic compounds vs other chemical agents.
104

2 MATERIALS AND THEORY
2.1 Catechol
Catechol is a benzenediol, which is a subset of the
phenol class of organic compounds. The chemical
formula of Catechol is C
6
H
4
(OH)
2
.

Following
oxidation, catechol loses its hydrogen atoms from
the hydroxyl groups and becomes an
orthobenzoquinone, more commonly referred to as
an o-quinone. The oxidation of catechol in the cell
creates free radicals which cause damage to vital cell
components such as lipids, proteins and DNA (Sies,
1997).
2.2 Chitosan
Chitosan is a unique material that is well suited for
biological micro-devices due to its ability to be
selectivity deposited and its high density of amine
groups, which provide active bonding sites. The
selective deposition occurs due to chitosans
insolubility above a pH of 6.5. At low pH, chitosan
is protonated and soluble in water. As the pH rises
above 6.5, the amines lose their net positive charge
and the chitosan becomes insoluble. By taking
advantage of this property, one can deposit a film of
chitosan onto a cathode during an electrochemical
reaction. The pH rises with increasing proximity to
the cathode due to the reduction of the hydrogen
ions. The chitosan forms as a thin film or hydrogel
over the cathode surface depending on the amplitude
of the applied current density.
Chitosan has an added advantage over other
polysaccharides because it contains nucleophilic
primary amine (NH
2
) groups at nearly every
repeating sugar residue in its structure. The o-
quinones, which are formed from the oxidized
catechol molecules, bind to the amine groups and
impart physical changes to the film, such as a
change in the optical absorbance (Wu et al., 2005).
2.3 Optics
Understanding the operating principle of the device
requires a more detailed understanding of light
propagation and absorption through a medium. The
absorbance can be related to the concentration of
absorbing species present as demonstrated by the
Beer-Lambert Law:
lc
I
I
A = = ) ( log
0
1
10
(1)

Fluid In Fluid Out
Light In
Light Out
SU-8
Oxide
Deposited Chitosan Film
Silicon
PDMS
Fluid In Fluid Out
Light In
Light Out
SU-8
Oxide
Deposited Chitosan Film
Silicon
PDMS

Figure 1: 3-d Schematic of the packaged device. The
device dimensions are 3.2 x 2.4 x 0.24 cm.
Where is the molar absorptivity, l is the path length
the light takes as it propagates through the absorbing
layer and c is the concentration. In our experiments,
the path length, l, is defined as the thickness of the
deposited chitosan film. O-quinones have been
reported to show a strong absorbance in the UV and
near UV range of the electromagnetic spectrum (Wu
et al., 2005). For this reason, a blue laser source at
472 nm was chosen for the optical measurements
taken with the MEMS sensor.
In our device, on-chip waveguides are patterned
from the polymer SU-8 as shown in the device
schematic (Figure 1). Blue light is coupled in and
out of the waveguides via multimode fibers with a
core diameter of 62.5 m. The cross sectional area
of the polymer waveguides is 100 m by 150 m.
The light propagates through a film of chitosan that
has been deposited onto a transparent, conductive
film of indium tin oxide.
Since the absorbance measurement is purely
related to the optical power being received, it is
important to understand the different optical loss
mechanisms through the device in order to achieve
an acceptable signal to noise ratio. The primary
sources of loss are caused by waveguide losses
which include material absorption and scattering,
divergence of the light crossing the channel, and
roughness associated with the waveguide facets.
Waveguide loss will occur due to the roughness
of the waveguide and any material absorption
through the SU-8. This attenuation was measured to
be 21.15 dB/cm at 472 nm for waveguides with our
dimensions using image processing software. A top-
down digital photograph was taken of fabricated test
waveguides coupled to a blue laser source. The light
intensity down the length of the waveguide was
analyzed to determine the attenuation coefficient.
Divergence of the light as it crosses the
microfluidic channel from one waveguide to the next
is another possible source of loss. The light
capturing efficiency from one waveguide to the next
is found by integrating the surface energy of a beam
CATECHOL DETETION USING AN OPTICAL MEMS SENSOR
105

with a Gaussian profile and is related to the width of
the beam waist as seen in (2).
2
0
2
0
) (
* ) ( 2
w z w
w z w
+
= (2)
Where w
0
and w(z) are the width of the beam
waist before and after it has traversed the channel.
For the dimensions used in our device, the coupling
efficiency of the light as it traverses the channel
from one waveguide to another is near unity ( =
0.99). The high coupling efficiency is a result of
using waveguides with large cross sectional areas.
With these considerations, the most significant
factor that contributes to optical loss in the device is
the roughness of the waveguide facets and sides. The
surface roughness of the waveguides is an inherent
limitation when using lithography and cannot be
completely avoided.
3 DEVICE FABRICATION
3.1 Wafer Level Processes
The MEMS sensor was fabricated using
conventional MEMS patterning techniques. Four
inch silicon wafers (<100>orientation) begin with a
one m thick thermal SiO
2
to act as a bottom
cladding layer for the waveguides. Layers of chrome
(20 nm) and gold (200 nm) are sputtered onto the
oxide coated wafers and patterned to create the
electrodes inside the microfluidic channels.
SU-8 was applied to the wafer and spun first at
600 RPM for 10 seconds followed by 1150 RPM for
30 seconds to achieve a final thickness of 100 m.
The pre-bake was performed on a hotplate at 55
o
C
for 2 hours with a temperature ramp of 5 degrees per
minute.
The SU-8 was exposed to UV light at a dose of
2500 mJ /cm
2
using a mask aligner system, and then
placed back on the hotplate to bake at 55
o
C for 90
minutes with a temperature ramp of 5 degrees per
minute. After post-baking, the SU-8 was developed
for 10 minutes.
A film 200 nm thick of indium tin oxide (ITO)
was deposited on the wafer using RF magnetron
sputtering. The sidewall patterning procedure of the
ITO using AZ9245 photoresist has been described
elsewhere (Powers et al., 2005). The wafers are
cleaned using acetone, methanol, isopropyl alcohol
and DI water, then diced into individual dies for
testing.
3.2 Die-level Processes
Medium molecular weight (~200 kDa) chitosan
flakes were purchased from Sigma Aldrich and
prepared using established methods resulting in a
solution with pH of 5.3 and w/v chitosan of 0.5%
(Yi et al., 2005). The chitosan solution was applied
to the active electrode area using a 100 l syringe
and biased at a current of 0.35 A, which
corresponds to a current density of 4 A/m
2
. This
procedure results in complete chitosan coverage of
the sidewall interface. The current was applied for
10 minutes, after which the device was rinsed
extensively with DI water and blown dry with
nitrogen.
The deposited chitosan films were measured to
be between 5 and 10 microns thick by measuring the
distance the chitosan extends from the sidewall
using an optical microscope. Following deposition
and rinsing, the chips are immersed in a 1 M
solution of NaOH for 5 minutes to neutralize the
chitosan film.
The fluidic channel was sealed using a thick (1
mm) flexible polymer, PDMS. PDMS curing agent
and polymer were purchased from Sigma Aldrich
and mixed in a 1:10 ratio. The solution was cured at
80
o
C for 25 minutes in a box furnace, and then cut
into smaller pieces to fit over the device. To position
the PDMS layer, methanol is applied to one side and
the PDMS is slid into place over the device.
Metal capillaries with OD 400 m and ID 200
m were inserted through the PDMS to create liquid
inlet and outlet ports. Multimode optical fibers were
aligned to the on-chip waveguides through the use of
patterned grooves in the SU-8 resist. Once aligned
by hand under a stereomicroscope, an adhesive was
used to secure the fiber. The final packaged device is
shown in figure 2.

Figure 2: Photograph of packaged device.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
106
Microsensor
472 nm Blue
Laser
XYZ Fiber
Aligner Stage
Syringe Pump
Photodetector
Computer
Software
Input Fiber Output Fiber
Microsensor
472 nm Blue
Laser
XYZ Fiber
Aligner Stage
Syringe Pump
Photodetector
Computer
Software
Input Fiber Output Fiber

Figure 3: Block Diagram of testing setup for the MEMS sensor.
4 TESTING AND RESULTS
Absorbance measurements from catechol oxidation
were taken using blue light (472 nm) coupled
through the MEMS device. Figure 3 displays a
graphic of the testing setup used. Light was
delivered from a free space blue laser (LaserMate,
Pomona, CA) operating in the continuous wave
mode and focused into a multimode fiber using a
manual alignment stage. Output light was coupled
via the optical fiber to a USB linked
spectrophotometer (Ocean Optics, Dunedin, FL)
which facilitated automated data collection using
software. Catechol flakes were purchased from
Sigma Aldrich and dissolved in a 20 mM phosphate
buffer at a pH of 5.3. Liquid was administered using
a GENIE PLUS syringe pump (Kent Scientific,
Torrington, CT) at a flow rate of 100 l/hr, which
translates to a linear flow velocity in the channel of
about 1 mm/s. All of the experiments were
performed at room temperature.
Figure 4 displays the change in measured light
intensity for three different catechol concentrations
after they were oxidized for 10 minutes at a current
density of 4A/m
2
. No decrease in intensity was
observed from the oxidation of the buffer solution or
-10
0
10
20
30
40
50
60
70
100 mM
Catechol
10 mM
Catechol
1 mM
Catechol
Buffer
Solution
100 mM
Ascorbic Acid
D
e
c
r
e
a
s
e

i
n

I
n
t
e
n
s
i
t
y

(
%
)
with chitosan
without chitosan

Figure 4: Measured decrease in light intensity at 472 nm
after 10 minute oxidation at 4 A/m
2
. A clear signal
increase is displayed for all catechol concentrations when
using chitosan in the device.
the common antioxidant, ascorbic acid. Also shown
are measurements for devices without the chitosan
film. The data clearly demonstrates the necessity of
the chitosan in order to detect smaller concentrations
of catechol. Devices with the chitosan film display a
3x and 7x signal increase vs. those without chitosan
for 100 mM and 10 mM catechol concentrations
respectively. At our lowest measured concentration
of 1 mM, no change in the light intensity is detected
for the device without the chitosan film. The
chitosan film amplifies the signal because it
effectively traps the generated o-quinones at the
sensing area of the device through covalent bonding.
It should be noted that these tests without chitosan
require the liquid flow to be stopped in the channel.
Any applied flow rate will cause the o-quinones to
be swept away from the sensing area, disallowing
any accumulation which would cause a detectable
change in absorbance. Measurement error was
calculated based on the observed fluctuations in the
intensity measurement due to either changes in the
laser power or noise effects in the detector.
The increasing absorbance change over the 10
minute period for each sample is displayed in figure
5. The accumulation of the o-quinones is roughly the
same for each concentration of catechol for the first
minute as the rate is primarily reaction limited. Over
-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0 100 200 300 400 500 600
Time (sec)
A
b
s
o
r
b
a
n
c
e
100 mM Catechol
10 mM Catechol
1 mM Catechol
100 mM Ascorbic Acid
Buffer Solution

Figure 5: Change in absorbance over time for each sample.
CATECHOL DETETION USING AN OPTICAL MEMS SENSOR
107

time, the absorbance rate becomes diffusion limited
as the catechol must traverse through the chitosan
film to reach the electrode.
5 CONCLUSIONS
We report the demonstration of on-chip catechol
detection using optical absorbance measurements.
The device successfully demonstrates selective
detection of phenolic (catechol) vs. non-phenolic
(ascorbic acid) compounds without the need of
enzymes. The device also exhibits good
differentiation for a wide range of catechol
concentrations. Since our device uniquely allows for
the collection of time resolved absorbance data,
calibration curves can be fit to different times in
order to achieve more accurate sensing of the
concentration. The analysis performed in this
research can hopefully help to provide the
groundwork for a device used for the detection of
catechol packaged in a low-cost, portable system.
ACKNOWLEDGEMENTS
The authors would like to thank the National
Science Foundation (NSF-EFRI) and the R. W.
Deutsch Foundation for funding this research.
REFERENCES
Abdullah, J ., Ahmad, M., Karuppiah, N., Heng, L. Y. &
Sidek, H. (2006) Immobilization of tyrosinase in
chitosan film for an optical detection of phenol.
Sensors and Actuators B, 114, 604-609.
Powers, M. A., Koev, S. T., Schleunitz, A., Yi, H.,
Hodzic, V., Bentley, W. E., Payne, G. F., Rubloff, G.
W. & Ghodssi, R. (2005) A fabrication platform for
electrically mediated optically active biofunctionalized
sites in BioMEMS. Lab on a Chip, 5, 583-586.
Shan, D., Zhu, M., Han, E., Xue, H. & Cosnier, S. (2007)
Calcium carbonate nanoparticles: A host matrix for the
construction of highly sensitive amperometric phenol
biosensor. Biosensors and Bioelectronics, 23, 648-
654.
Sies, H. (1997) Oxidative stress: oxidants and
antioxidants. Experimental Physiology, 82, 291-295.
Starek, A. (2003) Estrogens and Organochlorine
Xenoestrogens and Breast Cancer Risk. International
Journal of Occupational Medicine and Environmental
Health, 16, 113-124.
Wu, L.-Q., Ghodssi, R., Elabd, Y. A. & Payne, G. F.
(2005) Biomimetic Pattern Transfer. Advanced
Functional Materials, 15, 189-195.
Wu, X. J ., Choi, M. M. F. & Wu, X. M. (2004) An
organic-phase optical phenol biosensor coupling
enzymatic oxidation with chemical reduction. The
Analyst, 129, 1143-1149.
Yi, H., Wu, L.-Q., Bentley, W. E., Ghodssi, R., Rubloff,
G. W., Culver, J . N. & Payne, G. F. (2005)
Biofabrication with chitosan. Biomacromolecules, 6,
2881-2894.

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
108
CHITOSAN FOR MEMS
Demonstration of Micromechanical and Optical Biosensors
Stephan T. Koev, Peter H. Dykstra, Reza Ghodssi
MEMS Sensors and Actuators Lab, Institute for Systems Research, Departement of Electrical and Computer Engineering
University of Maryland, College Park, MD 20742, U.S.A.
[email protected], [email protected], [email protected]
Gary W. Rubloff
Institute for Systems Research, Department of Materials Science and Engineering
Univerisity of Maryland, College Park, MD 20742, U.S.A.
[email protected]
William E. Bentley
Fischell Department of Bioengineering, Univerisity of Maryland, College Park, MD 20742, U.S.A.
[email protected]
Gregory F. Payne
Center for Biosystems Research, University of Mayland Biotechnology Institute, College Park, MD 20742, U.S.A.
[email protected]
Keywords: Chitosan, MEMS, DNA hybridization, Microcantilever, Fluorescence.
Abstract: This paper presents the biological functionalization of MEMS sensors by using the polysaccharide chitosan.
Chitosan is a unique polymer due to its abundance of primary amine groups and its ability to be
electrodeposited with spatial and temporal control. Biomolecules such as DNA and proteins can be attached
to chitosan films by standard coupling chemistries. This biofunctionalization approach was demonstrated for
two different MEMS devices: a microcantilever and an optical sensor. The devices were coated with
chitosan and probe DNA and were used for detecting the hybridization with target DNA. Here, we describe
the design, fabrication procedure, and testing results for both of these biosensors.
1 INTRODUCTION
Functionalizing surfaces for the detection of specific
analytes is a common practice for many biosensor
devices. Electrochemical sensors, for example, often
employ hydrogels which selectively block the
diffusion of particular molecules or entrap enzymes
to enable specific detection (Geng et al., 2008). It is
a greater challenge to functionalize surfaces of
microelectromechanical systems (MEMS) sensors
due to their small size. MEMS sensors hold many
advantages over their macroscale counterparts,
including low cost due to batch fabrication
techniques, high-throughput screening ability, and
small required sample volumes. Several
functionalization schemes have been demonstrated
for MEMS devices (Mizutani, 2007), including the
self-assembly of thiol labeled molecules to gold
surfaces and of silane labeled molecules to silica
surfaces. These techniques require time consuming
laboratory procedures to ensure the integrity of the
biomolecules and offer limited control over their
patterning.
We report the use of an amine rich
polysaccharide, chitosan, to functionalize surfaces in
both mechanical and optical MEMS biosensors.
Chitosan can be selectively electrodeposited on
patterned conductive surfaces, and it has primary
amine groups at every repeating sugar unit of its
polymer structure (Yi et al., 2005). The amine
groups can be used for covalent attachment of
various biomolecules, making chitosan an excellent
interface between microfabricated devices and
biological components. Here, we present the
attachment of probe DNA to chitosan-coated sensors
used for the detection of target DNA.
109

2 CHITOSAN PROPERTIES
Chitosan is derived from the partial deacetylation of
chitin, an abundant material found in nature. At low
pH below a pKa value of 6.3, chitosan is cationic
and soluble in water. However, as the pH rises,
chitosan becomes protonated and insoluble. We take
advantage of chitosans pH dependent solubility to
electrodeposit a chitosan film with spatial and
temporal control in the MEMS sensors. In an
electrochemical reaction, the pH at the cathode
surface will rise due to the reduction of the hydrogen
ions. This rise in pH will cause a film of chitosan to
form over the conductive surface with a rate
dependent on the applied bias (Yi et al., 2005).
3 MICROMECHANICAL
SENSOR
3.1 Design
The cantilever sensor consists of layers of Si
3
N
4

(500nm thick), Au/Cr (100nm thick), and chitosan
with probe DNA (~100nm) on a silicon substrate as
shown in Fig. 1 (Koev et al., 2007). The cantilever
length and width are 100m and 40m respectively.
When exposed to target DNA with a complementary
sequence to the probe, the target binds to the probe
and causes two different effects that can be used for
detection. First, the mass of the cantilever is
increased, causing a drop in its resonant frequency
(dynamic mode detection). Second, the surface
stress is increased, causing the cantilever to deflect
(static mode detection). The cantilever displacement
and resonant frequency are measured with an optical
interferometer (Veeco NT1100). For resonant
frequency measurement, the device is
electrostatically actuated by applying a voltage
between the gold layer and the substrate.
3.2 Fabrication
The cantilever is fabricated on a 4 inch Si wafer with
two contact lithography steps. First, a layer of Si
3
N
4

is deposited by chemical vapor deposition (CVD),
and layers of Cr and Au are deposited by sputtering.
The metal is patterned by wet chemical etching, and
the Si
3
N
4
is patterned by reactive ion etching. The
cantilever is released by etching the Si substrate with
KOH. Chitosan is deposited on the fabricated device
by immersing it in an acidic chitosan solution and
applying a negative voltage. Amine-labeled probe
DNA is attached to the chitosan with glutaraldehyde
as a crosslinker.
Si
3
N
4
Si
Au/Cr Chitosan
SiO
2
Electrode for
chitosan
deposition and
electrostatic
actuation
Displacement
measured by
interferometry
Target DNA Probe DNA
Si
3
N
4
Si
Au/Cr Chitosan
SiO
2
Electrode for
chitosan
deposition and
electrostatic
actuation
Displacement
measured by
interferometry
Target DNA Probe DNA

Figure 1: Cross sectional schematic of microcantilever
sensor.
-0.2
0.8
1.8
2.8
3.8
0 20 40 60 80 100
Position along cantilever (m)
H
e
i
g
h
t


(




m
)





.
Before hybridization
After hybridization
After denaturation

Figure 2: Static response of cantilever to complementary
DNA hybridization and denaturation.
Frequency (kHz)
57 58 59 60 61 62 63 64
A
m
p
l
i
t
u
d
e

(
A
.

U
.
)
0.6
0.8
1.0
1.2
Before hybridization
After hybridization
After denaturation

Figure 3: Dynamic response of cantilever to DNA
hybridization and denaturation.
3.3 Testing and Results
The device is immersed in a complementary target
DNA solution for hybridization and in a urea
solution for denaturation (Koev et al., 2007). After
each step, the device is rinsed with deionized water
and dried. Then, the resonant frequency and the
bending profile of the dried cantilever are measured
by interferometry as described previously. Fig. 2
shows the static response to DNA hybridization, and
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
110

Fig. 3 shows the dynamic response. Both responses
demonstrate that biological recognition occurred and
was transduced to a mechanical signal.
4 OPTICAL BIOSENSOR
4.1 Design
Our design for the biophotonics platform is shown in
Fig. 4 (Powers et al., 2005). A thick photosensitive
polymer, SU-8, is used to define both the fluidic
channel and the optical waveguide on a pyrex
substrate. SU-8 is ideal for this application due to its
chemical inertness, ability to be spun very thick, and
high optical transmission for visible wavelengths.
Pyrex is chosen due to its smaller index of refraction
compared to the SU-8 (1.47 to 1.59). As shown in
the schematic, excitation light is applied through the
same optical fiber which collects the fluorescence
emission through the use of an off-chip bi-
directional coupler. In order to maximize the
collection efficiency at the waveguide facet, the
chitosan is deposited onto the sidewall of the SU-8
channel. A film of indium tin oxide (ITO) is chosen
to facilitate this deposition. Indium tin oxide is
conductive and mostly transparent to visible
wavelengths. The height of the SU-8 is chosen to be
110 m in order to adequately couple the emitted
fluorescence into a multimode fiber with a core
diameter of 62.5 m.
Optical
Path
Electrode (ITO)
Chitosan
Pyrex
Emission
Fluid Channel
Optical Fiber
Fiber Holder
Polymer (SU-8)
Waveguide
Excitation
Waveguide

Figure 4: Schematic of the optical biosensor.
4.2 Fabrication
Gold electrodes are initially patterned on a 4 inch
pyrex wafer using standard lithography techniques.
Prior to the addition of the SU-8, an adhesion
promoter AP300 (Silicon Resources, USA) is spun
on the wafer and baked on a hotplate in order to
improve the adhesion between the SU-8 and the
pyrex. SU-8 is spun to a thickness of 110 m and
patterned to create the waveguides and fluidic
channel. Indium tin oxide is deposited using
magnetron RF sputtering to achieve a final thickness
of 200 nm. The sidewall of ITO is created by
patterning AZ9245 photoresist over the features on
the wafer and etching away the exposed ITO in a 1:1
HCl:DI water solution. In order to functionalize the
ITO surface, the wafer is submerged in a chitosan
solution at a pH of 5.07 (Sigma-Aldrich, USA)
while a voltage bias of 2 VDC is applied to the
electrodes. After 15 minutes, a film of chitosan on
the order of a few microns thick is deposited onto
the ITO electrode. The wafer is rinsed with DI water
to remove any excess residues. An optical fiber is
aligned to the output waveguide by a patterned fiber
clamp structure and is glued in place using UV-
curable epoxy. The index-matching epoxy fills the
gap between the fiber and the waveguide to
eliminate losses due to reflections.
4.3 Testing and Results
The optical biosensor was tested in response to
attachment of probe DNA and to hybridization with
target DNA (Badilita et al., 2007). Amine labeled
fluorescent probe DNA is attached to the chitosan by
glutaraldehyde crosslinking. The optical output was
measured by using an optical spectrum analyzer.
Fig. 5 shows the spectrum of the output signal
collected through the waveguide from probe DNA
labeled with Alexa Fluor 633. The emission signal is
filtered by a band pass filter with cutoff at 660nm.
The signal was measured before and after DNA
attachment. There is a clear increase in the measured
intensity due to the fluorescence signal from the
attached probe DNA.
Spectral Response to DNA Functionalization
0
500
1000
1500
2000
2500
3000
3500
4000
4500
640 660 680 700 720 740 760
Wavelength (nm)
I
n
t
e
n
s
i
t
y

(
A
.

U
.
)
1) DNA Signals
2) Background Signals

Figure 5: Spectra of optical emission signals received both
1) after introduction of fluorescent DNA and 2) when no
fluorophores are present.

CHITOSAN FOR MEMS - Demonstration of Micromechanical and Optical Biosensors
111

For the DNA hybridization experiments, the
chitosan surface functionalized with probe DNA was
exposed to matching or mismatching DNA solutions
for 30 min at room temperature. In order to
emphasize the selectivity, the mismatching DNA
was twice as concentrated (8 M) as the matching
DNA (4 M). Both DNA sequences were labeled at
the 5 end with AlexaFluor 633 fluorophore. The
response was analyzed with a fluorescence
microscope and a spectrum analyzer.
Fig. 6 displays the intensity response to both
matching and mismatching DNA sequences. A
strong emission signal with a peak at about 650 nm
is demonstrated only for the matching DNA
sequence. The images on the right of the figure
display the fluorescence microscope images of the
sensor area after subjection to both the matching and
mismatching DNA. These results were also
demonstrated to be repeatable using the same device
by washing away the target DNA with a 4 M urea
solution at 95
o
C for 30 min and reintroducing the
DNA samples.
0
1000
2000
3000
4000
500 550 600 650 700 750 800
1-Matching DNA
1-Mis-Matching DNA
I
N
T
E
N
S
I
T
Y

[
a
.
u
.
]
WAVELENGTH [nm]
(a)
(b)
(c)

Figure 6: Collected output spectra demonstrating
successful DNA hybridization. Inset: fluorescence
microscope images of device: (a) after probe DNA
attachment (b) after mismatching target DNA (c) after
exposure to matching target DNA.
5 CONCLUSIONS
Chitosan enables a wide range of applications due to
its unique structure and relative abundance in nature.
We have reported the successful design, fabrication
and testing of two distinct MEMS biosensors which
utilize chitosan as a functionalization layer. Our
microcantilever has a deposited film of chitosan on
its surface to facilitate attachment of probe DNA. Its
structure allows for highly sensitive detection due to
mass loading. Our biophotonics sensor utilizes a
novel sidewall pattern of chitosan to improve
fluorescence emission capture into a planar
waveguide. DNA hybridization experiments have
been successfully performed with both devices. By
bridging the world of MEMS with the world of
biology through mechanical or optical detection,
chitosan forms the missing link allowing for more
robust and selective biological sensors.
ACKNOWLEDGEMENTS
This work was funded by the Laboratory for
Physical Sciences, the National Science Foundation,
and the R. W. Deutsch Foundation.
REFERENCES
Badilita, V., Shamim, I., Yi, H., Koev, S. T.,
Gerasopoulos, K. & Ghodssi, R. (2007) Chitosan-
Mediated Biomems Platform For Optical Sensing. The
14th International Conference On Solid-State Sensors,
Actuators, And Microsystems (Transducers '07). Lyon,
France.
Geng, R., Zhao, G., Liu, M. & Li, M. (2008) A Sandwich
Structured Sio2/Cytochrome C/Sio2 On A Boron-
Doped Diamond Film Electrode As An
Electrochemical Nitrite Biosensor. Biomaterials, 29,
2794.
Koev, S. T., Powers, M. A., Yi, H., Wu, L.-Q., Bentley,
W. E., Rubloff, G. W., Payne, G. F. & Ghodssi, R.
(2007) Mechano-Transduction Of Dna Hybridization
And Dopamine Oxidation Through Electrodeposited
Chitosan Network. Lab On A Chip, 7, 103-111.
Mizutani, F. (2007) Biosensors Utilizing Monolayers On
Electrode Surfaces. Sensors And Actuators B, 130, 14-
20.
Powers, M. A., Koev, S. T., Schleunitz, A., Yi, H.,
Hodzic, V., Bentley, W. E., Payne, G. F., Rubloff, G.
W. & Ghodssi, R. (2005) A Fabrication Platform For
Electrically Mediated Optically Active
Biofunctionalized Sites In Biomems. Lab On A Chip,
5, 583-586.
Yi, H., Wu, L.-Q., Bentley, W. E., Ghodssi, R., Rubloff,
G. W., Culver, J . N. & Payne, G. F. (2005)
Biofabrication With Chitosan. Biomacromolecules, 6,
2881-2894.

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
112
DROPLET MANIPULATION ON HIGH ADHESION
SUPERHYDROPHOBIC SURFACES
Daisuke Ishii, Masatusgu Shimomura
WPI-AIMR, Tohoku University,2-1-1 Katahira, Aoba-ku, Sendai 980-8577, Japan
CREST, Japan Science and Technology Agency, 4-1-8 Hon-cho, Kawaguchi 332-0012, Japan
[email protected], [email protected]
Hiroshi Yabu
IMRAM, Tohoku University,2-1-1 Katahira, Aoba-ku, Sendai 980-8577, Japan
CREST, Japan Science and Technology Agency, 4-1-8 Hon-cho, Kawaguchi 332-0012, Japan
[email protected]
Keywords: Superhydrophobicity, Microfluidics, Water droplet, Adhesion, Lotus effect.
Abstract: Micro droplet handling is very important for micro and nano fluidic devices and an intelligent bio interface.
Micro droplet transfer via high adhesion superhydrophobic surfaces has been reported in recent years. We
demonstrated water droplet adhesion controllable superhydrophobic metalpolymer surfaces. Moreover we
achieved micro droplet transfer between superhydrophobic surfaces by using different droplet adhesion
properties. Water micro droplets were transferred from a low-adhesive superhydrophobic surface to a
middle-adhesive superhydrophobic surface via a high-adhesive superhydrophobic surface without any mass
loss. After transferred droplet possessed high water contact angle over 150 degrees. These moving processes
were performed repeatedly. Droplet handlings on the adhesion superhydrophobic surfaces will be expected
for fluidic bio devises with energy saving.
1 INTRODUCTION
Droplet manipulations mimicking behaviours on
plant or insect surfaces such as lotus leaf effect are
now interesting because simple surface structures
provide amazing functionalities. Superhydrophobic
surfaces which have the water contact angle lager
than or near 150 are much paid attention, since its
good water repellent property is used various
applications in coating and electronic technologies
(Zhang, 2008). Many researchers have been reported
to obtain strong water repellent surface such as a
hydrophobic fractal surface (Onda, 1996) and a
nanopin array surface (Hosono, 2005). Recently
several reports were published about water droplet
adhesive superhydrophobic surfaces in mimicry of
geckos feet (Cho, 2008) and roses petals (Feng,
2008). These adhesion properties were caused by
van der Waals force on large real surface area
against small apparent surface area. It was difficult
to control the adhesion forces because the adhesion
was caused by the surface structures.
Herein we demonstrated that a superhydrophobic
metalpolymer (MP) surface with different droplet
adhesion properties. The adhesive superhydrophobic
surfaces were composed of hexagonally ordered
polymer pillar arrays made from a self-organized
honeycomb-patterned polystyrene film (Yabu, 2005)
and metal micro domes deposited by nickel
electroless plating (Ishii, 2008). The dome density
was changed by catalyzation process for electroless
plating.
Droplet manipulations such as a transfer were
achieved by using the MP surface possessing
different water adhesion force. Micro droplet
handling by control of wettability is important for
further understanding of superhydrophobic surfaces
and application in microfluidic bio devices.
2 EXPERIMENTAL
2.1 Preparation Method
The superhydrophobic metalpolymer surface (MP
surface) composed of hydrophobic polymer pillar
113

arrays and metal micro domes was fabricated by
electroless plating for honeycomb-patterned polymer
films and peeling process (See Figure 2).
According to our previous report (Karthaus,
2000), the honeycomb films were prepared by
casting a chloroform solution of 10:1 mixture of
polystyrene (PS; M
w
= 280 000 g mol
-1
) and
synthesized amphiphilic copolymer (CAP; M
w
=
270 000 g mol
-1
) on a glass substrate with
hexagonally condensed water droplet arrays. The
honeycomb film cut to 1 1 cm
2
was soaked in a
catalytic mixture solution of 6.0 ml containing 0.010
mol dm
-3
poly(allylamine hydrochloride) (PAH; M
w

=14 000 g mol
-1
) and 0.010 mol dm
-3
PdCl
2
at 25C.
The catalytic solution was gradually heated to 30C,
45C, and 60C, respectively, and kept for 10 min
under horizontal shaking at 10 rpm. Treated
honeycomb films were immersed in a nickel plating
bath (Ishii, 2006) at 25C containing 0.10 mol dm
-3

Ni(H
2
PO
2
)
2
6H
2
O, 0.19 mol dm
-3
H
3
BO
3
, 0.030 mol
dm
-3
CH
3
COONa and 0.0098 mol dm
-3
(NH
4
)
2
SO
4

without any rinse and drying. Then the plating bath
including the treated honeycomb film was heated to
70C and kept for 2h with no stirring. After rinsing
and drying, a nickel layer was covered on the
honeycomb film. After electroless plating, metallic
faces of the nickel-covered honeycomb films were
adhered on an acryl substrate by an epoxy resin.
After heating at 70C for 2h, a lower half layer of
the nickel-covered honeycomb film was peeled off
from the acryl substrate.
2.2 Physical Measurements
Surface structures of the MP surfaces were observed
by a scanning electron microscope (SEM; Hitachi S-
5200, J apan). A water contact angle (WCA) to 3 mg
water droplet on a surface was measured by contact
angle meter (Kyowa Interface Science DW-300,
J apan). A sliding angle (SA) was measured to tilt the
surfaces with a micro-droplet of 5 mg. Density of
the metal dome which is defined by division of the
*
*
0.8 0.2
N N O O
OH
O
n
*
* n
*
* n
NH
2
HCl
PS
CAP
PAH

Figure 1: Chemical structures used in this report.
70C
(a) Honeycomb film
30C, 45C, 60C
1) Catalyzation
2) Electroless plating
3) Rinse and drying
4) Peel off top layer
(b) Nickel-covered honeycomb film
(c) Metal polymer co-existing surfaces
70C
(a) Honeycomb film
30C, 45C, 60C
1) Catalyzation
2) Electroless plating
3) Rinse and drying
4) Peel off top layer
(b) Nickel-covered honeycomb film
(c) Metal polymer co-existing surfaces

Figure 2: Schematic illustrations of a preparation method
of MP surfaces. SEM images of top (left) and tilt (right)
views of (a) a honeycomb film, (b) a nickel-covered
honeycomb film fabricated by immersion in the catalytic
mixture solution at 45C, and (c) a MP surface fabricated
by peeling off a top layer of the nickel-covered
honeycomb film shown in Figure 2b. (Scale bar: 10 m).
number of metal domes by the number of
honeycomb holes was calculated by means of low-
magnified SEM images.
2.3 Droplet Manipulations
Droplet manipulations such as droplet transfer were
demonstrated by using the MP surfaces with
deferent adhesion properties. Figure 3 shows a
schematic illustration of water droplet transfer. The
5-mg water droplet was prepared on the MP surface
fabricated by using the catalytic mixture solution at
30C. Then the MP surface fabricated by using the
catalytic mixture solution at 60C was closed to the
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
114

water droplet from above and touched a little bit.
The upper MP surface was pulled up slowly from
the lower MP surface. Then the upper MP surface
catching the water droplet was closed and touched to
the other MP surface fabricated by using the
catalytic mixture solution at 45C. Finally the upper
MP surface was pulled up again.

Touch
High-adhesion
Low-adhesion
Middle-adhesion
Pull up
Pull up Touch
Transfer
Touch
High-adhesion
Low-adhesion
Middle-adhesion
Pull up
Pull up Touch
Transfer

Figure 3: Schematic model of a micro-droplet transfer by
using the MP surfaces with different adhesion properties.
3 RESULTS AND DISCUSSION
3.1 Superhydrophobic Metalpolymer
Surface
SEM images of a honeycomb film, a nickel covered
honeycomb film fabricated by immersion in the
catalytic solution at 45C, and a MP surface
fabricated by immersion in the catalytic solution at
45C are inserted in Figure 2. An average diameter
of a honeycomb hole was about 7 m. A nickel-
covered honeycomb film possessing some pores,
which were distributed in the honeycomb holes,
were obtained after electroless plating including
immersion in the catalytic mixture solution. A tilted
SEM image shown in a right column of Figure 2b
clears that the pores were openings of micro mono
vessels. When the temperature of the catalytic
mixture solution was changed low (30C) and high
(60C), the number of the vessels was decreasing
and increasing, respectively. In general, wettability
of all surfaces including the PS honeycomb film is
influenced by a solution temperature, because the
surface tension of all solutions is represented by
function of the solution temperature. This result
indicates that the number of the vessels was
dependent on wettability of the catalytic mixture
solution to the honeycomb film. In the case of
immersion in the catalytic solution at low
temperature, wettability of the honeycomb film was
low, so that the number of the vessels in the nickel-
covered honeycomb film was a few. On the other
hand, in the case of immersion in the one at high
temperature, the number of the vessels was much
because of good wettability to the honeycomb film.
The number of the vessels of the nickel-covered
honeycomb film was easily changed by the catalytic
mixture solution temperature.
The MP surfaces after peeling off the top half of
the nickel-covered honeycomb film were composed
of superhydrophobic PS pillar arrays and
hydrophilic nickel micro-domes as shown in Figure
2c. The nickel micro-dome was reverse side of the
micro mono vessel in the nickel-covered honeycomb
film. This result anticipates that density of the nickel
dome to the honeycomb hole is controlled indirectly
by temperature of the catalytic mixture solution.
Figure 4 shows SEM images of the MP surfaces
having different nickel dome density. The nickel
dome density of the MP surface prepared by
immersion in the catalytic mixture solution at 30C,
45C, and 60C was about 3%, 15%, and 25%,
respectively. The surface properties such as surface
wettability and droplet adhesion properties were
controlled easily, because hydrophilic-hydrophobic
balance was varied by difference of the nickel dome
density (See Table 1).
(a) 30C (b) 45C (c) 60C (a) 30C (b) 45C (c) 60C (a) 30C (b) 45C (c) 60C

Figure 4: SEM images of the superhydrophobic
metalpolymer surfaces fabricated by using the catalytic
mixture solution at (a) 30C, (b) 45C, and (c) 60C. The
black dots indicate the nickel domes. (Scale bar: 100 m).
Table 1: Surface properties of the MP surfaces.
Sample Density WCA SA Adhesion
30C 3% 155 <5 Low
45C 15% 150 30 Middle
60C 25% 145 N/A High
3.2 Micro-droplet Transfer
The water droplet adhesion properties were
measured by water contact angles (WCAs) and
sliding angles (SAs). The MP surface with nickel
dome density of 3% possessed a WCA of 155 and a
SA of less than 5, and was abbreviated as a low-
adhesion MP surface. The MP surface with dome
DROPLET MANIPULATION ON HIGH ADHESION SUPERHYDROPHOBIC SURFACES
115

density of 15% possessed a WCA of 150 and a SA
of 30 (a middle-adhesion MP surface). The MP
surface with dome density of 25% possessed a WCA
of 145 and a not measured SA because the droplet
adhered the surface when turned upside down (a
high-adhesion MP surface). As the dome density
was increasing, the WCA was decreasing and the SA
was increasing, which means that the hydrophilic
nickel domes gave the adhesion behaviors. This
result made clear that the adhesion property was
controlled by the quantity of the nickel dome easily
changed by the catalytic mixture solution
temperature for electroless plating.
155 150
Low-adhesion surface
HIgh-adhesion surface
Middle-adhesion surface
155 150
Low-adhesion surface
HIgh-adhesion surface
Middle-adhesion surface

Figure 5: Droplet transfer of 5.0-mg water droplet from
the low-adhesion MP surface to the middle-adhesion MP
surface via the high-adhesion MP surface.
Water micro droplet transfer was attempted by
using the MP surfaces with different adhesion
properties as shown in Figure 5. A water droplet of 5
mg on the low-adhesion MP surface was carried
with the high-adhesion MP surface by means of
pulling off after slight contact from above. However,
in the case of the middle-adhesion MP surface, a
water droplet did not remove from the low-adhesion
MP surface. On the other hand, the high-adhesion
MP surface was not caught a water droplet on the
middle-adhesion MP surface from above. These
results suggest that the adhesion force of the high-
adhesion MP surface was stronger than that of the
low-adhesion MP surface plus gravity on the water
droplet, and weaker than that of the middle-adhesion
MP surface plus gravity on the water droplet. By
using this difference, the water droplet was
transferred from the low-adhesion MP surface to the
middle-adhesion MP surface via the high-adhesion
MP surface. After transfer, the water droplet on the
middle-adhesion MP surface was sliding when the
surface was tilted at about 30. A droplet transfer
reported in the past is from a superhydrophobic
surface to a hydrophilic surface via an adhesion
superhydrophobic surface (Cho, 2008). Therefore,
after transfer, the water droplet is spreading, and is
unable to be handled. The novel transfer method in
this report remains a droplet shape after transfer, so
that the droplet was handily manipulated again.
These behaviors were useful to microfluidic devices,
bio interfaces, and micro-reactors.
4 CONCLUSIONS
We could fabricate water repellency and adhesion
properties of superhydrophobic metal-polymer
surfaces by electroless plating for self-organized
honeycomb films including immersion in a catalytic
Pd salt and a cationic polymer mixture solution. It
was found that a water contact angle and a water
droplet adhesion property were changed by metal
dome density which was easily controlled by the
temperature of the catalytic mixture solution.
Droplet transfer between superhydrophobic surfaces
was demonstrated by means of using the metal-
polymer surfaces with different adhesion properties.
REFERENCES
Cho, W. K., & Choi, I. S. (2008). Fabrication of hairy
polymeric films inspired by geckos: wetting and high
adhesion properties. Advanced Functional Materials,
18, 10891096.
Feng, L., Zhang, Y., Xi, J ., Zhu, Y., Wang, N., Xia, F., &
J iang, L. (2008). Petal effect: a superhydrophobic state
with high adhesive force. Langmuir, 24, 41144119.
Hosono, E.,Fujihara, S., Honma, I., & Zhou, H. (2005).
Superhydrophobic perpendicular nanopin film by the
bottom-up process. Journal of the American Chemical
Society, 127, 1345813459.
Ishii, D., Yabu, H., Shimomura, M. (2008). Selective
metal deposition in hydrophobic porous cavities of
self-organized honeycomb-patterned polymer films by
all-wet electroless plating. Colloid and Surface; A
313314, 590594.
Ishii, D., Udatsu, M., Iyoda, T., Nagashima, T., Yamada
M., & Nakagawa, M. (2006). Electroless deposition
mechanisms on fibrous hydrogen-bonded molecular
aggregate to fabricate Ni-P hollow microfibers.
Chemistry of Materials. 18, 21522158.
Karthaus, O., Maruyama, N., Cieren, X., Shimomura, M.,
Hasegawa, H., & Hashimoto, T. (2000). Water-
assisted formation of micrometer-size honeycomb
patterns of polymers. Langmuir, 16, 60716076.
Onda, T., Shibuichi, S., Satoh, N., & Tsujii, K. (1996).
Super-water-repellent fractal surface. Langmuir, 12,
21252127.
Yabu, H., Takebayashi, M., Tanaka, M., & Shimomura,
M. (2005). Superhydrophobic and Lipophobic
Properties of Self-Organized Honeycomb and
Pincushion Structures. Langmuir, 21, 32353237.
Zhang, X., Shi, F., Niu, J ., J iang, Y., & Wang, Z. (2008).
Superhydrophobic surfaces: from structural control to
functional application. Journal of Materials Chemistry,
18, 621633.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
116
POLYISOPRENE NANOSTRUCTURED CARBON COMPOSITE
(PNCC) MATERIAL FOR VOLATILE ORGANIC
COMPOUND DETECTION
Gita Sakale, Maris Knite, Valdis Teteris
Institute of Technical Physics, Riga Technical University, Azenes iela 14/24, Riga, Latvia
[email protected], [email protected]
Velta Tupureina
Institute of Polymer Materials, Riga Technical University, Riga, Latvia
[email protected]
Keywords: Polymer-high structured carbon black composite, Volatile organic compound sensors.
Abstract: Our scientific group has chosen the elaboration of conductive composite material, which could be useful for
volatile organic compound detection, as one of research areas. It was found out that the most sensitive
composite material consists of polyisoprene and 10 mass parts of nanostructured carbon black. The electric
resistance changes of the composite in presence of 10 different saturated organic solvents vapour were
measured. Results obtained form our mass-sorption experiments indicated that electrical resistance of the
composite increases because of volatile organic compound (VOC) molecule absorption in the composite
matrix material. We also evaluated VOC compatibility with PNCC matrix material and estimated how the
PNCC resistance change velocity (
R
v ) versus organic solvent vapour molecule diameter varies.
1 INTRODUCTION
Available statistical data evidence about people
exposed to organic solvent daily at their workplaces,
but there are no monitoring devices used to control
VOC concentration in the room. There is also a
necessity to protect environment and equipment
from exposure to VOC. Above mentioned denotes
that there is an urgent need for VOC sensor
materials.
Devices (sorbent polymer films, metal oxide
semiconductors, quartz microbalance (quartz
resonator), laser gas sensors ect.) in the market can
not still be compared with mammals olfactory
system. Scientists are trying to design a prototype of
sensor which in sensing capability of different gases
could be close to mammals olfactory system and
even could be improved for practical applications.
We think that the desirable result of VOC
detection can be achieved by using polymer
carbon black composites as gas sensor materials
because polymer matrix can be selected for direct
volatile compound detection and identifying.
In our opinion a candidate sensor material for gas
sensing should fulfil the following criteria: not
expensive constituent materials; simple production;
fast and reversible response; in-situ control of VOC;
small dimensions of sensing element and ability to
function for a long period of time.
The mechanism how polymer carbon black
composite detects VOC is as fallows: i) the sample
of the composite material is exposed to VOC,
molecules of VOC adsorbe on the surface of
composite and diffuse into the matrix material; ii)
intermolecular chains in the polymer matrix weaken
and form intermolecular chains between VOC
molecules and macromolecules of matrix material;
iii) the matrix material swells; iv) electroconducting
pathways break down because distance between
carbon black aggregates increases; v) at the same
time tunnelling currents between carbon aggregates
in thin layers of matrix decreases and the electrical
resistance of the composite increases.
2 EXPERIMENTAL
Polyisoprene nanostructured carbon composite
material was made by rolling highly structured nano-
size carbon black PRINTEX XE2 (specific surface
117
Figure 1: The change of relative resistance vs. time for the sample, held in saturated vapours of different solvents.
950 m
2
/g, mean diameter of primary particles 25 nm,
DBP absorption 380 ml/100g) and necessary
additional ingredients sulphur and zinc oxide
into a Thick Pale Crepe No 9 Extra polyisoprene
matrix using cold rolls. Then follows vulcanization
process (under 30BAR pressure at 150C for 15
minutes) when not only sulphur crossbonds form but
also possibly chemical bonds between carbon black
nanoparticles and matrix macromolecules form. As
better carbon black particle electrocodutive grid is
connected with polyisoprene macromolecules as
better sensing element to any kind of deformation is
achieved.
Preparation of samples and schematic illustration
of the experimental set-up is described in (Knite,
2007).
2.1 The Change of PNCC Electrical
Resistance Due to VOC Presence
Samples were exposed to vapour for 30 seconds and
then held in open air for relaxation processes to let
go on until the composite material reaches its initial
resistance R
0
. Then the measurement was repeated.
Composite material response to different saturated
organic solvents vapour can be seen in Fig. 1. In the
figure
R
v denotes electrical resistance change
velocity (K/min). It is obvious that the largest
resistance change velocity is obtained when PNCC
is exposed to tetrahydrofuran vapour followed by
benzene, ethylacetate and dichloroethane ect. In the
case of propanol vapour PNCC resistance did not
change at all. The composite can not sense the
presence of propanol vapour. In these experiments
PNCC samples with dimensions 50 x 5 x 1 mm
(length x width x thickness) were used.

Figure 2: Resistance change of PNCC vs. toluene vapour
concentration.
We carried out experiments with PNCC samples
with dimensions 50 x 5 x 0,25 mm also. The
composite capability to sense different toluene
vapour concentrations were tested (Fig. 2.).
Resistance of PNCC increases proportionally to
toluene vapour concentration. Here we would like to
mention that PNCC can sense concentrations which
are equal to or lower then TWA limits (TWA
indicates a time-weighted average concentration for
up to a 8-hour workday during a 40-hour workweek)
(http://www.cdc.gov). In the case of toluene vapour
TWA limit is 200 ppm.
2.2 The Relaxation Process of PNCC
Electrical Resistance
After samples were subjected to saturated organic
solvents vapour electrical resistance relaxation
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
118

process in air were carried out. Typical PNCC
sample with thickness 1mm relaxation process is
shown in Fig.3.
Experimentally obtained data were fitted with
theoretical curve which can be characterised by
equation:
0 1 2
1 2
exp( ) exp( )
t t
R R A A

= + +
(1)
, where
R
0
initial electrical resistance
R transient electrical resistance
A
1
, A
2
constants

1
,
2
relaxation time (h).

From equation (1) we calculated relaxation time
1
and
2
. For all tested organic solvent vapour the
relaxation process can be divided into fast and slow
processes.

Figure 3: Electrical resistance relaxation of specimen in
open air. The sample was exibited to tetrahydrofuran for
30s.
In the case of 250m thick samples electrical
resistance relaxation process is about one hour if the
sample has been exposed to 108,7 ppm toluene
vapour concentration.
2.3 Mass sorption Experiments
Experiments of the change of the sample mass as a
function of time the sample is kept in organic
solvent vapour were used to find out the mechanism
of the change of resistance. For example, specimens
of pure polyisoprene, PNCC, and a pellet of
compressed high-structured carbon black powder
were held in toluene vapour for ~48 hours and a
mass as a function of time (sorption curve) was
recorded (Fig. 4.). Sorption of vapour in carbon filler
in the initial period (first 15 min) is approximately
three times (approximately 1,5 times after up to 104
min) as big as that of pure polyisoprene rubber. Yet
the sorption of vapour in PNCC material in the
initial period is around 1,3 times shorter if compared
with pure polyisoprene rubber, although it seems
that carbon filler should increase the vapour sorption
in the PNCC material. That can be explained, firstly,
by the fact that in the PNCC composition there are
only 10 mass parts of carbon and, secondly, in
processing (mixing and vulcanizing) the PNCC
compounds insulating polyisoprene layers are
formed between the carbon nano-particles. So, even
near the percolation threshold (10 mass parts of
carbon), when electro-conductive channels are
formed, very thin polyisoprene intermediate layers
between the nano-particles still exist and tunnelling
currents may emerge between the channels.

Figure 4: The change of mass for three materials
(polyisoprene rubber, HSCB and its composite) as
functions of the time of exposure to toluene vapour.
Only after approximately 28 h the sorption of vapour
by the PNCC material noticeably exceeds the
sorption by polyisoprene rubber, which indicates
that only after this period of time the sorption in the
filler begins to play substantial role. Consequently,
1) the obtained result indirectly proves the existence
of quantum effect tunnelling currents in the PNCC
material, 2) sorption of vapour in the polyisoprene
matrix plays absolutely uppermost role in effect of
gas sensing (due to sorption of vapour molecules
and swelling of the matrix the distance between
carbon nano-particles increases and tunnelling
currents rapidly decrease) (Knite, Shakale 2007).
POLYISOPRENE NANOSTRUCTURED CARBON COMPOSITE (PNCC) MATERIAL FOR VOLATILE ORGANIC
COMPOUND DETECTION
119

3 DISCUSSION
3.1 Electrical Resistance as a Function
of VOC Molecule Diameter
Previously we have made positron annihilation
lifetime spectroscopy measurements for PNCC in
collaboration with scientists from Monash
University in Australia (Knite, 2006). The purpose
of these experiments was to evaluate free volume
cavities dimensions in the composite material when
it is stretched and in normal (unloaded) state. As it
can be seen from Fig. 5 free volume mean radius for
PNCC containing 10 mass parts of carbon black in
normal state is 3,305. Then diameter of free
volume cavities is 6,61. When molecule diameters
of VOC are smaller than the diameter of free volume
cavities in the composite material then there is no
need for extra activation energy for molecule
diffusion into the matrix material.

Figure 5: Mean radius of free volume cavities in PNCC as
a function of carbon black content in normal and stretched
(l =15 mm) states measured by PALS.
Molecule volume of VOC was calculated using
equation (Askadcli, 1983):

A i
i
N
M k
V
(2)
, where
k Atoms packing density
N
A
Avogadro number, mol
-1

i
i
V - Van der Valse volume of VOC molecule,
which consists of discrete atom volume sum, cm
3

M Molar mass of VOC, g/mol
Density of VOC, g/cm
3
.
After calculation of VOC molecule volume we
accepted approximation that all molecules take the
form of sphere. Finally we calculated diameters of
VOC molecules. We suppose with increasing
molecule diameter of organic solvent vapour the
maximum PNCC electrical resistance change
Rmax/R0 should be decreased.
In Fig. 6 we can see that above mentioned
realizes only partly. Rmax/R0 increases with
increasing molecule diameter starting from acetone
vapour until tetrahydrofuran vapour. In our opinion
this means that R
max
/R
0
is not dependent only from
organic solvent vapour molecule dimensions.

Figure 6: Electrical resistance change of PNCC vs. organic
solvents vapour molecule diameter.
R
max
/R
0
decreases with increasing molecule
diameter of organic solvents vapour in Fig. 6 from
tetrahydrofuran vapour until o- and p-xylene vapour.
In the case of toluene and chlorbenzene an exception
has to be made. Toluene vapour causes greater
R
max
/R
0
change than chlorbenzene vapour, while
molecule diameter of toluene molecule is lager than
that of chlorbenzene.
3.2 VOC Compatibility with Polymer
Matrix Material
The law like dissolves the like is already well
enough known. Polar solvents dissolve in polar
solvents and analogous non-polar dissolve in non-
polar solvents. This also can be attributed to organic
solvents and polymer materials.
To evaluate organic solvents compatibility with
polyisoprene matrix material we compared solvents
dielectric permeability () values to polyisoprene
value. As value of organic solvents is closer to
polyisoprene as the matrix material better absorbs
solvent molecules and then greater
R
v
change is
observed.
From literature data (Brandrup, 1989) we know
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
120

that polyisoprene dielectric permeability value is
2,68. Organic solvent vapour dielectric permeability
values are summarized in Table 1.
Table 1: VOC dielectric permeability values.
Substnace
Type of
solvent
P-xylene 2,2
Non-polar
Benzene 2,3
Toluene 2,3
O-xylene 2,5
Chlorobenzene 5,6
Weakly polar Ethyl acetate 6
Tetrahydrofurna 7,4
Dichloroethane 10,6
Polar Propanol 20,1
Acetone 20,7

From Fig. 7 we can see that PNCC resistance
response to organic solvent vapour is concentrated
into three groups. First let us begin with polar
organic solvents group. As we can see in this group
small electric resistance change is observed while
molecule dimensions of these solvents are the
smallest from all tested organic solvents. It can be
explained by acetone, propanol and dichlorethane
vapour non-compatibility with the composite matrix
material. Thus, we can conclude that resistance
change in this group is more dependent of organic
solvents vapour compatibility with the composite
matrix material.

Figure 7: PNCC resistance change vs. organic solvents
dielectric permeability values.
We will continue discussion with weakly polar
and non-polar solvents group. From Fig. 6 and Fig. 7
we can see that in these groups molecule dimensions
and dielectric permeability compensate each other
and both affect the composites response to organic
solvents vapour. For example, tetrahydrofuran
vapour molecule diameter is the smallest from these
two groups and the largest R
max
/R
0
value is
obtained, while value is 7,4.
Above we compared toluene and chlorbenzene
vapour caused electrical resistance change versus
vapour molecule dimensions. If we look at Fig. 7
then we can see that chlorbenzene belongs to weakly
polar solvents group. Chlorbenzene vapour is less
compatible with the composite matrix material. For
that reason PNCC resistance change in chlorbenzene
vapour is smaller then in toluene vapour. If we
compare mass of molecule for toluene and
chlorbenzene then for toluene it is 15,3010
-23
g and
for chlorbenzene it is 18,69
-23
g. From kinetic-
molecular theory we know that for molecules with
smaller mass the motion velocity is higher. This also
explains why resistance change of PNCC is larger in
the case of toluene vapour.

Figure 8: Relaxation time
1
vs. resistance change in
organic solvents vapour.
Composites organic solvent vapour sensing
mechanism is based on matrix swelling, which
causes increased distance between carbon black
aggregates and resistance change of the composite.
As larger is value R
max
/R
0
as larger is amount of
organic solvent vapour molecules the composite has
absorbed and as longer should be relaxation process.
According to afore said relaxation time
1
and
2

should be proportional to R
max
/R
0
change.
Concerning Fig. 8 we can say that
1
are nearly
proportional to R
max
/R
0
, but there are also some
exceptions which are related to previously described
molecule dimensions and organic solvents
compatibility with polymer matrix material. But
from Fig. 9 we can not find any mathematical
relationship between R
max
/R
0
change and
2
values.
We conclude that fast relaxation processes (
1
)
are determined by vapour molecule diffusion from
swelled matrix interior layers to the surface and
crosslinked macromolecule relaxation. Slow
relaxation processes (
2
) are related to
electroconductive grid relaxation made of carbon
black particles.

POLYISOPRENE NANOSTRUCTURED CARBON COMPOSITE (PNCC) MATERIAL FOR VOLATILE ORGANIC
COMPOUND DETECTION
121


Figure 9: Relaxation time
2
vs. resistance change in
vapour of organic solvents.
4 CONCLUSIONS
In conclusion we can say that for the production of
PNCC we used not expensive and available
materials. The production of the composite is rather
simple if the production procedure is strictly
performed. When PNCC is exposed to VOC the
electrical resistance of the composite increases
rapidly and the effect is reversible. Electric
resistance change velocity is dependent of both VOC
molecule diameters and VOC compatibility with
PNCC matrix material.
Better sensitivity PNCC exhibits when it is
exposed to non-polar or weakly polar solvent
vapour. So, we can declare that PNCC sensing of
VOC is selective.
Further, in our work, we are going to try to
decrease relaxation time of electrical resistance to
couple of minutes or even seconds. We suppose that
PNCC can be used for VOC detection with some
improvements, for example, reducing dimensions of
the composite to optimal size. Else, functioning of
PNCC needs to be evaluated in long time period.
REFERENCES
Knite, M., Klemenok, I., Shakale., G., Teteris, V., and
Zicans, J ., 2007. J . Alloys Comp.
Knite, M., Shakale, G., Klemenoks, I., Ozols, K., Teteris,
V., 2007. J ournal of Physics: Conference Series 93,
012031.
Askadcli, A. A., Matveev, J . I., 1983. Chemical structure
and physical characteristics of polymers, Kimija:
Moscow.
Knite, K., Hill, A.J ., Pas, S.J ., Teteris, V., Zavickis, J .,
2006. Materials Science & Engineering C, V 26.
Brandrup, J ., Immergut, E.H., 1989. Polymer handbook,
J ohn Wiley&Sons: New York, Chichester, Brisbane,
Toronto, Singapore, 3
rd
edition.

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
122
SHORT PAPERS
BIOSIGNALS WITH A FLOOR SENSOR
Near Field Imaging Floor Sensor Measures Impedance Changes in the Torso
Henry Rimminen and Raimo Sepponen
Department of Electronics, Helsinki University of Technology, Otakaari 7B, Espoo, Finland
[email protected], [email protected]
Keywords: Remote sensing, Non-contact measurement, Cardiac monitoring.
Abstract: We analyse biosignals recorded with a near field imaging floor sensor, using a test group of five people.
This human tracking system is capable of non-contact biosignal recording. A time domain integration
method is used to extract periodic cardiac waveforms from the raw signals, while an ECG signal is used as a
trigger for windowing. The most favourable posture for cardiac monitoring is when the test subjects are
lying prone on the sensor floor. A clear correlation between the test subjects can be found when waveforms
in the lying prone or supine postures are compared. The respiration monitoring capability is also discussed.
1 INTRODUCTION
In this study, we analyse the recorded biosignals of a
near field imaging floor sensor (Rimminen, 2008),
using a test group of five people. The main purpose
of this floor sensor is to track people walking on top
of it, but the proposed applications of the floor
sensor, such as care for the elderly and seclusion
monitoring, would most probably benefit from a
vital signs monitoring capability.
Recently, some promising results have been
presented regarding the remote sensing of the human
body. Using electric potential probes with input
impedances up to 10
15
, a clear cardiac signal has
been recorded from a distance of 3 millimetres
(Prance et al., 2000) and later from a distance of one
metre (Harland, 2001). Low impedance charge
amplifiers have also proved their strength in off-
body sensing. This technique has produced good
results from a recording distance of up to 10
millimetres (Smith, 2004). These methods measure
biopotentials produced by the human body, and no
electrical stimulus is generated by the measurement
system.
Some promising experiments have also been
made with electret films, which produce a charge
under pressure. These films produce clear cardiac
and respiratory signals when applied to the chest and
to the chair on which the subject is sitting
(Alamets, 2004). This method measures solely the
fine movements of the body.
Unlike the works discussed above, our method
uses a 90-kHz electrical stimulus to measure
impedance changes in the torso, with no galvanic
contact with the body. This kind of measurment is
often referred as Electric field tomograpy
(Korjenevsky, 2004), and has some applications
using planar electrode arrays (Tuykin, 2007) similar
to us. Instead of a spatial analysis, we sample the
signal from one electrode, and analyse the results in
the time domain. As far as we know, there is no
implementation of a biosignal monitor integrated in
a non-pressure-sensitive floor sensor.
The goal of this study is to analyse cardiac
activity in the biosignals recorded with the near field
imaging floor sensor. We also aim to find out which
the most favourable postures are for this kind of
recording. A secondary goal is to observe respiration
in the recorded signals.
2 MATERIALS AND METHODS
2.1 Measurement System Overview
The positioning system under study measures
impedance changes between conductive elements in
a thick film sensor matrix (Rimminen, 2008). The
rectangular sensor elements have a pitch of
approximately 50 cm x 25 cm. One measurement
unit can cover up to 255 sensor elements and 32 m
2

125

f =90 kHz
MUX
LO
G =70 dB 2 - 100 Hz
LO
ADC
Tuned transformer
Sensor matrix
under floor surface
Biosignal-
channel
ADC
DC-
channel
2.5 kHz
DAC
G =40 dB
Bias
removal
PSD

Figure 1: Block diagram of the measurement system. The
local oscillator (LO) signal is fed to one sensor element at
a time using an array of multiplexers.
of floor area. This means that small and medium size
rooms can be covered entirely. These sensor films
are installed under common dielectric floor
coverings with thicknesses of up to 10 mm. The
plastic floor covering in our test room is 3
millimetres thick. This justifies the use of the term
non-contact measurement.
The amplitude of the recorded biosignal is
assumed to be proportional to impedance changes in
the body. Because of this assumption, we refer to the
recorded samples as Z signals.
The Z recording is performed by feeding
alternating current to a single sensor element and
grounding the others (see Figure 1). The amplitude
of the current is measured using a tuned transformer
and phase sensitive detector (PSD). These structures
perform well in rejecting common mode EMI. After
this, the signal is fed through a band-pass filter and
amplified by 70 decibels. Then we use a 10-bit a/d
converter integrated in a microcontroller. The band-
pass filter has a 50-Hz twin T notch to reject
interference from the mains (National
Semiconductor, 1969) (see Figure 2). The DC
channel in Figure 1 is used for human tracking and
is not discussed in this study.
10
-2
10
-1
10
0
10
1
10
2
10
3
0
20
40
60
Frequency [Hz]
G
a
i
n

[
d
B
]

Figure 2: The simulated frequency response of the
Biosignal channel.

Figure 3: a) Sternal height recording b) Abdominal height
recording. The grey rectangles represent the floor sensor
elements.
2.2 Test Arrangement
We recorded 20 second samples of five different test
subjects in four different lying postures. The
postures were the following: prone, supine, left
lateral, and right lateral. Two recordings were taken
in each posture: from sternal height and from
abdominal height (see Figure 3). We selected a fixed
sensor element for the recording, which was marked
on the floor covering. The total number of 20-second
samples was 40. The test subjects were breathing
normally during the recordings.
An ECG signal between electrode locations V2
and V4 was recorded simultaneously during every
floor sensor recording. The ECG signal was
acquired using wet electrodes and an
instrumentation amplifier with an adjustable band-
pass filter (PRE AMP Model 5113, Princeton
Applied Research, Tennessee, USA).
The test group consisted of five M.Sc. students;
three males and two females. The average weight
was 74 kilograms, and the average age was 23 years.
The use of elderly people as test subjects was not
feasible because of the difficult postures on a hard
floor surface.
2.3 Pulse Integration
To extract a periodic waveform that is characteristic
for each test subject in each posture and recording
point, we integrate the Z signal in the time domain
by windowing it. The windowing is done by using a
simultaneously recorded ECG signal for timing. The
rising edge of the QRS complex is used to trigger
the start and stop of each window, which are then all
summed together. This helps us to find the periodic
cycles in the sometimes noisy Z signal. From now
on, we refer to them as the standard pulse
waveforms.
The analysis of the standard pulse waveforms is
performed by comparing the signal-to-noise ratios
(SNR) of different test subjects and postures. We are
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
126

Figure 4: Z signal and ECG signal recorded from test subject number 4 (male). The posture was prone and the recording
was performed at abdominal height. Cardiac activity is clearly visible. SNR is 8.64.

Figure 5: Z signal and ECG signal recorded from test subject number 3 (male). The posture was prone and the recording
was performed at abdominal height. Both cardiac activity and respiration are clearly visible. SNR is 3.77.
interested in finding out if some postures are more
favourable than others, and if some people produce
stronger cardiac signals than others. The SNR is
calculated by taking the true rms voltage of a
standard pulse waveform and dividing it by the rms
voltage of the base noise. The base noise is
measured by recording a 20-second sample while the
floor is empty. Correlation coefficients are also
calculated in order to find out if different people
produce similar signals in the same postures.
3 RESULTS
3.1 Raw Signals
Some of the raw samples are presented in Figures 4,
and 5. The first shows clear cardiac activity, and the
latter shows clear respiratory activity. The
simultaneously recorded ECG signals are plotted
under the Z signals. Dashed vertical lines represent
the windowing separators triggered by the rising
edges of the QRS complexes in the ECG signals.
3.2 Integrated Pulses
When the raw Z signals are integrated in time
using the pace provided by the ECG signal, we get
the standard pulse waveforms presented in Figure 6.
The dark grey traces represent individual test
subjects, and the light grey fill colour represents the
variation within the whole test group. The black
trace shows the average within the group.
It seems that the clearest peaks in the averaged
pulses are present in the prone posture. Also the
supine posture at abdominal measurement height
produces clear peaks.
Table 1 shows the SNR values of the standard
pulse waveforms in every posture and at every
recording point. The higher the SNR, the stronger
the cardiac signal. Respiration and other artefacts in
the signal do not affect the SNR value because of the
ECG-based windowing. The averaged SNR of each
test subject is shown in the last row, and the average
of each posture is shown in the right-hand column.
The rms base noise voltage used in the SNR
calculations was 14.03 mV.
The correlation coefficients of the postures and
recording points are presented in Table 2. These
values are averages of the correlation coefficients
between the test subjects in a certain posture and at a
certain recording point.
If the correlation is high, people produce similar
signals in the same posture and at the same
recording point. High correlation can also be seen as
a narrow grey area in the corresponding part of
Figure 6.
It seems that the correlation between the test
subjects is over 50 percent in the prone posture. It is
also noteworthy that the correlation reaches 33
percent in the supine posture at abdominal
measurement height.
BIOSIGNALS WITH A FLOOR SENSOR - Near Field Imaging Floor Sensor Measures Impedance Changes in the Torso
127

Figure 6: The standard Z pulse waveforms of all test subjects in each posture and at each recording point. The black trace
is an average of the test group. The grey fill colour represents the variation within the test group. Note that the y-scale in the
prone postures is larger (from -400 mV to +400 mV) than in the other postures (from -150 mV to +150 mV).
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
128
Table 1: Signal-to-noise ratio in different circumstances.
SNR Test subjects



Position
1

2

3

4

5

Average
Prone,
sternal ht.
1.79 1.34 7.92 4.75 3.80 3.92
Prone,
abd. ht.
1.89 0.85 3.77 8.64 3.20 3.67
Supine,
sternal ht.
3.77 1.18 0.96 2.52 1.27 1.94
Supine
abd. ht.
1.68 1.27 2.71 4.11 2.22
2.40
Left lateral,
sternal ht.
2.03 0.58 0.61 2.87 1.93
1.60
Left lateral,
abd. ht.
2.30 1.39 1.71 2.66 2.60 2.13
Right lat.,
sternal ht.
1.38 0.95 2.17 2.92 2.30 1.95
Right lat.,
abd. ht.
1.35 1.24 1.27 2.01 2.46
1.76
Average 2.03 1.10 2.70 3.81 2.47 2.42
Table 2: Correlation between the test subjects.
Posture and recording point Mean correlation
Prone, sternal ht. 0.555
Prone, abdominal ht. 0.500
Supine, sternal ht. 0.043
Supine, abdominal ht. 0.325
Left lateral, sternal ht. 0.093
Left lateral, abdominal ht. 0.265
Right lateral, sternal ht. 0.193
Right lateral, abdominal ht. -0.033
Average 0.243
4 DISCUSSION
The floor sensor produces a cyclic cardiac signal,
which has the same period as a simultaneous ECG
recording. The cardiac signal presumably originates
from changes in the blood concentration in the torso.
The relatively good conductance of blood reduces
the average tissue impedance seen with the sensor
elements. The cardiac waveform recorded with the
floor sensor resembles remotely the Z signal in
impedance cardiography (Patterson, 1989).
However, this similarity is present only in the prone
postures.
The results show that the cardiac signal is clear
when test subjects are lying prone on the sensor
elements. The sternal height recording point seems
to be slightly better than the abdominal height
recording point. The SNR values when they are
lying prone are significantly higher than in other
postures (see grey cells in Table 1). Postures other
than lying prone produce weaker signal amplitudes;
however, they are still mostly above the base noise
(SNR >1). The two females in this test group had
lower SNR values than the males.
The standard pulse waveforms show a clear
correlation between all the test subjects in both of
the prone recording points. This suggests that this
recording method could be reproducible and that
people produce similar waveforms in the prone
posture. Also the supine posture at abdominal
measurement height produces clear correlation (see
grey cells in Table 2).
In addition to the correlation values, the
recording points and postures have other similarities
to each other. Almost every part in Figure 6 has a
notch or a peak at 100 milliseconds and a second
notch/peak at 200 milliseconds. When observing the
notch/peak at 200 milliseconds, we notice that it
points upwards when the person is lying prone and
downwards when they are lying supine. The
behaviour of the 100-millisecond notch/peak is
similar but inverted.
Respiration is most often visible when the
recording is performed at abdominal height, while
the person is lying prone or supine. This suggests
that the respiratory signal originates from the
movements of the diaphragm. Observing Figure 5,
the respiratory activity seems to be similar to the
respiratory activity in bioimpedance signals recorded
with galvanic electrodes (Vuorela, 2008).
The near field imaging floor sensor under study
can not match the cardiac monitoring distance of the
ultra-high input impedance probes (Harland, 2001).
The body of the person must be in direct contact
with the insulating 3 mm floor covering. The bulk
impedance between the sensor elements of the floor
sensor system is approximately 650 at 90 kHz.
The human body is coupled parallel to these
elements and is dominant compared to the bulk
impedance only at very close ranges. This explains
the low cardiac monitoring distance, which most
probably causes the severe limitations in the
postures. However, as far as we know, there is no
implementation of probes with high input impedance
incorporated in a system capable of tracking people.
The fact that we can track a person and measure
vital signs of a fallen person on an arbitrary point in
the monitored space, makes this a novel method.
5 CONCLUSIONS
The floor sensor system under study was designed to
track people, but also shows promise in vital sign
BIOSIGNALS WITH A FLOOR SENSOR - Near Field Imaging Floor Sensor Measures Impedance Changes in the Torso
129

monitoring. Cardiac activity is clearly visible when a
person is lying prone on the floor.
Waveform correlation between all the test
subjects is clear when the recording posture is prone
(at both measurement heights) or supine (at
abdominal height). Respiration is most often visible
when recording is performed at abdominal height
while the person is lying prone or supine.
Combined with automatic fall detection, the vital
sign monitoring capability would be most useful in
many applications. These could include the care of
the elderly and seclusion monitoring. The limitations
in the favourable recording postures prevent the use
of this vital sign monitoring method in crucial
applications.
Our future work includes development of
algorithms for automatic detection of the best vital
sign recording point using the results obtained from
this study. We also aim to publish our existing
method for automatic fall detection.
ACKNOWLEDGEMENTS
This study was supported by the European Union and the
J enny and Antti Wihuri Foundation. The authors are also
grateful to UPM Corporate Venturing for providing the
necessary multi-layer thick film sensor laminates.
REFERENCES
Rimminen, H., Linnavuo, M., and Sepponen, R., 2008.
Human Tracking Using Near Field Imaging. In
Proceedings of the Second International Conference
on Pervasive Health, pp. 148-151. ICST.
Prance, R. J ., Debray, A., Clark, T. D., Prance, H., Nock,
M., Harland, C. J ., and Clippingdale, A. J ., 2000. An
ultra-low-noise electrical-potential probe for human-
body scanning. In Measurement Science and
Technology, Volume 11, Issue 3, pp. 291-297. Institute
of Physics Publishing.
Harland, C. J ., Clark, T. D., and Prance, R. J ., 2001.
Electric potential probes New Directions in The
Remote Sensing of The Human Body. In
Measurement Science and Technology, Volume 13,
Issue 2, pp. 163-169. Institute of Physics Publishing.
Smith, W. J ., and LaCourse, J . R., 2004. Non-Contact
Biopotential Recording from the Human Body Using a
Low-Impedance Charge Amplifier. In Proceedings of
the 30th Annual International Conference on
Bioengineering, pp. 31-32. IEEE.
Alamets, J ., Vrri, A., Koivuluoma, M., and Barna, L.,
2004. The Potential of EMFi Sensors in Heart Activity
Monitoring. In 2nd OpenECG Workshop "Integration
of the ECG into the EHR & Interoperability of ECG
Device Systems, pp. 1.-3.
Korjenevsky, A.V., 2004. Electric field tomography for
contactless imaging of resistivity in biomedical
applications. In Physiological Measurement, Volume
25, Issue 1. pp. 391-401. Institute of Physics
Publishing.
Tuykin, T.S., and Korjenevsky, A.V., 2007. Electric field
tomography system with planar electrode array. In
Proceedings of the 13th International Conference on
Electrical Bioimpedance and the 8th Conference on
Electrical Impedance Tomography, pp. 201-204.
Springer Berlin Heidelberg.
National Semiconductor Corporation, 1969. High Q Notch
Filter. In Linear Brief series, No. 5. National
Semiconductor.
Patterson, R.P., 1989. Fundamentals of impedance
cardiography. In Engineering in Medicine and Biology
Magazine, Volume 8, Issue 1, pp. 35-38. IEEE.
Vuorela, T., Vanhala, J ., Sepp, V.-P., Hyttinen, J ., 2008.
Two portable long-term measurement devices for ECG
and bioimpedance. In Proceedings of the Second
International Conference on Pervasive Health, pp.
169-172. ICST.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
130
VIBRATIONAL SPECTROSCOPY (FTIR-ATR AND FT-RAMAN)
A Rapid and Useful Tool for Phycocolloid Analysis
Leonel Pereira
IMAR-CMA, Institute of Marine Research
Department of Botany, FCTUC, University of Coimbra, P-3004-516 Coimbra, Portugal
[email protected]
Ana M. Amado
Molecular Physical Chemistry Group, Department of Chemistry, FCTUC, University of Coimbra
P-3004-535 Coimbra, Portugal
[email protected]
Paulo J . A. Ribeiro-Claro
Department of Chemistry CICECO, Universty of Aveiro, P-3810-193 Aveiro, Porugal
[email protected]
Fred van de Velde
Wageningen Centre for Food Sciences, P.O. Box 557, 6700 AN Wageningen
NIZO food research, Texture Department, Kernhemseweg 2, P.O. Box 20, 6710 BA Ede, The Netherlands
[email protected]
Keywords: Polysaccharides, Seaweed, Phycocolloid, Carrageenan, FTIR-ATR, FT-Raman.
Abstract: The wide industrial application of phycocolloids (e.g. alginates, agar and carrageenans) is based on their
particular properties to form gels in aqueous solution. Recently, new spectroscopic techniques have
provided more accurate identification of the natural composition of the polysaccharides produced by these
seaweeds. With the combination of two spectroscopic techniques (FTIR-ATR and FT-Raman) it is possible
to identify the principal seaweed colloids in ground seaweed samples as in extracted material. Since the
seaweed samples receive the minimum of handling and treatment (e.g. they are simply dried and ground),
the composition determined represents, as accurately as possible, the native composition of the
phycocolloids.
1 INTRODUCTION
Many seaweeds produce hydrocolloids, associated
with the cell wall and intercellular spaces. Members
of the red algae (Rhodophyta) produce galactans
(e.g. carrageenans and agars) and the brown algae
(Phaeophyceae) produce uronates (alginates).
Carrageenans represent one of the major texturising
ingredients used by the food industry; they are
natural ingredients, which have been used for
decades in food, excepients applications and are
generally regarded as safe (GRAS).
The phycocolloid carrageenin, as it was first
called, was discovered by the British pharmacist
Stanford in 1862 who extracted it from Irish moss
(Chondrus crispus). The name was later changed to
carrageenan so as to comply with the -an' suffix for
the names of polysaccharides. The modern
carrageenan industry dates from the 1940s, receiving
its impetus from the dairy industry where
carrageenan was found to be the ideal stabilizer for
the suspension of cocoa in milk chocolate.
The commercial carrageenans are normally
divided into three main types: kappa, iota and
lambda-carrageenan. Generally, seaweeds do not
produce these idealized and pure carrageenans, but
more likely a range of hybrid structures. Several
other carrageenan repeating units exist: e.g. xi, theta,
131

beta, mu and nu. The precursors (mu and nu), when
exposed to alkali conditions, are modified into kappa
and iota, respectively, through formation of the 3,6-
anhydro-galactose bridge (Rudolph, 2000).
Infrared (IR) spectroscopy was, until recently the
most frequently used vibrational technique for the
study of the chemical composition of phycocolloids.
This technique presents two main advantages: it
requires minute amounts of sample (milligrams),
and it is non-aggressive method with reliable
accuracy (Givernaud-Mouradi, 1992; Pereira et
al.,2003). However, conventional IR spectroscopy
requires laborious procedures to obtain spectra with
a good signal/noise ratio (Chopin and Whalen,
1993). This limitation was overcome with the
development of interferometric IR techniques
(associated with the Fourier transform algorithm),
known as FTIR spectroscopy (Fourier Transform
IR). More recently, Pereira and collaborators had
used a technique of analysis on the basis of FTIR-
ATR (from Attenuated Total Reflectance)
spectroscopy, allowing for the determination of the
composition of the different phycocolloids from
dried ground seaweed, without having to prepare
tablets of KBr (Pereira, 2006; Pereira and Mesquita,
2004).
In contrast to FTIR, the application of
conventional Raman spectroscopy was limited until
recently, due to need for an incident visible laser in
dispersive spectrometers: the visible laser light often
excites electronic transitions in biochemical
samples, which can lead to either sample
degradation or strong background signal from
unwanted laser-induced fluorescence. The use of
Nd:YAG lasers operating at 1064 nm (far from the
visible region) in interferometric spectrometers has
been generalized to decrease the fluorescence level
and avoid sample degradation. The modern FT-
Raman spectrometers have been used to produce
good quality Raman spectra from seaweed samples.
(Matsuhiro, 1996; Pereira et al., 2003).
In this work, a combined FTIR-ATR and FT-
Raman spectroscopy study were used to identify the
colloid produced by one of the principal source of
carrageenans, the red algae Chondrus crispus. Since
the analysis of ground seaweed samples required
minimal treatment (the seaweeds are simply dried
and ground), the determined composition represents,
as accurately as possible, the natural colloid
composition.

2 MATERIALS AND METHODS
2.1 Algal Material and Standard
Samples of Phycocolloids
Specimens of red algae (Rhodophyceae) Chondrus
crispus are collected in the central zone of the
western coast of Portugal (wild specimens) and other
are cultivated in Canada (lambda strain). Standard
samples were obtained from Sigma (type IV, C-
3889) and CP Kelco (pure lambda-carrageenan).
The sample composition and purity were
controlled by NMR.
2.2 Preparation of Ground Seaweed
Samples for FTIR-ATR and
FT-Raman
The seaweed samples were rinsed in distilled
freshwater to eliminate salt and debris from the
thallus surface and dried to constant weight at 60 C.
The dried seaweeds were finely ground in order to
render the samples uniform. For FTIR analysis the
samples do not need additional treatment. The
analysis by FT-Raman requires that these are
without pigmentation. The lack of pigmentation can
be achieved by sun drying (process used by
collectors/producers of commercial seaweeds) or by
pigment elimination in the laboratory by the addition
of acetone/methanol moisture (V/V) or by the
addition of calcium hypochlorite solution (4%, 30/60
s, 4 C) (Pereira, 2004).
2.3 Phycocolloid Extraction
Before phycocolloid extraction, the ground dry
material was rehydrated and pre-treated in acetone
followed by ethanol to eliminate the organosoluble
fraction (Zinoun and Cosson, 1996).
For extraction of the native phycocolloid, the
seaweed samples were placed in distilled water (50
ml/g), pH 7 at 85 C for 3 h. For an alkaline-
extraction (resembling the industrial method), the
samples were placed in a solution (150 mL/g) of
NaOH (1 M) at 80-85 C for 3-4 h according to
Pereira and Mesquita (2004), and neutralised to pH
6-8 with HCl (0.3 M).
The solutions were hot filtered, twice, under
vacuum, through cloth and glass fibre filter. The
extract was evaporated under vacuum to one-third of
the initial volume. The carrageenan was precipitated
by adding the warm solution to twice its volume of
ethanol (96 %).

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
132

Table 1: Carrageenan composition determined by vibrational spectroscopy (FTIR-ATR and FT-Raman) and NMR.
Species/Sample Lifecycle phase Origin
Carrageenan
Yield
1

Alkali
extracted
2
(%mol)
Native
3

Chondrus crispus
Female
gametophyte
Portugal
(Wild)
23.2 % 70.0, 28.0 - (/)
C. crispus Tetrasporophyte
Portugal
(Wild)
36.6 % 100.0
C. crispus Tetrasporophyte
Canada
(Cultivated)
43.6 % 100.0
Sigma - - - 100.0
CP-Kelco - - - 100.0
1
expressed in percentage of dry weight;
2
composition determined by
1
H-NMR;
3
composition determined by FTIR-ATR and FT-Raman
analysis of ground seaweed samples; the carrageenans are identified according to the Greek lettering system; the letters between
parenthesis ( ) correspond to the biological precursors of the carrageenans, present in native samples.

2.4 FTIR-ATR and FT-Raman
Analysis
The FTIR spectra of sample materials (ground dried
seaweed, native and alkali-modified carrageenan)
were recorded on an IFS 55 spectrometer, using a
Golden Gate single reflection diamond ATR system,
with no need for sample preparation. All spectra are
the average of two independent measurements with
128 scans each at a resolution of 2 cm
-1
.
The corresponding FT-Raman spectra were
recorded on a RFS-100 Bruker FT-spectrometer
using a Nd:YAG laser with an excitation wavelength
of 1064 nm. Each spectrum was the average of two
repeated measurements, with 150 scans at a
resolution of 2 cm
-1
.
2.5 NMR Analysis
1
H-NMR spectra were taken on a Bruker AMX600
spectrometer operating at 500.13 MHz at 65 C.
Typically 64 scans were taken with an interpulse
delay of 5 s (T
1
values for the resonance of the
anomeric protons of - and -carrageenan are shorter
than 1.5 s). Sample preparation for the
1
H-NMR
experiments involved dissolving the carrageenan
sample (5 mg mL
-1
) at 80 C in D
2
O containing 1
mM TSP (3-(trimethylsilyl) propionic-2,2,3,3-d
4
acid sodium salt) and 20 mM Na2HPO4, followed
by sonication for three times 1 h in a sonicator bath
(Branson 2510), according Pereira et al. (2007).
Chemical shifts () are referred to internal TSP
standard ( =-0.017 ppm) relative to the IUPAC
recommended standard DSS for
1
H according to van
de Velde et al. (2004). Assignments of the
1
H-NMR
spectra were based on the chemical shift data
summarized by van de Velde et al. (2002, 2004).
3 RESULTS AND DISCUSSION
The main results of the analyses are listed in Table
1. The assignments of the IR spectra were mostly
based on the previous work of Chopin et al. (1999)
and Sartori et al. (1997). The Raman spectra were
assigned based on the IR information and on the
comparison between samples of known composition,
controlled by NMR spectroscopy.
The carrageenans are identified by the Greek
lettering and by the letter code proposed by Knutsen
et al. (1994).
Figure 1 presents four different FT-Raman spectra
(Chondrus crispus, female gametophytes),
corresponding to the different tests of
depigmentation to reduce the background signal
from unwanted laser-induced fluorescence in Raman
VIBRATIONAL SPECTROSCOPY (FTIR-ATR AND FT-RAMAN) - A Rapid and Useful Tool for Phycocolloid Analysis
133

spectroscopy. The spectrum A corresponds to the
ground seaweed treated with a mixture of acetone
and methanol; this presents some fluorescence,
particularly in the spectral area 600-875 cm
-1
and the
peaks are ill-defined. The spectrum B corresponds to
the fresh seaweed treated with calcium hypochlorite
4% (30 s), then dried and milled. The spectrum C
concerns to the ground seaweed (obtained from a
herbarium sample) treated with calcium hypochlorite
4% (30 s). Finally, the spectrum D was obtained
from the native carrageenan (C. crispus water-
extracted) analysis. The last three spectra (B, C, D)
dont present fluorescence, with peaks well-defined
and without background noise.
600 700 800 900 1000 1100 1200 1300 1400
A
B
C
D
Wavenumber (cm
1
)
8
0
5
8
5
0
600 700 800 900 1000 1100 1200 1300 1400 600 700 800 900 1000 1100 1200 1300 1400
A
B
C
D
Wavenumber (cm
1
)
8
0
5
8
5
0

Figure 1: FT-Raman spectrum of ground seaweed (C.
crispus female gametophyte) treated with a mixture of
acetone and methanol (A). FT-Raman spectrum of fresh
seaweed treated with calcium hypochlorite 4% (30 s), then
dried and grounded (B). FT-Raman spectrum of ground
seaweed (obtained from a herbarium sample) treated with
calcium hypochlorite 4% (30 s) (C). FT-Raman spectrum
of C. crispus extracted carrageenan (D).
Since this algae produces a hybrid kappa/iota-
carrageenan the diagnoses peaks referenced in
Figure 1 are the 805 cm
-1
(DA2S),corresponding to
iota-carrageenan and 850 cm
-1
(G4S), corresponding
to kappa-carrageenan.
The FTIR-ATR and FT-Raman spectra of
commercial lambda-carrageenan (Sigma) and
ground C. crispus tetrasporophytes are shown in
Figure 2. These samples present high sulphate
content as indicated by the broad band between 820
and 830 cm
-1
in FTIR-ATR spectra. The C. crispus
and lambda-carrageenan FT-Raman spectra show
two combined peaks between 815 and 830 cm
-1
.
Figure 3 shows the FT-Raman spectra of
commercial sample (CP Kelco) of pure lambda-
carrageenan (A), alkali-extracted carrageenan (B) of
C crispus (tetrasporophyte) and ground seaweed
sample (C) of C. crispus (cultivated strain). The
spectrum of alkali-extracted carrageenan is similar
to that of commercial pure lambda-carrageenan. The
high sulphate content, typical of the lambda variant,
is patent in the spectra, with a presence of two
combined peaks at 815 cm
-1
(G/D6S) and 830 cm
-1

(G/D2S).
A
C
D
B
8
3
0
8
2
0
500 600 700 800 900 1000 1100 1200 1300 1400 1500
Wavenumber (cm
-1
)
A
C
D
B
8
3
0
8
2
0
500 600 700 800 900 1000 1100 1200 1300 1400 1500 500 600 700 800 900 1000 1100 1200 1300 1400 1500
Wavenumber (cm
-1
)

Figure 2: FT-Raman (A) and FTIR-ATR (B) spectra of
commercial lambda-carrageenan (Sigma); FT-Raman (C)
and FTIR-ATR (D) spectra of ground seaweed sample
(Chondrus crispus, tetrasporophyte).
4 CONCLUSIONS
The present work confirms the usefulness of FTIR
spectroscopy in the comparative study of
carrageenan types. However, it also shows that the
complementary use of IR and Raman spectroscopy
provides relevant additional information, allowing a
better interpretation of the vibrational spectra and a
more accurate identification of diverse colloids and
variants. In fact, due to the different selection rules,
bands of weak intensity or even absent in the IR
spectra may appear as sharp and intense bands in the
Raman spectra. This is particularly evident, for
instance, in the spectra of different fractions
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
134

belonging to the family of lambda-carrageenan
(Pereira et al., 2003) and the biological precursors of
kappa and iota-carrageenan (mu and nu,
respectively) (Pereira and Mesquita, 2004).
600 700 800 900 1000 1100 1200 1300 1400
A
8
1
5
8
3
0
Wavenumber (cm
-1
)
B
C
600 700 800 900 1000 1100 1200 1300 1400
A
8
1
5
8
3
0
Wavenumber (cm
-1
)
B
C

Figure 3: FT-Raman spectra: (A) commercial sample of
pure lambda-carrageenan (CP Kelco); Chondrus crispus
(tetrasporophyte) alkali-extracted carrageenan; (C) ground
seaweed sample (C. crispus, cultivated strain).
With the combination of these two spectroscopic
techniques (ATR-FTIR and FT-Raman), it is now
possible the rapid and reliable identification of all
major types of carrageenan, both extracted
carrageenan and grounded material samples. The
joint application of these spectroscopic techniques
has as main advantages:
a) It is a quick and simple methodology in
phycocolloid analysis. Only need few minutes,
instead of several days needed for the extraction of
colloids;
b) Requires small quantities of algal material (a
few grams of weight fresh or milligrams of dry
weight), allowing the analysis of herbarium samples,
even of algae or portions of algae with small size;
c) Since the seaweeds are subject to a process of
minimal manipulation and treatment (they are
simply dried and ground), the determined
composition represents, as accurately as possible,
the natural composition of phycocolloid produced.
Since the vibrational spectrometers are now
standard equipment in many Laboratories, the
techniques described in this work are useful for the
implementation of strategies of sustainable seaweed
harvest, the evaluation of the natural seaweed
composition with industrial potential, the evaluation
and control of the quality of the different batches of
algal material harvested and/or cultivated. These
spectroscopic techniques are also useful to analyze
the composition of pharmaceutical and cosmetic
excepients.
ACKNOWLEDGEMENTS
The authors acknowledge financial support from the
Portuguese Foundation for Science and Technology
IMAR-CMA (Institute of Marine Research),
Unidade de Qumica-Fsica Molecular and
Laboratrio Associado CICECO.
REFERENCES
Chopin, T., Kerin, B. F. and Mazerolle, R. (1999).
Phycocolloid chemistry as taxonomic indicator of
phylogeny in the Gigartinales, Rhodophyceae: A
review and current developments using Fourier
transform infrared diffuse reflectance spectroscopy.
Phycological Research, 47, 167-188.
Chopin, T. and Whalen, E. (1993). A new and rapid
method for carrageenan identification by ft-ir diffuse-
reflectance spectroscopy directly on dried, ground
algal material. Carbohydrate Research, 246, 51-59.
Givernaud-Mouradi, A. (1992). Recherches biologiques et
biochimiques pour la production d'agarose chez
Gelidium latifolium (Rhodophyces, Glidiales). UFR
des Sciences de la Vie et du Comportement (p. 351).
Caen: University de Caen.
Knutsen S.H., Myslabodski, D.E.,Larsen, B. and Usov,
A.I. (1994). A modified system of nomenclature for
red algal galactans. Botanica Mararina, 37(2), 163-
169.
Matsuhiro, B. (1996). Vibrational spectroscopy of
seaweed galactans. Hydrobiologia, 327, 481-489.
Pereira, L., van de Velde, F. and Mesquita, J . F. (2007).
Cytochemical studies on underutilized
carrageenophytes (Gigartinales, Rhodophyta).
International Journal of Biological and Biomedical
Engineering, 1(1), 1-5.
Pereira, L. (2006). Identification of phycocolloids by
vibrational spectroscopy. In A. T. Critchley, M. Ohno,
and D. B. Largo (Eds.), World Seaweed Resources -
An authoritative reference system: ETI Information
Services Ltd.
Pereira, L. and Mesquita, J . F. (2004). Population studies
and carrageenan properties of Chondracanthus teedei
var. lusitanicus (Gigartinaceae, Rhodophyta). Journal
of Applied Phycology, 16(5), 369-383.
Pereira, L. (2004). Estudos em macroalgas carragenfitas
(Gigartinales, Rhodophyceae) da costa portuguesa -
aspectos ecolgicos, bioqumicos e citolgicos. PhD
VIBRATIONAL SPECTROSCOPY (FTIR-ATR AND FT-RAMAN) - A Rapid and Useful Tool for Phycocolloid Analysis
135

Thesis. Departamento de Botnica - FCTUC (p. 293).
Coimbra: Universidade de Coimbra.
Pereira, L., Sousa, A., Coelho, H., Amado, A. M. and
Ribeiro-Claro, P. J . A. (2003). Use of FTIR, FT-
Raman and
13
C-NMR spectroscopy for identification
of some seaweed phycocolloids. Biomolecular
Engineering, 20(4-6), 223-228.
Rudolph, B. (2000). Seaweed products: red algae of
economic significance. In R. E. Martin (Ed.), Marine
& freshwater products handbook (pp. 515-529).
Lancaster, PA: Technomic Pub. Co.
Sartori, C., Finch, D. S., Ralph, B. and Gilding, K. (1997).
Determination of the cation content of alginate thin
films by FTir spectroscopy. Polymer, 38(1), 43-51.
van de Velde, F. and de Ruiter, G. A. (2002).
Carrageenan. In E. J . Vandamme, S. D. Baets, & A.
Steinbuchel (Eds.), Biopolymers v. 6.
Polysaccharides II, polysaccharides from eukaryotes
(pp. 245-274). Weinheim; Chichester: Wiley-VCH.
van de Velde, F., Pereira, L. and Rollema, H. S. (2004).
The revised NMR chemical shift data of carrageenans.
Carbohydrate Research, 339(13), 2309-2313.
Zinoun, M. and Cosson, J . (1996). Seasonal variation in
growth and carrageenan content of Calliblepharis
jubata (Rhodophyceae, Gigartinales) from the
Normandy coast, France. Journal of Applied
Phycology, 8(1), 29-34.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
136
BIODEVICES BASED ON SHAPE-MEMORY POLYMERS
Current Capabilities and Challenges
Andrs Daz Lantada, Pilar Lafont Morgado, Hctor Lorenzo-Yustos, Vicente Lorenzo Esteban
J ulio Muoz-Garca, J os Luis Muoz Sanz, J avier Echavarri Otero and J uan Manuel Munoz-Guijosa
Grupo de Investigacin en Ingeniera de Mquinas E.T.S.I. Industriales Universidad Politcnica de Madrid
Grupo POLmeros, Caracterizacin y Aplicaciones (POLCA) Universidad Politcnica de Madrid
C/ Jos Gutirrez Abascal, n 2. 28006 Madrid, Spain. (+34) 913364217
[email protected]
Keywords: Shape-memory polymers (SMPs), Biodevices, Synthesis, Characterization, Processing, Complete
development.
Abstract: Shape-memory polymers are active materials with thermomechanical coupling and a high capability to
recover from high levels of deformation, which, combined with their low cost and density has favoured the
appearance of numerous applications, particularly those linked to the Medical Industry. In many cases, these
materials are of medical standard, which increases the chances of obtaining biocompatible devices.
In the last decade enormous progress has been made on many areas, regarding these materials, such as
synthesis, characterization, activation and others, aimed at improving their applicability. However, various
spheres of action still remain that require more in depth research to promote the production start-up of
various shape-memory polymer-based devices that have had laboratory validation.
This work sets out the potential these materials provide for developing biodevices and the main advances
achieved. Also shown are various medical devices just being developed, as well current study needs and
trends.
1 INTRODUCTION TO
SHAPE-MEMORY POLYMERS
(SMPS)
Shape-memory polymers (SMPs) are materials that
show a mechanical response to external stimuli,
usually to changes of temperature. When these
materials are heated above their activation
temperature, there is a radical change from rigid
polymer to an elastic state that will allow
deformations of up to 400%. If the material is cooled
down after manipulation it retains the shape
imposed; the said structure is frozen and returns to
a rigid but non-equilibrium state. If the material is
again heated above its vitreous transition
temperature or activation temperature it recovers
its initial non-deformed state. The cycle can be
repeated numerous times without degrading the
polymer and most suppliers can formulate different
materials with activation temperatures ranging from
30 C to 260 C, depending on the application
required. Of
They are therefore active materials that present
thermomechanical coupling and a high capability for
recovery from deformation, (much greater than that
shown by shape-memory metal alloys), which
combined with their lower density and cost has
favoured the appearance of numerous applications.
Their properties permit applications for
manufacturing sensing devices or actuators,
especially for the aeronautics, automobile and
medical industry.
2 POTENTIAL FOR BIODEVICES
2.1 Some Advantages
As polymers, SMPs can be easily conformed into
different complex shapes and their properties
designed or adapted to specific applications and can
also be integrated with other microelectromechanical
sensors (MEMS) to produce intelligent
bioactuators and biodevices.
137

Compared to other shape-memory alloys used in
numerous medical devices, SMPs show a far greater
capability for changing their geometry during
activation.
They are also much cheaper to synthesise and
their large scale mass production costs are reduced
by using technologies such as injection moulding.
All this makes them very versatile active
materials with a high potential for industry, provided
they overcome some of the limitations set out in the
following sections.
2.2 Proposed Devices
Bellow are explained some specific proposals for
developing medical devices based on the use of
shape-memory polymers, most of which have
undergone in vitro laboratory testing. After
undergoing in vitro testing and meeting the
requirements for official approval, in some cases
their commercialisation is subject to their attaining
the goals described at the end of this paper.
Self-expanding Stents. Like the stent designed by
Boston Scientific Corporation using the polymer
from CRG Industries known as Veriflex under its
trade-name, to treat the problems arising when the
arteries become narrow or obstructed and also for
removing obstructions from other tube-shaped
body parts, like the uretheres and the bronchial
tubes. The stent is inserted in its temporary form
(reduced) and the bodys own heat causes it to dilate
and become attached to the artery.
They may be used to replace stents based on
shape-memory alloys such as Nitinol, once the
appropriate biocompatibility studies have been
carried out. Developments of self-expanding stents
have also been carried out by using injected
polyurethane (Wache, 2003).
Intelligent Sutures. Like those developed at the
Forschungszentrum in Karlsruhe by Lendleins team
and at the M.I.T. by Langers team, which have a
temporary linear shape and a permanent shape in the
form of a knot, with the change in geometry being
activated by the bodys own temperature. They have
numerous applications in minimally invasive surgery
and, as they are biodegradable, they have additional
advantages over the use of textile sutures and metal
clips (Lendlein, Kelch, Langer, 2002, 2005).
Thrombectomy Devices. With the recent discovery
that the thermal effect of shape-memory can be
activated by a laser, part of whose energy is
absorbed by the polymer, devices with special
geometries have been proposed for removing clots
(Wilson, 2006). The polymer is shaped in a spiral
mould and then heated and stretched to give it its
temporary shape. When the laser light passes
through the polymer, the shape-memory effect is
activated and the device recovers its spiral shape
trapping the clot which can then be removed.
Active Catheters. By using shape-memory
polymers for the distal point of catheters together
with a subsequent activation of the memory effect
by laser light or body heat, different drugs and
antitumoral agents can be released. The presence of
an active catheter point can also help reach zones
that are difficult to access in minimally invasive
surgery tasks (Yackaki, 2007).
Drug Release Devices. If biodegradable shape-
memory polymers are used for implantable medical
devices, drug supply reservoirs can be incorporated
into the device itself. After implant, the polymer
begins to be absorbed by the organism and the drug
is released. Patents have been taken out in this
respect for self-expanding coronary stents or intra-
urethral stents (Boston Scientific Co. and Surmodics
Inc.). The possibility of obtaining temporary
geometries with micro-reservoirs for drug storage
has also been studied. The drugs would then be
released on activation of the shape-memory effect by
body heat (Gall, 2004).
Active Annuloplasty Rings. Aimed at obtaining a
progressive postoperative treatment of mitral
insufficiency, they are based on the use of a
polymeric ring with heating resistances distributed
around the inside to activate the shape-memory
effect by Joule effect. This activation must allow the
cross section of the mitral ring to be gradually
reduced and, therefore, the mitral insufficiency
improved.
Figure 1 shows a schematic design of such a
device proposed by our group at Universidad
Politcnica de Madrid. Prototypes of these rings,
with different geometries and materials, have been
developed and tested in vitro in pigs hearts (Daz
Lantada, Lafont Morgado, 2008).
The different devices explained will provide
considerable therapeutic benefits compared to
conventional devices, due to their capability to act
inside the body thanks to the use of shape-memory
polymers.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
138


Figure 1: Active annuloplasty ring design. SMP with
internal heating resistances. Biodevices 2008.
3 CURRENT CAPABILITIES
Set out below are some of the main advances
achieved in the last decade concerning shape-
memory polymers and the technologies associated
with their use.
3.1 Synthesis
In recent years particular emphasis has been placed
on obtaining new formulations of polymers with
shape-memory properties, by changing the
proportions of monomers, including additives, and
inducing multiple crosslinkings and working on
previously known formulations whose shape-
memory properties have been boosted. (Lendlein,
2002, Liu, 2007).
Different prestigious laboratories have conducted
exhaustive work on synthesis and subsequent
classification in accordance with the molecular
structure of the polymers.
The recent synthesis of polymers capable of
remembering two pre-set shapes using two
programming stages (triple shape effect) has brought
new possibilities for future medical devices, due to
the fact that two postoperative changes can be made
to the geometry (Bellin, 2006).
In spite of the numerous formulations for the
shape-memory polymers that have been synthesised
recently, the main problems for obtaining
commercial biodevices based on these materials are:
the small number of commercial formulations, the
toxicity of many of them and problems regarding
thermomechanical properties, as will be explained in
the following sections.
3.2 Characterization
During the last decade most experiments linked to
characterizing different shape-memory polymer
properties have attempted to compare the
thermomechanical response of different
formulations.
At the Langley Research Centre the results of
tests using thermomechanical analysers (TMA) has
been compared with those obtained by using
differential scan calorimeters (DSC) to obtain
precisely the vitreous transition temperature in
shape-memory polymers (Volk, 2005). This research
also explains deformation recovery tests conduced
by heating under constant deformation and under
constant stress, for which MTS Alliance RT1
traction machines and a heating chamber are used.
Three-point bending tests have also been used in
heated chambers in order to evaluate the geometric
recovery capability of these materials subjected to
different levels of stress and deformation (Lendlein,
2002, Tobushi, 2008).
Dynamic mechanical analyses (DMTA) have
been used basically to evaluate the elastic modulus
of these materials according to temperature. They
also enable the vitreous transition temperatures of
the materials to be found (Mather, 2002, Liu, 2003,
2006, Huang, 2006, Yakacki, 2007). This is a
supplementary technique to DSC tests (which are
usually used for the study of vitreous transitions,
polymorphisms, crystallisations and aging).
All these experiments and many others have
helped to provide basic knowledge concerning the
thermomechanical behaviour of these materials,
which is decisive for future developments.
3.3 Processing Technologies
3.3.1 CAD-CAE-CAM Tools
Computer-aided design, calculation and
manufacturing technologies (CAD-CAE-CAM),
have become essential tools for product
development. They let 3D geometries and alternative
designs be obtained rapidly. Calculations on stress,
deformations, ergonomics, dynamic response and
other aspects including material comparison and
design can also be performed for design
optimization.
The numerous benefits of these technologies for
developing conventional products can also be
applied to the development of shape-memory
polymer-based medical devices.
The recent use of programs such as MIMICS for
processing medical image technique files (TAC,
RMN and others) enables biodevices to be made-to-
measure (Harrysson, 2007). With these programs
three-dimensional geometries of parts of the human
body can be obtained and exported to other CAD-
BIODEVICES BASED ON SHAPE-MEMORY POLYMERS - Current Capabilities and Challenges
139

CAE-CAM programs to perform the customised
designs and obtain prototypes by using techniques
that we will now explain. They also contribute new
possibilities to the design of customised implants
that benefit from the use of SMPs.
3.3.2 Rapid Prototyping Technologies
These new technologies mean that physical parts can
be obtained in a short time (days) directly from the
computer-aided designs. They are of great help in
optimising design iterations, improving end quality
and speeding up production.
Rapid prototyping systems first appeared in 1987
with the American company, 3D Systems
stereolithography, currently the most widespread
technology. It is based on being able to activate a
polymerisation reaction in a liquid state epoxy resin
by means of laser beam projection with a power and
frequency suited to the type of resin, which draws
the required geometry layer by layer. By using this
technology, epoxy resin prototypes with shape-
memory properties can be obtained directly.
Several results of our investigations at Product
Development Laboratory Universidad Politcnica
de Madrid related with the application of these
technologies to the development of SMP-based
devices are shown below.
Figure 2 shows how a pincer-shaped end for an
active catheter can be obtained from its 3D geometry
in a CAD file. An SLA-3500 machine was used to
polymerise a 3D Systems epoxy resin sold under the
trade name of Accura 60.
The pincer can be made to open by hot
deformation during the shape-memory training
process. By then heating it, the pincer closes, as
Figure 3 shows.
This ability can be used in similar devices, with
different geometries and materials, to extract foreign
bodies and in minimally invasive surgery.
Using silicone mould vacuum casting,
polyurethane prototypes with shape-memory can be
obtained. Figure 4 shows a ring device obtained with
this technology using a type of polyurethane sold
under the trade name of MCP 3115, whose
capability for recovering shape through heating is
also shown. Similar devices can be used to change
soft tissue geometry (annuloplasty, sutures,
cerclages and others).
The main benefits of these technologies are:
designs and functionalities can be efficiently
compared and the associated products more rapidly
developed. Being able to apply these technologies to
SMP-based devices is of great importance to their
becoming widespread in industry.

Figure 2: Active pincer design and prototypes.

Figure 3: Geometric activation in an epoxy resin pincer.

Figure 4: PU geometric cerclage activation.
3.3.3 Microfabrication
Being able to use microfabrication technologies with
SMPs provides a new line of use for these materials
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
140

inside the medical device industry, particularly in
lab-on-a-chip and controlled drug delivery systems.
Typical devices for these applications require
surface channels and microtextures with micrometric
geometries that can be obtained in polymeric
materials by using technologies such as hot-
embossing, micro injection moulding or LIGA.
To be precise, by applying hot-embossing to
SMPs, surface microtextures and microreservoirs
with temporary geometry can be achieved (Gall,
2004).
The possible use of physical or chemical vapour
deposition technologies, combined with the ability to
produce protective masks by applying UV
photolithography, enables surface embossing to be
performed on very different materials (metals, alloys
and ceramics) using shape-memory polymers as
substrate (Paumier, 2008).
Thus, by making connection microtracks, these
can be used to send an order to certain parts of a
device to activate a geometric change by heating an
adjacent resistance. In some cases, the connection
track itself can be used as a heating element if its
cross section is sufficiently small and its electrical
resistance, therefore, high enough.
This ability to manufacture by layers and
combine different materials enormously strengthens
the capability to integrate certain SMP-based active
parts into complex systems (such as implantable
medical devices).
3.4 Shape-memory Effect Training
The shape-memory effect training process is usually
conducted through heat deformation of the device
manufactured in SMP and subsequent cooling to
maintain the deformation, thereby obtaining the
temporary shape.
To increase the length of the temporary shape
devices traction machines with heated chambers are
used. To produce temporary surface marking hot
compression moulding presses are used.
Recently, the use of cone-shaped countershapes
has been proposed to obtain ring devices with a
temporarily enlarged diameter (Daz Lantada, Lafont
Morgado, 2008).
3.5 Activation
Another aspect where most progress has been made
is the activation of the memory effect by various
methods, especially:
Joule Effect Activation. Based on distributing
heating resistances at the core of the polymer where
the passing of an electric current generates the
necessary heat.
Light or Laser Activation. Based on projecting a
laser through a shape-memory material with a
similar absorption frequency to that of the laser
used, which produces heating (Lendlein, 2005,
Wilson, 2006).
Magnetic Activation. Based on heating by
induction of magnetic or metallic microparticles,
distributed at the core of the polymer while it is
being conformed to its shape (Buckley, 2006).
However, the biocompatibility of the associated
devices needs to be further optimised.
Support Technologies. Progress in the field of
wireless communications means that devices can
now be remotely activated, which is promoting the
appearance of new active implantable biodevices.
3.6 Commercial Formulations
The promising applications of these materials,
particularly in the field of medicine, together with a
growing industrial demand, has led to departments
dedicated to the synthesis of shape-memory
polymers being set up in large companies and the
appearance of some spin-off. The major ones are:
Mitsubishi Heavy Industries Ltd.
DIAPLEX.
mNemoscience GmbH.
CRG Industries LLC.
Most of these recently set up companies and
departments offer a la carte design work and
prototyping applications using SMP. They also
commercialise their developments, both synthesised
materials and products based on those materials.
4 CHALLENGES
This section deals with the main fields where more
in-depth study is particularly important, in order to
facilitate the industrial expansion of shape-memory
polymers as an integral part of active implantable
medical devices.

BIODEVICES BASED ON SHAPE-MEMORY POLYMERS - Current Capabilities and Challenges
141

4.1 Thermomechanical Response
Unfortunately, the shape-memory polymer materials
developed up to now only let forces of
approximately 3 MPa be withstood during
activation, which is insufficient for certain medical
applications intended for use as actuators, specially
when wishing to change the geometries of biological
tissues.
Enhancing the activation forces requires greater
understanding of the basic physical-chemical
principles of these phenomena. To this end,
computational models can be used that help apply a
combined knowledge of materials science,
thermodynamics, mechanics and heat transmission
(Conti, 2007).
4.2 Modelling and Simulation
Using the data obtained from the characterization
tests, (of materials and specific applications),
behaviour models can be obtained that facilitate the
development of new applications with the same
material or similar applications with other polymers.
The possibility of combining models that are
developed ad hoc and the multivariable simulations
that allow finite element calculation programs will
help simplify the design of training systems and the
heat activation of SMP-based devices.
4.3 Stability of Properties
In general, the variation in the properties of
polymeric materials through aging has major
economic implications as it affects in-service
performance. Particularly in the case of shape-
memory implants, any change in the vitreous
transition temperature (or activation temperature)
can cause problems when activating the necessary
geometric changes.
It is also necessary to study the changes to the
mechanical properties of these polymers (elasticity
modulus, hardness, and resilience), due to their
being implanted in the human body. Figure 5 shows
an example of an study (carried out by our group at
Universidad Politcnica de Madrid) on how the
hardness of an aging shape-memory polyurethane
evolves at 40C for 80 days. A Vickers
microhardness tester was used with a 0.98 N load
and a 15 s contact time.
114
116
118
120
122
124
126
128
130
132
-1 -0,5 0 0,5 1 1,5 2
log (time)
V
i
c
k
e
r
s

M
i
c
r
o
h
a
r
d
n
e
s
s

M
P
a

Figure 5: Evolution of hardness in a shape-memory
polyurethane.
Changes in the mechanical properties like that
shown can considerably affect the applicability of
devices based on these materials. The use of
additives and the synthesis of new formulations that
help minimise the changes to properties with the
passage of time will be highly useful for optimising
devices that can be commercialised.
However, the effects of moisture on changes in
the properties of SMPs may be determining factors
for making an application invalid. This effect has
been studied on shape-memory polyurethanes by
conducting DMTA and DSC tests with samples
submerged in water for different lengths of time to
age them (Yang, 2004).
4.4 Activation Temperature
Only on rare occasions, in the polymeric products
industry, have materials with vitreous transitions of
between 0 C and 50 C been used, as in-service
changes to properties are not usually desirable. For
this reason, it is difficult to find commercial
formulations for polymers with Tg in the 25 to 45 C
range.
However, for the development of shape-memory
polymer-based active implantable medical devices,
it is precisely temperatures near to the 37 C of body
heat that are sought. SMPs with a Tg of around 30
C may give rise to devices that change their
geometry on contact with the patients body. SMPs
with a Tg of around 45 C can be used to develop
implants intended for postoperative activation
through heating to induce geometric changes.
Some laboratories and companies achieve
noticeable changes in the Tg of dual component
SMPs by modifying the proportion of monomers and
additives for cross-linking that are used to synthesise
them. Most formulations still have activation
temperatures that are too high to be used in
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
142

implantable devices without causing damage to
surrounding tissues.
Table 1: Materials and acceptable Tg for biodevices.
Material
Vitreous transition
temperature
Reference
tBA-co-PEGDMA
40 - 52 C
(according to % of
cross-linking)
(Yakacki, 2008)
Polynorbornene
(Norsorex

)
Around 40 C
(Liu, Mather,
2003, 2007)
Polyurethane RoomTemp. 50 C (Tobushi, 2008)
Polyurethane Diaplex
MM5520
55 C
DIAPLEX Ltd.
(Small, 2005)
Poly(-caprolactone) 40 59 C (Lendlein, 2002)
Epoxy-based 35 105 C CRG Industries
Styrene-based 45 95 C CRG Industries
Fortunately, in the last 5 years new SMPs with a
Tg closer to body temperature have been synthesised
and could be used in conjunction with appropriate
protective coatings to develop percutaneous
implants. Table 1 shows some prime examples.
In addition, some considerations, which are set out
below, must be taken into account concerning the
feasibility of using these materials.
4.5 Security Issues
4.5.1 Biocompatibility Improvements
Starting up production of shape-memory-based
polymers is closer than ever as more emphasis is
being placed on improving the biocompatibility of
these devices.
Many SMP formulations are toxic; however,
some of them have been shown to be compatible
with human tissues (Cabanlit, 2007, Sokolowsky,
2007), which is hopeful for future developments. In
whatever case, the use of protecting coatings (using
PVD or CVD) may be of considerable help in
improving this aspect.
4.5.2 Sterilization
Before in vivo implantation the devices need to be
sterilised using some of the methods that are usually
applied to polymers (steam, ethylene oxide, gamma
radiation, low temperature plasma LTP or the
Noxilizer process).
In spite of the numerous methods that can be
used, it is preferable to choose low temperature
sterilisation (LTP, ethylene oxide or the Noxilizer
method) to avoid activating the memory effect
before implanting the devices.
The influence of these methods on toxicity and
thermomechanical response of these materials has
recently begun to be studied with promising results
(Yakacki, 2008).
4.5.3 Regulations
In order to optimise the safety of devices based on
these materials they must be in compliance with the
guidelines of the European Directive on Medical
Devices 93/42/EEC and the European Directive
on Active Implantable Medical Devices
90/385/EEC. It is also advisable to follow the
recommendations of Standard ISO 13485 on quality
in medical devices as well as specific legislation
concerning materials characterization tests (ISO and
ASTM Standards especially).
4.6 Structured Development Process
If the development of commercial medical
applications based on these devices is to be
promoted, it is important to increase the connection
between all the actors taking part in the different
development stages.
In this way, they could collaborate to establish a
structured design process to combine the tasks of:
synthesis, materials characterization and processing,
mechanical design, prototype manufacture, in
vitro and in vivo trials, official approval and
subsequent production start-up.
Similar proposals are being successfully applied
to promote developments based on other active
materials, such as electroactive polymers (EAPs),
(Bar-Cohen, 2002, 2006).
5 IMPROVING RESULTS AND
CONCLUSIONS
Shape-memory polymers have emerged with
enormous potential allowing the development of
medical devices with special features and
capabilities for activation hitherto unachievable.
The development of bioactuators based on these
materials at present requires progress in various
scientific-technological aspects to optimise their
possibilities. It will then be possible to obtain
commercialisable medical devices (diagnostic and
therapeutic) that fulfil all the mechanical,
therapeutic, stability and safety requirements.
Recent advances in issues of international co-
operation concerning active materials with the
setting up of specific forums like Scientific.net,
Biomat.net and others, are helping to disseminate
results and exchange opinions. However, it would be
BIODEVICES BASED ON SHAPE-MEMORY POLYMERS - Current Capabilities and Challenges
143

of great interest to create a specific forum on shape-
memory polymers and their applications, where
researchers, universities and enterprises could make
contact in order to fit technological supply with
market requirements, which is of particular
importance for the Medical Industry.
While the new capabilities brought by these
materials give rise to expectations that many medical
devices will become more effective, considerably
more effort still needs to be put into research and
development, so as to obtain robust and effective
actuators based on these materials.
REFERENCES
Lendlein, A., Kelch, S., 2002. Shape-Memory Polymers.
Angewandte Chemie International.
Lendlein, A., Kelch, S., 2005. Shape-Memory Polymers.
Encyclopedia of Materials: Science and Technology.
Liu, C., Mather, P., 2007. Review of progress in Shape-
Memory Polymers. Journal of Materials Chemistry.
Wache, H., 2003. Development of a polymer stent with
shape-memory effect as a drug delivery system.
Journal of Materials Science Materials in Medicine.
Lendlein, A., Kelch, S., 2005. Shape-Memory Polymers as
Stimuli-sensitive Implant Materials. Clinical
Hemorheology and Microcirculation.
Lendlein, A., Langer, R., 2002. Biodegradable, Elastic
Shape-Memory Polymers for Potential Biomedical
Applications. Science.
Wilson,T., et al., 2006. Shape-memory Polymer
Therapeutic Devices for Stroke. Lawrence Livermore
National Laboratory.
Small, W., et al., 2005. Laser-activated Shape-Memory
Polymer Intravascular Thrombectomy Device. Optics
Express.
Yakacki, C.M. et al., 2007. Unconstrained Recovery of
Shape-Memory Polymers Networks for
Cardiovascular Applications. Biomaterials.
Yakacki, C.M. et al., 2008. Deformation Limits in Shape-
Memory Polymers. Advance Engineering Materials.
Gall, K.; Kreiner, P. et al., 2004. Shape-memory Polymers
for MEMS Systems. Journal of
Microelechtromechanical Systems.
Daz Lantada, A., Lafont, P. et al., 2008. Treatment of
Mitral Valve Insufficiency by Shape-Memory Polymer
Based Active Annuloplasty. Biodevices 2008
International Conference onBiomedical Electronics
and Devices. INSTICC Press.
Lafont, P., Daz Lantada et al., 2006. Patent Document
P200603149: Sistema activo de anuloplastia para
tratamiento de la insuficiencia mitral y otras patologas
cardiovasculares. Oficina Espaola de Patentes y
Marcas.
Bellin, I., et al., 2006. Polymeric Triple-Shape Materials.
Proceedings of the National Academy of Sciences.
Volk, B. et al., 2005. Characterization of Shape-memory
Polymers. NASA Langley Research Centre. Texas
A&M University.
Tobushi, H. et al., 2008. Shape Recovery and
Irrecoverable Strain Control in Polyurethane Shape-
Memory Polymer. Science and Technology of
Advanced Materials.
Liu, C. and Mather, P., 2002. Thermomechanical
Characterization of a Tailored Series of Shape-
memory Polymers. Journal of Applied Medical
Polymers.
Liu, Y. et al., 2003. Thermomechanical Recovery
Couplings of Shape-memory Polymers in Flexure.
Smart Materials and Structures.
Huang, W. and Lee, C., 2006. Thermomechanical
Behaviour of a Polyurethane Shape-memory Polymer
Foam. Journal of Intelligent Material Systems and
Structures.
Liu, Y. et al., 2006. Thermomechanics of shape-memory
polymers: Uniaxial experiments and constitutive
modelling. Int. Journal of Plasticity.
Yakacki, C.M. et al., 2007. Unconstrained recovery of
shape-memory polymer networks for cardiovascular
applications. Biomaterials.
Harrysson, O., et al., 2007, Custom-designed orthopaedic
implants evaluated using FEM analysis of patient
computed tomography data. BMC Musculoskeletal
Disorders.
Paumier, G., et al., 2008. Thermoresponsive Polymer-
Based Microdevice for Nano-Liquid Chromatography.
Biodevices 2008 Int. Conference on Biomedical
Electronics and Devices. INSTICC Press.
Lendlein, A., et al., 2005. Light-induced shape-memory
polymers. Nature.
Buckley, P., et al., 2006. Inductively Heated Shape-
memory Polymer for the Magnetic Actuation of
Medical Devices. IEEE Transactions on Biomedical
Engineering.
Conti, S., et al., 2007. Modelling and Simulation of
Magnetic Shape-Memory Polymer Composites.
Journal of Mechanics and Physics of Solids.
Yang, B., et al., 2004. On the effects of moisture in a
polyurethane shape-memory polymer. Smart Materials
and Structures.
Yakacki, C.M. et al., 2008. Cytoxicity and
Thermomechanical Behaviour of Biomedical Shape-
Memory Polymer Networks Post-sterilization.
Biomedical Materials.
Cabanlit, M., et al., 2007, Polyurethane Shape-Memory
Polymers Demonstrate Functional Biocompatibility In
Vitro. Macromolecular Bioscience.
Sokolowsky, W., et al., 2007, Medical Applications of
Shape-memory Polymers. Biomedical Materials.
Bar-Cohen, Y., 2006. Artificial Muscles using
Electroactive Polymers (EAP): Capabilities,
Challenges and Potential. SPIE Press.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
144
INFRARED THERMOGRAPHY AS A SUPPORT TOOL FOR
DEVELOPING SHAPE-MEMORY POLYMER BIODEVICES
Andrs Daz Lantada, Pilar Lafont Morgado, Hctor Lorenzo-Yustos, J ulio Muoz-Garca
J os Luis Muoz Sanz, J avier Echavarri Otero and J uan Manuel Munoz-Guijosa
Grupo de Investigacin en Ingeniera de Mquinas E.T.S.I. Industriales Universidad Politcnica de Madrid
C/ Jos Gutirrez Abascal, n 2. 28006 Madrid, Spain. (+34) 913364217
[email protected]
Keywords: Infrared thermography, Shape-memory polymers, Biodevice development, Prototype testing.
Abstract: Infrared Thermography is a technique for carrying out inspections and non-destructive tests that can also be
used as a support tool for developing medical devices based on the use of shape memory polymer materials.
This paper sets out some of the opportunities and advantages provided by this technique for designing the
heating systems associated with these shape-memory polymer based devices. Its application for developing
an active pincer, whose geometry can be changed by heating, is explained in detail. Similar devices can be
used as active catheter ends for minimally invasive surgery tasks.
These thermography tests can also be used as a validation tool for the heating simulations, linked to
optimising this type of devices, and also during in vitro tests aimed at obtaining safer active implantable
devices.
1 INTRODUCTION TO
INFRARED THERMOGRAPHY
Infrared (IR) Thermography is a technique for
carrying out inspections and non-destructive tests
which has multiple applications in the development
of machines and products, equipment and facilities
maintenance, and troubleshooting.
Since all bodies emit (according to their
temperature) infrared radiation, which increases in
intensity as the temperature rises, variations in this
intensity can be detected by using infrared sensors.
Thermal cameras can detect radiation in the
infrared range of the electromagnetic spectrum
(usually between a 900 and 14000 nm wavelength,
instead of operating in the visible range of 450 to
750 nm) and can produce images of this radiation.
These cameras are fitted with a sensor matrix
(called microbolometer) that is sensitive to this
radiation. Depending on the intensity of the radiation
more or less current is sent to the cameras control
electronics, which with the aid of specific software
enables temperature maps to be obtained.
The spheres of application range from
Mechanical Engineering, Electrics, Electronics,
Aeronautics, Architecture and Engineering in
general, to Medicine or Veterinary Science and even
Art and Archaeology. The last decade has seen
enormous progress in the equipment available on the
market as well as more affordable prices that have
led to its expansion as a testing technique in all
kinds of industry.
Some of the fundamental advantages of the
technique are its speed and ease of use, easy to
interpret temperature map-based results and the fact
that it is a non-destructive technique that does not
damage the systems under study (Schindel, 2007,
Maldague, 2001).
Apart from these applications, its use as a
support tool for developing medical devices,
especially those based on the use of shape-memory
polymer (SMP) materials, is detailed in this work.
The main objectives of the study, as exposed in
the following chapters, are:
To show the importance of validating FEM
thermal simulations with trials.
To explain how IR thermography can be used in
such validating trials.
To make clear how both techniques should be
used in a combined way for improving SMP-
based medical devices development.
145

2 SHAPE-MEMORY POLYMERS
AND POTENTIAL BIODEVICES
Shape-memory polymers (SMPs) are materials that
show a mechanical response to external stimuli,
usually to changes of temperature. When these
materials are heated above their activation
temperature, there is a radical change from rigid
polymer to an elastic state that will sometimes allow
deformations of up to 400%. If the material is cooled
after manipulation it retains the shape imposed; the
said structure is frozen and returns to a rigid but
unbalanced state. If the material is again heated
above its glass transition temperature or activation
temperature it recovers its initial non-deformed
state.
The cycle can be repeated numerous times
without degrading the polymer and most suppliers
can formulate different materials with activation
temperatures of between 30 C y 260 C, depending
on the application required. Of all the polymers
developed that show shape memory properties, those
most worthy of mention are epoxy resins,
polyurethane resins, cross-linked polyethilene,
styrene-butadiene copolymers, polynorbornene and
other formulations (Lendlein, 2002, 2005, Liu,
2007).
They are therefore active materials that present
thermomechanical coupling and a high capability for
recovery from deformation, (much greater than that
shown by shape memory metal alloys), which
combined with their lower density and cost has
favoured the appearance of numerous applications.
Their properties permit applications for
manufacturing sensing devices or actuators,
especially for the aeronautics, automobile and
medical industry.
They have been proposed to develop numerous
medical devices such as self-expanding stents
(Wache, 2003), intelligent sutures (Lendlein, Kelch,
Langer, 2002, 2005), thrombectomy devices
(Wilson, 2006), active catheters (Yackaki, 2007),
drug delivery devices (Gall, 2004) or annuloplasty
systems (Daz Lantada, 2008).
As an example of the capability of these
materials to recover their geometry Figure 1 shows
the closure of a pincer manufactured in epoxy resin
(whose trade name is Accura

60) when its memory

is activated by heating (under forced convection
using a hot-air gun with air at 80 C).
Similar devices, introducing appropriate changes
to their geometry according to the application, could
be used as the active end of a catheter to remove
harmful particles, clots and other elements.

Figure 1: Shape-memory effect in an epoxy resin pincer.
The following section explains how infrared
thermography can help optimise the development of
shape memory polymer-based medical devices.
3 THERMOGRAPHY AND SMP
BIODEVICE DEVELOPMENT
3.1 Operational Considerations
Generally speaking, the total power emitted per unit
of area is given by Stefan-Boltzmanns Law, which
for a black body is expressed using the constant :
E
b
= T
4
(1)
It is important to point out that the concept of a
black body is an ideal concept as the real objects
are grey bodies, which means the concept of
emissivity has to be taken into account, giving
the equation:
E
b
= T
4
(2)
Where emissivity shows values in the 0 < <1
range and relates the radiation that would be emitted
by an ideal black body at the same temperature. This
constant is highly dependent on the materials
surface and its finish, and on the wavelength and
surface temperature, and can have a marked
influence on the results of the tests performed with
infrared thermography equipments.
So, when conducting thermography tests, the
value of the said emissivity needs to be selected
depending on the material of the object or device
under study. The camera itself incorporates typical
values for different metals, polymers, ceramics and
other materials.
For materials not included in the camera
software, the emissivity of the material can be
determined by painting a part of the object or device
with black optical paint, for example Nextel Black
Velvet, which gives an emissivity very close to 0.94.
5 mm
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
146

Then the value of the emissivity in the camera can
be changed for the original material surface until its
reading is adjusted to the pre-measured value for the
reference painted zone.
Figure 2 shows an example of the influence of
surface colour on the emissivity of the pincer
prototypes manufactured in shape-memory epoxy
resin. It shows how the different emissivity leads to
erroneous measurements of temperature, with
differences of around 3 C, even though all the
prototypes are the same temperature.

Figure 2: Importance of calibrating the emissivity of the
material.
Therefore, for the above prototypes, according to
the data in Figure 2, the emissivity values would be:
Black pincer.- 0.94
White pincer.- 0.92
Translucent pincer.- 0.86
This calibration of material emissivity is
particularly important when the device has to fulfil
its mission inside the human body, where
differences of 2 to 3 C can be decisive for avoiding
damage to tissue surrounding the end device.
Other environmental factors such as wind, rain
or snow, whose influence is usually listed in the
correction tables supplied by the thermography
equipment manufacturers, may be omitted if the
characterisation tests are conducted in a laboratory.


3.2 Characterization of Materials and
Applications
As already explained, in order to produce geometric
changes in active implantable devices based on
shape-memory polymers the temperature of the
material needs to be raised above its glass transition
temperature. In the development process of these
devices infrared thermography can be used as a
support tool for other characterization and design
technologies, particularly regarding:
Determining the activation temperature in
the different candidate materials.
Designing and optimising the heating
system to activate a change in the device
geometry.
Validating thermal simulations so that
design decisions can be made more quickly.
Carrying out the shape memory training
process at the most suitable temperatures.
Carrying out in vitro tests, as a prior step
to tackling in vivo tests.
The last four points put forward are set out in the
following sections, since determining the activation
temperatures of the different candidate materials is
usually performed with DSC or DMTA tests
(Mather, 2002, Volk, 2005).
3.3 Designing the Activation System
To make the preliminary design for the heating
system the number and value of the heating resistors
can be pre-selected (in case of Joule effect based
activation) in line with the following calculation
procedure.
If the resistors in the active device are required
to reach a steady state temperature above shape-
memory activation temperature, the power generated
must be equal to the leakage at that temperature. It
must be thus verified that:
q =n R I
2
=C S (T
biodevice
T
body
) (3)
With the following notation:
q.- Heat generated [W].
n.- Number of resistors connected in series.
R.- Resistance [].
I.- Intensity through the resistors [A].
C.- Heat transfer coefficient [W/(m
2
K)].
S.- External surface of the device [m
2
].
Having selected the heating resistors the
transient state can also be evaluated and the time
INFRARED THERMOGRAPHY AS A SUPPORT TOOL FOR DEVELOPING SHAPE-MEMORY POLYMER
BIODEVICES
147

estimated that the device will take to reach
activation temperature, as will now be explained:
q =n p =m c
p
dT/dt +C S (T T
body
) (4)
With the following notation:
m.- Device mass [kg].
c
p
.- Specific heat of the SMP [J /(kgK)].
T.- Device temperature [C K].
p.- Power generated by each resistor [W].
The differential equation is integrated between
the initial temperature, usually 37 C, and the final
temperature required for activation, which gives the
necessary activation time.
However, in order to optimise the design of the
heating system required to activate the geometric
change in a shape-memory material based device,
the combined use of infrared thermography tests and
simulations made applying the finite element
method is highly valuable.
This procedure will be explained below for the
shape memory pincers already mentioned, which
could be used as an active catheter end in minimally
invasive surgery tasks.
The properties of the epoxy resin used for
manufacturing the prototypes are listed in Table 1
and have been used when designing the device as
well as for the simulations performed.
Table 1: Properties of the material used.
Accura

60 Epoxy Resin
Density 1.21 g/cm
3
Tensile Strength 58 68 MPa
Tensile Modulus 2690 3100 MPa
Glass Transition (T
g
) 58 C
Hardness, Shore D 86
It is important to emphasise that although its
activation temperature of around 60 C could not
result in a safe intracorporeal device, there are
several shape-memory polymers whose activation
temperatures are closer to human body temperature
and which could be subjected to a procedure similar
to that set out in this work.
A heating resistor of 4.7 was chosen for the
tests and it was fitted into the device in an
specifically designed housing, which was finally
filled with additional epoxy resin. The prototype is
shown in Figure 3.


Figure 3: Epoxy resin pincer with built-in heating resistor
for activation.
This prototype was heated above the glass
transition temperature of the epoxy resin by using a
hot air gun with air at 80 C for forced convection.
The temporary shape was obtained by inducing
opening using traction perpendicular to the middle
plane of the pincer, just 2 mm above the resistor. In
this way, most of the temporary deformation is
produced in the zone near to the heating resistor,
which means that the initial shape can be recovered
by heating a small area of the device.
The low conductivity of the polymers means that
the use of several heating resistors is usually
required for devices with geometric changes in
different zones. The way of giving our device its
temporary shape means that the tests can be carried
out with a single resistor.
Figure 4 shows heat activation of the shape-
memory polymer based pincer. For the infrared
thermography tests, shown in Figure 5, a variable
voltage transformer was used. The first image shows
the situation under a steady state with a 0.6 W
heating power (geometric change activation is not
reached) while the second shows the situation under
a power of 1 W (which allows T
g
to be exceeded and
activates the device).

Figure 4: Activation by heating the pincer and recovery of
its geometry.
The ANSYS 9.0 finite element calculation
program was used to carry out the heating
simulations. Heating powers of 0.6 and 1 W were
used, similar to those used in the thermography tests,
with the purpose of comparing the results.
Additional data on the material and boundary
conditions used in these tests are set out in Table 2.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
148



Figure 5: Infrared thermograph of the pincer during the
heating process, below (54.7C) and above (92.5C)
activation temperature.
Table 2: Additional data for finite element simulations.
Thermal conductivity 0.2 W/(mK)
Specific heat 1.6 J /(gK)
Emissivity 0.86
Convection coefficient 15 W/(m
2
K)
Ambient temperature 26 C
Isotropic material hypothesis
The results of the simulations carried out are
shown in Figure 6 as steady state temperature maps
and its relation with the thermography tests
performed is also explained.
It is important to point out the similarity of the
results given by simulations and tests, which makes
us specially confident in the usefulness of our
simulations for future heating system optimisations.
For a 0.6 W heating, the steady state temperatures of
the hottest zone given by testing and simulation are
54.7 C and 52.8 C respectively. For 1 W of
heating, these temperatures reach 92.5 C in the test
and 92 C in the simulation. The temperature
distribution is also similar, with errors of less than
5% overall.
So, infrared thermography can be used as a tool
in non-destructive tests for validating the results of
simulations performed as part of the design process
for shape memory polymer-based devices.


Figure 6: Heating simulations. Solution of steady state
temperatures for heating powers of 0.6 and 1 W.
3.4 The Memory Effect Training
Process
To enhance the process of obtaining the temporary
shape of the shape memory polymer device, instead
of using oven convection heating or the help of a hot
air gun, the resistors in the activation system of the
device itself can be used.
In this way, with the help of associated control
electronics, the glass transition temperature can be
exceeded in specific zones of the device where the
geometry could be mechanically changed.
The use of thermography devices can be useful
for controlling and limiting the heating of the device
so that it will only exceed the T
g
in the zone whose
geometry is wished to change temporarily, the rest
of the structure remaining unaltered.
3.5 Carrying Out in vitro Tests
Infrared thermography is also useful when it comes
to carrying out in vitro tests for validating the
device before going on to the in vivo tests.
In a similar way to the examples shown, the
temperature reached by the surface of a prototype of
a SMP-based implantable device during activation
can be checked. It can be checked whether this
INFRARED THERMOGRAPHY AS A SUPPORT TOOL FOR DEVELOPING SHAPE-MEMORY POLYMER
BIODEVICES
149

temperature is harmful for the tissue that will be in
contact with the device. In addition the effectiveness
of using protective coatings can also be studied.
Indeed, by implanting the device in vitro and
taking a thermograph during the activation process,
the temperature of the tissue surrounding the device
can be measured at any time to ensure that no
harmful heating is being produced.
4 CONCLUSIONS
The work presented shows the use of infrared
thermography as a support tool for the development
of medical, surgical or implantable devices based on
the use of thermally activated SMPs.
This technology is extremely helpful for
designing the activation system of these devices and
for validating heating simulations aimed at
optimising their development. As an application,
tests and simulations carried out with active pincers
obtained by rapid prototyping are shown. With slight
changes to their geometry and size, such devices
could be used as active catheter ends for minimally
invasive surgery.
Thermography is also extremely useful for
carrying out in vitro tests and obtaining safer end
implantable devices, which would not potentially
harm the surrounding tissues by heating, as a result
of optimising the heating and isolating system using
combined FEM simulations and thermography trials.
As future challenges it would also be interesting
to apply such a procedure to medical devices based
on other active materials that can be thermally
activated. It could therefore be applied to improve
the response of actuators manufactured with shape
memory alloys or sensors based on pyroelectric
materials.
REFERENCES
Schindel, B., 2007. Thermographie in der Theorie und
Praxis. www.irPOD.net.
Maldague, X., 2001. Nondestructive evaluation of
materials by infrared thermography. John Wiley &
Sons Publishers.
Lendlein, A., Kelch, S., 2002. Shape-Memory Polymers.
Angewandte Chemie International.
Lendlein, A., Kelch, S., 2005. Shape-Memory Polymers.
Encyclopedia of Materials: Science and Technology.
Liu, C., Mather, P., 2007. Review of progress in Shape-
Memory Polymers. Journal of Materials Chemistry.
Wache, H., 2003. Development of a polymer stent with
shape memory effect as a drug delivery system.
Journal of Materials Science Materials in Medicine.
Lendlein, A., Kelch, S., 2005. Shape-Memory Polymers as
Stimuli-sensitive Implant Materials. Clinical
Hemorheology and Microcirculation.
Lendlein, A., Langer, R., 2002. Biodegradable, Elastic
Shape-Memory Polymers for Potential Biomedical
Applications. Science.
Wilson,T., et al., 2006. Shape Memory Polymer
Therapeutic Devices for Stroke. Lawrence Livermore
National Laboratory.
Yakacki, C.M. et al., 2007. Unconstrained Recovery of
Shape-Memory Polymers Networks for
Cardiovascular Applications. Biomaterials.
Gall, K.; Kreiner, P. et al., 2004. Shape Memory Polymers
for MEMS Systems. Journal of
Microelechtromechanical Systems.
Daz Lantada, A., Lafont, P. et al., 2008. Treatment of
Mitral Valve Insufficiency by Shape-Memory Polymer
Based Active Annuloplasty. Biodevices 2008
International Conference onBiomedical Electronics
and Devices. INSTICC Press.
Liu, C. and Mather, P., 2002. Thermomechanical
Characterization of a Tailored Series of Shape-
memory Polymers. Journal of Applied Medical
Polymers.
Volk, B. et al., 2005. Characterization of Shape-memory
Polymers. NASA Langley Research Centre. Texas
A&M University.
Paumier, G., et al., 2008. Thermoresponsive Polymer-
Based Microdevice for Nano-Liquid Chromatography.
Biodevices 2008 Int. Conference on Biomedical
Electronics and Devices. INSTICC Press.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
150
AN EXPLOITATION OF THE SELF-ORGANIZING MAP FOR
HUMAN MOTION ANALYSIS
W. Kurdthongmee
Division of Computer Engineering, School of Engineering and Resources, Walailak University
222 Thaibury, Tha-sa-la, Nakorn-si-thammarat 80160, Thailand
[email protected]
P. Kurdthongmee
Center for Scientific and Technological Equipment, Walailak University
222 Thaibury, Tha-sa-la, Nakorn-si-thammarat 80160, Thailand
[email protected]
Keywords: Kohonen Self-organizing map, Motion analysis, Gyroscope, Accelerometer.
Abstract: Falls are the most common type of home accidents among elderly people and are a major threat to their
health and independence. Evaluating the risk of falling is important because it enables the provision of
adapted assistance and of taking preventive measures with subjects deemed at risk of falling. Currently, the
risk of falling has been evaluated by using questionnaires with their associated problems of subjectivity and
limited accuracy in recall. The Kohonen Self-Organizing Map (SOM) has found applicability in a wide
range of application areas. Our research as a whole has a final objective to employ the concept SOM to
implement an adaptive fall risk detection and warning system. In this paper, we present the preliminary
results from our research to utilize SOM to analyze the motion parameters from a miniature sensor with
integrated gyroscopes and accelerometers attached to the chest of an individual. The results clearly indicate
that SOM can be successfully used to cluster the activities by means of their motion parameters. This is very
promising results to extend the concept to implement our final objective system.
1 INTRODUCTION
Falls are the most common type of home accidents
among elderly people and are a major threat to their
health and independence (Najafi, 2002). Thirty-two
percent of a sample of community dwelling persons
75 years and older fell at least once a year. Among
them, 24% sustained serious injuries (Tinetti, 1988)
In addition, falling can dramatically change an
elderly peoples self-confidence and motivation,
affecting their ability to function independently.
Considering the growing proportion of old people
(over 75) in the populations of industrial countries,
falls will be one of the major problems of this
important part of the population (Askham, 1990) In
2050, 16.4% of the world population and 27.6% of
the European population are projected to be 65 years
and above, and in 14 countries, including nine
European countries, more than 10% of the total
population will be 80 years or older. Most cases of
falls sustained by elderly people appear to result
from the cumulative effect of multiple specific
disabilities. Among these, balance and gait disorders
play a key role (Tinetti, 1986).
Our research has the following major goals: (1)
to investigate the changes of an individuals motion
parameters over a period of time, (2) to
experimentally proof that SOM can be used to learn
an individuals motion parameters and make the
decision for an unsafe motion that could be a fall
risk, and (3) to present the prototype of a SOM-
based adaptive system for monitoring and warning
of the fall risk motions. In this paper, we present the
results that fulfil part of the second objective. We
have a strong postulation that different persons have
different styles of motion, different activities to
perform in the daily life. In addition, it is very likely
that the styles of motion and activities are likely to
change over time. This calls for an automatic system
that is capable of learning along with the wearer and
warning if the motion parameters are out of normal
conditions previously and continuously learnt by the
system.
151

Evaluating the risk of falling is important
because it enables the provision of adapted
assistance and of taking preventive measures with
subjects deemed at risk of falling. The risk of falling
has generally been evaluated by using questionnaires
with their associated problems of subjectivity and
limited accuracy in recall (Cummings, 1988). Risk
of falls can also be evaluated by clinical and
functional assessment including posture and gait,
independence in daily life, cognition, and vision
(Tinetti, 1986). However, no simple objective
method is available.
A method of evaluating the characteristics of
postural transition (PT) and their correlation with
falling risk in elderly people is described in (Tinetti,
1988). With respect to the report, the time of sit-to-
stand and stand-to-sit transitions and their duration
were measured using a miniature gyroscope attached
to the chest and a portable recorder placed on the
waist. The comparison between two groups of
elderly subjects (with high and low fall-risk) showed
that the computed parameters were significantly
correlated with the fall risk as determined by the
record of falls during the previous year, balance and
gait disorders, visual disorders, and cognitive and
depressive disorders. From our point of view, the
drawbacks of the proposed system are three folds.
Firstly, the differences in the collected data among
different persons, or even within the same person,
but different time are not taken into account.
Secondly, the history of falls in the part was used as
an input parameter for the system. This could be an
incomplete data due to the limitation of memory of
the studied group. Finally, the proposed monitoring
and warning system is in a class of a pre-
programmed system. This is in contrast to our
proposed final system that will rely on using SOM to
make it adaptable. The exploitations of gyroscopes
and accelerometers for monitoring and warning
applications were also described in several patents
(Patent, 2008). They, however, lacked the
adaptability features.
J avanov et al. (J avanov, 2005) present a
prototype wireless body area network (WBAN)
system with unobtrusive, commercially available
sensor platforms that have minimum size and
weight. The prototype relies on using their proposed
algorithms to monitor wearers activity with a
minimal number of accelerometers to reduce the
price of future systems. The obvious difference
between this prototype and our proposed one is that
the prototype, itself, does not perform any adaptive
feature on board at all. Its main function is only to
transfer the accelerometer and electrocardiogram
(ECG) data to be analyzed by the remote server.
In (Hwang, 2004), a novel algorithm and real-
time ambulatory monitoring system for fall detection
in elderly people were described. The system
comprised of accelerometer to measure kinetic force,
tilt sensor and gyroscope to estimate body posture.
The BlueTooth module was integrated to the
system to send real-time data to a personal computer
for data analysis and warning. The system was
evaluated by attaching to the chest for fall detection
on three people aged over 26 years. The experiment
of four cases; forward fall, backward fall, side fall
and sit-stand, was repeated ten times and the
experiment in daily life activity was performed one
time to each subject. The results showed that the
system and the proposed algorithm could distinguish
between falling and daily life activity. Moreover, the
accuracy of fall detection is 96.7%. From our
opinion, the system required adjustment to be suited
for different persons since it lacks the feature to
automatically adapt the decision making rules.
The objective of this paper is to present the
experimental results of applying SOM to learn an
individuals motion parameters and correctly cluster
the motion parameters by means of the activities.
The remaining of this paper is organized as follows.
Our method is presented in Section 2, and
experiments and results are detailed and analyzed in
Section 3. Section 4 draws conclusions and
introduces future work.
2 METHODS
2.1 Data Gathering System
As mentioned earlier that individuals are likely to
have different styles of motion and different daily
activities to perform. In addition, it is very likely that
the styles of motion and activities change over time
and age. In our research, we developed an
experiment to confirm this postulation. A simple
low-cost data logger was developed in-house to
fulfil this purpose. The prototype of the system,
illustrated in Figure 1, consists of a microcontroller
(dsPIC30F2010), a couple of EEPROMs with total
capacity of 256 KB (2 of AT24C1024) and a special
purpose chip to serve as a voltage level converter
between TTL and RS232 standards. The RS232 is
used as a communication channel between a
personal computer and the data logger. The Analog
Devices Incs ADIS16350 tri-axis inertial sensor
(triple axes gyroscopes plus triple axes
accelerometers) sensor is used to measure both
angular and linear accelerations.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
152

A Windows based application software on a
personal site was specially developed to control the
operations of the data logger; i.e. time stamp setting,
sample time setting and data capturing for analysis.
Figure 2 shows the screenshot of the results from the
data logger during the period of 12:43:30
12:49:00. The first three columns are the time stamp
of the recorded data which are arranged in the
following format: the angular accelerations in the x,
y and z axes and the linear accelerations in the x, y,
and z axes. It is noted that the data logger was
programmed to sample data every 1 second period
but it is capable of immediate recording the new
incoming data if and only if all parameters change
more than the predefined acceptable delta which are
6 units in this case. Also, it is noted that in order to
get the true value of both angular and linear
accelerations the recorded values must be multiplied
by 0.07326
o
/s
2
and 2.522 mg, respectively.

Figure 1: A photograph of the data logger system (right)
and a miniature gyroscopes and accelerometers (left).

Figure 2: The screenshot of the results obtained from the
data logger during the period of 12:43:30 12:49:00.
Figure 3 show two snapshots of the experimental
results from recording the motion, time and activity
of a male volunteer of 39 years old. The legends of
the activities are as follows: (0) Walking, (1) Sitting,
(2) Walking Downstairs, (3) Walking Upstairs, (6)
J ogging and (7) Sleeping. It is noted that during the
period of our experimentations, these activity types
were approximately inserted offline after the data
had already been transferred to a personal computer.
This was done in order not to interrupt and disturb
the normal activities of our volunteers. It can be
observed that there are some patterns in the graphs
which have some relations with the activities. These
are very promising input data for training SOM.
Raw Data from Accelerometers at Different Times & Activities
-300
-200
-100
0
100
200
300
400
500
3:43 PM 3:50 PM 3:57PM 4:04PM 4:12PM 4:19PM
Time Stamp
Accel X
Accel Y
Accel Z
Activity

Raw Data from Gyroscopes at Different Times & Activities
-1500
-1000
-500
0
500
1000
1500
3:43PM 3:50PM 3:57PM 4:04PM 4:12PM 4:19PM
Time Stamp
Gyro X
Gyro Y
Gyro Z
Activity

Figure 3: The snapshots of the experimental results from
recording motion, time and activity of a volunteer from the
accelerometers (top) and the gyroscopes (bottom).
2.2 A Brief Introduction to SOM
A 2-D map is defined by k locations or k cells
arranged as a 2-D lattice. Each location contains an
n-dimensional model vector which comes to
resemble n-dimensional input (teaching) data during
the unsupervised learning process, the self
organization (J outsiniemi, 1995). As a result of
SOM process, the distribution of the model vectors
in the n-dimensional space will approximate the
probability distribution of the input vectors. The
topographic organization of the map will also
approximate the matrix ordering relations in the
input space. Thus, similar inputs project near each
other onto the map. Increasing the number of
locations k increases the accuracy of the
approximation, which means that k should be chosen
according to the computing power available. In our
experiment, we chose k =500. (a hexagonal lattice
with dimensions 25 and 20).
In our experimentation, the maps were
initialized, taught, and evaluated using the routines
in the SOMPAK package. The learning consisted of
two phases. In the first phase, the learning
coefficient (t) decreased from 1.0 to 0 in 100,000
AN EXPLOITATION OF THE SELF-ORGANIZING MAP FOR HUMAN MOTION ANALYSIS
153

steps, while the radius of the neighbourhood
decreased from 15 to 1. In the second phase, (t)
decreased from 0.125 to 0 in 10,000 steps, while the
radius of the neighborhood decreased from 3 to 1.
The data retrieved from the sensors were modified to
have zero mean and one standard deviation
(Zachrison, 2006). This was down in order to
restrain one of the input dimensions from becoming
too dominant. The time stamps were not used
directly for self organization process but they were
instead transformed into the period of change of a
consecutive pair of motion parameters. By doing this
way, it made the resulting map to be in a more
readable form with highly cluster groups of data.
3 RESULTS AND DISCUSSIONS
The resulting map after training process with an
individuals input data of 7 dimensions, which are
the period of change, the angular accelerations
and the linear accelerations in the x, y, and z
axes, is shown in Figure 4(a) (lines separating
different clusters and labels within the clusters were
manually inserted). In total, the data used during the
learning stage consisted of 1,044 records. In order to
get as general results as possible with the limited
data, a leave-one-out procedure was used. This was
done by taken out 10 percents of the records of the
motion parameters for each activity from the training
dataset and analyzed using a map trained with the
parameters of the remaining records. The activities
of the former set of data records were known to us
but they were invisible with respect to SOM. It can
be seen that the self-organization process produced
the final map with different distinguishable clusters.
In Figure 5(a), the clusters are grouped and the
corresponding activities are labeled manually with
the guidance of the data resulting from the labeling
process of the SOMPAK program.
It is noted that with respect to the results after the
SOMPAK labeling process, some cells in the map
were labeled by many activities. Also, some cells
were completely left unlabelled. For the former case,
this means that some activities share common
motion parameters. These activities are walking
upstairs and downstairs. For the latter case, such
cells could be matched to a group of motion
parameters which could be a fall risk. The
investigation of the properties of these cells is
beyond the scope of this paper.
At this point, consider the leave-one-out testing
results which was performed by querying the trained
SOM with an unknown motion parameters with
respect to the SOM (the one which was taken out
from the training dataset), the results are presented
in the middle column of Table 1 (within group). It is
clear that, with respect to the proposed set of motion
parameters for SOM training, all activities could be
correctly classified with the mean and the standard
deviation of 73.45 percents and 16.08, respectively.
It is noted that the hundred percents could not be
achieved because some activities were matched to a
group of cells of the SOM map which were
previously labeled by many activities.
Table 1: The results from the leave-one-out testing
procedure: within group and cross validation.
Activities
Percent Correctness
Within
Group
Cross
Validation
Walking (W) 61.54 56.24
Sitting (S) 86.67 62.00
Walking Downstairs (D) 64.29 42.85
Walking Upstairs (U) 61.54 50.00
J ogging (J ) 50.00 42.71
Sleeping (SL) 100.00 87.00
Table 2: The probability of being matched of all activities
after presenting to SOM with the known activity motion
parameters (Use the abbreviations from Table 1).
Act
The Probabilities of Matched to Activities
W S D U J SL
W 0.62 0.07 0.07 0.08 0.10 0.00
S 0.15 0.87 0.29 0.00 0.20 0.00
D 0.18 0.05 0.64 0.08 0.15 0.00
U 0.00 0.00 0.00 0.62 0.05 0.00
J 0.03 0.00 0.00 0.15 0.50 0.00
SL 0.03 0.02 0.00 0.08 0.00 1.00
The rightmost column of Table 1 presents the testing
results after performing the similar testing
procedures with the dataset of the motion parameters
from different volunteer: the cross-validating test. It
was expected prior to performing the test that the
results could not be as good as the previous testing.
The outcomes show that the mean of correctness is
only 58.97 percents with a standard deviation of
15.19. This confirms that different persons have
their own set of motion parameters. To correctly
classify the activities of an individual, SOM is
required to be trained with the motion parameters of
the individual.
There are some interesting results if we
reconsider the probability of being matched to all
activities after presenting the unknown motion
parameters to SOM. The results are shown in Tab. 2.
Column 2 7 are the types of activity whose motion
parameters were presented to SOM for querying and
all rows are the resulting probabilities that SOM

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
154


(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)
Figure 4: (a) The resulting map after clustering and labelling, the projection maps in (b) (h) show the relationships
between the activities and the training parameters: period of change, angular and linear accelerations in the x, y, and z axes.
matched those parameters to the activities. It can be
seen that (walking, walking downstairs),
(sitting, walking downstairs) and (jogging,
walking downstairs, sitting) share some
common motion parameters. It is quite clear for the
first group but question could be arisen for the
second and third one. This can be explained that the
volunteer could be moving while sitting.
After all activities were clustered, it is now
possible to analyze the relationships between
different training parameters and the activities.
Figure 5(b) (h) show the projection maps with
respect to individual motion parameters. Consider
the map of the x and y axes of gyroscope, Figure
5(c) and (d), both maps show almost the similar
clusters and the scales of the motion parameters
which are reflected from their colour appearances
(see the legend on the right side of the maps.) It is
surprised to see that only the activity of type
walking can make a prominent high value to the
sensors in these axes. This could come from the fact
that our volunteers walking style causes a more
forward movement comparing to jogging and
walking upstairs and downstairs. Also, the
projection map of the z-axis gyroscope in Figure
5(e) which clearly shows different clusters of
activities, comparing to the previous 2 maps,
indicates that walking and jogging cause this
sensor to output a very high level.
The maps with respect to the accelerometers
shown in Figure 5(f) (h) obviously indicate that
SOM is capable of clustering sleeping which
differs from the rest activities in the way that the
motion parameters in all axes are completely
changed. Also, the maps in Figure 5(f) which
projects onto the x axis accelerometer clearly show
the clusters of walking. There are some variations
in the motion parameters of sitting with respect to
the accelerometers which could be interpreted that
our volunteer could be moving while sitting. It is
noted that these variations could not be correctly
recorded during the period of our experimentations.
AN EXPLOITATION OF THE SELF-ORGANIZING MAP FOR HUMAN MOTION ANALYSIS
155

Lastly, consider the last projection map, the period
of change, in Figure 5(b). The map reveals some
interesting points in the jogging cluster on the
bottom left part, also the sitting, the walking
upstairs and the sleeping clusters. These points
can be interpreted that there were suddenly changed
of the motion parameters during performing such
activities. It is easy to understand these cases for
jogging and walking upstairs which always
cause the sudden changes of these motion
parameters. For sleeping and sitting, such the
changes could be resulted from the immediate
change of motion patterns; i.e. move backward and
change to forward immediately during sitting or
quickly lie before sleeping.
4 CONCLUSIONS
In this paper, we presented the results that partly
fulfilled an objective of our overall research project
which is intended to develop an adaptive system to
detect motion parameters that are fall-risk. We
experimentally proved that SOM could be trained
with an individuals motion parameters: the period
of change of a consecutive pair of parameters, the
angular and linear accelerations in (x, y, z), resulting
in clustering of similar motion parameters. Also,
SOM could match between normal activities and the
clusters of motion parameters on the maps with as
high as 73.45 percents of correctness. However, the
matching between abnormal motion parameters that
could be a fall risk still needs more efforts to pursue.
From the experiment results, it can be concluded that
different activities of an individual have different
motion parameters (period of change is also
included). SOM can successfully and correctly
cluster these activities in relation to the motion
parameters. It is worth noting that in order to
classify the activities of a person with a high degree
of correctness, SOM needs to be trained with the
motion parameters of that person. With positive
experimental results, we expect that SOM can be
utilized to make the decision for an unsafe motion
that could be a fall risk in an adaptive way.
ACKNOWLEDGEMENTS
The author would like to thank Analog Devices Inc
for providing ADIS16350 sample. This work is
supported by the Assistive Technology Program of
Thailands National Electronics and Computer
Technology Centre (NECTEC).
REFERENCES
J . Askham, E. Glucksman, P. Owens, C. Swift, A. Tinker,
and G. Yu, 1990, Home and leisure accident research:
A review of research on falls among elderly people,
Age Concern Institute of Gerontology, Kings
College, London, UK.
S. R. Cummings, M. C. Nevitt, and S. Kidd, 1988,
Forgetting falls: The limited accuracy of recall of falls
in the elderly, J. Amer. Geriatr. Society, vol. 36, pp.
613616.
J . Y. Hwang, J .M. Kang, Y.W. J ang, H. C. Kim, 2004,
Development of Novel Algorithm and Real-time
Monitoring Ambulatory System Using Bluetooth
Module for Fall Detection in the Elderly, Proc. of the
26th Annual Int. Conf. of the IEEE EMBS, San
Francisco, CA, USA.
E. J ovanov, A. Milenkovic , C. Otto, P. de Groen , B.
J ohnson, S. Warren, G. Taibi, 2005, A WBAN
System for Ambulatory Monitoring of Physical
Activity and Health Status: Applications and
Challenges, Proc. of the 27th Annual Int. Conf. of the
IEEE Engineering in Medicine and Biology Society,
Shanghai, China.
S. L. J outsiniemi, S. Kaski and T. A. Larsen, 1995, Self-
Organizing Map in Recognition of Topographic
Patterns of EEG Spectra, IEEE Trans. on Biomedical
Engineering, vol. 42, no. 11, pp. 1062 - 1068.
B. Najafi, K. Aminian, F. Loew, Y. Blanc, and P. A.
Robert, 2002, Measurement of StandSit and Sit
Stand Transitions Using a Miniature Gyroscope and
Its Application in Fall Risk Evaluation in the Elderly,
IEEE Trans. on Biomedical Engineering, Vol. 49, No.
8, pp. 843 851.
M. Oja, S. Kaski, T. Kohonen, 2002, Bibliography of Self-
Organizing Map (SOM) Papers: 1998-2001
Addendum, Neural Computing Surveys, 3, 1-156,.
Patent, 2008, Patents: 5823845, 7141026, 6165143,
6095991, 6059576, 5919149, Available from
http://www.patentstorm.us (accessed on 16/06/2008).
M. E. Tinetti, T. F. Williams, and R. Mayewski, 1986, Fall
risk index for elderly patients based on number of
chronic disabilities, Amer. J. Med., vol. 80, pp. 429
434.
M. E. Tinetti, M. Speechley, and S. F. Ginter, 1988, Risk
factors for falls among elderly persons living in the
community, N. Eng. J. Med., vol. 319, pp. 17011707.
A. Zachrison and M. Sethson, 2006, Detection of System
Changes for a Pneumatic Cylinder Using Self-
Organizing Maps, Proceedings of the 2006 IEEE
Conf. on Computer Aided Control Systems Design,
Munich, Germany, pp. 2647 2652.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
156
SURFACE MODIFICATION OF DENTAL DEVICES
Surface Analysis of Plasma-based Fluorine and Silver Ion Implanted & Deposited
Acrylic Resin
Yukari Shinonaga, Kenji Arita and Milanita E. Lucas
Department of Pediatric Dentistry, Institute of Health Biosciences, The University of Tokushima Graduate School
Tokushima, Japan
[email protected], [email protected], [email protected]
Keywords: Surface modification, Dual ion implantation & deposition, Dental device, Fluorine, Silver, XPS, Contact
angle, Surface free energy, Abrasion.
Abstract: The aim of this study was to modify acrylic resin (PMMA) with fluorine (F) and silver (Ag) dual ions by a
hybrid process of plasma-based ion implantation & deposition (PBII-D), and to enable the surface of the
devices inhibition of bacterial adhesion. The surface characteristics, hydrophobic property and brushing
abrasion resistance were evaluated by XPS analysis, contact angle measurement and brushing abrasion test.
F and Ag were implanted-deposited and have formed carbon-fluoride and Ag-deposited layer on the surface
of the PMMA plate. The contact angle of F+Ag implanted-deposited PMMA was increased compared with
non-treated control and F only deposited PMMA. After 60,000 brushing strokes, the contact angle of
modified PMMA remained to be higher than that of the control PMMA. This study indicated that F+Ag
implantation-deposition has improved the hydrophobic property of acrylic resin and was sustained even
after routine tooth brushing.
1 INTRODUCTION
Bacterial colonization and subsequent device
infection are common complications of medical and
dental devices (Gristina, 1987). Especially, acrylic
base plates for prosthodontics, orthodontics and
pedodontics are exposed to oral microbial flora to
include bacteria, viruses, and fungi, and are
susceptible to adhesion of bacterial plaque. Oral
bacteria can be released from denture plaque into
salivary secretions and then aspirated into the lower
respiratory tract causing pneumonia (Sumi, 2007).
Among patients, the most general method for the
removal of denture plaque is brushing. However,
effective plaque removal requires a degree of
manual dexterity that is often lacking especially
among elderly and individuals with disabilities.
Therefore, it is important to modify the acrylic resin
denture surface to enable inhibition of oral bacterial
adhesion.
Plasma-based ion implantation (PBII) is a
promising method for the surface modification of
three-dimensional materials (Conrad, 1987). In
particular, the ion deposition with simultaneous ion
implantation (plasma-based ion implantation and
deposition: PBII-D) is desirable for efficient
processing and has an advantage over conventional
methods (Kuze, 2002).
Recently, several researchers have carried out the
surface modification using fluorine (F) ion and
found it to be a useful means of inhibiting bacterial
adhesion (Zhao, 2007 and Nurhaerani, 2007). In
addition, it is well known that silver (Ag) possesses
the antibacterial property without any toxic effects in
comparison to other heavy metal ions. We
developed the new technology to simultaneously
implant and deposit both F and Ag ions into dental
and medical devices using PBII-D. The aim of this
study was to examine the effectiveness of both F and
Ag ions implanted-deposited into acrylic resin by
evaluating the surface characteristics and brushing
abrasion resistance.
2 MATERIALS AND METHODS
2.1 Preparation of Specimens
Poly methyl methacrylate (PMMA) (Clarex 000,
Nitto J ushi Kogyo, Co., Ltd., Tokyo, J apan) plates
157

with measurements of 10mm10mm1mm were
used. The PMMA plates were modified by plasma-
based ion implantation-deposition equipment at
Plasma Ion Assist Co., Ltd., Kyoto, J apan. Fluoride
gas used for F ion implantation-deposition was
perfluoropropane (C
3
F
8
). For Ag ion implantation-
deposition, a 99.8 % Ag mesh cover was set 10mm
above the plates and sputtered by C
3
F
8
gas. The
conditions of plasma-based F and Ag dual ion
implantation-deposition are shown in Table 1.
Table 1: The condition of fluorine and silver ion
implantation-deposition into PMMA plates.
Group Conditions
Implantation
process
Deposition
process
Control
Voltage (keV)
Time (min)
Ag mesh
0
0
unused
0
0
unused
F deposited
Voltage (keV)
Time (min)
Ag mesh
0
0
unused
-0.5
60
unused
F+Ag
implanted-
deposited
Voltage (keV)
Time (min)
Ag mesh
-5
30
used
-0.5
60
used
2.2 Surface Shemical Analysis by XPS
The surfaces of the control, F deposited and F+Ag
implanted-deposited PMMA were characterized by
X-ray photoelectron spectroscopy (XPS). XPS
spectra were obtained using an X-ray photoelectron
spectrometer (ESCA-850, Shimadzu Co., Kyoto,
J apan) with Al-K radiation operated at 30 mA
current and 7 kV accelerating voltage. Specifically,
depth profile analysis for the F+Ag implanted-
deposited PMMA was performed using Ar etching
under the pressure of 510
-4
Pa. The Ar etching rate
was approximately 6 nm/min on Ag.
2.3 Contact Angle Measurements
Specimens were ultrasonically cleaned in distilled
water for 10 minutes, and then dried at room
temperature before contact angle measurement.
Static contact angle measurements were conducted
by the sessile drop technique using a contact angle
meter (CA-DT, Kyowa Kaimenkagaku Co. Ltd.,
Saitama, J apan) with three test liquids: distilled
water, diiodomethane and ethylene glycol at room
temperature. One point per specimen was measured
(N=5/group, i.e. 5 points/group). The surface free
energies of specimens were calculated from the
contact angles with the three test liquids (Liu, 2005).


2.4 Brushing Abrasion Test
The brushing abrasion test machine (MANA-63S,
MASUDA Co., Osaka, Japan) was used. A
commercial toothbrush (Dr. Bee Young S, Bee
Brand Medico Dental Co., Ltd., Osaka, J apan) was
attached to the toothbrush holder in contact with the
F deposited and F+Ag implanted-deposited PMMA
set on the sample holder. Distilled water without
dentifrice was then poured into the vessel, and the
machine was run at 80 rpm with a 200 g load. The
water contact angle measurements of the F deposited
and F+Ag implanted-deposited PMMA were done
every 10,000 strokes up to 60,000 strokes. In the
control PMMA, the contact angle was measured
without brushing. Five points per specimen were
measured.
2.5 Statistical Analysis
The results of the contact angle measurements
before and after the brushing abrasion test were
expressed as the mean standard deviation. The
data were analyzed using one-way ANOVA and
Scheffes multiple comparison tests ( =0.05).
3 RESULT
3.1 XPS Analysis
XPS wide-scan spectra of the control, F deposited
and F+Ag implanted-deposited PMMA are shown in
Figure 1. The peaks of C1s and O1s were detected in
the wide-scan spectrum of the control PMMA.
Moreover, F1s peak appeared on the F deposited
PMMA surface; and F1s, Ag3d and Ag3p peaks
appeared on the F+Ag implanted-deposited PMMA
surface. The F1s and Ag3d XPS depth profiles of the
800 600 400 200 0
Binding Energy (eV)
1000
F+Ag
Implanted/deposited
F dposited
Control
F1s
F1s
Ag3p
Ag3d
O1s C1s
C
o
u
n
t

(
a
r
b
i
t
r
a
r
y

u
n
i
t
)

Figure 1: XPS wide-scan spectra of PMMA specimens.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
158

F+Ag implanted-deposited PMMA are shown in
Figure 2. The chemically shifted peaks of F1s from
surface to 7 minutes Ar etching depth are observed
in the higher binding energy region (689-690 eV).
The higher values was close to the reported binding
energy of p-(CF
2
=CF
2
) (689 eV) (Briggs and Seah,
1990).The F1s peak decreased as the Ar etching
increased and was not detected after 15 minutes of
Ar etching. Ag3d peak was also detected after Ar
etching for about 15 minutes in Ag3d depth profile.
In F1s and Ag 3d depth profiles, the peaks of AgF
were not detected.
700 696 692 688 684 680
BindingEnergy (eV)
380 376 372 368 364 360
A
r
e
t
c
h
i
n
g

t
i
m
e

(
m
i
n
)
0
5
10
15
20
BindingEnergy (eV)
(A) F1s (B) Ag3d
p-(CF2=CF2) F
Ag
AgF
AgF
C
o
u
n
t
(
a
r
b
i
t
r
a
r
y

u
n
i
t
)

Figure 2: F1s (A) and Ag3d (B) XPS depth profiles of the
F+Ag implanted-deposited PMMA. Dashed lines show the
binding energy value of p-(CF
2
=CF
2
): 689.0 eV, F: 685
eV and AgF: 682.7 eV in the F1s depth profile (A), and
Ag: 368.2 eV and AgF: 367.7 eV in the Ag3d depth
profile (B). The values were based from Briggs and Seah,
(1990).
3.2 Surface Energy Analysis
The contact angles of three test liquids: distilled
water (W), diiodomethane (Di) and ethylene glycol
(EG) and the calculated surface free energy values
of control, F deposited and F+Ag implanted-
deposited PMMA are shown in Table 2. The contact
angles of all test liquids on the F deposited and
F+Ag implanted-deposited PMMA were
significantly higher than that of the control PMMA
Table 2: Contact angle and surface free energy of PMMA
specimens.



(p<0.01). Moreover, the contact angle of all test
liquids on the F+Ag implanted-deposited PMMA
was higher than that of the F deposited PMMA
(p<0.001). The surface free energy of the F+Ag
implanted-deposited PMMA was lower than that of
the control and F deposited PMMA.
3.3 Brushing Abrasion Test
In the F deposited and F+Ag deposited PMMA, the
contact angles after 10000, 30000 and 60000
brushing strokes were significantly lower than
before brushing (p<0.001). The contact angle of the
F+Ag implanted-deposited PMMA was not
significantly different compared with that of the F
deposited PMMA for the same number of brush
strokes. However, the contact angles of the F
deposited and F+Ag implanted-deposited PMMA
after the brushing abrasion test resulted in a
significantly higher contact angle compared to that
of the control PMMA (p<0.001).
0 10000 30000 60000
Number of brush strokes
50
60
70
80
90
100
110
C
o
n
t
a
c
t

a
n
g
l
e

(
d
e
g
r
e
e
)
120
Control
F deposited
F+Agimplanted/deposited
not significant
ANOVA/Scheffe, =0.05

Figure 3: Contact angles of distilled water on the F
deposited and F+Ag implanted-deposited PMMA
before/after the brushing abrasion test. Horizontal lines
indicate no significant differences (ANOVA/ Scheffe,
=0.05).
4 DISCUSSIONS
In this study, we have attempted to modify the
surface of PMMA by the simultaneous F and Ag ion
implantation-deposition by PBII-D process.
The results have shown that F and Ag were
detected on the surface of the F+Ag implanted-
deposited PMMA by XPS analysis (Figure 1). The
existence of carbon-fluoride complexes, such as p-
(CF
2
=CF
2
), and Ag were also detected on the F+Ag
implanted-deposited PMMA surface. Moreover, F
and Ag were deposited on the surface and implanted
Group
Contact angle (degree)
Surface free
energy
(mJ /m
2
)
W

Di

EG

Control 64.4 36.3 50.8 44.05
F deposited 100.0 52.2 62.7 33.77
F+Ag
implanted/
deposited
123.9 82.3 95.2 20.65
SURFACE MODIFICATION OF DENTAL DEVICES - Surface Analysis of Plasma-based Fluorine and Silver Ion
Implanted & Deposited Acrylic Resin
159

to about 90 nm depth (Figure 2). All these suggested
that both F and Ag ion implantation-deposition
method by PBII-D was applicable to PMMA. In ion
implantation-deposition of insulating materials such
as PMMA, electric charge-up could be a serious
problem, which damages the specimens. In this
study, it was thought that Ag mesh was used not
only for providing Ag ions but also for decreasing
the electric charge-up on the specimens by providing
electrons.
It was reported that some properties are specific
to the inert surface, such as the surface free energy,
surface charge, hydrophobic property, surface
roughness and surface chemistry (Perini, 2006). One
approach in the attempt to reduce the bacterial
colonization is to modify the surface free energy and
chemistry. The contact angle is characteristic of the
surface energy of a solid surface, and has been used
for determining the wettability and hydrophobic
property of various solid materials. Bacterial
adhesion is energetically unfavourable, if the solid
surface free energy is less than 50 mJ /m
2
(Busscher,
1984). In this study, the surface free energy of the
F+Ag implanted-deposited PMMA was 20.65 mJ /m
2
(<50mJ /m
2
), which may imply potential inhibition
of bacterial adhesion by hydrophobic mechanism. In
addition, the negative relationship between the
contact angle and the bacterial adhesion property
was reported (Zhao, 2007, and Nurhaerani, 2007). In
the present study, the contact angle of the F+Ag
implanted-deposited PMMA after 60,000 brushing
strokes was significantly higher than that of the
control PMMA. Kanter et al. (1982) estimated that
20,000 brushing strokes were the equivalent to
approximately 5 years of brushing. The present
study confirmed that the high contact angle of the
F+Ag implanted- deposited PMMA could remain
after the equivalent of 15 years of brushing with a
toothbrush. These results suggested that dual F and
Ag implantation-deposition could possibly inhibit
the bacterial adhesion to the PMMA devices by
increasing the contact angle and decreasing the
surface free energy.
These results suggested that both F and Ag ion
implantation-deposition by PBII-D process was the
superior surface modification method for acrylic
materials to inhibit bacterial adhesion.
5 CONCLUSIONS
In this study, PMMA plates were simultaneously
implanted-deposited with both F and Ag ions by a
hybrid process of PBII-D. The F+Ag implanted-
deposited PMMA surface has obtained low surface
free energy (20.65 mJ /m
2
) and the presence of
carbon-fluoride complexes and Ag on the surface
was indicated. Moreover, due to the presence of both
F and Ag ions, the hydrophobic properties remained
after brushing with a toothbrush. These results
suggested that both F and Ag ion implantation-
deposition by PBII-D has the potential to give the
medical and dental devices antibacterial qualities.
ACKNOWLEDGEMENTS
This study was supported by a Grant-in-Aid for
Scientific Research (C) (No. 20592300) from J apan
Society for the Promotion of Science (J SPS).
REFERENCES
Briggs, D., Seah, MP., 1990. Practical surface analysis 2
nd

edition., Volume 1 Auger and X-ray Photoelectron
spectroscopy, 595-634.
Busscher, HJ ., Weerkamp, AH., Mei, HC., Pelt, AWJ .,
J ong, HP., 1984. Appl. Environ. Microbiol. 980-983.
Conrad, J R., Radtke, J L., Worzala, FJ ., 1987. J. Appl.
Phys. 62, 4591-4596.
Grastina, A. G., 1987. Science.237, 1588-1595.
Kanter, J ., Koski, RE., Martin, D., 1982. J. Prosthet. Dent.,
47, 505-513.
Kuze, E., Teramoto, T., Yukimura, K., Maruyama, T.,
2002. Surf. Coat. Technol, 577-581.
Liu, Y., Zhao, Q., 2005. Biophys. Chem. 39-45.
Nurhaerani., Arita, K., Shinonaga, Y., Nishino, M., 2007.
Dent.Mater. J, 684-692.
Perini, CI., Zhao, Q., Liu, Y., Abel, E., 2006. Colloids.
Surf. B: Biointerfaces., 143-147.
Sumi, Y., Miura, H., Michiwaki, Y., Nagaosa, S., Nagaya,
M., 2007.. Arch. Gerontol. Geriatr. 44, 119-124.
Zhao, Q., Liu, Y., Wang, C., Wang, S., Peng, N., J eynes,
C., 2007. Appl. Surf. Sci, 8674-8681.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
160
MICROCOMPUTERIZED SYSTEM TO ASSESS THE
PERFORMANCE OF LUNG VENTILATORS
Daniel Marinho Silva
Biomedical Engineering Institute, Federal University of Santa Catarina, 88040 900, Florianpolis, SC, Brazil
[email protected]
Maurcio Campelo Tavares
Biomedical Engineering Laboratory, Catholic University of Pelotas, campus I, 96010-000, Pelotas, RS, Brazil
[email protected]
Raimes Moraes
Electrical Engineering Department, Federal University of Santa Catarina, 88040 900, Florianpolis, SC, Brazil
[email protected]
Keywords: IEC 60601-2-12, NBR 13763, Bluetooth, Clinical engineering, Maintenance.
Abstract: Lung ventilators may harm patients if they are not properly calibrated. Therefore, they must be periodically
assessed to verify if the supplied volume and pressure match the ventilatory settings. This paper presents a
system based on a microcomputer to assess the performance of lung ventilators. The hardware of the
developed system contains three modules: transducer, conditioning and acquisition. The transducer module
converts, to electrical signals, the pressure and flow waveforms supplied to a lung simulator by a lung
ventilator. It also supplies digital measurements of temperature and relative humidity (RH). The
conditioning module amplifies and filters the pressure and flow signals. The acquisition module reads the
digital measurements (temperature and RH) and carries out the analog to digital conversion of the
conditioning module outputs, sending these data to the microcomputer via radio-frequency. Software
written in C++shows the acquired waveforms on the PC screen and calculates the parameters required by
the IEC 60601-2-12. Data on the lung ventilator model, the sampled waveforms and the calculated
parameters are stored in a database, allowing the equipment follow-up. Comparative result of tests carried
out with the developed system and with commercial equipment is presented.
1 INTRODUCTION
Lung ventilators (LVs) supply air to patients that are
unable to breathe spontaneously, for instance, due to
chronic obstructive pulmonary disease, acute lung
injury, anesthesia or neurological disorder.
Therefore, they are widely used in intensive therapy
units (Pierce, 1995).
The LVs are built according to the IEC 60601-2-
12 that establishes the requirements needed to
minimize patients and operator risks (International
Electrotechnical Commission, 2001).
Since the LVs are used in critical clinical
situations, they must be periodically assessed to
verify if their performances were not degraded over
time, that is, if the supplied values match the
ventilatory settings on the LV. Besides, lungs may
be harmed by high airway pressure and high tidal
volume (Ricard et al., 2003; Wrigge et al., 2004;
Fernndez-Prez et al., 2006).
Therefore, it is very important to implement a
quality control program for LVs in order to avoid
patient injury.
Nevertheless, quality control programs in
developing countries are hampered by the high cost
of performance analyzers. Very often, heavy taxes
make the importation of these analyzers prohibitive,
preventing proper maintenance.
This work describes a lower cost LV analyzer
based on a PC microcomputer. It consists of an
electronic device to sample pressure, flow,
temperature and humidity. The sampled data are
sent, via radio-frequency, to a computer where the
waveforms are shown in real time. Parameters to
161

evaluate the LV are calculated. All data are stored in
a database.
2 MATERIALS AND METHODS
The Figure 1 depicts a block diagram of the
developed system and shows how it is connected to
the LV and lung simulator (LS). The LS acts as a
physiological load (resistance and compliance) for
the LV under assessment.
The developed system consists of an electronic
device that periodically measures the temperature
and relative humidity (RH) of the air inside the duct
that connects the LV to the LS as well as the flow
and the pressure waveforms generated by the LV.
The acquired data are sent to a PC microcomputer
via radio frequency (RF). The PC shows the sampled
data on the screen as well as parameters required by
the IEC 60601-2-12.

Figure 1: Block diagram of the system developed to assess
LV performance. It also shows how the system is
connected to the LV and LS.
The next sections describe each block of the
developed system.
2.1 Transducer Module
The transducer used to measure temperature and
relative humidity (RH) is the SHT75 (Sensirion Inc,
2007). It measures temperature from -40 to 123.8C
(accuracy: 0.5C; resolution: 0.01C) and RH from
0 to 100%RH (accuracy: 1.8%RH; resolution:
0.03%RH). These transducers have a calibration
certificate issued by the manufacturer.
Due to its small size (0.42 x 4.88 x 2.5 mm), it is
possible to insert the transducer into the air duct that
connects the LV to the LS.
The SHT75 yields the measurements in digital
format (14 bits) via a 2-wire protocol. This is a
bidirectional protocol that allows the sensor to
receive data such as commands to carry out the
measurements.
Before connecting the LV to the LS, the sensor is
exposed to the environment, allowing the system to
register the local temperature and RH.
To sample the flow and pressure produced by the
LV, two DC030NDC4 pressure transducers are used
(Honeywell Inc., 2008). The DC030NDC4 measures
the differential pressure applied to its inputs in a
range of 76.2cmH
2
O. It has a sensitivity of
52.36mV/cmH
2
O, producing a voltage output of
2.25V 2.0V.
An acrylic apparatus containing an obstacle is
placed between the LV and LS to create resistance to
the gas flow (pneumotacograph - PT). The pressure
drop across the resistance, measured by one of the
transducers, is proportional to the flow velocity
(Doeblin, 1990). The Figure 2 shows how the
transducer inputs are connected to the PT apertures
as well as a front view of the flow resistance. The
pressure drop is positive for inspiratory flow and
negative for expiratory flow.
To relate the A/D converter voltage input (that is,
the amplified and filtered differential pressure
transducer voltage output) to flow, 40 different flow
rates (20 positive and 20 negative) were applied to
the PT and to a calibrated flow meter (Fluke
Biomedical VT-Plus; uncertainty of 1,1l/min for
the -70 to +70l/min range). They were connected in
series to allow the comparison of their
measurements. An illustration of the experimental
setup is shown in Figure 3. A polynomial of seventh
order was fitted to the experimental points (voltage
input versus flow rate measured by the calibrated
meter) to allow inferring measures for flow rates not
evaluated. Using the polynomial, the flow
measurements obtained with the developed system
have an uncertainty of 4.4l/min.
The second transducer, connected to a third
aperture of the acrylic device, measures the
difference between the atmospheric pressure and the
one within the air duct.
To calibrate this transducer, 35 pressure values
(from 0 to 37.1cmH
2
O) were applied to the
transducer and, in parallel, to a calibrated meter
(Fluke Biomedical BP-Pump 2; uncertainty:
0.2cmH
2
O for a range from 0 to 120cmH
2
O). The
conditioned voltage output of the transducer (as
supplied to an A/D input) and the pressure readings
obtained from the calibrated meter were annotated.
From these values, a first order polynomial
between voltage and pressure was obtained.


BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
162


(a)

(b)
Figure 2: Diagram that shows how the pressure transducer
is connected to sample the flow waveform (a). Below (b),
there is a front view of the air resistance that provides a
pressure drop proportional to the flow velocity.

Figure 3: Experimental setup to calibrate the flow
measurements obtained with the developed system.
2.2 Conditioning Module
The electrical outputs of the pressure transducers
require further processing before being sampled.
Both signals are applied to second order Butterworth
low pass filters that have cut-off frequencies of
40Hz. For the expected ranges (pressure: 0 to
40cmH
2
O; flow: 70 to +70l/min), the output
amplitudes are adjusted to the input range of the
acquisition module (0 to 2.5V). For that, they are
amplified and shifted to positive values. The
achieved resolution is 50.0mV(cmH
2
O)
1
for the
pressure and 17.9mV(l/min)
1
for the flow.
2.3 Acquisition Module
A microcontroller (ADuC841, Analog Devices,
2003) samples the flow, pressure, temperature and
RH. The ADuC841 is an optimized single-cycle 20
MHz 8052 core that has an 8-channel analog input
multiplexer that feeds a 420 kSPS 12-bit ADC.
In this development, the ADuC841 timer is
programmed to sample the flow and pressure
waveforms at 160SPS.
For each set of 480 flow and pressure samples
acquired (3s), the microcontroller gets
measurements of temperature and RH from the
SHT75.
Since cables could cause difficulties for the
equipment handling, the sampled data are sent to the
computer via RF. For that, Bluetooth protocol was
chosen since it operates on the ISM band and does
not produce interferences in medical equipment
(J ones and Conway, 2005; Wallin and Wajntraub,
2004). The Bluetooth module employed is the KC-
21 (KCwirefree, 2007), configured for a 115.2kbps
transmission rate. The ADuC841 communicates
with the KC-21 via its serial interface. A transceiver
(SN74LVC1T45, Texas Instruments) was used to
convert the ADuC841 TTL levels to the low voltage
logic (3.3V) used in the KC-21 bus.
2.4 PC Software
The software, developed in C++ for Windows,
receives data from the electronic device described
above and calculates parameters that are stored in a
database.
The communication between the PC and the
device is established by means of another Bluetooth
module (KC-210) inserted into a USB port
(Kcwirefree, 2006).
When the user starts the program, a form is
launched to be filled up with data on the LV
(manufacturer, serial number, model and others).
This form also receives information on the
environment under which the test is being performed
(temperature, RH, atmospheric pressure, measured
power supply) as well as on the qualitative
assessment of the equipment (maintenance condition
of power cord, switches, alarms and others).
All inserted data is stored into an open source
relational database system (PostgreSQL -
http://www.postgresql.org/) that has native
programming interfaces for C++.
Pneumotachograp
Calibrated
Regulating Valve
Gas Source Pressure
MICROCOMPUTERIZED SYSTEM TO ASSESS THE PERFORMANCE OF LUNG VENTILATORS
163

The database implemented has two tables: one to
keep the information on the LV assessed and another
to store the test data.
Using two tables, different tests carried out with
the same LV can be stored on the database without
the need of reinserting data for each test performed.
By gathering the data on a same database, the
operator can promptly identify any performance
change since the previous results are available.
After registering the equipment in the database, it
is necessary to click on a button to proceed to the
next form page where the LV operating settings are
typed. This second form also requires information
on the range and resolution of the meters available in
the LV control panel.
Another button is shown that, when clicked on,
sends a command to the acquisition module to start
the data sampling.
The software was developed according to the test
procedure adopted by this laboratory (Tolotti, 2004).
The LV is switched on to work in volume cycled
mode during 40 minutes. After that, three set of
measurements carried out at intervals of 5 minutes
are recorded.
From the received flow waveform, the software
computes the volume supplied to the LS. Three
curves (pressure, flow and volume) are shown on the
screen in real time while temperature and RH
measurements are updated at 3s intervals. To
accomplish that, the software employs the graphic
library Graphics32 that provides fast operations with
pixels and graphic primitives
(http://www.graphics32.org/).
At the end of each respiratory cycle, the software
calculates the following indexes: breathing
frequency (BF), inspiratory time (IT), expiratory
time (ET), inspiratory/expiratory ratio (IER), peak

inspiratory pressure (PIP), positive end-expiratory
pressure(PEEP), mean airway pressure (MAP), peak
inspiratory flow (PIF), peak expiratory flow (PEF),
tidal volume (Vt) and minute volume (Vm).
After 5 minutes, the software automatically
records the values of all these parameters into the
database and warns the operator by means of a pop-
up window. When this occurs, the operator has to
read the LV meters and type these measurements to
be stored. This procedure is repeated twice at 5
minute intervals.
All these parameters and the three curves (one
minute interval) are stored into the database.
The Figure 4 gives an example of the waveforms
plotted on the screen as well as the fields filled up
with the data on the LV and indexes calculated by
the software.
A LV belonging to a public hospital was tested
using the developed system and a calibrated
commercial analyzer.
The test was carried out according to the
procedure proposed by Tolotti (2004). The LV was
working in volume-cycled mode (10 breaths/minute;
IER: 1:2; PEEP: 4cmH
2
O; Vt: 500ml).
3 RESULTS
Table 1 shows the average of three sets of
measurements obtained with developed system (DS)
and with a calibrated commercial analyzer (CA).
The relative error was calculated using the CA
values as reference. Since the flow resistance may
affect the results, it was not acceptable to connect
the two equipments in series to the LV and LS.
Therefore, the measurements were separately carried
out with each analyzer.
Table 1: Average of three set of measurements obtained
with the developed system (DS) and the commercial
analyzer (CA). For the other acronyms, refer to the text.
Indexes
Averaged
Measurements
(CA)
Averaged
Measurements
(DS)
Relative
Error
BF(min
-1
) 10 10 0.00%
IT(s) 2.023 2.05 1.33%
ET(s) 3.963 3.953 0.25%
IER 1:1.953 1:1.93 1.19%
PPI (cmH
2
O) 27.2 26.957 0.89%
MAP(cmH
2
O) 8.8 8.843 0.50%
PEEP(cmH
2
O) 4.03 3.94 2.23%
PEF(l/min) 48.34 54.89 13.54%
PIF(l/min) 20.61 20.63 0.10%
Vt(ml) 552.53 551.33 0.22%
Vm(l/min) 5.537 5.517 0.36%
Figure 4 exemplifies how the measurements and
sampled waveforms are obtained.

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
164

Figure 4: Screen presented by the developed software after performing a second set of measurements. A pop-up window
asks to insert the LV meters readings. In the first column, the operator has to inform the LV settings and, in the second one,
the resolution of the available LV meters. The fourth, sixth and eighth columns will be automatically filled up by the
measured and calculated indexes. Each one is obtained 5 minutes apart from each other. The third, fifth and seventh are
filled by the operator with the readings from the available LV meters. The indexes, from the first to last row, are:
Temperature, RH, BR(min
-1
), IE, IT, ET, IER, PIP, MAP, PEEP, PEF, PIF, Vt, Vm. The presented curves from top to
bottom are: pressure, flow and volume. The red line is a cursor that shows where the previously acquired waveforms are
being overwritten by the new samples.
4 DISCUSSION
Table 1 points out that, except for the PEF value, the
developed system has a good performance.
The error for the PEF can be explained by the
differences between the mechanical resistances of
both systems. During the expiratory phase, the LV
just opens a valve to empty the LS without
controlling the flow. Therefore, if the analyzers have
different flow resistances, they will produce
different PEF measurements. It should be mentioned
that the IEC 60601-2-12 does not establish a flow
resistance value for the analyzers. Therefore, to
analyze the LV performance over the time based on
this index, a same analyzer model shall be used.
5 CONCLUSIONS
A micro-computerized system to analyze the
performance of lung ventilators was successfully
implemented.
It has some better characteristics when compared
to commercial equipments.
A database is integrated to the analysis software.
For each test carried out, it is possible to store the
date, sampled waveforms and measurements under
the same equipment record. Therefore, it is easy to
follow up the LV performance along the time.
Besides, the database can provide information on the
life expectancy and the average number of repairs
for a given LV model, assisting the purchase of new
equipments.
RF communication between the acquisition
module and microcomputer provides comfort to the
operator. The computer can be placed by the LV
control panel to facilitate the registration of its meter
readings as required by the adopted procedure.
As the equipment interface and indexes
calculation are provided by the microcomputer
(usually available to the clinical engineering staff),
the hardware module has lower design complexity
when compared to similar analyzers and, therefore,
lower cost.
MICROCOMPUTERIZED SYSTEM TO ASSESS THE PERFORMANCE OF LUNG VENTILATORS
165

ACKNOWLEDGEMENTS
The authors thank CNPq (507363/2004-3) and
FAPESC (CON14598/2005-0) for financial support.
REFERENCES
Analog Devices 2003, ADuC841/ADuC842/ADuC843
microcConverter 12-Bit ADCs and DACs with
embedded high speed 62-kB Flash MCU datasheet.
USA.
Doeblin, E. O., 1990. Measurement Systems: application
and design. Chapter 7, McGraw-Hill, Singapore.
Fernndez-Prez, E. R., Keegan, M. T., Brown, D. R.;
Hubmayr, R. D., Ognjen G. 2006. Intraoperative Tidal
Volume as a Risk Factor for Respiratory Failure after
PneumonectomyAnesthesiology, 105: 14-18.
Honeywell Inc., 2008. DC030NDC datasheet.
http://sensing.honeywell.com/index.cfm?ci_id=14030
1&la_id=1&pr_id=82004
International Electrotechnical Commission, 2001. IEC
60601-2-12: Medical electrical equipment Part 2-12:
Particular requirements for the safety of lung
ventilators Critical care ventilators. Switzerland.
J ones, R. P., Conway D. H, 2005. The effect of
electromagnetic interference from mobile
communication on the performance of intensive care
ventilator, European Journal of Anaesthesiology; 22:
578583.
Kcwirefree, 2006. KC-210 embedded bluetooth USB
serial adapter datasheet. USA.
Kcwirefree, 2007. KC-21 OEM bluetooth module
datasheet. USA.
Pierce, L. N. B., 1995. Guide to Mechanical Ventilation
and Intensive Respiratory Care, W. B. Saunders
Company, USA.
Ricard, J . D., Dreyfuss, D., Saumon, G., 2003. Ventilator-
induced lung injury, Eur Respir J , 22: Suppl. 42, 2s
9s.
Sensirion, 2007. SHT1x / SHT7x humidity & temperature
sensor datasheet v3.01, Switzerland.
Tolotti, A. A., 2004. Desenvolvimento de procedimento
para realizao de ensaio de desempenho em
ventiladores pulmonares, Monograph, Clinical
Engineering Graduate Course, Federal University of
Santa Catarina, Florianpolis, Brazil.
Wallin, M. K. E. B., Wajntraub S., 2004. Evaluation of
Bluetooth as a Replacement for Cables in intensive
Care and Surgery, Anesth Analg; 98:763767.
Wrigge, H., Uhlig, U., Zinserling, J . , Behrends-Callsen,
E., Ottersbach, G., Fischer, M., Uhlig, S., Putensen,
C., 2004. The Effects of Different Ventilatory Settings
on Pulmonary and Systemic Inflammatory Responses
During Major Surgery, Anesthesia & Analgesia, 98:
775-781.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
166
MODELING OF MERIDIAN CHANNELS
Zimin Wang
1,2
, Yonghong Tan
2
1
School of Electronic Engineering, Xidian University, Xian, 710071, China
[email protected]
Miyong Su
1
2
College of Mechanical and Electronic Engineering, Shanghai Normal University, Shanghai, 201814, China
*Corresponding author: [email protected]
Keywords:
System identication, Modeling, Meridian systems, Acupuncture points.
Abstract:
In this paper, the auto-regressive and moving average (ARMA) models are constructed for the meridian chan-
nels based on the measuring data obtained from the acupoints in the Lung Channel of Hand-Taiyin (LU),
the Pericardium Channel of Hand-Jueyin (PC). For comparison, the ARMA models for the contrastive non-
acupoints around the measured acupoints are also obtained. Then, the analysis based on the zeros-poles
distribution of the obtained models is implemented.
1 INTRODUCTION
With the development of the research in the tradi-
tional Chinese medicine, many research reports had
shown great interest in the acupuncture-therapy. It
has been found that the acupuncture points (for ab-
breviation, we call them as acupoints) are distributed
in the meridian system of the human body. Moreover,
meridian system is an independent system which ex-
ists in parallel with neural systems and blood circu-
lation systems. The acupuncture points distributed
in the meridian system possesses many distinctive
ways for transferring signals and processing infor-
mation including electrical information (J. Julia and
FACOG, 1998). Moreover, the experimental results
have shown that the meridian system is an optimum
path for transferring information and works with a
close relation to the cerebral cortex and whole neu-
ral systems.
So far, the investigations on the meridian system
for acupuncture points are mainly concentrated on
the measurement of the resistance and potential, ca-
pacitance and conductivity distribution between acu-
points (F. III John and Erlichman, 2005; Yang, 1997;
W.Zhang and Zhu, 1999). However the research
on the features as the electrical signal transmission
through the meridian system with acupuncture points
has seldom been involved. Motivated by the phe-
nomenon that the meridian system is an optimumpath
for transferring information, in this paper, we use a
sequence of pseudo-random signal to stimulate the
Lung Channel of the Hand-Taiyin (LU) and the Peri-
cardium Channel of the Hand-Jueyin (PC). Then the
corresponding responses of the acupoints on those
channels are measured. Based on the obtained data,
the corresponding ARMA models are constructed by
using least squares algorithm. Those derived ARMA
models can be used to analysis the static and dynamic
characteristics of the meridian channels.
The paper is organized as follows. Section II
describes the experimental conguration for electric
stimulation and measurement of the corresponding re-
sponses of the acupoints and non-acupoints. Then the
AMRA models obtained based on measured data are
illustrated in Section III. In Section IV, the conclusion
will be given.
2 EXPERIMENTAL
CONFIGURATION
In this paper, a method based on three detecting elec-
trodes is used to measure the stimulation and the
corresponding response of the acupuncture points on
meridian systems. The architecture of the measure-
ment for meridian signal is shown in gure 1. The
167
three-electrode method of detecting the acupoint sig-
nal has the advantage of good operability and high
signal accuracy. The signal captured was affected by
skin moisture and electrode contact pressure. In order
to reduce the impact of test environment, we keep the
same test condition on every volunteer.
Figure 1: Acupoint signal measuring method using three-
electrodes.
The stimulation signal was a sequence of pseudo-
random current signal produced by a computer. This
signal is shown in gure 2.
Figure 2: The stimulation signal fed to acupoints.
In this experiment, the pseudo-random currents
through two electrodes stimulate acupoint 1 and acu-
point n of the measured meridian. Then detecting
electrode was used to measure the response of acu-
point m located in between acupoint 1 and acupoint
n. The outputs of the measured acupoints were trans-
ferred through a current/voltage conversion circuit
then sampled by an analog / digital convertor. The
sampled signals were sent to the computer for fur-
ther processing. There were 20 healthy volunteers ac-
cepted the test. Before the test, the volunteers were
relaxed to avoid the strenuous disturbance. Based on
the theory of Chinese medicine, there are 11 acupoints
in the Lung Channel of Hand-Taiyin. In this experi-
ment, the stimulating acupoints are Tianfu (LU 3) and
Taiyuan (LU9) which is considered as the ground, the
detecting acupoints are Xiabai(LU 4), Chize (LU 5),
Kongzui (LU 6), Lieque (LU 7) and Jinqu (LU 8) re-
spectively. The Pericardium Channel of Hand-Jueyin
includes a total of 9 acupoints. In the experiment,
Tianquan (PC 2) and Laogong (PC 8) are respectively
the stimulating point and ground. The detecting acu-
points are respectively Quze (PC 3), Ximen (PC 4),
Jianshi (PC 5), Neiguan (PC 6), and Daling (PC 7).
To test the signal of non-acupoint, ve contrast points
of non-acupoints are selected. All the acupoints and
non-acupoints used for experiment in this paper are
shown in Figure 3.
Figure 3: The acupoints of LU, PC and non-acupoints.
Due to the limited space, we only illustrate one
of the measuring results of the responses of acupoints
and non-acupoints here. The response measured from
acupoint Chize (LU 5) is shown in Figure 4. On the
other hand, the corresponding response of the non-
acpoint 1 which is 3cm away from acupoint Chize
(LU 5) is illustrated in Figure 5. The purpose of this
investigation is to look for some differences between
the meridian system and the contrast tissue around the
meridian. From gures 4 and 5, it can be seen that
there are some differences between the responses of
the acupoint and that of the non-acupoint. In our pre-
vious works, we have applied technique of wavelet
transform to the derived signals to nd the different
characteristic of those two kinds of signals. In the fol-
lowing, we will analyze the AMRA model parameters
of the acupoint signals and contrastive non-acupoint
signals.
3 PARAMETER MODEL OF
MERIDIAN CHANNELS
According to the theory of Chinese medicine, the
meridian system contains different channels. There
are several acupoints distributed in each channel. Nat-
urally, it motivates us to investigate the characteristic
of these channels when electric signals pass by. In
this section, the auto-regressive and moving average
(ARMA) model is utilized to describe the dynamic
features of the meridian channels. It is known that
the autoregressive part of the ARMA model with a
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
168
non-trivial denominator polynomial A(z) (all-pole
model) is very appropriate to describe spectra with
high and narrow peaks (Korosec, 2000), each sharp
spectrum peak corresponds to the pole located close
to unit circle at certain frequency.
Figure 4: The response of acupoint Chize (LU 5).
Figure 5: The response of non-acupoint 1.
The moving average part of the model is that,
for which a numerator polynomial B(z) exists. Pre-
dominant features of their spectra are valleys, cor-
responding to the 0 near the unit circle at the spe-
cic frequencies. On many occasions, a series is
not exclusively tted to one model, but rather various
models may equally well t the series (I. Rojasa and
M.Pasadasb, 2007; Hwang, 2001). If we follow the
norms of Box and Jenkins, the model chosen is nearly
always the simplest one, i.e. that involving the fewest
terms (GEP Box and Reinsel, 1994). The ARMA
model used to describe the dynamic behaviour of the
meridian channels is described by
y(t) +a
1
y(t 1) +... +a
n
a
y(t n
a
)
= b
1
u(t 1) +... +b
n
b
u(t n
b
+1) +e(t) (1)
where y(t) is the output at time t, n
a
and n
b
are
orders of the polynomials A(z
1
) = 1 +a
1
z
1
+ +
a
n
a
z
n
a
and B(z
1
) = b
1
z
1
+ +b
n
b
z
n
b
, and e(t) is
the white-noise disturbance value. To determine the
orders of the ARMA model, the Akaike Information
Criterion (AIC) (Akaike, 1969) to perform a relative
comparison of models with different structures is ap-
pled. Smaller value of AIC indicates a better model.
Figure 6: Order identication for acupoint signal using
AIC.
By comparing the AIC values between the differ-
ent orders of ARMA model, nally na and nb, are
respectively set to na=5 and nb=4 for the model to de-
scribe the behaviour of the transmission channel be-
tween acupoints Tianfu (LU3) and Chize (LU 5).
Figure 7 demonstrates the model validation re-
sult of the channel between acupoints Tianfu (LU3)
and Chize (LU 5). Figure 8 shows the correspond-
ing model residual. It can be seen that the obtained
model can describe the dynamic characteristic of the
measured meridian channel quite well.
Figure 7: Model validation result of the model.
The identied model coefcients are illustrated in
Table 1. Based on the obtained model, the corre-
sponding poles-zeros distribution chart is shown in
Figures 9. We can see that this channel is a sta-
ble system since all the poles are located within the
unit circle. However, the response of this channel
may demonstrates some oscillation since the complex
poles are included in this model.
For comparison, the measured point 3cm away
fromacupoint Chize (LU5) is dened as non-acupoint
MODELING OF MERIDIAN CHANNELS
169
Figure 8: Residual of the obtained model.
Figure 9: Poles and zeroes of the model for the channel
between LU3 and LU5.
Figure 10: Poles and zeroes of the model for the channel
between LU3 and non-acupoint 1.
1. The measured data from this point is also used
to identify the model between the stimulated point
(LU3) and non-acupoint 1. The identication proce-
dure is similar to what has been shown-above. The
obtained model parameters can also be seen in Table
1.
Figure 10 shows the corresponding zeros and
poles distribution chart of the model to describe the
dynamic feature beyween the stimulation point LU3
and the non-acupoint 1.
From Figures 9 and 10, we note that the models,
which respectively to describe the behaviour of the
channels on the meridian system and non-meridian
system, have shown quite different characteristics.
We also identify the channel between the stimu-
lation point LU3 and the measured acupoint Xiabai
(LU4) with an ARMA model. The corresponding
zeros-poles distribution chart is shown in Figure 11.
By comparing with Figure 9, it is seen that the model
between LU 3 and LU5 and the model between LU3
and LU4 have the rather similar zeros- poles distribu-
tions.
Moreover, the locations of zeros-poles shown in
Figure 11 are also rather different from those shown
in Figure 10.
Figure 11: Poles and zeroes of the model between LU3 and
acupoint Xiabai.
The zeros-poles distribution charts of the model of
the channel between Tianquan (PC 2) and Quze (PC
3), as well as the model of the channel between Tian-
quan (PC 2) and Ximen (PC 4) are shown in Figures
12 and 13.
Figure 12: Poles and zeroes of the model for the channel
between PC2 and PC4.
Moreover, the zeros-poles distribution chart of the
model between Tianquan (PC 2) and non-acupoint 5
is illustrated in Figure 14.
From the zeros-poles distribution charts shown-
above, we know that there exist some differences be-
tween the models on the meridian channels and the
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
170
Figure 13: Poles and zeroes of the model for the channel
between PC2 and PC5.
Figure 14: Poles and zeroes of the model between PC2 and
non-acupoint 5.
models on non-acupoints.
The obtained models can also demonstrate the dy-
namic performance of the meridian channels when
electric signals pass through. Based on the obtained
models, one can further analyze the characteristic of
the meridian systems. Especially, they would be a po-
tential ways when we intend to use them for disease
diagnosis and treatment.
Table 1: The comparison of the ARMA parameters between
acupoint and contrast non-acupoint signal.
ARMA Acupoint Non-Acupoint
Parameters signals signals
a1 -0.3819 -0.2639
a2 -0.0800 -0.4880
a3 -0.0490 -0.0687
a4 -0.0175 0.0043
a5 -0.0046 0.0131
b0 1.1158 1.2334
b1 -0.3857 -0.1057
b2 -0.0526 -0.5601
b3 -0.0721 -0.1239
4 CONCLUSIONS
In this paper, based on the electric-stimulation, the
signal characteristics of the meridian system in the
human body are presented. Based on the measured
data, the corresponding ARMA models for the merid-
ian channels between the stimulating points and the
measured acupoints are constructed. Then models be-
tween the stimulating points and the measured non-
acupoints are also identied in order to make compar-
ison.
According to the zeros-poles charts of the ob-
tained models, we can see some difference existing
in the meridian channel models and the non-accupoint
models. Some similarity seems to be observed among
the meridian channel models.
Those phenomena may have some potential to nd
a new way for disease diagnosis and treatment.
However, what we present in this paper is just a
primal investigation. The further research should be
implemented to see more details of the meridian sys-
tems.
ACKNOWLEDGEMENTS
This research is partially supported by Guangxi Sci-
ence Foundation (GXSF Grant No.: 0728210) and
Guilin University of Electronic Technology Research
Foundation (Z20507).
REFERENCES
Akaike, H. (1969). Fitting autoregressive models for pre-
diction. In Annals of the Institute of Statistical Math-
ematics.
F. III John, Trentini, B. T. and Erlichman, J. S. (2005). The
antinociceptive effect of acupressure in rats. In The
American Journal of Chinese Medicine. World Scien-
tic Publishing Company.
GEP Box, G. J. and Reinsel, G. (1994). Time Series Anal-
ysis: Forecasting and Control. Prentice-Hall, Engle-
wood Cliffs, New Jersey.
Hwang, H. (2001). Insights into neural-network forecasting
time series corresponding to arma (p,q) structures. In
Omega 29.
I. Rojasa, O. Valenzuelab, F. R. A. G. L. H. H. P. L. and
M.Pasadasb (2007). Soft-computing techniques and
arma model for time series prediction. In Neurocom-
puting.
J. Julia, M. T. and FACOG (1998). A modern interpreta-
tion of acupuncture and the meridian system. In 2nd
International Conference on Bioelectromagnetism.
MODELING OF MERIDIAN CHANNELS
171
Korosec, D. (2000). Parametric estimation of the continu-
ous non-stationary spectrum and its dynamics in sur-
face emg studies. In International Journal of Medical
Informatics.
W.Zhang, R. X. and Zhu, Z. (1999). The inuence of
acupuncture on the impedance measured by four elec-
trodes on meridians. In Acupunct and Electro-Therap
Res. Int.J.
Yang, H. (1997). The research and application of the dy-
namic testing systemfor point skin resistance. In Jour-
nal of Biomedical Engineering.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
172
MAGNETOMETRY USING ELECTROMAGNETICALLY INDUCED
TRANSPARENCY IN A ROOMTEMPERATURE VAPOUR CELL
Developing an Optical Magnetometer that Utilises the Steep Dispersion Curve
Observed in EIT to Detect Time Varying Magnetic Fields
Melody R. Blackman and Benjamin T. H. Varcoe
School of Physics and Astronomy, University of Leeds, Leeds LS2 9JT, U.K.
[email protected], [email protected]
Keywords:
Magnetometry, EIT.
Abstract:
The physiological importance of magnetic signals within biological systems has been investigated with ever in-
creasing sensitivities over the last decade. Currently superconducting quantum interference devices (SQUIDs)
are at the forefront of bio-magnetic diagnostics. In this research we aim to build an optics based magnetometer
that can compete with the sensitivity of the SQUID but that runs at a lower start up and operational cost. To
do this we intend to use the steep dispersion curve observed in the atomic physics effect electromagnetically
induced transparency. This magnetometer can operate at room temperature, its design is a convenient method
for monitoring bio-magnetic elds, making this technology an affordable technique for further bio-magnetic
diagnostics.
1 INTRODUCTION
Bio-magnetism is a fast developing eld of research
as its non-invasive applications make it a very desir-
able diagnostic tool. The presence of time varying
electric elds in the human body has been greatly
studied over the last hundred years. Electrocardio-
graphs (ECG) measure these elds as potential dif-
ferences on the skins surface, with emphasis on the
torso (Nash et al., 2002; Ramanathan et al., 2004).
However each of these electric elds is accompanied
by a weak magnetic eld that also holds information
on the condition of the organ. Unlike electric elds
the magnetic, do not suffer from varying attenuation
at they travel through the different types of tissue. An-
other advantage of measuring the magnetic eld is
the ability to make direct measurements of an organs
eld map instead of the potential difference between
two points.
The largest magneto-physiological signal in the
body is created by the hearts QRS peak which corre-
sponds to the contraction and relaxation of the cardiac
muscle during a single heart beat. The magnetic elds
associated with this process are of the order of tens
of picotesla, although the more clinically important
signals that are unobservable with ECG are formed
by the cardiac conducting system, which are super-
imposed on the eld and are less than 1 pT (Fenici
et al., 1983).
Currently at the forefront of magnetometer clini-
cal trials are Superconducting Quantum Interference
Devices (SQUIDs), which have displayed sensitivities
as high a few fT/

Hz (Vodel and Makiniemi, 1992).

Since their implementation they have been used to
form dynamic mapping of both the brain and hearts
magnetic elds (H am al ainen et al., 1993; Fagaly,
2006). The limitations with SQUIDs come from their
operational requirements, starting with their need for
a cryostat, crucial for the 4 Kelvin cooling of the su-
perconducting materials that make up the Josephson
junctions used in the detector heads. This causes the
overall cost of the SQUIDs to be very high (in the
order of hundreds of thousands of pounds) meaning
there is a place in the market for cheaper alternatives.
It is in the eld of optical magnetometry where
the most promising competition to SQUIDs can be
found (Bloom, 1962; Cohen-Tannoudji et al., 1969;
Nagel et al., 1998; Budker et al., 2000; Bison et al.,
2003; Kominis et al., 2003) and it is in this eld that
our research is based. These methods allow for low
cost, non-invasive, non-contact highly sensitive mag-
netic eld detectors that can be made and operated
at a fraction of the cost of a SQUID. The ability to
measure bio-magnetic signals without contact makes
173
it very useful for fetal cardiac diagnostics (Comani
et al., 2004) as well as the more standard magneto-
cardiographs (MCG) and in the case of burns victims
where contact is not an option.
When making observations of magnetic elds us-
ing atom-light interactions one may start by thinking
of the Zeeman effect. The Zeeman effect explains the
perturbation of atomic energy levels by means of an
applied magnetic eld, which couples to the magnetic
moment of the atomic electrons. Here we still utilise
the Zeeman effect but we need to go further into the
detail of the energy level shifts by discussing the ac-
Stark shift (also called the Autler-Townes effect) as
these are essential to the mechanism needed for mak-
ing our magnetometer more sensitive than a device
utilising just the Zeeman effect.
In this experiment we use a form of coherent pop-
ulation trapping (CPT) (Bel et al., 2007) called Elec-
tromagnetically Induced Transparency (EIT) (Boller
et al., 1991; Scully, 1991; Harris, 1997). EIT is
a quantum interference effect whereby an opaque
medium is made transparent to a probe laser under
certain resonance conditions. To understand how the
many mechanisms that create EIT work together we
shall start by considering a three level system in the
conguration. A depiction of the transition used in
this experiment is displayed in gure 1. In this case
the ground states are degenerate, resulting in a sym-
metric transition so the coupling and probe beams can
be resonant with either m
F
= 1 levels by switching
the sign of the circular polarisation (
+/
).
The intense coupling beam drives the transition
into the excited state; at this point an ac Stark shift
occurs at the upper energy level, whereby the atomic
absorption line splits into an Autler-Townes doublet,
symmetric about the transitions unshifted resonance.
We now consider the processes of EIT in more de-
tail. We are fortunate that a thorough mathematical
understanding of EIT is available allowing us to sim-
ulate the system before building an experiment. This
is presented in the following paragraphs.
The following time dependent interaction Hamil-
tonian describes the atom-light coupling for a system
with EIT;
H
int
=

2
[
p
(t)
eg
1
e
i
1
t
+
c
(t)
eg
2
e
i
2
t
+H.c]
(1)
where the
c/p
(t) are the Rabi frequencies of the
coupling and probe elds respectively,
i j
= |i j| is
the atomic projection operator, is the detuning and
H.c is the Hermitian conjugate. It is this Hamilto-
nian that is inserted into the Master equation and al-
lows us to calculate the atomic density operators ().
The solutions of the Master equation are fundamental
Figure 1: The conguration for our
87
Rb transition show-
ing the possible detuning () of the m
F
levels in the pres-
ence of a coupling and probe beam.
+/
denotes the sign
of the circular polarisation. As the applied magnetic eld
oscillates our system will rotate through these three sys-
tems.
to understanding how the photons are no longer ab-
sorbed at the resonance peak, showing us that when
the coupling and probe beams interact they can cancel
each others absorptive terms producing a dark state by
means of destructive quantum interference. How this
affects the absorption can be seen by looking at the
real and imaginary parts of the linear susceptibility
(), given by;
(
p
,
p
) =
|
13
|
2

N
atom
V

4(||
2
4) 4
2
g
1
g
2
|||
2
+(
eg
1
+i2)(
g
1
g
2
+i2)|
2
+i
8
2

eg
1
+2
g
1
g
2
(||
2
+
g
1
g
2

eg
1
)
|||
2
+(
eg
1
+i2)(
g
1
g
2
+i2)|
2

(2)
where the two photon detuning, =
1

2
=

g
1
g
2
(
p
c) and is derived from the single
photon detuning, =
1
=
eg
1

p
. The coherent
decay between states is , in this case
g1

g2
= 0. The
probe beam term (
e

g2
) is neglected from the equa-
tion as it is assumed it has no observable effect on the
atoms behaviour, this is in part due to the two photon
Raman resonance which occurs in the presence of the
dark states (Fleischhauer et al., 2005).
The linear susceptibility, as seen in equation 2
holds all the important information about the atoms
absorption and refractive index and as these are both
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
174
1 0.5 0 0.5 1
0.02
0.01
0
0.01
0.02
Linear susceptibility () of a system with EIT
/
31
R
e
[

]
1 0.5 0 0.5 1
0
0.01
0.02
0.03
/
31
I
m
[

]
Figure 2: This gure shows rst the real Re[] then the
imaginary Im[] parts of the susceptibility calculated for the
D1 line of
87
Rb using equation 2.
altered in a EIT system it gives us a clear picture of
how the atom-light interaction has changed. Figure 2
displays howall these effects alter the real (Re[]) and
the imaginary (Im[]) part of .
B
V
V
Figure 3: The top line is a representation of Re[] for a
standard saturation spectroscopy dispersion curve, while the
lower is Re[] in the presence of EIT. The gure displays
how the sensitivity to magnetic eld is increased propor-
tionally to the gradient of the dispersion curve, where B is
the amplitude of the input and V is the amplitude of the
response.
The key to why a system with EIT makes for a
good magnetometer is the steep dispersion curve. Fig-
ure 3 shows a representation of the difference that the
steep dispersion curve has on the atoms response to
magnetic elds. Therefore the steeper one can make
the dispersion curve the better the magnetic detector
will be produced from it.
2 EXPERIMENTAL SETUP
The experiment currently runs using free space op-
tics and a diode laser. The laser is tuned to
794.985nm, which is the transition frequency of
the D1 (5
2
S
1/2(F=2)
5
2
P
1/2(F=1)
) line of rubidium 87
(
87
Rb). The
87
Rb is contained within vapour cell with
a buffer gas of, neon at 30Torr. We use
87
Rb as it al-
lows us the run the experiment at room temperature
with low powered laser beams. Figure 4 displays the
apparatus used in this experiment.
The laser is a commercially available diode laser,
from which beams are tapped off and sent to the di-
agnostic and stabilisation section of the experiment.
Contained within this setup is the reference
85/87
Rb
vapour cell, arranged in a saturation spectroscopy
retro-reection conguration. Using this along with
a wavemeter and a lock-in amplier we can lock the
laser to the resonance peak of the transition, stabilis-
ing the laser system.
Laser
Amplifier
Lockin
Wavemeter
Lockin
Oscilloscope
Signal
Generator
Diagnostics
&
Laser Stabilisation
50:50
PBS
50:50
Rb cell
ND filter
Data Collection
/2
/4
ND
filter
Amplifier
3 layer
metal
Rb cell
2 solenoids
Figure 4: A schematic of the apparatus used in this experi-
ment.
The main design of the experiment is a Sagnac in-
terferometer, which starts a single beam that is then
split, these two beams then counterpropagate around
a ring. The beams return to their entry point and leave
the interferometer where they interact and produce in-
terference fringes.
Before the light enters the interferometer it is cou-
pled into a single mode optical bre to clean the mode
MAGNETOMETRY USING ELECTROMAGNETICALLY INDUCED TRANSPARENCY IN A ROOM
TEMPERATURE VAPOUR CELL - Developing an Optical Magnetometer that Utilises the Steep Dispersion Curve
Observed in EIT to Detect Time Varying Magnetic Fields
175
for the experiment. We use a series of optics to pre-
pare the polarisations before splitting the circular po-
larised light with a 50:50 beam splitter into the in-
terferometer. It is here we obtain the
+/
terms as
displayed in gure 1. Before entering the vapour cell
one of the beams is attenuated with a neutral density
lter (ND lter), making it into our weak probe beam.
Two solenoids surround the vapour cell within
three layers of -metal shielding. These are driven by
an arbitrary signal generator. The rst (modulation)
solenoid is driven with a square wave correspond-
ing to approximately 1nT of between 250-300KHz,
which is used as the reference for the signal recovery
lock-in amplier. The second solenoid is also con-
nected to the arbitrary signal generator and is driven
with a sine wave with frequencies between 2-8Hz.
The human heart has a frequency range of 1-1.67Hz,
with higher frequency components.
The raw signal is collected via a photodiode and
then connected to the input of the signal recovery
lock-in amplier. The lock-in amplier allows very
small signals to be extracted from the large amounts
of noise, using this system we have gained sensitiv-
ities in the range of a fetal heart beat. An example
of test data taken using a simulated heartbeat is dis-
played in gure 5.
Figure 5: A sceen dump of data take on the experiment.
The centre trace is the raw input signal, a simulated QRS,
the lower trace is the raw data from the photodiode and the
upper trace is the 20 trace avarage.
3 CONCLUSIONS
With current data we can measure magnetic elds of
the order of a fetal heart beat. The experiment is
still at a very early stage of development, therefore
the current data acquisition method is not suitable for
clinical applications. However even at this stage the
experiment has shown a great deal of promise as a
potential MCG device. We expect, with some calibra-
tion, to obtain at least femtotelsa sensitivities (Fleis-
chhauer and Scully, 1994), with the possibility of us-
ing more optical bres and replacing the modulation
solenoid with an alternative scheme such as an acous-
tic or electro-optic modulator (AOM/EOM). These
improvements will move this research closer to being
a competitive device for performing clinical MCG tri-
als.
ACKNOWLEDGEMENTS
This work is funded by an EPSRC DTA.
REFERENCES
Bel, J., Bevilacqua, G., Biancalana, V., Dancheva, Y., and
Moi, L. (2007). All optical sensor for automated mag-
netometry based on coherent population trapping.
Bison, G., Wynands, R., and Weis, A. (2003). Dynami-
cal mapping of the human cardiomagnetic eld with a
room-temperature, laser-optical sensor. Opt. Express,
11(8):904909.
Bloom, A. L. (1962). Principles of operation of the rubid-
ium vapor magnetometer. Appl. Opt., 1(1):6168.
Boller, K. J., Imamolu, A., and Harris, S. E. (1991). Obser-
vation of electromagnetically induced transparency.
Physical Review Letters, 66(20):2593+.
Budker, D., Kimball, D. F., Rochester, S. M., Yashchuk,
V. V., and Zolotorev, M. (2000). Sensitive magne-
tometry based on nonlinear magneto-optical rotation.
Physical Review A, 62(4):043403+.
Cohen-Tannoudji, C., Dupont-Roc, J., Haroche, S., and
Lalo e, F. (1969). Detection of the static magnetic eld
produced by the oriented nuclei of optically pumped
he3 gas. Physical Review Letters, 22(15):758+.
Comani, S., Mantini, D., Alleva, G., Di, L. S., and Romani,
G. L. (2004). Fetal magnetocardiographic mapping
using independent component analysis. Institute of
Physics Publishing Physiological Measurement.
Fagaly, R. L. (2006). Superconducting quantum interfer-
ence device instruments and applications. Review of
Scientic Instruments, 77(10).
Fenici, R., Romani, G., and Ern e, S. (1983). High-
resolution magnetic measurements of human cardiac
electrophysiological events. Il Nuovo Cimento D,
2(2):231247.
Fleischhauer, M., Imamoglu, A., and Marangos, J. P.
(2005). Electromagnetically induced transparency:
Optics in coherent media. Reviews of Modern Physics,
77(2).
Fleischhauer, M. and Scully, M. O. (1994). Quantum
sensitivity limits of an optical magnetometer based
on atomic phase coherence. Physical Review A,
49(3):1973+.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
176
H am al ainen, M., Hari, R., Ilmoniemi, R. J., Knuutila, J.,
and Lounasmaa, O. V. (1993). Magnetoencephalog-
raphy&#151;theory, instrumentation, and applications
to noninvasive studies of the working human brain.
Reviews of Modern Physics, 65(2):413+.
Harris, S. E. (1997). Electromagnetically induced trans-
parency. In Quantum Electronics and Laser Science
Conference, 1997. QELS 97., Summaries of Papers
Presented at the, page 25.
Kominis, I. K., Kornack, T. W., Allred, J. C., and Romalis,
M. V. (2003). A subfemtotesla multichannel atomic
magnetometer. Nature, 422(6932):596599.
Nagel, A., Graf, L., Naumov, A., Mariotti, E., Biancalana,
V., Meschede, D., and Wynands, R. (1998). Experi-
mental realization of coherent dark-state magnetome-
ters. Europhysics Letters, pages 3136.
Nash, M. P., Bradley, C. P., Kardos, A., Pullan, A. J., and
Paterson, D. J. (2002). An experimental model to cor-
relate simultaneous body surface and epicardial elec-
tropotential recordings in vivo. Chaos, Solitons &
Fractals, 13(8):17351742.
Ramanathan, C., Ghanem, R. N., Jia, P., Ryu, K., and Rudy,
Y. (2004). Noninvasive electrocardiographic imag-
ing for cardiac electrophysiology and arrhythmia. Nat
Med, 10(4):422428.
Scully, M. O. (1991). Enhancement of the index of refrac-
tion via quantum coherence. Physical Review Letters,
67(14):1855+.
Vodel, W. and Makiniemi, K. (1992). An ultra low noise dc
squid system for biomagnetic research. Measurement
Science and Technology, 3(12):11551160.
MAGNETOMETRY USING ELECTROMAGNETICALLY INDUCED TRANSPARENCY IN A ROOM
TEMPERATURE VAPOUR CELL - Developing an Optical Magnetometer that Utilises the Steep Dispersion Curve
Observed in EIT to Detect Time Varying Magnetic Fields
177
DEVELOPMENT OF AN ELECTRICAL STIMULATION DEVICE
FOR OSSEOINTEGRATED AMPUTEES
A Novel Approach for Expediting Skeletal Attachment and Rehabilitation
Brad Isaacson
1,2
, J eroen Stinstra
3
, Rob MacLeod
2,3
and Roy Bloebaum
1,2,4

1
Department of Veteran Affairs, Salt Lake City, UT, U.S.A.
2
Department of Bioengineering, Univerisity of Utah, Salt Lake City, UT, U.S.A.
3
Scientific Computing Institute, University of Utah, Salt Lake City, UT, U.S.A.
4
Department of Orthopedics, University of Utah, Salt Lake City, UT, U.S.A.
Keywords: Osseointegration, Electrical stimulation, Osteogenesis, Percutaneous, Amputation.
Abstract: The projected number of American amputees is expected to rise to 3.6 million by 2050. Many of these
individuals depend on artificial limbs to perform routine activities, but prosthetic suspensions using
traditional socket technology can prove to be cumbersome and uncomfortable for a person with limb loss.
Moreover, for those with high proximal amputations, limited residual limb length may prevent
exoprosthesis attachment all together. Osseointegration technology is a novel operative procedure that
allows integration between host tissue and an orthopaedic implant and has been shown to improve clinical
outcomes by allowing direct transfer of loads to a bone-implant interface. However, the associated surgical
procedures require long rehabilitation programs that may be reduced through expedited skeletal attachment
via electrical stimulation. To determine optimal electrode size and placement, we have developed a system
for computational modeling of the electric fields that arise during electrical stimulation of residual limbs.
Three patients with retrospective CT scans were selected and three dimensional reconstructions were
created using customized software (Seg3D and SCIRun). These software packages supported the
development of patient specific models and allowed for interactive manipulation of electrode position and
size; all variables that could affect the electric fields around a percutaneous osseointegrated implant.
Preliminary results of the electric fields at the implant interface support the need for patient specific
modeling in order to achieve the homogenous electric field distribution required to induce osteoblast
migration and enhance skeletal fixation.
1 INTRODUCTION
Osseointegration implant technology is a novel
surgical procedure that provides direct skeletal
attachment between an implant and host tissue and
can significantly increase the quality of life for
amputees (Albrektsson & Albrektsson, 1987;
Branemark, 1983). However, one challenge with
using natural biological fixation is attaining a strong
skeletal interlock at the implant interface, a
prerequisite for long-term implant function
(Albrektsson, Branemark, Hansson, & Lindstrom,
1981). Therefore, we propose a means to accelerate
osteogenesis through external electrical stimulation
and present a simulation approach for which to
develop such a system.
Veterans with combat related injuries form an
especially relevant population that requires the
development of new tools to enhance the success of
osseointegration, due to their limited residual limbs
caused by explosive devices. Improvements in
medical care and evacuation strategies have led to an
increase in survival rates, resulting in an elevated
number of veterans with amputations that require
follow-up care and extensive rehabilitation. The
relative youth and otherwise good health of these
amputees make them an ideal population for
aggressive rehabilitation but also reveal the
limitations of current technologies of prosthetic
attachment (Hagberg & Branemark, 2001). Physical
limitations with warrior amputees using sockets
include heat/sweating in the prosthetic socket, skin
irritation and inability to walk on challenging terrain
(Hagberg & Branemark, 2001). In addition, a
significant number of returning service men and
178
women have short residual limbs for which socket
technology is not suitable.
Utilizing metallic implants as a means of
biological fixation has been the objective of
orthopedic surgeons over the past two centuries
(Williams, 1982). However, controlling osteogenesis
at the implant interface, which is essential for
providing strong skeletal fixation, remains
challenging. Regulated electrical stimulation has
proven effective in fracture healing and non-
traumatized bone models (Brighton, 1981;
Friedenberg, Zemsky, Pollis, & Brighton, 1974), but
has not been investigated in a percutaneous
osseointegrated implant system. One advantage of
this patient population is that an orthopedic implant
protrudes from the residual limb functioning as an
exoprosthesis attachment and a potential cathode for
an external electrical stimulation device.
By understanding the method of current injection
with varying electrode size and placements, an
electric field on the magnitude of 1-10 V/cm may be
established at the implant interface, capable of
inducing osteoblast migration and improving
skeletal attachment (Ferrier, Ross, Kanehisa, &
Aubin, 1986). An electric field of this degree may
increase the quantity and quality of bone at the
implant interface, and thus would improve the
prospects for accelerated rehabilitation and skeletal
fixation for an amputee. Figure 1 contains a
schematic of a stimulation apparatus that we
envision.





Figure 1: Representative model of the electrical
stimulation device proposed for amputees with a
percutaneous osseointegrated implant.
The initial approach to develop an electrical
stimulation osseointegration system is to simulate
electric field strength on a patient specific basis
using a computational model. Because variations in
electrode size and placement could inhibit or
expedite bone growth, the modeling approach
includes both the electrode placement for
stimulation at the skin as well as patient specific
anatomy to determine an accurate estimate of the
underlying electrical simulation fields.
The objective of our research is to develop an
alternative means of prosthetic suspension superior
to the traditional socket method for patients with
amputations. This goal is strongly motivated by the
growing rate of amputations occurring annually
from vascular occlusive diseases, diabetes, and
traumatic injury. While an amputation can be a life
saving surgical procedure, many interpret the
operation as a significant loss, which alters the
perception of an individual of his or her degree of
physical, psychological and social well-being and
the effects that illness and treatment have on daily
life (Hagberg & Branemark, 2001). We
demonstrate here the feasibility of a computational
approach to developing such a system.
2 METHODS
2.1 Overview
In order to simulate the electric field that is present
at the bone-implant interface, three patient specific
models were created. A volume conductor model
was developed to compute realistic electric fields
from computer tomography (CT) scans of the limbs
of these patients by assigning tissue conductivities
during segmentation. Using this model and
assuming that the electrical field can be calculated
using a quasi-static approach, the electrical potential
was computed by solving Laplaces equation for
each tissue type.
0 = (1)
In this model the boundary conditions were
formed by the electrodes that injected currents and
the guideline that current remained within the body.
Since the electrodes and the implant had a much
larger conductivity than the surrounding tissues, it
was assumed that the implant (cathode) was at a
constant potential, likewise the surface electrodes
were modeled with a constant potential difference
from the percutaneous implant.
Numerical simulation was used to compute the
electric potential through the CT scans of the
patients residual limb. To evaluate the efficacy of
electrode configuration and sizing, patient specific
models were developed and the electrical potential
around the implant interface was used to determine
localized electric field strengths.


MedullaryCanal




DMM
Cathode
Anode
Rk
Power
Supply


Residual Limb
Muscle
Implant
Polyurethanefoamsaturatedwithsalinesolution
Electrode
DEVELOPMENT OF AN ELECTRICAL STIMULATION DEVICE FOR OSSEOINTEGRATED AMPUTEES - A Novel
Approach for Expediting Skeletal Attachment and Rehabilitation
179
2.2 Image Acquisition
CT images were obtained retrospectively from the
radiology department at the University of Utah in
accordance with Institutional Review Board (IRB)
approval. Femoral slice thicknesses ranged from 600
m to 1 mm and of the 50 patients examined, 3 were
selected based on predetermined demographics and
absence of metallic implants, which cause image
artifacts. Table 1 lists patient specifics.
Table 1: Patient Demographics.

To determine the variability amongst patients,
only one amputee was selected from the population
(Patient 3). The remaining two models were
generated from subjects in the general population
who were made into artificial amputees from
segmentation using computer software. To account
for natural anatomical differences in patient limbs,
wide variation in age (SD =26.2), height (SD =
17.8) and weight (SD =18.7) were selected.
Establishing accurate tissue differentiation was
performed using the Seg3D (www.seg3d.org)
software. The tissue boundaries of the bone, bone
marrow and adipose tissue were generated by
thresholding the CT images interactively. The
musculature was obtained by manually setting seed
points inside the muscle tissue and using a
confidence connected filter to find all the tissue
connected to the seed points. Finally, the skin, which
was impossible to discern reliably from the CT
images, was generated by dilating the outermost
tissue 2 millimeters based on average skin thickness
to produce a layer of homogeneous thickness that
surrounded the full model (Tortora & Nielsen,
2008). Segmentations were manually inspected,
corrected to ensure accuracy and combined in a
hierarchy into a single label map required for finite
element analysis. An example of a segmentation,
consisting of skin, adipose tissue, muscle, bone and
bone marrow is depicted in Figure 2.
2.3 Electrode Placement & Design
Since there are no preliminary results using such a
system for a percutaneous osseointegrated implant,
we selected four widely variable electrode
configurations for testing. Small electrodes were
designed with patient compliance in mind, since the
device should not restrict ambulation or daily
activity.


Figure 2: Hierarchical label maps constructed in Seg3D
for Patient 2. The completed femoral mapping was
composed of skin (A), adipose tissue (B), musculature
(C), bone (D) and bone marrow (D).
The SCIRun (software.sci.utah.edu) software
package was utilized for electrode design because it
supports interactive electrode placement and
simulation. The configurations selected consisted of
a one patch electrode, two patch electrodes, one
continuous band and two continuous bands (Figure
3). External electrode bands were applied on the
residual limb of the patient and were 1.6 cm in
thickness. Electrode patches were placed as a strip
covering approximately half the diameter of the leg
and were 3 cm in thickness. Lastly, the percutaneous
implants were set to match the size of the cavity of
the bone marrow to represent proper implant fit and
fill, since gaps in excess of 50 m may lead to
fibrous encapsulation without bone ingrowth
(Bloebaum, Bachus, Momberger, & Hofmann, 1994;
Hofmann, Bachus, & Bloebaum, 1993).







Figure 3: Electrode configurations modeled to determine
optimal performance for amputees. The electrode
configuration is shown from patient 2. A two patch setup
was changed to a one patch setup by removing the medial
electrode (A). A double band electrode was altered to a
single band configuration by removing one band and
centering it amongst the implant area (B). Electrode size
and position are illustrated with arrows.
Patient Sex Amputee Age
Height
[cm]
Weight
[kg]
1 M No 60 185.4 79.9
2 F No 28 157.5 50.1
3 F Yes 80 152.4 45.5
A B
C D
B A
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
180
2.4 Finite Element Analysis
In order to predict electric fields from exogenous
voltage potentials, label maps and electrodes were
modeled using a hexahedral mesh that consisted of
approximately 1.7 million elements. The elements
were treated as piecewise homogenous, ohmic and
isotropic, and were assigned conductivities using
previously published values and estimations by the
investigators (Chiu & Stuchly, 2005; Gabriel, Lau,
& Gabriel, 1996; Stinstra et al., 2007) (Table 2).
Because tissue behaves inherently electrolytic,
treating the models with DC conductivities was
considered to be an important factor (Grimnes &
Martinsen 2008).
Electrodes were incorporated in the finite element
meshing and assigned a constant potential difference
of 9 volts between the skin electrode and
osseointegrated implant, a selection based on
expected tissue resistivity. Using an iterative solver,
the potentials in the finite element models were
computed for the three patients and four electrode
configurations.
Table 2: Conductivity values assigned to segmentations.
Tissue Type Conductivities [S/m]
Skin 0.26
Muscle 0.25
Adipose 0.09
Organ 0.22
Cortical Bone 0.02
Bone Marrow 0.07
2.5 Data Analysis
In order for an electrode configuration to be deemed
acceptable, a uniform homogenous field around the
implant interface was required for bone growth.
Therefore, histograms were computed for about
6000 elements in the immediate area encompassing
the bone implant interface. The results were
analyzed using maps of electric field strength that
show the variability of the electric field within the
leg.
3 RESULTS
Interactive placement of electrodes allowed for
various computational simulations. Figure 4
illustrates one example of the differences between
patients. The figure depicts cross sections for three
different patients, where the color scale indicates the
strength of the local field. The corresponding
histogram (right) represents the electric field
strength of the 6000 elements surrounding the
implant site and shows the homogeneity of the field.
The histograms showed a broad variation among
patients, some producing normal Gaussian
distributions and others with broad or skewed peaks
and thus a higher than random level of homogeneity.
The complete set of histograms used for analysis is
listed subsequently in Figure 5.
4 DISCUSSION
The necessity for patient specific models with a
percutaneous electrical stimulation device was
confirmed in the study. The distribution of electrical
potentials at the implant-bone interface varied across
subjects due to variations in anatomy and the
presence of an amputation. While creating artificial
amputees using a segmentation program was
straightforward and permitted robust computation,
histograms of electric field strength confirmed that
electrical metrics changed dramatically when
compared to a known amputee. While the shape of
the histograms tended to be bell shaped, the position
of the peaks were often skewed depending on the
electrode configuration and patient. The results
clearly showed that the 1 patch electrode generates
the smallest electric field in the bone-implant
interface, while the 2 band electrode configuration
generated the highest field for the same applied
potential, suggesting that proper electrode placement
could improve efficiency.
Due to the limited quantity of patients in the
study, a strong correlation was not established
relating patient demographics with voltage
potentials. However, the highest voltage gradients
mapped during simulations were consistently from
subject 2, a patient who was in the best physical
condition. The increased electrical field was likely
caused by the reduction in the diameter and
thickness of adipose tissue in the subjects residual
limb, since adipose tissue would raise resistivity and
impede current flow.
While an optimal electrical configuration may
not have been established for the patient population
collectively, two bands appeared to produce the
most homogenous electrical field distributions
between 1-10 V/cm. Minor adjustments would be
required if the device were used clinically to account
for the varying anatomy of patients, spatial location
of topical electrodes and may be confirmed with CT
files and computational modeling.

DEVELOPMENT OF AN ELECTRICAL STIMULATION DEVICE FOR OSSEOINTEGRATED AMPUTEES - A Novel
Approach for Expediting Skeletal Attachment and Rehabilitation
181
ELECTRIC FIELD





Figure 4: Sample distributions of the electric field strength surrounding the implant. The color map reflects the strength of
the electric field in a cross section through the lower part of the limb. External electrode placements were illustrated as
rectangular objects on the outside of the residual limb. The results were for an 18 centimeter percutaneous implant that was
interactively inserted into the medullary canal of adult patients. The histograms on the right represent the distribution of the
electric field strength in the volume next to the implant.
While the initial target of the exogenous
electrical stimulation system utilizes an orthopedic
implant as a functional cathode, it may also reduce
the potential for superficial and deep infections by
preventing additional surgical procedures to remove
implanted devices as seen with older fracture
healing models (Lavine & Grodzinsky, 1987).
However, the success of the system is dependent on
numerous factors including hydration levels,
quantity of soft tissue and the material type selected
for the orthopedic implant. In order for this novel
technology to be beneficial, a balance must be
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
182
Electrode Type Patient 1 Patient 2 Patient 3
1 Band Electrode

2 Band Electrodes
1 Patch Electrode

2 Patch Electrodes
Figure 5: Comparison between patients and electrode configurations. The results confirm the requirement for individual
patient modeling. Distributions of electric fields were not homogenous in each case and would require manipulation of the
applied voltage potential to attain uniform bone ingrowth.
attained between obtaining a desired electric field
and the host tissue reaction which may occur with
varying implanted metals. Titanium alloy was
selected as the cathode in this model (3x10
6
S/m)
since it is regarded as the one of the most
biocompatible material type for total joint
replacements and has low thermal conductivity to
protect tissue necrosis from heat generation (Agins
et al., 1988; Beder & Eade, 1956; Emneus &
Gudmundsson, 1967). However, if clinicians and
engineers require an altered rate of electron
propagation, the material can be exchanged
(Grimnes & Martinsen, 2008) or porosity altered but
careful attention should be paid to ensure the
material does not illicit a foreign body response.
Utilizing electrical stimulation for older
amputees is a critical aspect which must be explored
as well. Bone mass is maximum a decade after
skeletal growth ceases but decreases significantly by
the eighth and ninth decade (Buckwalter, Glimcher,
Cooper, & Recker, 1995). As long bones change
confirmation with age, the endosteal diameter tends
to increase more rapidly than the periosteal diameter
which could lead to implant loosening (Lane &
Vigorita, 1983). This problem coupled with the
reduction of strain on bones by weaker muscles may
DEVELOPMENT OF AN ELECTRICAL STIMULATION DEVICE FOR OSSEOINTEGRATED AMPUTEES - A Novel
Approach for Expediting Skeletal Attachment and Rehabilitation
183
contribute to debilitating diseases such as
osteoporosis and osteopenia (Lane & Vigorita,
1983) and require additional treatment options.
However, controlled electrical stimulation and
mechanical loading may act as a synergistic catalyst
of bone ingrowth (Spadaro, 1997) and maintain host
bone bed integrity with elderly patients using an
osseointegrated electrical implant system.
Establishing tools for enhancing skeletal
attachment may assist with reducing the length of
rehabilitation required for an osseointegrated
procedure. Current programs require 2 to 24 months
of rehabilitation (Branemark, Branemark, Rydevik,
& Myers, 2001), a lengthy time period to ensure
uniform ingrowth. Because of the slow biological
process of skeletal attachment, loading at the
implant interface would be restricted for several
months after the operative procedure (Hofmann,
Bloebaum, & Bachus, 1997). However, using an
electrical stimulation system may enhance ingrowth
and allow patients to return to earlier ambulation.
4.1 Limitation
Since the conductivity of a titanium implant
significantly exceeded that of cortical bone, the
current densities at the implant interface should be
modeled to ensure localized tissue heating does not
occur, which may lead to patient discomfort or
potential tissue necrosis. Computational modeling of
current density fields are attainable with the given
software package and will be utilized in future work.
Additional efforts with be paid to altering the
porosity of the titanium implant and determining the
effect on the predicted electric fields since porosity
and conductivity are inversely related and may
affect the model as well (Ke et al., 2007).
5 CONCLUSIONS
The simulations developed for the proposed
biomedical device may have the capabilities of
expediting skeletal attachment by increasing
osteoblast migration. Computation modeling has
effectively shown that 1-10 V/cm electric fields may
be generated using the implant as a functional
cathode and topically applied anode band and
patches. Implementing computational models may
be the first step to resolving the classic problem with
electrical stimulation which is the inability to define
current pathways in the human body (Chakkalakal &
J ohnson, 1981; Noda & Sato, 1985).
Patient specific modeling was effective for
attaining values that may be osteogenic at the
implant site, but wide variations in electric field
distributions shown in histograms reaffirm the need
to evaluate each case specifically. However, in order
to determine the accuracy of finite element analysis,
the quantity of subjects will be increased in the
future work to determine if an electrode
configuration could be optimized for patients with
percutaneous osseointegrated implants. Additional
model validation of electrically enhanced
osseointegration will be assessed using a small in
vivo animal model based on computational evidence
in future work.
ACKNOWLEDGEMENTS
This material is based upon research supported by
the Office of Research and Development,
Rehabilitation R&D Service, DVA SLC Health Care
System, Salt Lake City, Utah; the Albert & Margaret
Hofmann Chair and the Department of
Orthopaedics, University of Utah School of
Medicine, Salt Lake City, Utah. Additional technical
support for the simulations was provided by the
Center for Integrative Biomedical Computing of
Scientific Computing and Imaging Institute and was
made possible in part by software from the
NIH/NCRR Center for Integrative Biomedical
Computing, P41-RR12553-07.
REFERENCES
Agins, H. J ., Alcock, N. W., Bansal, M., Salvati, E. A.,
Wilson, P. D., Pellicei, P. M., et al. (1988). Metallic
wear in failed titanium-alloy total hip replacements. J.
Bone Joint Surg. [Am.], 70-A(3), 347-356.
Albrektsson, T., Branemark, I.-I., Hansson, H.-A., &
Lindstrom, J . (1981). Osseointegrated titanium
implants. Acta Orthop Scand, 52, 155-170.
Albrektsson, T., & Albrektsson, B. (1987).
Osseointegration of bone implants. A review of an
alternative mode of fixation. Acta Orthop Scand,
58(5), 567-577.
Beder, O. E., & Eade, G. (1956). An investigation of
tissue tolerance to titanium metal implants in dogs.
Surgery, 39(3), 470-473.
Bloebaum, R. D., Bachus, K. N., Momberger, N. G., &
Hofmann, A. A. (1994). Mineral apposition rates of
human cancellous bone at the interface of porous
coated implants. J Biomed Mater Res, 28(5), 537-544.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
184
Branemark, P. I. (1983). Osseointegration and its
experimental background. J Prosthet Dent, 50(3), 399-
410.
Branemark, R., Branemark, P. I., Rydevik, B., & Myers,
R. R. (2001). Osseointegration in skeletal
reconstruction and rehabilitation: a review. J Rehabil
Res Dev, 38(2), 175-181.
Brighton, C. T. (1981). The treatment of non-unions with
electricity. J Bone Joint Surg Am, 63(5), 847-851.
Buckwalter, J . A., Glimcher, M. J ., Cooper, R. R., &
Recker, R. (1995). Bone biology. J Bone Joint Surg
Am, 77(2), 1276-1289.
Chakkalakal, D. A., & J ohnson, M. W. (1981). Electrical
properties of compact bone. Clin Orthop Relat
Res(161), 133-145.
Chiu, R. S., & Stuchly, M. A. (2005). Electric fields in
bone marrow substructures at power-line frequencies.
IEEE Trans Biomed Eng, 52(6), 1103-1109.
Emneus, H., & Gudmundsson, G. (1967). Final report on
clinical testing of titanium. Acta Orthop Scand, 372-
373.
Ferrier, J ., Ross, S. M., Kanehisa, J., & Aubin, J . E.
(1986). Osteoclasts and osteoblasts migrate in
opposite directions in response to a constant electrical
field. J Cell Physiol, 129(3), 283-288.
Friedenberg, Z. B., Zemsky, L. M., Pollis, R. P., &
Brighton, C. T. (1974). The response of non-
traumatized bone to direct current. J Bone Joint Surg
Am, 56(5), 1023-1030.
Gabriel, S., Lau, R. W., & Gabriel, C. (1996). The
dielectric properties of biological tissues: III.
Parametric models for the dielectric spectrum of
tissues. Phys Med Biol, 41(11), 2271-2293.
Grimnes, S. & Martinsen O. (2008). Bioimpedance and
Bioelectricity Basics. Amersterdam: Academic Press.
Hagberg, K., & Branemark, R. (2001). Consequences of
non-vascular trans-femoral amputation: a survey of
quality of life, prosthetic use and problems. Prosthet
Orthot Int, 25(3), 186-194.
Hofmann, A. A., Bachus, K. N., & Bloebaum, R. D.
(1993). Comparative study of human cancellous bone
remodeling to titanium and hydroxyapatite coated
implants. J. Arthroplasty, 8(2), 157-166.
Hofmann, A. A., Bloebaum, R. D., & Bachus, K. N.
(1997). Progression of human bone ingrowth into
porous-coated implants. Acta Orthop. Scand., 68(2),
161-166.
Ke, Z., Cheng-Feng, L., & Zhen-Gang, Z. (2007).
Measurement of Electrical Conductivity of Porous
Titanium and Ti6Al4V Prepared by the Powder
Metallurgy Method.. Chin. Phys. Lett., 24(1), 187-190.
Lane, J . M., & Vigorita, V. J . (1983). Osteoporosis. J
Bone Joint Surg Am, 65(2), 274-278.
Lavine, L. S., & Grodzinsky, A. J . (1987). Electrical
stimulation of repair of bone. J Bone Joint Surg Am,
69(4), 626-630.
Noda, M., & Sato, A. (1985). Appearance of osteoclasts
and osteoblasts in electrically stimulated bones
cultured on chorioallantoic membranes. Clin Orthop
Relat Res (193), 288-298.
Spadaro, J . A. (1997). Mechanical and electrical
interactions in bone remodeling. Bioelectromagnetics,
18(3), 193-202.
Stinstra, J . G., J olley, M., Callahan, M., Weinstein, D.,
Cole, M., Brooks, D. H., et al. (2007). Evaluation of
different meshing algorithms in the computation of
defibrillation thresholds in children. IEEE Engineering
in Medicine and Biology Conference.
Tortora, J . & Nielsen M. (2008). Principles of Human
Anatomy (11
th
ed). United States: J ohn Wiley & Sons.
Williams, D. F. (1982). Biocompatibility of Orthopedic
Implants. Volume I. In D. F. Williams (Ed.), CRC
Series in Biocompatibility (pp. 141-195). Liverpool:
CRC Press, Inc.
DEVELOPMENT OF AN ELECTRICAL STIMULATION DEVICE FOR OSSEOINTEGRATED AMPUTEES - A Novel
Approach for Expediting Skeletal Attachment and Rehabilitation
185
BRAIN COMPUTER INTERFACE
Feedback Effect Analysis by Comparison of Discrimination Capability of On-line
and Off-line Experimental Procedures based on LDA
Jos e Luis Martnez P erez and Antonio Barrientos Cruz
Grupo de Rob otica y Cibern etica, Universidad Polit ecnica de Madrid, C/Jos e Gutierrez Abascal 2, Madrid, Spain
[email protected], [email protected]
Keywords:
Electroencephalography, Brain Computer Interface, Linear Discriminant Analysis, Spectral Analysis,
Biomedical signal detection, Pattern recognition.
Abstract:
This paper analyses the users feedback inuence in the mental task discrimination capability through the
comparison of results obtained from Off-line and On-line Brain Computer Interface experimental procedures.
Experiments performed under these two paradigms were carried out by ve male volunteers. In order
to develop a wearable BCI device only two electrodes in C3 and C4 zones have been used for electro-
encephalographic signal acquisition. These procedures apply seven different types of preprocessing windows
and Linear Discrimination Analysis technique to reduce the dimension of the feature space before the quan-
tication of the discrimination capability between the proposed mental activities.The discrimination capability
is quantied through statistical analysis, based on bilateral contrast test, between the population of the LDA
transformed feature vectors.
1 INTRODUCTION
The objective of Brain Computer Interface technol-
ogy is the direct communication of users mind with
external devices, it uses the encephalographic signal
as primary source of commands for the external de-
vices (Wolpaw et al. 2000), (Birbaumer, N. et al.,
2000), (Wolpaw, 2007). A variety of methods for
monitoring brain activity might serve in BCI tech-
nology: electroencephalography (EEG), magnetoen-
cephalography (MEG), positron emission tomogra-
phy (PET), functional magnetic resonance imaging
(fMRI), and optical imaging. At present, only EEG
meets the requirements of short time constant, afford-
able cost, and it is relatively simple to implement.
In order to control an external device using
thoughts, it is necessary to associate some mental pat-
terns to device commands, so an algorithm that de-
tects, acquires, lters and classies the human elec-
troencephalographic signal is required (Wolpaw et
al., 2002), (Vidal, 1973), (Kostov and Polak, 2000),
(Pfurtscheller et al. 2000b). Usually all BCI systems
are compounded of the following blocks:
Signal Acquisition. In this block the signal is ac-
quired by the recording electrodes, amplied, and
digitalised. BCI devices can be categorised by
the different approaches they use for the signal
acquisition: non-invasive recordings with stan-
dard scalp electrodes, and invasive recording with
epidural, subdural, or intracortical electrodes.
Signal Processing: Feature Extraction. The dig-
italised signals are subjected to feature extrac-
tion procedures, such as spatial ltering, volt-
age amplitude measurements, spectral analysis, or
single-neuron separation (Lopes da Silva, 1999).
Signal Processing: The Translation Algorithm.
It translates the signal features into device
commands-orders that carry out the users intent.
The Output Device. Generally the output device
is a computer screen and the output is the selec-
tion of targets, letters, or icons presented on it.
Initial studies are exploring BCI control of a neu-
roprothesis that provides hand closure to people
with cervical spinal cord injuries(Pfurtscheller et
al., 2000a).
The Operating Protocol. It is the protocol that
guides the operation of the BCI device.
BCI devices fall into two classes: dependent and
independent (Chiappa, 2006). A dependent BCI does
not use the brains normal output pathways to carry
the message, but activity in these path-ways is needed
to generate the brain activity that does carry it. An
186
independent BCI does not depend in any way on the
brains normal output pathways.
This paper focus on the users feedback inu-
ence in the discrimination capability of three different
mental activities, it analyses the applicability of LDA
to BCI and how the windowing effect affects the dis-
crimination capability of the brain proposed activities.
Section 2 describes the Off-line and On-line ex-
perimental procedures applied to evaluate the users
feedback inuence (Martinez and Barrientos, 2007);
because the main changes in brain activity are associ-
ated to changes in the power amplitude of frequency
bands, spectrograms based on FFT are used to obtain
initial feature vectors. To minimise the leakage effect
seven types of preprocessing windows has been con-
sidered: rectangular, triangular, Blackmans, Ham-
mings, Hannings, Kaisers and Tukeys (Proakis and
Manolakis, 1997), (Allen and Rabiner, 1977). The
evidence of statistical difference in the feature popu-
lations associated to different brain activities has been
previously shown (Martinez and Barrientos, 2006). In
the experiments considered for this report a low num-
ber of scalp-electrodes has been used to capture the
electroencephalographic signal, in order to facilitate
the use of this technology it is important to make it
easy to use, as the fewer of electrodes used, the higher
the comfort, (Wolpaw, 2007).
Section 3 describes the LDA technique used to
combine these initial features in order to reduce the
dimensionality of the input space (Ripley, 2000).
Section 4 explains the bilateral contrast test used
to determine the discrimination power between the
proposed cerebral activities and the effect of the pre-
processing windows, the results of each contrast is
both qualitative and quantitative, qualitative in or-
der to accept or reject the null hypothesis of equal-
ity in the population of features, quantitative in order
to compare the discrimination power through signi-
cance contrast level =1 p =2.5%.
Sections 5 and 6 present and analyse the results.
Section 7 is devoted to conclusions.
2 EXPERIMENTAL
PROCEDURES
Off-line and On-line tests were carried out on ve
healthy male subjects, one of them has been trained
before, but the other four were novice in the use of
the system. The Off-line tests have been carried out
before On-line tests in order to have data to allow the
training procedure of a simple classier. The subjects
were sat down in front of the acquisition system mon-
itor, at 50 cm from the screen, their hands were in
Figure 1: Diagram of the Off-line experiment realization.
a visible position, the supervisor of the experiments
controlled the correct development of them (Neuper,
2001), (Penny et al., 2000).
2.1 Procedure for Off-line Experiments
The experimental Off-line process is shown on g.1.
Test of system devices. Checks the correct level of
battery, and the correct state of the electrodes.
System assembly. Device connections: super-
cial electrodes (Grass Au-Cu), battery, bio-amplier
(g.BSamp by g.tec), acquisition signal card (PCI-
MIO-16/E-4 by National Instrument), computer.
System test. Veries the correct operation of the
whole system.
Subject preparation for the experiment. Applica-
tion of electrodes on subjects head. It is veried that
electrode impedance was lower than 4 KOhms.
System initialisation and Experiment setup. Ver-
ication of data register. The supervisor sets-up the
number of replications, N
rep
= 10, and the quantity
of different mental activities, N
act
= 3. The duration
of each mental activity, a trial, is t = 7s, the acquisi-
tion frequency is f
s
= 384Hz. The system randomly
suggests the mental activity to think about.
2.2 Procedure for On-line Experiments
In these tests, a cursor in the centre of the screen and
a square goal are shown to the subject, the square goal
appears half the trials on the left of the screen and the
other half on the right. The subject shall try to move
the cursor towards the goal thinking in the cerebral ac-
BRAIN COMPUTER INTERFACE - Feedback Effect Analysis by Comparison of Discrimination Capability of On-line
and Off-line Experimental Procedures based on LDA
187
Figure 2: Diagram of the On-line experiment realization.
tivities proposed in the Off-line experiments.The ex-
perimental On-line process is shown on g.2.
Experiment set-up. In this phase it is determined
which cerebral activities are used to move the cursor
to the left and to the right, the number of trials and the
time for each trial.
Display initialisation. It initialises the display, for
even trials the goal is on the right, for odds on the left.
Data acquisition. In this phase 128 samples per
channel are acquired at f s =384Hz.
Record samples. The previous samples are recorded
for a posterior analysis.
Feature extraction. A vector of features is extracted
from the acquired samples.
Classication. The vector of features is classied as
belonging to one of the previous cerebral activities,
and the associated movement is performed.
Figure 3: Electrode placement.
2.3 Position of the Electrodes and
Description of Cerebral Activities
For both types of experimental procedures, the elec-
trodes were placed in the central zone of the skull,
next to C3 and C4, two pair of electrodes were placed
in front of and behind of Rolandic sulcus, this zone is
one with the highest discriminant power, it takes sig-
nal from motor and sensory areas of the brain (Penny
et al., 2000),(Pfurtscheller et al. 2000b). Reference
electrode was placed on the right mastoid, two more
electrode are placed near to the corner of the eyes to
register blinking.
The supervisor of the experiment asks the subject
to gure out the following mental activities, these ac-
tivities will be the tasks to differentiate among them.
Activity A. Mathematical task. Recursive subtraction
of a prime number from a big quantity.
Activity B. Motor imagery. The subject imagines
moving their limbs or hands, but without the mate-
rialisation of the movement.
Activity C. Relax. The subject is relaxed.
2.4 Feature Selection
For Off-line experiments the registered signal is
chopped in packages of samples, similar to the bun-
dles of samples obtained from the acquisition card in
the On-line cases. Each package has 128 samples,
acquired at f
s
= 384Hz. A vector of six features is
extracted from each package, see table 1, this vector
is made up as the mean of the amplitudes of the fre-
quency bands (Proakis and Manolakis, 1997), (Neu-
per, 2001).
Because the frequency of normal human brain is
under 40-50Hz, only frequencies between 6 and 38Hz
have been considered.
Table 1: Feature vector.
Index Denomination. Frequency (Hz).
1 . 6 - 8
2
1
. 9 - 11
3 2. 12 - 14
4
1
. 15 - 20
5
2
. 21 - 29
6
3
. 30 - 38
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
188
3 LINEAR DISCRIMINANT
ANALYSIS PROCEDURE
3.1 Introduction
Supposed C classes of observations, Linear Discrimi-
nant Analysis is a preprocess technique that nds the
transformation matrix W which separates in an opti-
mal way two or more classes. It is used in machine
learning as linear classier or as a technique to re-
duce the feature space dimension before the classi-
cation process. LDA considers maximising the fol-
lowing objective:
J(W) =
W
T
S
B
W
W
T
S
W
W
(1)
where S
B
is the between classes scatter matrix and S
w
is the within classes scatter matrix, the denitions of
the both matrices are:
S
B
=

c
N
c
(
c
x)(
c
x)
T
(2)
S
W
=

c

ic
(x
i

c
)(x
i

c
)
T
(3)

c
=
1
N
c

ic
x
i
(4)
x =
1
N

i
x
i
=
1
N

c
N
c

c
(5)
and N
c
is the number of samples in class c.
Because J is invariant to rescaling of the vectors
W W, hence it is possible to choose W such that
the denominator is W
T
S
W
W = 1. So the problem
of maximising J can be transformed to the following
constrained optimisation problem,
min
W

1
2
W
T
S
B
W (6)
s.t. W
T
S
W
W = 1 (7)
corresponding to the Lagrangian,
L
P
=
1
2
W
T
S
B
W +
1
2
(W
T
S
W
W 1) (8)
With solution (the halves are added for convenience):
S
B
W =S
W
W S
1
W
S
B
W =W (9)
This is a generalised eigen-problem, and using the
fact that S
B
is symmetric positive denite and can
hence be written as S
1
2
B
S
1
2
B
, where S
1
2
B
is constructed
from its eigenvalue decomposition as S
B
=UU
T

S
1
2
B
=U
1
2
U
T
. Dening V = S
1
2
B
W it is get
S
1
2
B
S
1
W
S
1
2
B
V =V (10)
this is a regular eigenvalue problem for a symmetric
positive denite matrix, with solutions
k
as eigen-
values and V
k
as eigen-vectors, which leads to solu-
tion:
W = S

1
2
B
V (11)
Plugging the solution back into the objective J(W), it
is found that the desired solution which maximise the
objective is the one with largest eigenvalues.
3.2 Operational Procedure
1. Samples from each mental tasks are obtained.
X
a
Mathematical Activity.
X
b
Movement imagination.
X
c
Relax.
2. Population statistical denitions: (i = a, b, c).

i
= E[x
i
] S
i
= E[(x
i

i
)(x
i

i
)
T
] (12)
3. Calculation of the scattering matrices (eq.2 & 3).
4. Application of LDA optimising criterion (eq.10).
5. Calculation of the transformation matrix, W
(eq.11), formed by the eigen-vectors, V
k
, which
eigen-values are bigger than 110
4
ordered form
high to low magnitudes.
6. Transformation of the data sets: (i = a, b, c).
X
i
X

i
=W
T
X
i
(13)
4 STATISTICAL ANALYSIS
PROCEDURE
Bilateral contrasts between two population are used
to determine if there is statistical evidence of dif-
ference between the population of features obtained
from each mental activity. Each component of the
vector is considered to determine its signicance and
separability power. Bilateral contrast makes use of
population variance, if the equality of both population
variances is rejected it is necessary to apply a correc-
tion factor in the degrees of freedom. These contrasts
were done for each type of ltering window.
Bilateral contrast of two independent normal and
homocedastic populations.
Null hypothesis H
o
vs. alternative hypothesis H
1
.
n
1
: sample size of the rst population.
n
2
: sample size of the second population.

S
1
: variance estimation of the rst population.

S
2
: variance estimation of the second population.
T = Student distribution.
H
o
:
1

2
= vs. H
1
:
1

2
= (14)
The variances of the both population are equal
BRAIN COMPUTER INTERFACE - Feedback Effect Analysis by Comparison of Discrimination Capability of On-line
and Off-line Experimental Procedures based on LDA
189
but unknown.
T
Exp
=
(

X
1

X
2
) (
1

2
)

S(
1
n
1
+
1
n
2
)
(15)
In which

S is the pseudo-variance of

S
1
and

S
2

S =
(n
1
1)

S
1
+(n
2
1)

S
2
n
1
+n
2
2
(16)
The zone of H
o
acceptance is:
T
Teo
=t
(n
1
+n
2
2,1

2
)
(17)
If |T
Exp
| T
Teo
then H
o
is accepted, on the con-
trary H
1
is accepted and H
o
is rejected.
5 RESULTS
In gures 4 to 9 are represented the results of the
bilateral contrast test for the transformed coordinate
X
1
considering the Off-line and On-line experiments.
The gures show for each channel (C3-C3 and C4-
C4), and for each type of preprocessing window, the
results p of the associated probability of the bilateral
contrast tests between the mental tasks. In order to
represent the dispersion of the results the mode value
and bars from 15th to 85th percentile have been used.
6 DISCUSSION
The comparisons between the discrimination capabil-
ities of On-line and Off-line experiments are shown
in the gures 4 to 9. From the bilateral contrast
test carried out with a signicant level of = 2.5%,
= 1 p, it is obtained that in almost all cases the
null hypothesis H
o
, which maintains the equality in
the populations of the features associated to the men-
tal tasks, shall be rejected for both types of experi-
ments; in the comparison of mathematical task versus
motor imagery, p values are lower for the On-line case
in both channels and with all types of preprocessing
windows than the ones obtained for the Off-line case;
the dispersion of the results is similar in both experi-
ments. It is also shown that for X
1
, channel C4-C4
performs better than C3-C3. The best results are
obtained for X
1
with Tukeys and Kaisers windows.
The highest contrast power is obtained in the com-
parison of Motor imagery vs. Relax, it is followed by
Mathematical task vs. Relax, and the lowest is for
Mathematical task vs. Motor imagery.
In all cases only two eigen-values have got sig-
nicant magnitudes, so only two eigen-vectors have
been considered in the transformation matrix. This
causes that LDA technique had projected the orig-
inal six dimensional feature space over a bidimen-
sional space, weighting the power amplitude of the
frequency bands and maintaining the intrinsic charac-
teristics of each cerebral activity.
7 CONCLUSIONS
In this paper has been estudied the users feedback in-
uence in BCI technology by analyzing the discrimi-
nation capability obtained under the Off-line and On-
line experiments carried out with ve male volunteers,
the results indicate that it is possible to differentiate
between the proposed mental tasks under the On-line
paradigm, but also that the discrimation capability is
a bit lower (< 0.3%) than the one obtained under the
BCI Off-line case, (Pineda et al., 2003).
Because the experiments have been carried out
only with ve volunteers these preliminary conclus-
sions have to be corroborated with more tests.
REFERENCES
Allen, J. B. and Rabiner, L. R. (1977). A unied approach
to short-time fourier analysis and synthesis.
Birbaumer, N. et al. (2000). The thought translation de-
vice (TTD) for completely paralyzed patients. IEEE
TRANS. ON REH. ENG.. 8(2):190193.
Chiappa, S. (2006). Analysis and Classication of EEGSig-
nals using Probabilistic Modles for Brain Computer
Interfaces. PhD thesis, Ecole Polytechnique Federale
de Lausanne, Lausanne EPFL.
Kostov, A.; Polak, M. (2000). Parallel man-machine
training in development of EEG-based cursor control.
IEEE TRANS. ON REH. ENG., 8(2):203205.
Lopes da Silva, F. (1999). Electroencephalography: ba-
sic principles, clinical applications and related elds.
MD: Williams and Wilkins, Baltimore.
Martinez, J.L.; Barrientos, A. (2006). The windowing Ef-
fect in Cerebral Pattern Classication. An Application
to BCI Technology. IASTED Biomedical Engineering
BioMED 2006, pages 11861191.
Martinez, J.L.; Barrientos, A. (2007). Linear Discriminant
Analysis on Brain Computer Interface. 2007 IEEE. In-
ternational Symposium on Intelligent Signal Process-
ing. Conference Proceedings Book., pages 859864.
Neuper, C.; et al. (2001). Motor Imagery and Direct Brain-
Computer Communication. Proceedings of the IEEE,
89(7):11231134.
Penny, W. D.; et al. (2000). EEG-based communication: A
pattern recognition approach. IEEE TRANS. ON REH.
ENG., 8(2):214215.
Pfurtscheller et al. (2000a). Brain oscillations control
hand orthosis in a tetraplegic. Neuroscience Letters,
1(292):211214.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
190
Pfurtscheller et al. (2000b). Current trends in Graz brain-
computer interface (BCI) research. IEEE TRANS. ON
REH. ENG., 8(2):216219.
Pineda, J.A. et al. (2003). Learning to Control Brain
Rhythms: Making a Brain-Computer Interface Pos-
sible. IEEE TRANS. ON REH. ENG., 11(2):181184.
Proakis, J. G. and Manolakis, D. G. (1997). Tratamiento
digital de seales : [principios, algoritmos y aplica-
ciones]. Prentice-Hall, Madrid.
Ripley, B. (2000). Pattern Recognition and Neural Net-
works. Cambridge University Press, London, 2nd edi-
tion.
Vidal, J. J. (1973). Toward direct brain-computer commu-
nication.
Wolpaw, J. R. (2007). Brain-computer interfaces as new
brain output pathways. THE JOURNAL OF PHYSI-
OLOGY.
Wolpaw, J. R., Birbaumer, N., McFarland, D. J.,
Pfurtscheller, G., and Vaughan, T. M. (2002). Brain-
computer interfaces for communication and control.
Wolpaw, J.R.; et al. (2000). Brain-Computer Interface Tech-
nology: A Review of the First International Meeting.
IEEE TRANS. ON REH. ENG., 8(2):164171.
APPENDIX
Figure 4: Off-line. Math task vs. Motor imagery. Coordi-
nate X1.
Figure 5: Off-line. Math task vs. Relax. Coordinate X1.
Figure 6: Off-line. Motor imagery vs. Relax. Coordinate
X1.
Figure 7: On-line. Math task vs. Motor imagery. Coordi-
nate X1.
Figure 8: On-line. Math task vs. Relax. Coordinate X1.
Figure 9: On-line. Motor imagery vs. Relax. Coordinate
X1.
BRAIN COMPUTER INTERFACE - Feedback Effect Analysis by Comparison of Discrimination Capability of On-line
and Off-line Experimental Procedures based on LDA
191
DATA ACQUISITION ELECTRONICS FOR PET
MAMMOGRAPHY IMAGING
Carlos Leong
1,2
, Pedro Machado
1,2
, Vasco Bexiga
1,2
, J . Paulo Teixeira
1,2
, Isabel C. Teixeira
1,2

1
INESC-ID, Lisboa, Portugal
2
Instituto Superior Tcnico, UTL, Lisboa, Portugal
[email protected], [email protected], [email protected], [email protected], [email protected]
J oel Rego
3
, Pedro Neves
3
, Fernando Piedade
3
, Pedro Lous
3

Pedro Rodrigues
4
, Andreia Trindade
4
, R. Bugalho
4
, J . F. Pinheiro
4
, M. Ferreira
4
, J oo Varela
2,4

3
INOV, Lisboa, Portugal
[email protected], [email protected], [email protected], [email protected]
4
LIP-Lisboa, Lisboa, Portugal
[email protected], [email protected], [email protected], [email protected], [email protected], [email protected]
Keywords: PET mammography, Breast cancer, Data communication, Synchronism, Multi-clock domains.
Abstract: The purpose of this paper is to present recent developments in the off-detector electronics of a PET
(Positron Emission Tomography) system for mammography imaging. In particular, problems and solutions
associated with the integration of its Data Acquisition Electronics are targeted. Synchronism is a critical
issue in the DAE system. A resynchronization module is proposed to solve communication problems with
the internal asynchronous busses with limited degradation of the communication rate, as compared with
fully synchronous solutions. Data processing refers to 64 dual-channels, 11bit/channel, at a frequency of
100MHz. The maximum of the DAE output rate is 220 MB/s corresponding to 1 MCoincidence/s. The
robustness of the proposed solutions has been validated with software simulation and hardware
implementation. Results of test and validation on FPGA, boards and buses are presented.
1 INTRODUCTION
PET (Positron Emission Tomography) systems are
among the most effective imaging-based
technologies for medical diagnosis. In this
contribution, the final development of a PET system
for mammography - PEM (Positron Emission
Mammography) dedicated to the analysis of the
women breast (Santos, 2004) is described. Emphasis
is given to the integration aspects of the Data
Acquisition Electronic (DAE), which is responsible
for the off-detector acquisition and processing of the
PEM system.
Main aspects of the development of the entire
system, namely, the high level architecture, the
study of the crystal detectors (Lecoq, 2002) (Abreu,
2006), the transducers to sense the light and to
transform it into energy pulses, the associated Front-
end Electronics (Albuquerque, 2006), the off-
detector electronics (Bento, 2006) (Leong, 2006),
(Varela, 2005), the imaging reconstruction
procedures and, of course, the mechanical aspects of
the robot have been reported earlier. In the
identification of system requirements, software
models have been used and extensive Monte Carlo
simulations (Agostinelli, 2003) (Rodrigues, 2004)
(Trindade, 2004) have been performed.
This paper focuses on the difficulties associated
with system integration, and how those difficulties
have been overcome.
Main problems have been found in implementing
adequate communication infrastructures, satisfying
synchronism and performance requirements. In fact,
correct individual sub-systems functionality and
performance was verified in stand-alone test.
Nevertheless, several modifications were required
when integration took place, specifically on sub-
systems that communicate with different clock
domains. A typical case regards subsystems that
access the buses. The purpose of this paper is to
present innovative solutions which have been
designed and implemented to solve those problems.
192
The paper is organized as follows. In section 2,
the generic architecture of the PEM system is
provided to highlight the communication
infrastructure. In section 3, the DAE functionality
and architecture are described, highlighting the
critical sub-modules, in terms of the communication
environment. Section 3 also contains the proposal of
new architectures for modified modules that are
responsible for the communication between the
DAE and the Front End electronics and between the
DAE and the imaging reconstruction computer. In
section 4, details of the physical implementation are
provided. Section 0 presents the hardware validation
results. Finally, in section 6, the main conclusions
and future work are outlined.
2 PEM SYSTEM
The main objective of the system is to identify the
presence of cancer cells in women breasts. Imaging
reconstruction is used for the purpose. As it is well
known, image reconstruction quests for large
amounts of data. Therefore, main characteristics of
PEM system are high data volumes and high data
rates.
A key aspect is the need to guarantee that
meaningful data is unequivocally identified. In order
to understand what we refer as meaningful data, it is
necessary to understand PEM underlying physics.
Human cells emit rays when a radioactive
substance is injected into the human blood stream.
When this occurs, 2 ray photons are emitted in
opposite directions over a linear trajectory. Emission
sources are detected by the intersection of
trajectories. 2 planes of scintillant crystals detect
the emitted ray photons. Crystals emit light that is
afterwards converted into electric signals by
Avalanche Photo Diodes (APD). Crystal arrays in
the PEM scanner are organized in modules and sub-
modules in a hierarchical structure (Amaral, 2007)
(Matela, 2004). Data are captured through 12288
readout channels, organized in 296 identical detector
modules and distributed by the two crystal planes.
Potential meaningful data correspond to the
simultaneous detection of ray photons in both
crystal plans. In this case, we consider that a
coincidence has occurred. The underlying principle
of PEM systems behavior is the identification of
coincidences. PEM scanner is a high-resolution
system, capable of detecting breast tumors with
diameters down to 2 mm (Albuquerque, 2006).
In Figure 1, the top-level architecture of the
Clear-PEM system is depicted. Three main sub-
systems can be identified, namely, the scanner,
constituted by crystals and associated Front-End,
FE, electronics, the off-detector Data Acquisition
Electronics, DAE, and finally, an external computer,
PC (or array of computers) performing data storage
and image reconstruction.
The FE electronics is an analogue/mixed signal
system, responsible for first conditioning of the
signal generated by the APDs, for the
analogue/digital conversion and for the
communication with the off-detector DAE system.
DAE is responsible for digital data processing, in
order to identify coincidences and to send
meaningful data to the computer where image
reconstruction takes place.
Front End
Front End
D
A
E ADC
PEM
ADC
Control
Control
0011010

Figure 1: Top-level architecture of PEM system.
Data integrity must be kept during processing
and communication phases. Otherwise, the diagnosis
result would be unreliable, leading to false positive
or false negative results. This justifies the critical
relevance of the problems tackled in this
contribution.
3 DAE SYSTEM
3.1 DAE Requirements and
Functionality
The main functionality of the DAE is the
identification of relevant data and the transmission
of that data to the image reconstruction computer.
DAE specific requirements are as follows. The
system should support a data acquisition rate of 1
million events per second, under a total single
photon background rate of 10 MHz (Albuquerque,
2006). An event or hit (photoelectric event or
Compton - according to the associated energy) is
defined as the interaction of a ray with a crystal.
Data to be analyzed and processed correspond to
the hitting energy in the different crystals, as a
consequence of these interactions. Relevant data is
associated with relevant events (coincidences).
Hence, a relevant event is characterized by the
simultaneous occurrence of hits in both crystal
planes.
DATA ACQUISITION ELECTRONICS FOR PET MAMMOGRAPHY IMAGING
193
3.2 Communication Challenges
Data is transmitted from the FE to the DAE by a
large number of connecting cables, which introduce
various delays, noise, and possibly, signal
degradation. Therefore, physical interconnections
pose a challenge to system integration.
Another challenge is to guarantee the correctness
of data communication between DAE and the data
storage and image reconstruction computer, PC.
A third challenge is to guarantee error-free
communication among DAE boards and chips
through the DAE internal buses, while not degrading
system performance. In the DAE, the minimum
required transmission rate is 1MCoincidence/s at
100 MHz, which corresponds to 220MB/s bit rate.
3.3 DAE Architecture
A comprehensive description of the DAE system
(Figure 2) has been published elsewhere. However,
for the sake of clarity, a brief description of the
constituting elements is provided here.

Figure 2: Top-level architecture of the DAE electronic
system.
DAE maps the organization of crystals in the
scanner. As a consequence, DAE is constituted by 4
DAQ (Data Acquisition) boards and 1 TGR/DCC
(Trigger/Data Concentrator) Board, communicating
among them through buses.
The DAQ boards carry out the first data filtering
to identify probably useful data, out of all the data
that floods from the FE. As mentioned, the criterion
for this classification is the detection of coincidence,
interactions of photons with crystal pairs within a
given discrete time interval, to which an identifying
Time Tag is associated. Data is classified either as
(1) useful and stored, or as (2) noise, in which case it
is discarded.
Due to its flexibility, from the technological
point of view, main functionality is implemented
using reconfigurable devices, namely FPGA
technology. Each DAQ board houses two 4-million
gates DAQ FPGA, each containing a read out
controller - the DAQ ROC module. At present, the
DAE houses 4 DAQ and one TRG/DDC boards.
The TGR/DCC board is responsible for the
identification of coincidences. When a coincidence
is detected, TGR/DCC generates a trigger signal that
notifies the DAQ boards of the situation, picks up
the corresponding data and concentrates it according
to a given protocol, to be sent to the external PC for
image reconstruction. The DAE/PC communication
is based on a commercial Bus.
The 4 DAQ and the Trigger boards communicate
through buses, namely, the GBUS (Generic Bus) and
the DBUS (Dedicated Bus) (Figure 2).
3.4 Synchronism between FE and DAE
To guarantee that a detected coincidence is
effectively a coincidence, it is mandatory to know,
without ambiguity, at what time a given data have
been generated. For this, it is mandatory to
guarantee system synchronism. The PEM system is
a GALS (Globally Asynchronous, Locally
Synchronous) system, although FE and DAE are
driven by a system clock.
Data that leaves the FE, at the same time, refer to
the same temporal mark, that is, the same Time Tag.
Moreover, all clock signals driving the system
behavior should have the same frequency and phase.
However, channel links introduce delays. In
order to deal with this situation, data is transmitted
from the FE with the corresponding clocks.
Although, identical cables and components have
been used in identical paths, there is no guarantee
that the clock phase is identical for all local
components where clock is restored. In fact, small
physical differences in the cables, in local PLL,
delays induced by variable thermal maps, clock jitter
and clock skew may result in small phase
differences in the clocks arriving through the
different cables. If these clock signals are very close
to the internal clock edge, it may occur that one is
activated in a given clock cycle and the other be
activated in the following clock cycle.
To solve the FE/DAE synchronization problem,
DAE uses a specific signal, Sync, to guarantee data
synchronism and an auxiliary synchronization
module to guarantee that the synchronism is
restored. Sync behaves as a feedback loop that
makes possible to detect and to synchronize all the
active inputs. The description of this sub-system is
outside the scope of this paper, as it has been
published elsewhere.

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
194
3.5 Internal Buses Communication
As mentioned, another challenge in the integration
phase has been the need to guarantee robust
communication among DAE sub-systems. In fact,
each sub-system (e.g., an FPGA) is a clock domain
with its own local clock. Many solutions may found
in the literature, regarding GALS (Xin, 2005),
(Beigne, 2006), (Ogras, 2007).
In our system, data rate transmission is, at least,
220 MB/s. An adequate solution for dealing with
high data rates would be to use a synchronous bus
(Lee, 2007). However, to guarantee robustness on
data transfer, the asynchronous solution is
preferable. Therefore, a trade-off between speed and
robustness is required.
In PEM system, a slightly lower data rate
transfer is acceptable in order to achieve the required
robustness. This means that a small performance lost
(when compared with standard synchronous
solutions) in conjunction with a small increase on
silicon area is acceptable.
Therefore, in the PEM system, subsystems
communicate through asynchronous busses, which
guarantee correct communication at the highest data
rate among these multi-clock domains.
3.6 Resynchronization Module
In the DAE, DBus (Dedicated Bus) and GBus
(Generic Bus), Figure 2, are asynchronous buses. In
this case, Figure 3, two pipeline stages (to guarantee
data synchronization), constituted by 2 chains of 2
Flip-flops each; namely, D1aD1b and D2aD2b.
Additional flip-flops are used at the input of each
clock domain in order to assess if arriving data is
reliable. In this case, an event is generated and data
communication is started.

Figure 3: Bus Resynchronization block.
The Bus resynchronization timing diagrams
(Figure 4), show that the communication rate is not
as significantly decreased as it would be in a fully
asynchronous solution. This comes from the fact that
each data carries, with it, the corresponding
clock/strobe, avoiding the need to wait for hand
shake protocol as associated with asynchronous
communication.
Moreover, clock domains can work in parallel
with the communication procedure, since they are
locally synchronous domains. By doing so, the speed
of synchronous communication is almost reached.
Bus Data A B
Bus Clk
D1a A
D2a B
Int Clk
C1
C2
C3
G1 (En1)
G2 (En2)
D2a A
D2b B
Data Out A B
Data Ready

Figure 4: Bus Resynchronization timing diagram.
As would in an asynchronous solution,
communication is correctly performed regardless the
bus physical length (limited by the line drivers
capability). This is a considerable advantage of our
methodology.
3.7 DAE/PC Interconnection
Three solutions have been analyzed and
experimented for the communication link between
DAE and external PC in order to meet speed and
data rate requirements, namely, the PCI, the USB
and the SLINK Fed Kit (CERN). Finally, a mixed
solution based on the USB and the SLINK protocols
has been adopted.
The USB solution is based on the (Cypress
CY7C68013A) microcontroller. Although, the
bandwidth is limited to 60MB/s (maximum
theoretical data rate in USB 2.0 protocol), the USB
link provides a bidirectional communication, able to
handle command requests and consequent replies
(all of each are limited to 64 bits package).
The SLINK (CERN S-LINK Fed Kit) is a
proprietary unidirectional protocol that,
theoretically, achieves up to 800MB/s at full speed.
The SLINK guarantees raw data transmission from
the DAE to the PC.
This USB+SLINK solution allows simultaneous
communication through both links. Separating data
and command communication channels minimizes
the resources needed on the acquisition PC.
DATA ACQUISITION ELECTRONICS FOR PET MAMMOGRAPHY IMAGING
195
4 DAE IMPLEMENTATION
Transmission between FE and DAE is carried out at
frequency of 300MHz (3.3ns) using 32 cables.
Correction is 1,7ns (1/2 cycle).
Some electrical problems have been faced, in the
choice of the interconnecting cables. The ideal
would be to use a twisted pair, individually shielded
cable. However, the cross section and the flexibility
of such a set of cables are incompatible with the
mechanical robot. Therefore, a trade-off between
electrical characteristics with mechanical
requirements was necessary.
Main electrical problems that have been faced
are related with the physical distance between the
FE electronics and the DAE (around 6 meters). Most
relevant electrical problems are skew and DC
balance.
Only Amphenol SpectraStrip cables respond to
the system requirements and therefore have been
used in all tests. Nevertheless, some modifications
have still been carried out in this cable in order to
improve electromagnetic characteristics.
DAQ functionality is implemented with 8
Xilinx xc2v4000-4bf957 FPGA (2 FPGA per
DAQ Board), i.e., 4 million equivalent gates and 957
pins FPGA. The TGR/DCC functionality is
implemented with one Xilinx xc2v3000-4bg728, 3
million equivalent gates and 728 pins FPGA.
The FE/DAE communication between the FE
and DAE subsystems is carried out through LVDS
(Low Voltage Differential Signaling) channel links.
LVDS channel links de-serializer convert the high
speed, serialized, long distance communication lines
into the standard LVTTL (Low Voltage TTL
(Transistor Transistor Logic)) electrical signals at
the input/output of DAQ boards logic components.
The DAE system is shown in Figure 5. As
mentioned, the DAE is constituted by 5 boards. 4
DAQ boards and 1 TGR/DCC board communicating
among themselves by 2 internal buses (DBus and
GBus).

Figure 5: DAE physical implementation.
Transceivers are used as FPGAs gateways to the
internal BUSes. Transceivers serve also as buffers
for these BUSes.
5 VALIDATION RESULTS
Validation conditions are as follows. Input data
(input events) are provided by a previously
developed system that emulates the Front End (FE)
functionality. Output data has been obtained with the
DAE hardware.
The test strategy is as follows. The electrical test
of the DAE boards has been carried out prior to the
FPGAs functional test. The functionality of the
FPGAs has been validated in silicon (DAE system
and board test).
In Figure 6, the different buses event rate is
represented. It is worth to notice that, these results
have been obtained with a working frequency of
50MHz (limited by current FE hardware version).
This means that if we work at 100MHz, the event
rate duplicates. Therefore, validation results are
beyond system specifications.
0E+00
1E+05
2E+05
3E+05
4E+05
5E+05
6E+05
7E+05
8E+05
9E+05
1E+06
0 500000 1000000 1500000 2000000 2500000
input (FE-EMU) event rate
Dbus rate
Gbus rate
SLink rate
USB rate
Event/s

Figure 6: Event rate (experimental result).
The GBus shows a linear dependence of the
event rate on the input rate until almost 700Kevent/s.
After this value, starts to flare until reaching a
maximum transmission rate of 800Kevent/s. After
that, it starts dropping. This is caused by the
inefficiency of the arbiter which is now under
revision. In the linear region, this inefficiency is
masked by the fact that the required bandwidth is
below the maximum.
It is expected that with a more efficient arbiter,
the linearity of the GBus is verified until higher
values and will remain almost constant after that.
SLINK performance in this test is only limited
by the GBus throughput.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
196
6 CONCLUSIONS AND FUTURE
WORK
In this paper, some problems faced during the
integration of the PEM DAE system have been
reported. Emphasis has been put on communication
related problems, particularly, those related with
synchronism aspects.
Some aspects related with the communication
between DAE and FE electronics and the DAE and
the imaging reconstruction computer are provided.
More detail has been given to the solution
proposed for guaranteeing the correctness of the
communication through the DAE internal buses.
With this solution, the robustness of asynchronous
communication and the speed of synchronous
solutions are almost reached.
Experimental results show that the achieved
performance surpassed system specifications.
Additional performance improvements will be
obtained with the review of the GBus arbiter, which
is currently under development.
System architecture is also under review in order
to allow the integration of more data acquisition
boards. This will allow more flexibility in the
scanner geometry, thus allowing extension to the
system applicability for medical imaging of other
regions of the human body. Results will be reported
in the future.
ACKNOWLEDGEMENTS
The work reported in this paper has been partially
supported by FCT (Portuguese Foundation for
Science and Technology), partially supported by
ADI (Portuguese Innovation Agency) in the scope of
the PET II Project 70/00327 and partially supported
by PETSys (Medical PET Imaging Systems, S.A.).
REFERENCES
Abreu, M.C. et al., 2006. Design and Evaluation of the
Clear-PEM Scanner for Positron Emission
Mammography. In IEEE Transactions on Nuclear
Science, vol. 53, N. 1, pp. 71-76.
Agostinelli, S. et al., 2003. GEANT4 A simulation
toolkit, Nuclear Instrum. Meth. A. vol. A 506.
Albuquerque, E. et al., 2006. The Clear-PEM Electronics
System. In IEEE Transactions on Nuclear Science,
vol. 53, n. 5, pp. 2704-2711.
Bento, P. et al., 2006. Performance Simulation Studies of
the Clear-PEM DAQ/Trigger System, IEEE
Transactions on Nuclear Science, vol. 53, n. 4, pp.
2102-2111.
CERN S-LINK, http://hsi.web.cern.ch/hsi/s-link/
Leong, C. et al., 2006. Design and Test Issues of an
FPGA Based Data Acquisition System for Medical
Imaging using PEM, IEEE Transactions on Nuclear
Science, vol. 53, n. 3, pp. 761-769.
Varela, J . et al., 2005. FPGA Based Architecture for the
Data Acquisition Electronics of the Clear-PEM
System, Proc. Int. Workshop on Applied
Reconfigurable Computing (ARC 2005), pp. 131-138.
Rodrigues P. et al., 2004. Geant4 applications and
developments for medical physics experiments, IEEE
Trans. On Nuclear Science, vol. 51, pp. 1412-1419.
A. Trindade et al., 2004 .Design and Evaluation of the
Clear-PEM Detector for Positron Emission
Mammography, IEEE-Medical Imaging Conference,
Rome.
Santos, A. I. et al, 2004. Simulation studies and
mechanical system for axillary lymph nodes
examination with Clear-PEM, Medical Imaging
Conference MIC/IEEE, Rome, Italy.
Amaral, P., et.al. 2007. Performance and quality control of
Clear-PEM detector modules. In Nuclear Instrum.
Meth. Vol. A 580 pp. 1123-1126.
Beigne, E., Vivet, P., 2006. Design of on-chip and off-chip
interfaces for a GALS NoC architecture. In 12th IEEE
International Symposium on Asynchronous Circuits
and Systems.
Lecoq, P., et al., 2002. Clear-PEM, A dedicated PET
camera for mammography, In Nuclear Instrum. Meth.
vol. A 486, pp. 1-6.
Lee, H.G., et. al., 2007. On-Chip Communication
Architecture Exploration: A Quantitative Evaluation
of Point-to-Point, Bus, and Network-on-Chip
Approaches. In ACM Transactions on Design
Automation of Electronic Systems, Vol. 12,No. 3.
Matela, N. et al, 2004. System matrix calculation for
Clear-PEM using ART and linograms. In Medical
Imaging Conference MIC/IEEE.
Ogras, Umit Y., et. al., 2007. Challenges and Promising
Results in NoC Prototyping Using FPGAs. In IEEE
Computer Society.
Santos, A. I. et al, 2004. Simulation studies and
mechanical system for axillary lymph nodes
examination with Clear-PEM, In Medical Imaging
Conference MIC/IEEE.
Xin, J ia, Vemuri, R., 2005. Using GALS architecture to
reduce the impact of long wire delay on FPGA
performance. In Design Automation Conference.
DATA ACQUISITION ELECTRONICS FOR PET MAMMOGRAPHY IMAGING
197
DEVELOPMENT OF STRATHCLYDE UNIVERSITY DATA
LOGGING SYSTEM (SUDALS) FOR USE WITH FLEXIBLE
ELECTROGONIOMETERS
Vivek Padmanaabhan Indra Mohan, G. Valsan and P. J . Rowe
Health Qwest, Bioengineering Unit, University of Strathclyde, Wolfson Centre, Glasgow, G4 0NW, U.K.
[email protected] , [email protected], [email protected]
Keywords: Flexible electrogoniometer, Activities of daily living (ADL), User-friendly system, Remote control,
Wireless data transmission.
Abstract: We have developed a 6 channel battery operated remote control microprocessor based system that collects
data from flexible electrogoniometers and force sensing resistors attached to the lower extremities of the
body. During functional activities, the user-friendly system stores the data from these transducers and
transfers the same to a PC at the end of the recording period via a bluetooth connection. Software on the PC
then displays the angular displacement and allows visual inspection of the entire sequence of recordings or
particular events of interest. This system was tested on 10 normal subjects and the pattern pertaining to the
flexion/extension of knee during range of activities of daily living (ADL) such as walking, ascending and
descending stairs, in and out of a chair and deep squatting were recorded and found to be reproducible and
similar to those reported in the literature.
1 INTRODUCTION
Normal lower limb activity and goal directed
movements are essential for the well-being of an
individual. However, such movements and efficient
functioning of the lower limb can be seriously
affected for a variety of reasons and one such
common cause is Osteoarthritis. This degenerative
joint disorder is often disabling and is characterized
by pain and physical limitation. (Rowe et al, 2005).
As a result, functions of lower limb are affected,
causing individuals to have problems with ADL
such as; walking, climbing stairs, getting in and out
of chair, getting in and out of bath etc. At this point
of time, rehabilitation and health care professionals
play a very crucial role by improving the functional
ability of their clients. Periodic assessment of the
individuals is necessary to aid the health professions
in assessing efficiency of their interventions.
Currently, two types of assessment techniques are
available for this purpose: questionnaire based
assessment and assessment based on clinical gait
analysis.
The former technique makes use of knee scoring
questionnaires such as the Western Ontario and
McMaster Universities Osteoarthritis Index
(WOMAC) and Knee Society Clinical Rating
System. (Rowe et al, 2005). Even though, these
questionnaires are popular, easy to administer and
characterize the overall performance of an
individual, research reveals that they are highly
subjective and do little to reveal any objective
information regarding the actual restoration of the
knee function required by an individual to perform
ADL.
On the other hand, clinical gait analysis is an
expensive and time consuming process. (Rowe et al,
2005). Alternatively, researchers have started using
electrogoniometry to record the dynamic knee joint
movement during a range of functional activities due
to its simpler, cheaper and reproducible nature.
(Rowe et al, 2005, Rowe et al, 2001) Mostly, such
body mounted transducers are used in combination
with information storage devices known as Data
Loggers. (Rowe et al, 2005).The role of these
devices is not merely to store the data collected from
these transducers but also to convert the signals
obtained from the transducers to an understandable
form. Many such devices have been developed in the
past and are also being currently used along with a
wide range of transducers such as flexible
electrogoniometers, accelerometers and strain
198
gauges for mobility assessment, recording of plantar
pressure etc. (Zhu et al, 1991). Wireless
communication is finding its way into various
medical technological applications (Zhang.Z &
Liu.P, 2004), but most data loggers remain
hardwired. It was the premise of this work that the
data logger currently used with flexible
electrogoniometers needs further improvement in
functionality so that, the process of collecting a large
stream of data and extracting the relevant sections
could be carried out more efficiently. Further, such a
system should be able to be used by any allied health
professional in a multi centered clinical trial
evaluating post-operative rehabilitation. The lack of
such a system merits the development of a user
friendly system, whereby pushing a button would
start, stop, collect multiple data sets and transmit the
same without any physical contact between the
subject and the operator.
We have developed a portable, battery operated,
remote control microprocessor based system that
allows recording, deleting and transmitting the data
obtained from two flexible electrogoniometers and
four force sensing resistors. The data is stored in
static random access memory (SRAM) and can
subsequently be transferred via Bluetooth to a PC
which processes and analyzes the data.
2 METHOD
2.1 Overall System
The flexible electrogoniometer consists of a strain
gauged shim (a thin flexible strip) which runs the
length of the device. Damage to the device and
injury to the test subjects is prevented by enclosing
the shim in a spring. To facilitate the attachment of
device to the subjects, two light weight plastic plates
are fastened to the ends of the shim. The resulting
transducer does not have a specific centre of rotation
and is flexible in both medio-lateral and anterior-
posterior directions. Each electrogoniometer was
attached using double sided medical grade tape
laterally to the shank and thigh of individuals via
two flexible plastic strips adjusted to the length of
their shank and thigh. In addition to this, light
weight force sensing resistors (FSR) or footswitches
were attached to the first metatarsal area of the toe
and to the heel for marking the events by indicating
the contact between the foot and the floor. Since the
transducer was mounted in the sagittal plane of the
knee, the output of the device represented the
flexion-extension angle of the knee. Both the
electrogoniometers and footswitches were interfaced
to SUDALS via thin flexible cables.
2.2 Hardware Design
The entire prototype was built on an evaluation
board Eval ADUC7026 which consists of a 12 bit
successive approximation type Analog to Digital
converter (ADC), with an on chip 32 bit
microcontroller. The microcontroller provides both
high performance and low power consumption. The
microcontroller has several on chip facilities
including programmable watchdog timer and 12
channel multiplexer. Hence, an additional
multiplexer or a sample or hold circuit was not used
in our system. The ADC chip analogue input range
is 0 to 2.5V DC, whereas, the output of the flexible
electrogoniometer is a differential voltage. As a
result, the voltage signals from these transducers
were conditioned using high precision
Instrumentation amplifier (INA101) with suitable
gain resistors so as to make these signals compatible
with the input range of ADC. Due to the low
temperature drift feature of this amplifier, the system
will not be significantly affected by ambient
temperature. Six 1.2V alkaline AA batteries are used
for powering the evaluation board, which is
regulated via an on chip voltage regulator to 3.3V.
This is used to drive the digital side of the board and
the same voltage is being filtered by the on chip
features to drive the analog side of the board. In
addition to this, the output from the batteries is
stepped down to +/- 5V via DC-DC converter to
power the amplifiers, the transducers and other
signal conditioning components on board. The data
from two flexible electrogoniometers and from the
four force sensing resistors are sampled at 50 HZ
and the digital values are stored in a 32KB x 16
static RAM an external memory chip interfaced
via the footprint provided on the evaluation board.
To the same memory, the data from the FSR
channels are compressed to on/off data and saved as
a single byte.
The data from the external memory is transferred
to a personal computer via a bluetooth transmitter
(HDWBTRS232 wireless RS232 Transceiver)
interfaced to the Eval ADUC7026 via the universal
asynchronous transmitter (UART) terminal provided
on board and a transmitter line driver ADM202. The
transmitter works on a voltage range of 5V-9V DC,
which is being provided using the same AA batteries
on board. Due to high power draining application
(wireless transmission), batteries chosen for the
operation of this system has a high power rating of
DEVELOPMENT OF STRATHCLYDE UNIVERSITY DATA LOGGING SYSTEM (SUDALS) FOR USE WITH
FLEXIBLE ELECTROGONIOMETERS
199
2400 mah. A Baud rate of 19200 was used to
transmit the collected data to a personal computer
(PC) in less than a minute. At the PC end a software
code written in MATLAB is used to receive the
transmitted data and store the data in the format of
excel files which are analyzed further depending
upon the user requirement. We simultaneously
measured the knee flexion / extension angles during
activities of daily living such as; walking, ascending
and descending the stairs, sitting in and out of a
chair and deep squatting and established that the
portable system could faithfully reproduce the
signal. The obtained data is analyzed for maximum
and minimum knee flexion / extension during these
activities and the results are compared against the
normal knee range of motion during these activities
published in the literature. The overall block
diagram of the system is as shown in Figure 1.

Figure 1: System block diagram.
2.3 System Functionality
The system is aimed to perform 5 functions
corresponding to data collection. These include;
recording a test, scrapping a failed test, transmitting
the collected data from the electrogoniometers and
footswitches and resetting the entire system. In
addition to this, the system also facilitates zeroing of
the electrogoniometers prior to each recording
depending upon the user requirements. This is
accomplished by making use of an Operational
amplifier 3240 at the hardware end.
Initially, when the system is connected to the
sensors and switched on, the system is ready to zero
the sensors and to record the data corresponding to
the knee flexion/extension. Once, a singe recording
is completed, then the system facilitates the user to
make use of other functions such as; scrapping the
recorded data if needed, else transmit the collected
data via wireless and reset the entire system for next
set of data collection. Each of these above
mentioned functions can be accomplished by
providing interrupt service routines (ISR) to the
microcontroller to start or stop that specific function
via an Infra red remote transmitter and receiver
interfaced with the microcontroller, when the LED
corresponding to that function illuminates. Since, all
these functions operate within a loop arrangement;
the user can perform single recording or multiple
recording, store it in the external memory and then
transmit it to the PC via wireless. The functional
flow chart of this system is as shown in Figure 2.
2.4 System Evaluation
The system was evaluated by carrying out a pilot
study, during which the data pertaining to the
flexion/extension of the knee of the 10 young
normal healthy subjects who volunteered for this
study was collected via the flexible
electrogoniometer interfaced with this portable unit.
All the 10 subjects were asked to perform the
following 6 activities Walking, In and Out of a
Chair, Stair ascent, Stair descend and deep squat
corresponding to daily living.
Start and stop commands were given at the
beginning and completion of each task and the
subjects were asked to repeat these tasks three times
for reproducibility and repeatability purposes.
Further, the event marking was taken into account
by the FSRs attached to the toes and heels of each
subject.
After data collection a 4
th
order low pass
Butterworth filter at a cut-off frequency of 6 Hz was
used to eliminate the noise present in the data. The
data collected during these activities were averaged
for each subject individually and were analyzed for
maximum and minimum knee flexion. The
excursion of the knee during these activities for each
individual was obtained by calculating the difference
between the maximum angle and minimum angle.
This procedure was carried out for both the left and
right knees and was then averaged to provide the
group mean. The excursion of the knees from
SUDALS is as shown in Table 1. Table 2 shows the
maximum knee flexion angle reached during each of
these ADL. The results were compared against the
values published in the literature as shown in Table
3.
The mean normalized gait cycle obtained by
SUDALS during the experimentation is shown in
figure3.

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
200


Figure 2: System functional flowchart.
Table 1: Knee ROM during ADL from SUDALS.
ADL SUDALS - excursion
Left Knee Right Knee
Gait 66.0 66.4
Stair up 71.3 71.8
Stair
Down
69.7 61.5
Chair in 101.7 100.9
Chair
Out
102.4 107.1
Squat 116.4 111.7


Table 2: Knee Flexion angles during ADL from SUDALS.

Table 3: Knee flexion angles from literature.

2.5 Discussion
The system described here was able to record, store
and transmit the data corresponding to the ADL.
Our results indicate that, the average maximum
knee flexion angle for all the 10 subjects during the
above mentioned ADL lies within the results
published in the literature as shown in Table3.
Transmit
Start
Switch
on
system
Record
/Zero?

Zeroing Recording
Enable trigger via
Remote Control to
Start performing
desiredfunction.
Enable trigger to stop
performingthefunction
A
Is1recording
complete?
Scrap/
transmit/reset?
Scrap? Transmit? Reset?
Reset
Data
Transmit
Data Scrap
Data
Hasall
recordings
transmitted?
Switch
off the
System.
Retransmit
Data
A
A
A
A
Reset
N
Y
N
Y
N
Y
N
Y
Y N
DEVELOPMENT OF STRATHCLYDE UNIVERSITY DATA LOGGING SYSTEM (SUDALS) FOR USE WITH
FLEXIBLE ELECTROGONIOMETERS
201

Figure 3: Normalized average Gait Cycle.
However, the knee flexion angles obtained
during squatting seem to be a little lower than the
values published in the literature. One of the
possible reasons for this could be the way in which
the subjects performed this activity. Though, the
subjects were shown what they were suppose to
perform during the process of recording, certain
subjects were unable to completely squat as it was a
difficult task and required a lot of effort. Due to this,
certain subjects performed half squat instead of a
complete squat. As a result, the knee flexion angle
recorded during this activity would be different from
those reported by Wyss.U et al - 2003, where the
subjects have performed a complete squat.
Moreover, most of the authors, other than
Huddleston.J et al - 2006, havent used flexible
electrogoniometer for measuring the knee flexion
angles. At the same time, though these authors have
reported the maximum knee flexion angles during
ADL, none of them have reported the knee
excursion of the subjects during these activities. On
the other hand, most of the studies by Rowe.P.J et al
2005, have reported the obtained knee excursion
during various ADL, but most of his studies are
concerned with the follow-up of TKA and elderly
population. Hence, the results from this study of
young healthy subjects were unable to be compared
with those published by Rowe et al. Most of the day
to day activities can be accomplished in less than a
minute.
Evidently, during our experimentation, we
noticed that the time taken to complete a single trial
of all the above mentioned six ADL by all the
subjects was less than a minute. Henceforth, despite
the usage of a 512 KB SRAM as prescribed by the
manufacturers of the evaluation board, we were able
to record, store and transmit the biomechanical
motions corresponding to six ADL with a little
difficulty. This ability to record, store and then
rapidly transmit the data facilitates data collection in
a free living environment and enables the user to
check whether the data recorded is reliable or not. In
this case, the user can re-record the activity
immediately unlike the commercially available
Biometrics data acquisition systems with flash
memory, where the user has to wait until the entire
data collection process is completed to check for
reliability and reproducibility of the data.
Currently, most of the portable data acquisition
systems that are used with flexible
electrogoniometer do not facilitate remote control
operation. Consequently, every time the users have
to physically change the settings of the data logger
such as starting, stopping or resetting; once its
being worn by the subjects. However, we were able
to control the entire process of data collection by
staying at a convenient distance of less than a foot
from the subjects. This would not only avoid any
physical contact with the subjects, but at the same
time, it would also minimize the degree of
inconvenience to the subjects. None of our subjects
reported any discomfort with SUDALS during the
process of data collection. Due to high power
draining application (wireless transmission of data),
we have used 6 x 1.2 V AA high wattage batteries of
2400 mah in our system. Further, unlike the
commercially available systems, SUDAL also has a
provision similar to the car battery charger for
recharging the batteries without removing them from
the system. Charging these batteries for 2 hours
enables us to use the system for more than 8 hours.
Though our system doesnt facilitate real time
waveform display, simultaneous data collection and
transmission almost replicates those systems with
real time waveform display. Thus, the users would
be able to analyze the transmitted data stored in the
form of excel files. Figure 4 shows the usage of
SUDALS in an experimental set up.
3 CONCLUSIONS
In summary, the system worked without any
technical difficulties and was able to accurately
measure the knee flexion/extension during activities
of daily living in healthy subjects. The results of the
present study in conjunction with the literature
review support the use of SUDALS together with
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
202
flexible electrogoniometers as a complimentary
instrument along with other functional assessment
questionnaires in providing objective data to the
clinicians. This would in turn help the rehabilitation
professionals to improve their intervention. In the
future, we plan to validate SUDALS against vicon
system (Gold standard) and use the system to
evaluate the functional outcomes of patients
following Total Knee Arthroplasty. The system can
be further developed to interface with mobile
devices and provide real time display of data
collected. Thus, a flexible, compact, powerful and
portable multi-channel data collecting system of
flexible electrogoniometry has been designed and
developed.

Figure 4: SUDALS mounted on a subject during
experimentation.
ACKNOWLEDGEMENTS
I would like to thank the University of Strathclyde
for funding my PhD with the Overseas research
student award (ORSA) and I would also like to
thank Mr. J ohn Mcclean (Technician
Bioengineering Unit, University of Strathclyde),
who has dedicated his time in assisting us with all
technical issues that were present during the
development of the system.



REFERENCES
Andriacchi.T.P. et al, vol 62-A, no: 5, 749-757, 1980. A
study of lower-limb mechanics during stair climbing,
J ournal of bone and joint surgery.
Costigan.P.A et al, vol16, 31-37, 2001. Knee and hip
kinetics during normal stair climbing. Journal of Gait
and posture
Huddleston.J et al, 3:21, 2006. Ambulatory measurement
of knee motion and physical activity: preliminary
evaluation of a smart activity monitor. Journal of
neuroengineering and rehabilitation
J evsevar.D.S et al, vol73, no: 4, 1993. Knee kinematics
and kinetics during locomotor activities of daily living
in subjects with knee arthroplasty and in healthy
control subjects. Journal of physical therapy
Kettlekamp et al, vol 52, 775-790, 1970. An
electrogoniometric study of knee motion in normal
gait. Journal of bone and joint surgery
Lark S.D et al, vol 91, 287-295, 2004. Knee and ankle
range of motion during stepping down in elderly
compared to young men. European journal of applied
physiology
Protopapadakki.A. et al, vol22, 203-210, 2007, Hip, Knee,
ankle kinematics and kinetics during stair ascent and
descent in healthy young individuals, Clinical
Biomechanics.
Rowe P.J et al, vol 87, no: 9, 479 487, 2001. Validation
of FEG as a measure of joint kinematics. Journal of
Physiotherapy
Rowe P.J et al, vol13, no: 2, 131-138, 2005.The effect of
TKA on joint movement during functional activities
and joint range of motion with particular regard to
higher flexion users. Journal of Orthopedic surgery
Wyss.U et al, 2003, High range of motion activities of
daily living: Differences in the kinematics between
Hong Kong and Chennai, India Subjects. ISB XXth
Congress ASB 29
th
Annual Meeting, Ohio
Zhu.H et al, vol 38, No: 7, J uly 1991.A Microprocessor-
Based Data acquisition system for measuring plantar
pressures from ambulatory subjects. IEEE
TRANSACTIONS ON Biomedical Engineering
Zhang.Z&Liu.P, 2004. Application of bluetooth
technology in ambulatory wireless medical
monitoring. 4th international conference on
microwave and millimeter wave technology
proceedings
DEVELOPMENT OF STRATHCLYDE UNIVERSITY DATA LOGGING SYSTEM (SUDALS) FOR USE WITH
FLEXIBLE ELECTROGONIOMETERS
203
ELECTRONIC DEVICE FOR SEISMOCARDIOGRAPHY
Noninvasive Examination and Signal Evaluation
Zdenek Trefny
1
, Milan Stork
2
and Martin Trefny
1

1
Cardiological laboratory in Prague U Pruhonu 52, 17000 Praha, Czech Republic
2
Univeristy of West Bohemia, Univerzitni 8, 30614 Plzen, Czech Republic
[email protected], [email protected]
Keywords: Quantitative seismocardiography, Heart rate, Systolic force, Direct digital Synthesis, Analog/digital
converter, Savitzki-Golay filter.
Abstract: The Quantitative seismocardiography (Q-SCG) opens a new field of cardiovascular dynamics examination.
Using this absolutely non-invasive method, a new field of monitoring heart rate variability was opened up.
Systolic forces as well as heart rate variability in relation to changes in external stimuli are registered. Q-
SCG probably offers a more complex view of both isotropic and chronotropic heart functions. It will be
suitable for: examining operators exposed to stress; for assessing the effect of work, fatigue and mental
stress; for monitoring persons as part of disease prevention; for determining a persons ability to carry out
their duties both on the ground and in the air. An electronic system for acquisition of data for noninvasive
Q-SCG and signal processing is also presented. The measuring system is based on analog filter,
analog/digital converter, microcontroller and personal computer. A special digital smoothing polynomial
filter is used for signal processing. The example of real measured and evaluated signal is also shown.
1 INTRODUCTION
1.1 Balistocardiography
In balistocardiography (BCG), the body vibrations
caused by the heart activity are registered.
Balistocardiography is a non-invasive method
enabling the examination of the cardiovascular
dynamics. This field has a longer history than is
commonly known. William Harvey (1578-1657)
who discovered blood circulation called his work,
published in 1628, Exercitatio anatomica de motu
cordis et sanguinis in animalibus. As the title
suggests, this work covers two main topics:
a) Movement of the heart
b) Movement of the blood
Harvey also states that movement is one of the basic
functions that sustain circulation. This process
requires impulse and force (impetus et violentia),
which are produced by the heart (impulsor). The
heart itself may produce force and impulse, while
blood is propelled and forced to leave its source and
home, towards the peripheral parts of the body.
In 1936, Starr took part in a conference held by the
American Society of Physiology which dealt with
methods of determining cardiac output. For this
purpose, he used a bed with tight springs, whereby
by the movement against these springs increased the
instruments natural frequency to values higher than
the heart rate. Thus began the era of high-frequency
balistocardiography, which lasted approximately 15
years. Other types of instruments were developed
later on which measured the displacement, velocity
or acceleration of a body lying on a bed. Later
studies showed that there are difficulties when
comparing records registered on different
apparatuses. This is mainly caused by two factors:
a) The instruments natural frequency
b) The instruments damping
The instruments natural frequency lies within the
range of the frequencies caused by the cardiac
activity that we wish to observe. This leads to
interference and the subsequent recording is a
summation of the oscillations of the instrument and
those of the heart. The other factor that significantly
affects the shape of the registered curve is the
damping installed in these instruments (which are
basically oscillatory systems) in order to prevent the
periodic oscillations of the instruments themselves.
Records of heart activity are therefore deformed.
204

Figure 1: Principle of the noninvasive quantitative
seismocardiography measuring: PT - piezoelectric
transducer, ES - electronic system, PC - personal
computer.
1.2 Quantitative Balistocardiography
Following the critical evaluation of all these facts, in
1952 it was begun with our own experiments related
to the construction of an apparatus which would lack
the aforementioned shortcomings. Thus, over the
years, an apparatus was constructed whose
advantages lie not only in the simplicity of its
design, but also in its important functional qualities.
The properties of the pick-up device and bearing
structure, the subjects sitting position in close
contact with the seat and an amplifier with a
sufficiently long time constant reduce the possibility
of shape, phase and time deformation of the records.
All this enabled to conduct a physical and
mathematical analysis of the balistocardiographic
system and to calibrate our instrument. Based on
these processes, the apparatus was designated a
quantitative balistocardiograph. This was chiefly to
distinguish it from previous instruments that
registered displacement, velocity and acceleration
and were designed to determine cardiac output on
one hand, and also because our instrument was
calibrated so that force expressed in Newton's
registers an amplitude measurable in mm, whereby
the relationship between the size of the active force
and the registered amplitude is linear, on the other
hand. The quantitative balistocardiographic method
enabled to introduce two characteristic quantities:
systolic force (F) and minute cardiac force (MF),
thus using quantitative balistocardiography in an
exact manner when studying cardiovascular
dynamics at rest and during stress. Current
applications of quantitative balistocardiography (Q-
BCG) in papers published to date the fact that the
relationship between the force acting on the pick-up
device and the amplitude of the balistocardiographic
curve is linear was proved. This enabled to study the
evolution of systolic force in relation to age and
ageing, the influence of hypoxia and hyperoxia. It
was also possible to follow the changes in Q-BCG
indices at rest and under workload in various groups
of volunteers, and to determine the linear
relationship between the skeletal muscle force and
systolic force, and determine changes in Q-BCG
indices in various pathological states. Our
parameters, determined by Q-BCG, with parameters
determined using other non-invasive methods were
compared. (Trefny at all, 1996).
1.3 Quantitative Seismocardiography
During a visit to the Flight Psychophysiology
Laboratory at Wright-Patterson Airforce Base, a new
application field for Q-BCG emerged. This made
use of the fact that our method enables the recording
of force applied without phase or time deformation.
Thus, heart rate may be monitored and analyzed
using the method of heart rate variability. The
method of Q-BCG was designated by the laboratory
employees as absolutely non-invasive, as the persons
examined did not have any electrodes attached to the
body surface and was not connected by cables to the
registering instrument. This new field of monitoring
heart activity, whereby we determine both
amplitude-force and time-frequency relationships, is
termed Quantitative Seismocardiography (Q-SCG).
(Trefny at all, 1998). Thus, one may determine the
force-response of the cardiovascular system to
changes in external stimuli, as well as the
autonomous nervous system regulation of the
circulation and the activity of the sympathetic and
parasympathetic systems. The basic part of the Q-
SCG is a rigid piezoelectric force transducer resting
on steel chair. The examined person sits on the seat
placed on the transducer and force caused by the
cardiovascular activity is a measured (Figure 1). The
natural frequency of the chair is higher then 1 kHz
so that there is no interference with the vibrations
caused by the heart activity. Neither damping nor
isolation from building vibrations are necessary.
These properties enabled to calibrate
seismocardiographic system and determine the
absolute value of force acting upon the pick-up-
device. (Trefny at all, 1999).
The system described in the present study enable
better signal evaluation based on high resolution
analog/digital converter (ADC), digital filtration and
digital correction of nonlinearities and noise
suppression by means of personal computer (PC).
The heart rate (HR), systolic force (F), minute
ELECTRONIC DEVICE FOR SEISMOCARDIOGRAPHY - Noninvasive Examination and Signal Evaluation
205


cardiac force (MF) and breathing frequency (BF) is
non-invasively measured.

Figure 2: The simplified block diagram of the electronic
system for Q-SCG measuring. The main parts of the
system are: Piezoelectric force transducer, Analog Filter,
PGA - programmable gain amplifier, Sigma-Delta A/D
converter, Digital Filter, microcontroller and Personal
Computer connected to system by means of USB
(Universal serial bus).

Figure 3: The frequency and phase responses of analog,
electronically controlled filter.
2 MATERIALS AND METHODS
The electronic system used for data acquisition
consists of a piezoelectric force transducer (PT),
analog front end for low frequency measurement
applications (containing ADC), microcontroller and
PC. The block diagram of the whole system is
shown in Figure 2. It is important to note, that
amplitude of measured signal from PT is sometimes
under 1 mV (depend on subject heart activity) and
desired frequency spectrum is lower then 30 Hz.
The measured signal is corrupted by strong noise,
baseline wander, etc. therefore the analog and digital
signal processing (DSP) are used for signal
denoising. The frequency and phase responses of
electronically controlled analog bandpass filter are
shown in Figure 3. The analog front end of A/D
converter can accept either 2 low level input signals
(10 mV to 1.225 V, depends on PGA setting) and
produce serial digital output. (AD7707, 2000). It
employs a sigma-delta conversion technique to
realize up to 16 bits of no missing codes
performance.

Figure 4: The frequency response of on-chip digital filter
in A/D converter.
The sigma-delta modulator output is processed by an
on-chip digital filter. The first notch of this digital
filter can be programmed via an on-chip control
register allowing adjustment of the filter cutoff (1.06
Hz to 131 Hz) and output update rate (4.054 Hz to
500 Hz). The -3 dB frequency f-3dB is determined
by the programmed first notch frequency according
to the relationship (1):
f
-3dB
= 0.262 f
FN
= 0.262 f
s
[Hz] (1)
where f
FN
is filter first notch frequency and f
s
is
output update rate (sampling rate). The AD7707s
digital filter is a low-pass filter with a (sinx/x)
3

response (also called sinc
3
). The transfer function for
this filter is described in z-domain by:
and in the frequency domain by:
3
1
1 1
( ) .
1
N
z
H z
N z

(2)
and in the frequency domain by:
3
sin( / ) 1
( ) .
sin( / )
s
s
Nf f
H f
N f f

= [Hz]
(3)
where N is the ratio of the modulator rate to the
output rate (modulator rate is 19.2 kHz for
Xtal=2.4576 MHz).
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
206
The frequency responses of the digital filter are
shown in Fig. 3 and Fig. 4. Phase response is given
by (4):
Phase(f) = -3 (N -2) f/f
s
[Rad] (4)
The data from A/D converter are next filtered also
by Savitzky-Golay Smoothing filter (SG).

Figure. 5: Record of the Q-SCG in normal man, age 45
years, 78 kg, after lowpass and highpass filtration. Raw
signal is filtered by SG filter and Remez, finite impulse
response (FIR) filter; 250 samples =1 sec.
Savitzki and Golay defined a family of filters which
are suitable for smoothing and/or differentiating
sampled data (commonly called Savitzki-Golay,
DISPO - Digital Smoothing Polynomial, or least-
square smoothing). (Savitzki and Golay, 1994). The
data are assumed to be taken at equal intervals. The
smoothing strategy is derived from the least squares
fitting of a lower order polynomial to a number of
consecutive points. (Madden, 1998). For example, a
cubic curve which is fit to 5 or more points in a least
squares sense can be viewed as a smoothing
function. (Bromba, Ziegler, 1998), (King at all, 1999).

Figure 6: Breathing frequency derived from raw signal by
means of two SG filters.

Figure 7: The heart rate variability can be also detected
from Q-SCG signal.
An example of Q-SCG measurement is illustrated in
Figure 5. The output update rate was 250 Hz, f
-3dB

was 62.5 Hz. The tree SG filters with different
window length were used for Q-SCG and breathing
signal processing. Data on Y axis are decimal values
of A/D converter. The breathing signal detection is
shown in Figure 6.
The heart rate variability (HRV) can be also
evaluated from Q-SCG signal. The signal processing
example for beat to beat detection is shown in short
time slice of signal in Figure 7. After calibration (Y
axis in Newton), the systolic force F and minute
cardiac force MF can be computed according (5) and
(6):
F = (F
HI
+ F
IJ
+ F
JK
)/3 [N]
(5)
MF= F * HR [N. beats/min]
(6)
where HR is heart rate and F
HI
, F
IJ
, F
JK
can be find
according Figure. 8. The systolic force represent the
force response caused by the heart activity and is
expressed in units of force [Newton]. For the total
intensity of the heart activity is introduced the
minute cardiac force which equals the systolic force
multiplied by the HR.

Figure 8: The systolic force (F) determination from Q-
SCG measured signal from points: H, I, J K.
ELECTRONIC DEVICE FOR SEISMOCARDIOGRAPHY - Noninvasive Examination and Signal Evaluation
207


3 DISCUSSION
The anatomy and function of single organs of human
organism are in correlation. This is true for muscle
mass, the body weight and the muscle force too. The
reason of this fact is that higher body weight needs
for the defined movement greater force, which
cannot be realised but by the development of the
skeletal musculature. Consequently greater
musculature needs more energy which is transported
and distributed by the cardiovascular system. In
addition, the increased performance of the
cardiovascular must be adjusted by the heart muscle.
From these relationship it can be concluded that
there must be not only the correlation between the
skeletal muscle force and the heart mass but also
between the skeletal muscle force and the systolic
cardiac force as it was observed in the present study.
According to our opinion these results may be
extrapolated generally for healthy men without
pathological changes in cardiovascular system.
4 CONCLUSIONS
The principles of Q-SCG, measuring system for
noninvasive measuring of heart activity, breathing
and heart rate variability was presented. Also signal
processing for Q-SCG evaluation was described.
ACKNOWLEDGEMENTS
Zdenek Trefny preparing this paper has been
supported by: Grant Eureka E! 2249.
Milan Stork's participation has been supported by:
Department of Applied Electronic and Tele-
communication, University of West Bohemia and
from GACR (grant No. 102/07/0147).
REFERENCES
Trefny Z., Trefny M., David E., Machova J ., Svacinka J .,
1996. Some physical aspects in cardiovascular
dynamics, J . Cardiovaasc., Diag. and Procedures, 13,
No: 2, pp. 141-45.
Trefny Z., Svacinka J ., Trefny M., Trojan S., Slavicek J .,
Kittnar O., 1998. Relation between cardiac force and
maximal sceletal muscle force, Journal of Physiology,
Cambridge University Press.
Trefny Z., Svacinka J ., Trefny M., David E., David V.,
Trojan S., Slavicek J ., Kittnar O., 1999. Noninvasive
method - quantitative balistocardiography (Q-BCG)
and its value, XIII. Congres of the Cardiovascular
System Dynamics Society, Gent., Belgium.
AD7707, 3-Channel 16-bit, Sigma-Delta ADC, Analog
Devices, 2000, Internet site address:
http://www.analog.com.
Savitzki and Golay, 1994. Smoothing and Differentiation
of Data by Simplified Least Squares Procedures,
Analytical Chemistry, vol. 36, no. 8, pp. 1627-38.
Madden H., 1998. Comments on Smoothing and
Differentiation of Data by Simplified Least Square
Procedure, Analytical Chemistry, vol. 50, no. 9, pp.
1383-86.
Bromba U., Ziegler H., 1998. Application Hints for
Savitzki-Golay Digital Smoothing Filters, Analytical
Chemistry, vol. 53, no. 11, pp. 1583-86.
King R., Ruffin C., LaMastus F., Shaw D., 1999. The
Analysis of Hyperspectral Data Using Savitzki-Golay
Filtring - Practical Issues (Part 2), IEEE IGARSS 99
proceedings.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
208


AN ASIC SOLUTION FOR INTELLIGENT ELECTRODES AND
ACTIVE-CABLE USDED IN A WEARABLE
ECG MONITORING SYSTEM
Geng Yang, J ian Chen, Fredrik J onsson, Hannu Tenhunen and Li-Rong Zheng
School of Information and Communication Technology
Royal Institute of Technology (KTH), Forum 120, SE-16440 Kista, Stockholm, Sweden
[email protected], [email protected], [email protected], [email protected], [email protected]
Keywords: Wearable ECG, ASIC, Active-Cable, Intelligent Electrode.
Abstract: This paper describes a digital CMOS Application Specific Integrated Circuit (ASIC) solution with the
complete data acquisition and transmission for the use in a wearable electrocardiography (ECG) monitoring
system. The main particularity of this system is related to the proposed reconfigurable microchip
architecture for an intelligent electrode. The chip area is 2.3 mm
2
in a standard 0.18 m CMOS technology.
The chip is operating at 24 MHz system clock with 3.3 V power supply for I/O cells and 1.8 V for the core
circuit respectively. The estimated dynamic power dissipation is only 857 W. The post-layout simulation
results show that the microchip embedded inside an intelligent electrode features ultra low power
consumption and is quite feasible for a hand-held Personal Health Assistant (PHA) which uses a battery as
energy source.
1 INTRODUCTION
The global demographic trend towards an ageing
population is leading to higher probability and
earlier onset of the heart disease, which is the main
cause of death in most countries. In modern
medicine, there are sorts of methods to diagnose
heart disease, such as electrocardiography (ECG),
ultrasound, computerized tomography and so on.
Among these methods, ECG diagnosis can be used
in a wide area due to the advantages of convenience
and low cost (Dong, Zhang and J ia 2008). However,
several unsolved problems still exist in current ECG
monitoring systems, such as high power
consumption, low signal quality and too many
cables.
In this research, we develop a digital CMOS
application specific integrated circuit (ASIC) with
the characteristics of reconfigurable architecture,
high resolution ECG data, ultra low power
consumption and tiny package size. Such an ASIC is
suitable to be embedded inside a paper plaster to
form an intelligent electrode. Moreover, a 3-lead
wearable ECG monitoring system is developed
based on three intelligent electrodes and one single
Active-Cable, with the capability of continuous non-
intrusive ECG signal sampling and real-time ECG
data processing.
Instead of multiple cables in traditional ECG
applications, one single Active-Cable (Yang et al.
2008), with serial communication enabled by
microchips embedded inside intelligent electrodes, is
used to connect all intelligent electrodes together.
ECG data with minimal noise interference can be
achieved due to the new architecture of the
intelligent electrode which enables synchronous
analog/digital (A/D) conversions and signal
processing performed directly on electrodes rather
than inside a portable device (Hung, Zhang and Tai
2004). Prolonged lifetime is achieved due to the
power efficient ASIC architecture for the intelligent
electrode. This solution does not only dramatically
increase the system energy efficiency and patients'
comfort but also provide high-quality ECG data.
This is an industry-relevant biomedical
application, under cooperation with the paper
industry and the semiconductor industry, which is
still in the research and development phase. The aim
of this design is to develop low-cost disposable
paper-based intelligent electrodes which could be
used both in hospital and home healthcare scenarios.
209


(a) (b)
Figure 1: Block diagram of the device.
2 FUNCTIONAL DESCRIPTION
The block diagram in Figure 1 shows the ASIC
structure. In this system, three intelligent electrodes
are connected serially by the Active-Cable.
Biological potential differences between specific
electrodes applied on the patients skin are captured,
pre-processed and digitized directly on the
intelligent electrodes. These obtained signals are
collected via the Active-Cable and finally
transmitted to a hand-held personal health assistant
(PHA) using IEEE 802.11b standard by a Bluetooth
module (Tejero-Calado et al. 2005). Upon the
patients need, the PHA can process the data, display
the ECG waveforms and other parameters on its
LCD screen as a real-time feedback. In addition, the
data can be stored in a local non-volatile memory
card for a further diagnosis. If required, the data can
be sent to the hospital server via GPRS
communication networks (Rasid and Woodward
2005). In the following parts of this paper, we will
describe the principle and the methodology of the
intelligent electrode in details.
2.1 ECG Signal Acquisitions Overview
The signal from the electrode is an analog signal
with low amplitude ranging from 0.1 mV to 5 mV.
The bandwidth of the ECG signal is only 250 Hz.
The electrodes which are positioned on the skin
could generate a polarization voltage of several 100
mV. The amplifiers, which gain the electrode
signals, have to be controlled if the voltage reaches
their limits. A high energy defibrillation impulse of
4.5 kV applied to the patients chest should not
destroy the IC.
According to the functionality shown in Figure
1b, the ECG chip could be divided into analog part
and digital part. The analog part performs the
functions of ECG signal capturing, amplification,
filtering, conditioning, and consequently digitizing
with an integrated 16-bit Delta-Sigma A/D converter
at a sample rate of 1000 samples per second. The
digital part carries out the tasks of digital signal
filtering (Lian and Yu 2005), ECG data store and
transmission.
In order to have an easy functional verification,
we prefer to tape out the analog part and the digital
part individually. The analog part is implemented
and verified on another chip. In this paper, we only
discuss the digital part of the ECG chip.
2.2 Reconfigurable ASIC
As shown in Figure 1b, each microchip mainly
consists of three blocks: Master-block, Slave-block
and M/S Selection Logic. There are two working
modes available for each intelligent electrode.
According to the system definition, each intelligent
electrode can be configured either as a Master-
electrode or a Slave-electrode by the M/S Selection
Logic module. The Master-block works when the
intelligent electrode is configured as a Master-
electrode, otherwise this block keeps at an idle state
to minimize the power consumption. The same
circumstance applies to the Slave-block.
Different from the traditional ECG signal
sampling approach, where the electrode works only
as an electric conductor suffering from many kinds
of noise, in this design, the ECG signal is directly
captured and digitized inside the Slave-electrode
with minimal noise interference. The function of
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
210


Slave-block is to obtain the digitized ECG signal
through A/D interface and store them into an on-
chip DPRAM temporarily. As illustrated in Figure
1b, each Slave-electrode has two pieces of built-in
DPRAMs with 512 words each. At any given time,
one DPRAM is used for the write operation (save
the ECG data); the other DPRAM is used for the
read operation (transmit the ECG data). The finite
state machine (FSM) running on the Slave-block
could guarantee that the ECG data stored in one
DPRAM would be read out before being overwritten
by the upcoming ECG data.
In this system, the intelligent electrode placed on
the patients lower-belly is assigned as the Master-
electrode, which is connected with the other two
Slave-electrodes serially by the Active-Cable. The
Master-electrode takes charge of the whole ECG
monitoring system. The FSM running inside
Package Controller can issue a series of commands
according to the specific system status stored in
local registers. By generating these commands, the
Master-electrode is able to collect ECG data from a
target Slave-electrode, or check the current status of
a certain Slave-electrode. Compared with a Slave-
electrode, a Master-electrode has a larger DPRAM,
because it has to provide enough memory space for
the ECG data from all Slave-electrodes for an
interval of half a second. ECG data collection
process is initiated by the Master-electrode with a
time interval of half a second. During this collection
process, all ECG data stored inside a certain Slave-
electrode will be transferred to the Master-electrode
via the Active-Cable. Finally, all collected ECG data
will be sent to the PHA by the Bluetooth module.
2.3 Active-cable Architecture
All ECG data and command packages are
transmitted over the Active-Cable which is an
indispensable part of this wearable ECG monitoring
system. As illustrated in Figure 1a, it is a thin, soft
and dedicated cable composed of five metal wires,
which are named REF, SCK, SDA, VDD and VSS,
respectively.
REF is an analog reference signal used for A/D
conversions on all Slave-electrodes. ECG data are
obtained by digitizing the electric potential
difference between a local measurement point and
the REF signal. In this design, by controlling a
switch embedded inside Analog Front End block,
the Master-electrode provides its local electric
potential as the system REF signal. The ECG data
transmission reliability is guaranteed by using a 2-
wire bus which is composed of SCK and SDA in the
Active-Cable. SCK carries the serial clock, while
SDA transmits commands and digital ECG data. The
ECG data are transmitted at a bit rate of 0.9 Mbits/S.
The last two wires in the Active-Cable are VDD and
VSS providing 3.3V system power and digital
ground respectively. In order to minimize the
electrical interference induced by neighbouring
wires and achieve high-quality ECG data, REF is
properly shielded with metal foil.
2.4 Slave-chain Scan Process and ECG
Data Collection Methodology
When the system is powered up, the Master-
electrode should know the exact addresses of Slave-
electrodes active in this system. In the current
design, a unique four-bit vector is assigned to each
Slave-electrode as its address.
In order to get a thorough knowledge of the
whole system, the Master-electrode initiates a slave-
chain scan process after power up, shown in Figure
2. All Slave-electrodes that are connected to the
Active-Cable will be scanned during this process.
The Master-electrode sends out a command frame
containing the target Slave-electrode address to the
Active-Cable. Simultaneously, a built-in timer starts
up. An acknowledgement will be sent back to the
Master-electrode if the target Slave-electrode exists.
Subsequently, this target address is saved into the
address-table, meaning that this Slave-electrode is
online and the Master-electrode will talk to it later.

Figure 2: Slave-chain scan process.
AN ASIC SOLUTION FOR INTELLIGENT ELECTRODES AND ACTIVE-CABLE USDED IN A EARABLE ECG
MONITORING SYSTEM
211


Otherwise, there will be no acknowledgement if the
target Slave-electrode does not exist. When the timer
overflows, the Master-electrode asserts that this
Slave-electrode does not exist in the system. It will
try the next Slave-electrode address until the last
Slave-electrode is reached and Slave-chain scan
process stops here.
To analyse the ECG signals, it is important that
they should be stored simultaneously (Desel et al.
1996). In this design, all A/D converters located
inside Slave-electrodes would not start up until the
Synchronous Sampling Command is issued by the
Master-electrode. In other words, this broadcast
command makes synchronous A/D conversions start
to sample skin electrical potential simultaneously.
During the ECG data transmission, time division
multiplexing mode is employed. The Master-
electrode initiates this process every half second,
and visits all Slave-electrodes in a serial way with
one time slot each.
3 LAYOUT AND RESULTS
The embedded microchip is implemented using a
standard 0.18 m cell library for UMC 0.18 m
Mixed-Mode and RF_CMOS process. A photograph
of the ASIC is shown in Figure 3. The Slave-block
is located on the left side; the Master-block is on the
right side. The chip summary is shown in Table 1.
The microchip has a core area of 1000 m
1000 m and a die size of 1525 m 1525 m. It
consists of 121,514 gates excluding memories and
operates at a clock frequency of 24 MHz. The total
number of I/O pins is 54. And hence, a 64-pin CQFP
package is adopted for this design. The package
bonding diagram is shown in Figure 4. Even though

Figure 3: Photography of the ASIC.
the estimated dynamic power dissipation is 857 W
with cell leakage power of 6 W excluding the
power consumption of memories, the actual power
consumption would be far less than this estimated
value because the microchip consumes much less
when it stays at the idle state. The I/O cell power
supply is 3.3 V and the core circuit power supply is
1.8 V. The I/O cell power supply is splitted into 3
2 different power ports to suppress crosstalk.
Table 1: Embedded microchip summary.
Microchip Summary
Technology UMC 0.18 m Mixed-Mode and
RF_CMOS. 1.8 V core, 3.3 V I/O
Die size 1525 m 1525 m
I/O pin number 54
Package 64-pin CQFP
Gate count 121,514
Clock 24 MHz
Dynamic power 857 W
4 CONCLUSIONS
In this paper, we present a novel ASIC architecture
for the intelligent electrode. The chip is taped out
using a standard 0.18 m CMOS process with an
area about 2.3 mm
2
. The post-layout simulation
result shows that this proposed microchip is featured
by ultra low power consumption, high sample rate
and high ECG data resolution. It is quite feasible for
the battery powered long term ECG monitoring and
analysis system. This intelligent electrodes based
wearable ECG monitoring system is capable of
capturing, pre-processing and digitizing of two
analog signals simultaneously. This system has also

Figure 4: Package bonding diagram.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
212


the capability of polling and digitizing analog
signals from additional twelve intelligent electrodes
for a more complicated ECG study.
In the next release of this ASIC design, the
number and size of Dual Port Memories will be
optimized. A switched DPRAM structure will be
adopted to make more chip area for other enhanced
functionalities. Moreover, the analog front end and
A/D converter will be integrated in the next tape out.
ACKNOWLEDGEMENTS
The authors would like to thank iPack Centre at
Royal Institute of Technology, Sweden, who
supports the above project. We would also like to
thank Ms. Zhangwei Yu for her comments.
REFERENCES
Desel, T., Reichel, T., Rudischhauser, S., Hauer, H., 1996.
A CMOS nice channel ECG measurement IC. In 2nd
International Conference on ASIC.
Dong, J ., Zhang, S., and J ia, X., 2008. A portable
intelligent ECG monitor based on wireless internet and
embedded system technology. In International
Conference on BioMedical Engineering and
Informatics.
Hung, K., Zhang, Y.T., and Tai, B., 2004. Wearable
medical devices for tele-home healthcare. In 26th Ann.
International Conf. of the Engineering in Medicine
and Biology Society, IEEE-EMBS.
Lian, Y., and Yu, J ., 2005. A low power linear phase
digital FIR filter for wearable ECG devices. In 27th
Ann. International Conf. of the Engineering in
Medicine and Biology Society, IEEE-EMBS.
Rasid, M. and Woodward, B., 2005. Bluetooth
telemedicine processor for multichannel biomedical
signal transmission via mo-bile cellular networks. In
IEEE Transactions on Information Technology in
Biomedicine, vol.9, no.1 pp.3543.
Tejero-Calado, J ., Lopez-Casado, C., Bernal-Martin, A.,
Lopez-Gomez, M., Romero-Romero, M., Quesada, G.,
Lorca, J ., and Rivas, R., 2005. IEEE 802.11 ECG
monitoring system. In 27th Annual International
Conference of the Engineering in Medicine and
Biology Society, IEEE-EMBS.
Yang, G., Chen, J ., Cao, Y., Hannu, T., and Zheng, L-R.,
2008. A Novel Wearable ECG Monitoring System
Based on Active-Cable and Intelligent Electrodes. In
IEEE Proc of 10th International Conference on e-
Health Networking, Applications and Services,
Singapore.

AN ASIC SOLUTION FOR INTELLIGENT ELECTRODES AND ACTIVE-CABLE USDED IN A EARABLE ECG
MONITORING SYSTEM
213
ROBUST EAR LOCATED HEART RATE MONITOR
L. Rossini, R. Vetter, C. Verjus, P. Theurillat, P. Renevey, M. Bertschi and J. Krauss
Swiss Center for Electronics and Microtechnology, Jaquet-Droz 1, Neuch atel, Switzerland
[email protected], [email protected], [email protected], [email protected], [email protected], [email protected], [email protected]
Keywords:
Ear located heart rate monitor, Strapless heart rate monitor, Ear clip, Earphone, Portable audio players.
Abstract:
We have developed a device for heart rate estimation with the optical sensing unit integrated in a classic
media player earphone. The sensing principle is based on an optical infrared measurement directly on the ear
lobe whereas the heart rate estimation is obtained utilizing robust model-based signal processing techniques
valid even for quasi periodic activities such as running. Nevertheless, the remaining problem related to these
strategies are statistically undetectable inter-beat during short term sporadic activities. In this paper, we present
a novel robust inter-beat discarding method based on an activity related modeling of the expected heart rate
dynamics that incorporates a simple cardiovascular model to reduce related inaccuracies in the heart rate
estimation. A validation protocol has been designed and 9 subjects were asked to carry out their daily normal
ofce activities for a time length ranging from 1 to 2 hours. A global absolute relative error mean of 0.9%
between the estimated heart rate and a reference device and a sensitivity above 90% demonstrate encouraging
performances of the proposed device.
1 INTRODUCTION
The advent of portable audio players is continuously
increasing the popularity of exercising with music.
Besides enjoying their favorite songs, people may nd
themselves often more motivated toward the achieve-
ment of a given training load (Elliott et al., 2005;
Schie et al., 2008). However, to obtain a desired
training effect such as weight-loss or improvement of
the cardiovascular performance, the exercising person
should observe and respect its personal heart rate tar-
get zones (Noakes, 2003). This requires supplemen-
tary equipment such as a heart rate monitor with its
associated chest-strap ECG sensing. Often, due to
the tightness of the strap, the heart rate monitor and
chest-strap sensor may be perceived by users as op-
pressive and may consequently diminish the exhila-
rating and motivating feeling provided by music. In
order to avoid this degradation, a device for heart rate
estimation based on a sensing unit directly located in
a classical audio player earphone has been presented
recently in (Celka et al., 2004; Verjus et al., 2003).
The sensing is based on an infrared measurement at
the ear cartilage whereas the signal processing is di-
rectly performed in the audio player unit where audi-
tory user feedback may be achieved. A critical aspect
of this approach is the signal processing. Adaptive
model-based enhancement of infrared signals is per-
formed to obtain a robust heart rate estimation even
for quasi periodic activities such as for example run-
ning. A remaining problem of this device is a reli-
able management of heart rate estimation during short
term sporadic activities. Indeed, in this case, model
based IR signal enhancement may not operate accu-
rately due to the convergence time of the enhancement
model (Haykin, 1991). Consequently, residual move-
ment related artifacts might induce erroneous inter-
beat detections that may not be detected with a sta-
tistical assessment (Vaseghi, 1996). In order to cope
with this problem, we have developed a reliable dis-
carding of erroneous inter-beat intervals based on an
activity related modeling of the expected heart rate
dynamics. The exploitation of previous biomedical
knowledge allowed us to develop a simple cardiovas-
cular model to reduce inaccuracies in the heart rate
estimations.
214
2 METHODS
2.1 Background
Optical probes for sensing biological tissue proper-
ties based on photoplestysmography (PPG) have been
widely used over the past years for the estimation of
cardiovascular parameters such as for example pulse
oximetry and heart rate (Webster, 1997), (Tremper
and Barker, 1989). Corruption of the PPG signal
arises fromthe inuences of ambient light and subject
motion (Tremper and Barker, 1989) (Trivedi et al.,
1997). Processing of ambient light artifacts is not crit-
ical since the inuence can be measured using multi-
plexing techniques and an artifact free PPG signal can
be restored using subtractive-type techniques (Trivedi
et al., 1997). Various methods for improving the PPG
technique during motion artifacts and low perfusion
of the tissue have been designed (Coetzee and Elg-
hazzawi, 2000). A very sound and robust approach
of motion artifacts removal in PPG measured sig-
nals has been recently addressed (Celka et al., 2004).
The parametric signal enhancement method exploits
the information contained in a motion reference sig-
nal generated by a two-dimensional accelerometer in
order to obtain a robust PPG heart rate estimation.
Very reliable heart rate estimations have been ob-
tained even under intense physical activity such as for
instance running. However, because of the slow con-
vergence of the enhancement algorithm, the heart rate
estimation may become erroneous during sporadic,
short, and transitory activities. The convergence time
of the enhancement algorithm is in the range of the
few seconds that are necessary to obtain sufcient ac-
curacy of the enhancement parameters.
2.2 Sensor and Processing Device
In this paper we propose a fully integrated heart rate
measurement device with sensing located at the ear
and providing reliable estimates of the heart rate that
are robust against short sporadic or transitory activi-
ties. This systemis based on infrared optical measure-
ment of the sub-cutaneous blood ow by transillumi-
nation, together with an integrated two-dimensional
accelerometer (see Figure 1). The chosen optical
wavelength is 875 nm. The emitter is a light emit-
ting diode (LED) and the receiver is a photodetector
(PD). The light wave is sent trough the ear cartilage
and penetrates the skin and blood vessels to nally
reach the PD. The PD transforms the received light
intensity I(t) into a current that is then transformed
into a voltage. Subsequently, as shown in Figure 2, a
compensation of the ambient light is performed using
Figure 1: Ear located sensor device with portable process-
ing unit.
a subtractive multiplexing technique (Trivedi et al.,
1997).
The resulting signal IR(t) together with the two ana-
log acceleration signals a
1
(t) and a
2
(t) are then con-
ditioned through amplication and a 2nd order But-
terworth bandpass lter between 0.5 Hz and 3.5 Hz.
Finally, an analog-to-digital (ADC) conversion is per-
formed on these signals at a sampling frequency of
20 Hz. Signals are processed in the portable unit and
the resulting heart rate estimations are displayed to
the user. A direct auditory user feedback through the
sensing earphone is an attractive option for a future
development.
Figure 2: The optical concept and electronic signal con-
ditioning of the proposed ear located heart rate estimation
device.
2.3 Physical Principle
The principle of the proposed method resides in in-
jecting an optical infrared (IR) signal at the surface
of the body tissue and measuring the resulting op-
tical signal. This signal propagates through the ear
tissue where it is subject to modications due to re-
ection, refraction, scattering and absorption. The
resulting signal is captured by one or multiple opti-
cal sensors distributed on the earphone. For the near
IR wavelength, the light propagation into the tissue
ROBUST EAR LOCATED HEART RATE MONITOR
215
is primarily governed by scattering and absorption
(Cheong et al., 1990). The Beer-Lambert equation
is generally used to describe the phenomenon of light
absorption in biological tissue (Coetzee and Elghaz-
zawi, 2000) relating the injected light I
i
to the output
light I
o
. However, motion artifacts affect the compo-
nents of the Beer-Lambert equation. Under this con-
ditions, the received intensity can be written in terms
of the major contributions:
I
o
(t) = I
i
(t)
tissue

pulse
(t)
motion
(t) (1)
where
tissue
is the static attenuation due to the tissue,

pulse
(t) is the pulsatile component due to variations
in the sub-cutaneous blood ow, and
motion
(t) is the
contribution due to dynamic changes of the tissue in-
duced by movements of the head. The contribution
of
pulse
(t) in equation 1 is of pivotal interest for the
heart rate estimation. On the other hand, the time-
invariant term
tissue
is of no interest and can therefore
be removed using low-pass ltering. Adaptive sig-
nal processing enhancement techniques of the signal
I
o
(t) to cope with long term harmonic contribution of

motion
(t) have been previously successfully presented
in (Celka et al., 2004). The technique developed dur-
ing the present activity addresses the problem of the
heart rate estimation during short and sporadic activi-
ties.
2.4 Proposed Algorithm
2.4.1 Concept
The concept of the proposed ear located heart rate
estimation device is shown in Figure 3. It mainly
consists of an enhancement of the motion corrupted
IR signals, an inter-beat interval (IBI) extraction on
the enhanced IR signal, a discarding of unreliable
IBIs and a nal estimation of the most likely heart
rate through histogram clustering. The enhance-
ment method that receives one part of its parameters
from the activity analysis has been presented in detail
in (Renevey et al., 2001). The histogram clustering is
based on a standard method of statistical signal pro-
cessing (Gersho and Gray, 1992). The new features
that we introduce address the optimal sensor place-
ment strategy, the IR pulse-amplitude reliability as-
sessment, the activity analysis and the estimation of
a prior of heart rate using a simple model of the car-
diovascular dynamics. This prior estimation of heart
rate and its associated condence intervals together
with the reliability index of the IR-pulse-amplitude
are then used to discard unreliable IBIs.
2.4.2 IR-Pulse-Amplitude Reliability
The reliability index of IR-pulse-amplitude describes
the likelihood that a given IBI has been extracted from
an IR signal with an amplitude that corresponds to
the expected amplitude of pulse contributions. It is
obtained by evaluating the probability that the instan-
taneous IR amplitude has been generated by a pro-
cess with mean
IR,re f
and standard deviation
IR,re f
,
which are continuously re-estimated during periods
without any movement activity.
Figure 3: The proposed heart rate estimation algorithm
based on motion artifact removal using accelerometer sig-
nals and discarding of unreliable inter-beat intervals using
activity related cardiovascular modeling.
2.4.3 Optimal Sensor Placement Strategy
Due to peripheral vasoconstriction and under-optimal
sensor placement, IR pulse contribution may be very
low. Therefore, we introduce an optimal sensor place-
ment procedure that evolves during the initialization
phase of the device:
The signal amplitude is fed back to the user by
means of a quality index.
As long as the quality index is below a given
threshold, the user is asked to adjust the sensor
placement to improve the signal quality.
We observed that this strategy leads to a sufcient
quality of the IR pulse contribution. In about 2 %
of the cases only, the subject was unable to properly
adjust the device in an appropriate position. This pro-
cedure could also be processed fully automatically if
a multi-sensor approach is used.
2.4.4 Activity based Model Heart Rate Model
The proposed ear located sensor is based on IR sig-
nals and as such is prone to movement related arti-
facts. During transitory periods when the enhance-
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
216
ment method cannot converge, these artifacts may in-
duce erroneous IBI detections. To discard such er-
roneous detections we improve the statistical robust-
ness of the proposed heart rate estimator by introduc-
ing a modeling of the expected heart rate based on
accelerometer measurements. The study and devel-
opment of cardiovascular models has attracted a wide
spreading interest in the scientic community (see (Le
et al., 2008) and references therein). In our appli-
cation, we are particularly interested in the descrip-
tion of the heart rate dynamics with respect to a given
input activity. Such an approach has been recently
proposed for cycling (Le et al., 2008). In particular,
the instantaneous heart rate is modeled as a weighted
function of its past value and the power generated by
the cyclist that is estimated using commercial devices.
Notice, however, that for activities like running and
walking the generated power output is not obviously
measurable. Nevertheless, during running and walk-
ing the exercisers power output is approximately re-
lated to his speed and a method for speed estimation
using core body located accelerometers has been re-
cently presented (Vetter et al., 2008). Even though
core body accelerometer measurements are not iden-
tical to ear located accelerometer measurement due to
slight head movements during each stride cycle, we
may assume that they are strongly related. We exploit
the method presented in (Vetter et al., 2008) to obtain
a rough measurement of the power output of a run-
ner P(n) as a function of the maximal eigenvalue of
stride-wise speed variations.
The heart rate model estimation HR
model
(n) at
sample n is a combination of the HR value before the
exercise HR
rest
(n) and a heart rate increase HR(n)
HR
model
(n) =HR
rest
(n) +HR(n) (2)
The dynamic behavior of the heart rate variation un-
der physical effort is described by
HR(n) =HR(n 1) +(1 )HR
ATR
(n) (3)
where determines the temporal behavior of our
model. Specically, we use
act
when the activity is
increasing or stationary and
recov
during a recovery
phase. For this model, the evolution of the heart rate
is a function of both its past values and the innovation
term HR
ATR
(n). The innovation term combines the
power output of an aerobic effort P(n) with the differ-
ence between the cardiac anaerobic threshold HR
AT
and the heart rate at rest HR
rest
(n).
HR
ATR
(n) =
P
[HR
AT
HR
rest
(n)] P(n) (4)
The main goal of our model is to obtain a rough es-
timation of the heart rate in order to discard non-
plausible instantaneous estimations. In this sense,
to the various unknown parameters
P
,
act
,
recov
and HR
AT
of our model, we apply commonly utilized
sport physiological values (Noakes, 2003). These
values are then updated during the use of the device
so as to correspond to a users specic prole.
2.4.5 HR Reliability
Sporadic motion artifacts augment the number of dis-
carded marginal values, reducing the reliability of the
estimated HR. For this reason, together with the esti-
mated HR, the user is provided with a HR reliability
index dened as
HR
rel
(n) =
N
IBI,T
(n)
T HR
T
(n)
(5)
where T is the window length multiple of the sam-
pling time, N
IBI,T
(n) and HR
T
(n) are respectively
the number of valid IBIs and the expected mean HR
at the sample n over the time span T.
3 RESULTS
In order to validate the developed estimation tech-
nique, nine subjects were requested to follow an ad-
equate experimental protocol. Subjects were asked
to carry out their daily normal ofce work for a to-
tal time length comprised between 1 to 2 hours and
to carry out 2 to 3 high intensity activities lasting ap-
proximately 10 to 20 seconds. In addition to the de-
veloped ear located heart rate device, subjects were
equipped with a commercial chest-strap based heart
rate monitor (POLAR, RS800) employed as a refer-
ence.
To illustrate the performances of the proposed algo-
rithm, we present quantitative and qualitative results
comparing the heart rate estimated by the proposed
method to the POLAR reference. The most com-
monly applied quantitative assessments are based on
the Mean Absolute Relative Error (MARE) and Mean
Absolute Error (MAE):
MARE = 100
1
N
N

n=1

HR(n) HR
polar
(n)

HR
polar
(n)
(6)
MAE =
1
N
N

n=1

HR(n) HR
polar
(n)

(7)
Furthermore, we compute a mean reliability index
(R
HR
) associated to each measure. R
HR
is obtained
by adding all the HR reliabilities greater than 0.5 and
normalizing by the measurement length N, where HR
ROBUST EAR LOCATED HEART RATE MONITOR
217
reliabilities have been previously dened in equation
5.
R
HR
= 100
1
N
N

n=1
(HR
rel
(n) > 0.5) (8)
The promising performances relative to the entire
database are presented in Table 1. Indeed, for the
whole database, the mean and standard deviation
associated to the MARE are 0.9 and 0.6 respectively.
It is important to observe that the objective of this
validation is to demonstrate the ability of our device
to accurately estimate the heart rate under baseline
resting conditions without concentrating on heart rate
variabilities (HRVs) (of the European Society of Car-
diology et al., 1996). For this reason, we ltered the
heart rate provided by the POLAR to eliminate HRVs
above 0.04 Hz. Therefore, HRVs associated to the
sympathetic and parasympathetic nervous system are
not retained in the validation.
Table 1: Mean Absolute Relative error (MARE), Mean Ab-
solute Error (MAE), and reliability for 9 subjects in baseline
conditions with intermittent sporadic activities.
Subject MARE [%] MAE [bpm] R
HR
[%]
1 0.4 0.3 98
2 0.3 0.6 95
3 1.2 1.0 92
4 1.2 0.8 94
5 0.6 0.8 94
6 1.4 1.0 90
7 0.7 0.5 99
8 2.8 2.4 93
9 1.6 0.9 93
0.9 1.1 94
0.6 0.8 2.8
Although the MARE and MAE provide an infor-
mation about the average estimation performance of
the developed device when compared to the POLAR
RS800, they both only partially describe the impres-
sion perceived by the user. Indeed, the estimation per-
formance may be excellent for long periods despite
being erroneous for some short time intervals. There-
fore, to take into account this temporal variation, we
also apply an analysis of the specic threshold sensi-
tivity, namely, an assessment of the performance as
the percentage of time where the absolute value of
the error is lower than a given threshold . The spe-
cic threshold sensitivity of the algorithmis presented
in Table 2. The percentage of time where the error
in heart rate estimation is lower than the indicated
threshold highlights that low sensitivities are ob-
tained for a threshold of 1% and 1bpm respectively.
However, since the accuracy of the reference heart
rate monitor is about 1%, the analysis may not be
Table 2: Percentage of time where the error in heart rate
estimation is lower than the indicated threshold .

% bpm
Subject 1 3 5 1 3 5
1 92 99 100 95 100 100
2 90 95 97 93 97 99
3 56 88 96 62 90 97
4 68 89 95 76 92 97
5 74 94 98 82 97 99
6 48 85 94 62 91 97
7 77 98 100 84 99 100
8 33 68 83 46 80 91
9 62 83 92 72 92 97
81 90 95 75 93 97
19 10 5 16 6 3
of high statistical relevance. A validation with such
an accuracy should be performed using medical refer-
ence devices. In contrast, sensitivities for a threshold
of 3% and 5% are above or equal to 90%.
Figure 4 illustrates qualitative performances for
the rst subject in the database. One can observe the
high accuracy of the proposed method. Notice that
at the beginning of the recording, the IR signal qual-
ity was insufcient to provide a valid heart rate esti-
mation because the user was unable to properly ad-
just the ear located sensor. However, once the sen-
sor placement strategy is successfully achieved, heart
rate estimation converged as conrmed by a MARE
of 0.4% and relative heart rate estimation error lower
than 1bpm for 99% of the time (see subject number 1
of Table 1 and Table 2).
In Figure 5 we depict the result relative to subject
8. A closer visual inspection on the data highlights
that there are mainly three segments where the results
of the POLAR and of our device diverged. To illus-
trate, we observe two segments at the beginning of the
recording where the heart rate estimated by POLAR
is about 170 bpm despite weak accelerometer signals
corroborating that the subject was in a resting posi-
tion. This unlikely estimation may be the result of
an incorrect manipulation of the chest strap reference
device such as for instance an insufcient humidica-
tion. Finally, for the third segment at about 1300 sec-
onds where we notice a difference between the two
heart rate estimates there is an insufcient reliability
HR
rel(n)
of the heart rate provided by our device.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
218
Figure 4: Heart rate estimation for subject 1 in baseline
condition with sporadic intermittent activities using the pro-
posed algorithm.
Figure 5: Heart rate estimation for subject 8 in baseline con-
dition with sporadic intermittent activities.
4 CONCLUSIONS
In this paper we have presented an innovative ap-
proach for heart rate estimation using infrared opti-
cal measurements of the tissue at the ear lobe. The
sensing device integrated in a classical audio player
earphone improves the users comfort with respect to
commonly used chest-strap based heart rate monitors.
The challenge connected to statistically undetectable
inter-beat during short term sporadic activities has
been approached using a novel robust inter-beat dis-
carding method. The latter incorporates a simple car-
diovascular dynamic model to reduce related inaccu-
racies in the heart rate estimation. Nine subjects were
asked to participate in the validation protocol of the
proposed device. The validation resulted in positive
performances with an absolute relative error mean of
0.9% and a threshold sensitivity above 90% relative
to the chest-strap heart reference monitor.
REFERENCES
Celka, P., Verjus, C., Vetter, R., Renevey, P., and Neuman,
V. (2004). Motion resistant earphone locatedinfrared
based heart rate measurement device. In IASTEDCon-
ference on Biomedical Engineering (BioMED 2004),
pages 582585.
Cheong, W.-F., Prahl, S., and Welch, A. (1990). A review of
the optical properties of bilogical tissues. IEEE Jour-
nal Quantum Electronic, 26:21662185.
Coetzee, F. M. and Elghazzawi, Z. (2000). Noise-resistant
oximetry using a synthetic reference signal. IEEE
Trans. on Biomedical Engineering, 47(8):10181026.
Elliott, D., Carr, S., and Omre, D. (2005). The effect of mo-
tivational music on sub-maximal exercise. European
Journal of Sport Medicine, 5:97106.
Gersho, A. and Gray, R. (1992). Vector Quantization and
Signal Compression. Kluwer Academic Publishers.
Haykin, S. (1991). Adaptive Filter Theory. Prentice Hall.
Le, A., Jaitner, T., Tobias, F., and Litz, L. (2008).
A dynamic heart rate prediction model for train-
ing optimization in cycling. In The Engineer-
ing of Sport(ISEA2008), volume 1, pages 425433.
Springer.
Noakes, T. (2003). Lore of Running. Ed. Champaign.
of the European Society of Cardiology, T. F., the North
American Society of Pacing, and Electrophysiology
(1996). Heart rate variability - standards of measure-
ment, physiological interpretation, and clinical use.
Circ., 93:10431063.
Renevey, P., Vetter, R., Krauss, J., Celka, P., and De-
peursinge, Y. (2001). Wrist-located pulse detectionus-
ing ir signals, activity and nonlinear artifact cancella-
tion. In IEEE EMBS Conference.
Schie, N., A.Stuart, Becker, P., and Rogers, G. (2008). Ef-
fect of music on sub-maximal cycling. South African
Journal of Sport Medicine, 20:2831.
Tremper, K. and Barker, S. (1989). Pulse oximetry. Anes-
thesiology, 70:98108.
Trivedi, N., Ghouri, A., Shah, N., Lai, E., and Barker, S.
(1997). Effect of motion, ambient light, and hypoper-
fusion on pulse oximeter function. Journal of Clinical
Anesthesia, 9:179183.
Vaseghi, S. (1996). Advanced signal processing and digital
noise reduction. Wiley Teubner.
Verjus, C., Vetter, R., Celka, P., and Renevey, P. (2003).
Equipement portable destine a la mesure et/ou la
surveillance de la frequence cardiaque. In European
Patent EP 1 374 763, US Patent 7175601.
Vetter, R., Onillon, E., and Bertschi, M. (2008). Estimation
of a runners speed based on chest-belt integrated iner-
tial sensors. In The Engineering of Sport(ISEA2008),
volume 1, pages 151159. Springer.
Webster, J. (1997). Design of pulse oximeters. IoP Publish-
ing.
ROBUST EAR LOCATED HEART RATE MONITOR
219
A WIRELESS EMBEDDED DEVICE FOR PERSONALIZED
ULTRAVIOLET MONITORING
Navid Amini, J errid E. Matthews, Foad Dabiri, Alireza Vahdatpour
Hyduke Noshadi and Majid Sarrafzadeh
Computer Science Department, University of California, Los Angeles, CA 90095, U.S.A.
{amini, matth122, dabiri, alireza, hyduke, majid}@cs.ucla.edu
Keywords: Wireless, Embedded Systems, Ultraviolet, Erythema, Skin Cancer.
Abstract: The skin care product market is growing due to the threat of ultraviolet (UV) radiation caused by the
destruction of the ozone layer, increasing demand for tanning, and the tendency to wear less clothing.
Accordingly, there is a potential demand for a personalized UV monitoring device, which can play a
fundamental role in skin cancer prevention by providing measurements of UV radiation intensities and
corresponding recommendations. This paper highlights the development and initial validation of a wireless
and portable embedded device for personalized UV monitoring which is based on a novel software
architecture, a high-end UV sensor, and conventional PDA (or a cell phone). In terms of short-term
applications, by calculating the UV index, it informs the users about their maximum recommended sun
exposure time by taking their skin type and sun protection factor (SPF) of the applied sunscreen into
consideration. As for long-term applications, given that the damage caused by UV light is accumulated over
days, it displays the amount of UV received over a certain course of time, from a single day to a month.
1 INTRODUCTION
The skin is the largest organ of the body in both
mass and surface area and skin cancer is the most
common cancer among all existing cancers.
Unfortunately, the incidence of skin cancer has been
increasing dramatically (Ferguson 2005, Sue 2002)
and more than 1 million cases of skin cancer are
reported annually in the United States. There are
three different forms of skin cancer: Basal cell
carcinoma, squamous cell carcinoma and melanoma.
Basal cell carcinoma is cured by a local operation,
but squamous cell carcinoma and melanoma are
more dangerous, as they metastasise to other parts of
body. Melanoma is considered as the most lethal
form of skin cancer and its incidence and mortality
rates have increased dramatically in the past few
decades in the United States (Boscoe 2006). In the
year 2008, about 62,480 persons are expected to be
diagnosed with melanoma resulting in the death of
an estimated 8,420 individuals (ACS 2008).
Alarmingly, the incidence of melanoma is increasing
rapidly in children (Strouse 2005).
Overexposure to solar radiation, especially the
ultraviolet (UV) region (wavelengths 280400 nm)
of the solar spectrum, is the predominant risk factor
for the development of all forms of skin cancer (Hu
2004, J emal 2000).
While some sunlight is needed to synthesize
vitamin D, which is necessary for human health,
increased exposure to UV radiation is harmful; it is
well known that apart from the skin damage, the
solar UV radiation is extremely injurious regarding
the eyes (DeFabo 1983). During a field study
involving 94 volunteer subjects, the UV exposure of
seven anatomical sites during six different outdoor
activities was investigated (Herlihy 1994). The
results of this study verify the importance of UV
monitoring during outdoor activities to avoid skin
and eye damage.
Currently there is an internationally accepted
parameter, UV index, for measuring the intensity of
UV radiation (Wong 1995). A ground-based
instrument that measures the amount of UV light
from the sun at 5 different wavelengths between 306
and 320 nm is Brewer Spectrophotometer (RMI
2008) which is very widely-used to calculate UV
index. However, the price is too expensive and for
the correct operation an expert technician is
required. Another method used is the solar light 501
biometer which is a wide-band UV radiation
measurement device (Solar 2008). Although it is
220

easy to use, its high energy consumption and heavy
weight prevent it from being exploited in wireless
and small embedded devices.
The use of lightweight embedded systems to
assist in biomedical applications is being pursued by
many researchers and will become available as soon
as required sensors are made available. Hence, given
that the damage caused by UV light is accumulated
over days, there is a potential need for an embedded
device that monitors the amount of UV received
over a certain course of time, from a single day to a
month and gives corresponding advice. In recent
years, the concern over personal exposure to solar
UV has led to the development of numerous
personal (e.g., lapel or wristband) or hand-held
devices. These come with a variety of features, but
in general require input on skin type and degree of
protection and subsequently provide an audible
warning when sun exposure should cease. The
aforementioned devices solely show the current UV
index and they are not able to make accurate
measurements (OS 2008, EPP 2008). Further, none
of them has taken the received amount of UV during
for instance, one week, into consideration.
This work highlights the development and initial
validation of a wireless and portable embedded
device for personalized UV monitoring which is
based on a novel software architecture, written in
Python and is compatible with all Symbian OS cell
phones, a high-end UV sensor, ML8511 introduced
in 2008 by OKI Semiconductor (OKI 2008), and
conventional PDA (or a cell phone). The main
market for this wireless embedded device includes:
1- Children, since reducing UV exposures in early
life is fundamental to skin cancer prevention in
children. 2- Sun protection of outdoor workers in
occupational situations. 3- Sun protection of the
general population in recreational situations. 4-
Beach users and 5- Protection of photosensitive
patients.
This paper will start with preliminary notions
concerning the UV monitoring embedded device in
Section 2. Afterwards, the hardware and software
structure of this device will be presentred in Section
3. An example of the gathered data in a typical day
will be demonstrated in Section 4. Finally, Section 5
concludes this paper and discusses future research
directions.




2 PRELIMINARIES
This section presents the definition of related
concepts used in the design of the personalized UV
monitoring device.
2.1 UV Index
The Global Solar UV index (UVI) (Maddodi 2008,
HKO 2008) describes the level of solar UV radiation
at the Earths surface. Generally, providing the
public with an easy-to-understand daily forecast of
UV intensity is the main purpose of the UV index.
The values of the index range from zero upward
the higher the index value, the greater the potential
for damage to the skin and eye, and the less time it
takes for harm to occur. An index of 0 corresponds
to zero UV irradiation (darkness). The UV index is
an open-ended linear scale defined as follows:
,
400
250

=

nm
nm
er er
d ) ( s . E . k UVI

(1)
where E

is the solar spectral irradiance (see the

previous section) expressed in W/(m
2
nm
1
) at
wavelength and d is the wavelength interval used
in the summation. s
er
() is the erythema reference
action spectrum, and k
er
is a constant equal to 40
m
2
/W.
The determination of the UVI can be through
measurements or model calculations. Two
measurement approaches can be taken: The first is to
use a spectroradiometer and to calculate the UVI
using the above formula. The second is to use a
broadband detector that has been calibrated and
programmed to give the UVI directly.
Although the weather stations assign a unique
UVI to a large area, for several reasons, the
measured noon UV index can be different from the
forecast, sometimes by as much as 100%; the
forecasted UV index does not include the effects of
atmospheric pollutants or haze which can
substantially decrease UV intensity, especially in
urban areas. On the other hand, the forecast does not
take into account variable surface reflection (e.g.,
sand, water, or snow), which can substantially
increase individuals exposure at the beach or on
ski-slopes. The following facts demonstrate how the
environment or terrain affects the level of UV
radiation that we are exposed to (UNEP 2008):
1- Wet fresh snow can reflect as much as 85% of
UV radiation this means that snow reflection can
double overall UV exposure. Similarly, white
A WIRELESS EMBEDDED DEVICE FOR PERSONALIZED ULTRAVIOLET MONITORING
221

water and sand can intensify the UVI by up to
50% and 20% respectively.
2- Every 100 meters increase in altitude results
in the UVI increase of 100%. This is because at
higher altitudes a thinner atmosphere absorbs
less UV radiation.
3- 25% of the UV is reflected from white-water
reflection.
4- 80% of UV rays pass through a cloud.
Therefore, even with cloud cover, the UVI can
be adequately high.
5- Concrete buildings reflect 15% of the received
UV.
6- Shade can reduce UV by 50% or more.
The foregoing facts show that quantitative
measurements and research on UV radiation in
different environments and settings are vital in
developing and assessing UV-preventative strategies
for the reduction of skin cancer and other UV-
related problems for humans.
The UV index measurements and forecasts for
cities in many countries around the world are now
routinely posted on the internet by various
meteorological agencies (see Table 1 for UV index
categories). In summer in Europe, the solar UV
index typically peaks at values from 5 at high
latitudes (Scandinavia) to around 7 in central regions
such as the United Kingdom, France or central
Europe and up to 9 or 10 in Southern Europe. In the
United States maximum measured UV indices range
from 10 to 12 in the Southern continental United
States to 5 in Alaska while in Canada peak values
reach 8 in the southern cities. Generally, the highest
reported UV index measurements in the Northern
hemisphere have been recorded at high altitude. For
example, Bogota in Colombia, at over 2500m above
sea level, has registered a UV index of 16 and for
Mauna Loa volcano in Hawaii, at 3400m above sea
level, a UV index of 17 has been reported (Parisi
2000).
Table 1: UV index intensity.
UV index Extent
0-2 Low
3-5 Moderate
6-7 High
8-10 Very high
11+ Extreme
2.2 Sensor Characteristics
The spectral sensitivity characteristics of photo
diode used in ML8511, which are based on thin-film
Silicon-on-Insulator Technology (SOI), are shown in
Figure 1. Although this is a silicon photo diode, it is
highly sensitive and at the same time is selective
only to the UV-A (wavelengths of 320 to 400 nm)
and UV-B (wavelengths of 280 to 320 nm); this is
because of its SOI structure.
0
0.25
0.5
0.75
1
1.25
280 320 360 400 440 480 520 560 600
Wavelength (nm)
S
e
n
s
i
t
i
v
i
t
y

(
R
e
l
a
t
i
v
e

V
a
l
u
e
)
Figure 1: Spectral sensitivity characterisitics of the
ML8511.
3 THE UV MONITOR DEVICE
In this section the hardware and software structure
of the personalized UV monitoring device is
explained.
3.1 System Design
Figure 2 shows the schematic diagram of our
personalized UV monitoring system. The UV sensor
shown in the figure is a photo diode that uses a thin-
film SOI. With an additional filter, the accordance
with the erythema action spectrum curve of the
human skin has been further improved. A voltage
proportional to the amount of electrical current is
output by a current-to-voltage conversion amplifier,
comprised of an operational amplifier and resistor,
Rf. The output voltage can be linked directly to the
analog-to-digital converter (ADC), where it is
converted into a digital signal and input into the
processor of Atmel ATmega 128L microcontroller
via an interface (I/O). The resulting digital signal is
processed by the microcontroller to determine the
current UV index (see Table 2 and (2)) and
thereafter it sends the UV index data to the RN-24
Bluetooth adapter. The Bluetooth adapter sends its
received data without any changes to the Nokia N95
cell phone where the main software of our
embedded device is running. The software was
written in Python and is compatible with all
Symbian OS cell phones. As mentioned before, the
software has two parts. First, it shows the current
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
222

Figure 2: Schematic diagram of the personalized UV monitoring system.
UV index (from 0 to 20) and maximum exposure
time based on the skin type of the user, SPF of the
applied sunscreen and current UV index. Second, it
shows the UV history of the person together with the
amount of UV (UV dose) that the person received
during the past day/week/month. The software also
saves the average of UV indices during one hour of
operation.
The calculation of the UV index is performed
using equation (2); in the darkness, the sensor output
and corresponding output of the ADC are equal to
0.993 V and 320 respectively. Therefore, in order to
derive the UV index, it is required to subtract 320
from the current ADC output and scale it as shown
in (2). In this equation, the ceiling function and the
multiplication by 0.2 show that our accuracy in UV
index calculation is not better than 0.2.
The software takes the skin type of the user and
the SPF number (from 0 to 100) of the applied
sunscreen (0 in case of no sunscreen), respectively
as the input.
. .
ADC
UVI
OUTPUT
2 0 )
5
320
(

=

(2)
The information in the Table 3 were used in the
software to calculate how long one can stay in the
sun without sunscreen before he/she starts to burn
(i.e., before minimum erythemal dose, occurs).
Table 3 also shows the tolerable MEDs with respect
to solar light for each of four possible skin types.
The flowchart of the software is shown on Figure
3. In addition to the UV history, both the current UV
index and maximum exposure time are shown on the
cell phone screen.




Table 2: UV indices corresponding to sensor and ADC
outputs (Vcc =3.0 V).
Sensor output voltage ADC output UV index
0.993 320-345 0
1.073 345-370 1
1.153 370-395 2
1.233 395-420 3
1.313 420-445 4
1.393 445-470 5
1.473 470-495 6
1.553 495-520 7
1.633 520-545 8
1.713 545-570 9
1.793 570-595 10
1.873 595-620 11
1.953 620-645 12
2.033 645-670 13
2.113 670-695 14
2.193 695-720 15
2.273 720-745 16
2.353 745-770 17
2.433 770-795 18
2.513 795-820 19
2.593 820-845 20
3.2 Exposure Time
At present, the majority of countries, on the basis of
the recommendations of the COST-713 (Vanicek
2000), have adopted four skin types as a function of
tanning capacity.
A WIRELESS EMBEDDED DEVICE FOR PERSONALIZED ULTRAVIOLET MONITORING
223


Figure 3: Software operation (it receives the new UV
index each 15 seconds and stores to average for each
hour).
The principal characteristics of these skin types,
defined by the DIN 5050 standard, were shown in
Table 3, which also indicates the dose (in J /m2)
needed to produce one MED.
According to Table 3, a person with a skin type
of 3, in a UV index of 10, will start to sunburn after
just 20 minute of unprotected exposure to the sun:
[200 (min) / 10 (UVI) =20 min], (3)
and by using an SPF 30 sunscreen this becomes 600
minutes, or 10 hours:
[20 (min) 30 (SPF) =600 min].
(4)
4 EXPERIMENTAL RESULTS
In this section we see an example of the gathered
data in a typical day. Figure 4 and Table 4 show an
example of our personal cell phone software. The
data were gathered on a partially cloudy day in J une
in Los Angeles and we kept the device fixed on the
roof of an eight-story building. Therefore, it can
represent the amount of UV that an individual has
absorbed in a typical day.
Table 3: Skin types and corresponding tolerated MEDs
and maximum exposure time.
Skin
type
Color, burning
and tanning in
the sun
Tolerable
MEDs
Maximum
exposure
time
1
White, always
burns, never tans
2 hecto
J /m
2
67 min /
UVI
2
Yellow and white,
usually burns,
sometimes tans
4 hecto
J /m
2

100 min /
UVI
3
Yellow and black,
sometimes burns,
usually tans
5.75 hecto
J /m
2

200 min /
UVI
4
Black, rarely
burns, always tans
8.5 hecto
J /m
2

300 min /
UVI

In Table 4, as it can be seen, the first row values are
the UV indices at the exact hours of 8:00, 9:00, etc.
However, the second row shows the average amount
of UV index during each hour.
Moreover, for each day we calculate the
parameter UV dose that is:
UV Dose =Average UV Exposure time. (5)
While the UV index is a measure of UV
intensity, it is the accumulative dose that is
important for human exposures to solar UV.
5 CONCLUSIONS
In this paper we adopt OKI ML8511 UV sensor to
implement a real-time and personalized UV
monitoring.
The personalized UV monitoring device
explained in this paper can be put on the hat, can be
a necklace, skin patch and even a clip-on. Therefore,
it is not obtrusive and can decrease the incidence of
the skin cancer in an efficient way. Since the system
is small and accurate, it proved to be a very feasible
commercial product.
This device will enable users to embed UV
sensor functions in a variety of portable devices,
offering them the ability to check their UV exposure
wherever they are.


BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
224
0.0
2.0
4.0
6.0
8.0
10.0
12.0
5
:
0
0
5
:
4
5
6
:
3
0
7
:
1
5
8
:
0
0
8
:
4
5
9
:
3
0
1
0
:1
5
1
1
:0
0
1
1
:4
5
1
2
:3
0
1
3
:1
5
1
4
:0
0
1
4
:4
5
1
5
:3
0
1
6
:1
5
1
7
:0
0
1
7
:4
5
1
8
:3
0
1
9
:1
5
2
0
:0
0
2
0
:4
5
Time
U
V

I
n
d
e
x

Figure 4: Example of personal UV monitor software (UV index was updated each 15 seconds throughout a partially cloudy
day).
Table 4: Example of personal UV monitor software (average data).
Time 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00
UV index
(ending on the hour)
1.2 2.2 4 5.4 6.2 9 5.6 5.8 4.6 2.4 1.2 0.0
Hourly mean UV
index
1.6 3.1 4.2 5.4 7.8 7.4 6.1 5.5 3.9 2.2 0.8 0.0

REFERENCES
Ferguson, K., 2005, Melanoma (Clinical updates). Journal
of Continuing Education in Nursing, Vol. 36, No. 6,
pp. 242243.
Sue, N., Roberta, S., 2002, The management of high-risk
melanoma: Staging, treatment, and nursing issues,
J ournal of Dermatology Nursing, Vol. 14, pp. 363.
Boscoe, F., Schymura M., 2006, Solar ultraviolet-B
exposure and cancer incidence and mortality in the
United States, BMC Cancer Journal, Vol. 6, pp. 264.
ACS, 2008, American Cancer Society, Cancer facts and
figures, http://www.cancer.org/.
Strouse, J . J ., Fears, T. R., Tucker, M. A., Wayne A. S.,
2005, Pediatric melanoma: Risk factor and survival
analysis of the surveillance, epidemiology and end
results database, Journal of Clinical Oncology, Vol.
23, pp. 47354741.
Hu, S., Ma, F., Collado-Mesa, F., Kirsner, R. S., 2004, UV
radiation, latitude, and melanoma in US Hispanics and
blacks. Archives of Dermatology, Vol. 140, pp. 819
824.
J emal, A., Devesa, S. S., Fears, T. R., Hartge, P., 2000,
Cancer surveillance series: Changing patterns of
cutaneous malignant melanoma mortality rates among
whites in the United States, Journal of the National
Cancer Institute, Vol. 92, 811818.
DeFabo, E.C., Noonan F.P., 1983, Mechanism of immune
suppression by ultraviolet radiation in vivo. I.
Evidence for the existence of a unique photoreceptor
in skin and its role in photoimmunology, The Journal
of Experimental Medicine, Vol. 158, pp. 84-98.
Herlihy, E., Gies, H.P., Roy, C.R. J ones, M., 1994,
Personal dosimetry of solar UV radiation for different
outdoor activities, Journal of Photochemistry and
Photobiology, Vol. 60, pp. 288-294.
Wong, C.F., Toomey, S., Fleming, R.A., Thomas, B.W.,
1995, UV-B radiometry and dosimetry for solar
measurements, Journal of the Health Physics Society,
Vol. 68, No. 2, pp. 175184.
RMI, 2008, The Royal Meteorological Institute of
Belgium, http://www.meteo.be/.
Solar, 2008, Solar Light Company, 501 UV-Biometer,
http://www.solar.com.
OS, 2008, Oregon Scientific Company, UV Monitor with
Exposure Timer, http://www.oregonscientific.com.
EPP, 2008, ePlus premium Ltd., UV Index Meter,
http://www.eplus-premium.com
Maddodi, N., Setaluri, V., 2008, Role of UV in Cutaneous
Melanoma, Journal of Photochemistry and
Photobiology, Vol. 84, No. 2, pp. 528536.
HKO, 2008, Hong Kong Observatory, UV index forecast,
http:.www.hko.gov.hk.
FUNEP, 2008, United Nations Environment Programme,
UV index, http://www.unep.org/.
Vanicek, K., Frei, T., Litynska, Z., Schnalwieser, A.,
2000, UV-Index for the Public, COST-713 Action
(UV-B Forecasting), Office for Official Publications of
the European Communities, pp. 27.
Parisi, A. V., Kimlin, M. G., Wong J . C. F., Wilson, M.,
2000, Diffuse component of the solar ultraviolet
radiation in tree shade, Journal of Photochemistry and
Photobiology B: Biology, Vol. 54, pp. 116120.
OKI, 2008, OKI Semiconductor, ML8511 UV sensor,
http://www.oki.com/.

A WIRELESS EMBEDDED DEVICE FOR PERSONALIZED ULTRAVIOLET MONITORING
225
IMMUNOSENSORS FOR ATRAZINE DETECTION
IN RED WINE SAMPLES
Enrique Valera, ngel Rodrguez
Micro and Nano Technologies Group (MNTg), Departament dEnginyeria Electrnica
Universitat Politcnica de Catalunya,C/. Jordi Girona 1-3 Campus Nord, Mdul C4, Barcelona 08034, Spain
[email protected], [email protected]
J avier Ramn-Azcn, Francisco J . Sanchez, M.-Pilar Marco
Applied Molecular Receptors Group (AMRg), IIQAB-CSIC
CIBER of Bioengineering, Biomaterials and Nanomedicine, Jordi Girona 18-26, 08034 Barcelona, Spain
Keywords: Immunosensor; Interdigitated -electrodes, Atrazine; Impedance spectroscopy, Conductive measurements,
Wine matrix effect, Food safety.
Abstract: Two novel immunosensors, one impedimetric and other one conductimetric, for atrazine detection in red
wine samples have been developed. Impedimetric immunosensor is based on an array of interdigitated -
electrodes (IDEs) and bioreagents specifically developed to detect this pesticide. Conductimetric
immunosensor incorporates additionally gold nanoparticles. Bioreagents were covalently immobilized on
the surface of the electrodes (interdigital space). In both cases the biochemical determination of atrazine is
possible without any redox mediator. For the case of the impedimetric immunosensor, the detection method
is based on impedimetric measurements (in a wide range of frequencies), whereas in the case of the
conductimetric immunosensor the detection method is based on conductimetric measurements (DC
measurements).The potential of the impedimetric immunosensor to analyze atrazine in complex sample
matrices, such as red wine, have been evaluated. This immunosensor can detect atrazine with limits of
detection in the order sub-ppb, far below the maximum residue level (MRL) (50 g L
1
) established by
European Union (EU) for residues of this herbicide in the wine grapes.
1 INTRODUCTION
In the recent years, modern chemical analysis has
been revolutionized by the electrochemical
biosensors because of their accuracy, technical
simplicity, high efficiency, possibility of portability
and miniaturization, and because they offer fast
(instantaneous) response times, allow a rapid and
permanent control and a direct transduction of the
biomolecular recognition event into electronic
signals (Murphy, 2006; Pumera, 2007; Wang, 2006).
An important disadvantage of the
electrochemical sensors is that the impedance
changes due to biomolecular recognition are
generally very small and it can not reach the
necessary detection limits required by the
legislation. However, this disadvantage can be
solved by applying techniques such as
electrochemical impedance spectroscopy (EIS) or
including labels that amplify the signal.
By means of EIS is possible to record
information on biorecognition events, occurring at
the electrode surfaces, inducing impedance changes
(Guan et al., 2004; Katz and Willner, 2003), that can
be directly measure, allowing the development of
label-free biosensing devices.
On the other hand, many types of labels are used
to amplify biorecognition events. Between the
labels, the gold nanoparticles are some of the most
recently used (Pumera, 2007; Zhang, 2007), because
its unique properties at nanoscale dimension.
In this work we report the description of two
immunosensors for atrazine detection. One
immunosensor is label-free, whereas the other one is
labelled with gold particles (40 nm). In the case of
the label-free immunosensor, impedimetric
measurements (impedance spectroscopy) are used as
detection method. In the second case, due the
presence of the gold particles, conductimetric
measurements (DC measurements) could be used.
226

In order to demonstrate the applicability of the
devices developed, wine production has been
selected as scenario for the proof-of-concept study.
Reasons for this selection are because wine is a high
value product (with a great economic relevance in
EU) and because their strong matrix effect. In fact, if
the sensors are demonstrated to red wine samples,
their use can be extrapolated to many other matrices.
Likewise, we will prove that the impedimetric
immunosensor can respect the Maximum Residue
Level (MRL) established by EU (European Union)
for residues of this herbicide in wine grapes (50 g
L
-1
).
2 EXPERIMENTAL
2.1 Instrumentation
Impedimetric and conductive measurements were
carried out at room temperature in a probe station
(Faraday cage) KARL SUSS. Impedance analyses
were performed using an Agilent 4294A Precision
Impedance Analyzer and conductive measurements
were performed using an Agilent 4156C
Semiconductor Parameter Analyzer. The
competitive curves were analyzed with a four-
parameter logistic equation using the software
SoftmaxPro v2.6 (Molecular Devices) and GraphPad
Prism version 4.00 for Windows (GraphPad
Sofware, San Diego California USA). Data shown
correspond to the average of at least three replicates
per concentration of atrazine.
2.2 Arrays of Interdigitated
-electrodes
The immunosensors developed are based on a two
coplanar non-passivated interdigitated metallic -
electrodes. Thin Au/Cr (200 nm thickness)
interdigitated electrodes with 10 m pitch were
patterned on a Pyrex 7740 glass substrate (purchased
from Przisions Glas&Optik GmbH, 0.7 mm (0.05)
thickness). For the immunosensor measurements,
electrode arrays were constructed consisting on six
IDEs organized on a 0.99 cm
2
area. The Pyrex
substrate was first cleaned using absolute ethanol.
Following metal deposition was performed by
sputtering and the interdigitated -electrodes were
patterned by a photolithographic metal etching
process. The chromium layer, much thinner than the
gold layer, was deposited prior the gold to improve
adhesion to the Pyrex substrate.
2.3 Immunosensors Functionalization
Biofunctionalization with 2d-BSA was done
selectively on the surfaces of the gold electrodes.
The chemical recognition layer was covalent
immobilized on the interdigital space. For this
purpose the array of IDEs were treated as follows.
2.3.1 Surface Cleaning
Before functionalization, the IDEs samples were
first cleaned with a combination of ethanol:Mili-Q
water 70:30, absolute ethanol absolute, piranha
solution and absolute ethanol.
2.3.2 Surface Activation
After the pre-treatment explained above, surface
activation took place in two steps to modify
selectively the gold electrodes and the Pyrex
substrate.
Activation of gold surfaces was readily and
specifically performed using thiol-chemistry. N-
acetylcysteamine was used to cover the gold
electrodes and to protect the sensor from undesired
non-specific absorptions. Thus, the surface texture
of the IDE defines the template for deposition of
layers, since the gold fingers have been deposited on
a solid support such as glass with the necessary
controlled geometry. In a second step, the Pyrex was
derivatized with 3-glycidoxypropyl trimethoxysilane
(GPTS). The epoxy group provided the necessary
reactivity for further attachment of the bioreagents
through a nuchelophylic attack of functional groups
of the biomolecule such as the amino groups of the
lysine residues (Luzinov, 2000). The immunosensor
surface functionalization is schematically shown in
Figure 1.
2.3.3 Antigen Immobilization
Covalent immobilization of the pesticide antigen 2d-
BSA was performed on the interdigitated -
electrodes surface via the amino groups of the lysine
residues by reaction with the epoxy groups of the
surface.
2.4 Competitive Assay
The assay of detection, common for both
immunosensors, relies on the immunochemical
competitive reaction between the atrazine residues
and the immobilized antigen on IDEs for a small
amount of the specific antibody (primary antibody).
The detection of a small number of molecules of
IMMUNOSENSORS FOR ATRAZINE DETECTION IN RED WINE SAMPLES
227

atrazine is performed under competitive conditions
involving the competition between the free pesticide
(analyte) and a fixed amount of coated antigen for a
limited amount (low concentration) of primary
antibody (Ab
1
). At the end of the reaction the
amount of Ab
1
captured on the IDE surface and
hence the free antigen (analyte) is determined.
Finally, and only in the case of the
conductimetric immunosensor, a secondary labelled
with gold antibody (Ab
2
) is captured in order to
amplify the conductive signal. The immunosensor
assay is schematically shown in Figure 2.
2.5 Impedance Measurements
Impedance measurements were carried out at room
temperature. No redox mediator was used in the
devices presented in this work. The two electrodes
were covered by a diluted PBS solution with a
conductivity of 1.6 S cm
1
and connected to the
input of an Agilent 4294A Precision Impedance
Analyzer by means of standard probe tips.
Measurements were taken in the 40 Hz to 1MHz
frequency range using 0V of polarization potential
and a modulation voltage of 25mV amplitude. All
impedance measurements were performed in a
Faraday cage.
2.6 Conductance Measurements
As for impedance measurements, the conductance
measurements were also carried out at room
temperature without any redox mediator and in a
Faraday cage. The two electrodes were covered by a
diluted PBS solution with a conductivity of 1.6 S
cm
1
and connected to the input of an Agilent 4156C
Semiconductor Parameter Analyzer by means of
standard probe tips. Conductivity was measured to
+25 mV, from +22.5 to +27.5 mV sweep bias. These
conductive measurements were performed after the
incubation step of the secondary antibody labelled
with gold.
3 ATRAZINE DETECTION
In the case of the impedimetric immunosensor, the
quantitative tool that seems adequate to provide
sensitivity graphs is the impedance measurement in
a wide frequency range and the fitting of the Nyquist
plots of impedance spectra to an equivalent circuit.
Then, the atrazine concentration should finally
be related to the values of at least some of the
parameters of the equivalent circuit. The equivalent
circuit used was previously reported (Valera, 2007)
and the resistance of the solution (Rs) was chosen as
parameter of analysis.
Interdigitated -electrode (IDE)
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
IDEs protection with N-Acetylcysteamine
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
inter-digits space functionalization with GPTS
Si
CH2
O O
O
CH2
CH2
O
CH2
HC
H2C
O
Si
CH2
O O
O
CH2
CH2
O
CH2
HC
H2C
O
Si
CH2
O O
O
CH2
CH2
O
CH2
HC
H2C
O
Si
CH2
O O
O
CH2
CH2
O
CH2
HC
H2C
O

Figure 1: Schematic diagram of: i) protection of
interdigitated -electrodes with N-acetylcysteamine; and
ii) immunosensor surface functionalization with GPTS.
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2 SO2NH N NH2 SO2NH N
NH2 SO2NH N NH2 SO2NH N
NH2 SO2NH N NH2 SO2NH N
NH2 SO2NH N
NH2 SO2NH N
NH2 SO2NH N NH2 SO2NH N NH2 SO2NH N
NH2 SO2NH N NH2 SO2NH N
NH2 SO2NH N NH2 SO2NH N
NH2 SO2NH N
NH2 SO2NH N
NH2 SO2NH N
coated antigen
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
primary
specific
antibody
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
O
Si
CH2
O
O
CH2
CH2
HC
H2C
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH
H2C
CH2
S
O
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
NH2
SO2NH N
secondary
antibody
labelled with
gold
nanoparticles

Figure 2: Schematic diagram of the complete assay system
performed on the IDEs for the immunosensors
developed.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
228

For the conductimetric immunosensor, the
measurement of the conductance after the capture of
the secondary labelled with gold antibody could be
used as detection method, as it was previously
demonstrated using PBS in the competitive assay
(Valera, 2008). Then, the atrazine concentration
would be related to the amount of gold nanoparticles
present on the immunosensor.
The impedimetric response of the immunosensor
in red wine is shown in Figure 3.
To validate the sensing approach, our
immunosensors were also characterized by means of
chemical affinity methods. In Figure 4 the results
obtained from the immunosensor are compared with
the results of an ELISA assay on the same IDEs
devices but using a chemically raised colorimetric
signal. In order to comparatively show the
immunosensor performance, in Figure 4 the
normalized values of the change in the Rs as a
function of the atrazine concentration as well as the
normalized results of the ELISA assay are plotted.
It is important to remark that the ELISA assay
performed on the IDE device show a similar
analytical profile that the obtained using microtitrer
plates. Therefore, we could be confident that our
immunosensor really reflects the selective binding
event. The more important analytical features of the
atrazine assays (immunosensors and ELISA) are
shown in Table 1.
As it can be seen in Table 1, using the
impedimetric immunosensor is possible to detect
atrazine in sub-ppb concentrations (0.19 g L
1
).
These good results are directly related to the
advantages of the impedimetric device presented
such as the use of IDEs and the competitive assay
based on the antibodies variation.
Comparing these results with the recent
literature, the impedimetric immunosensor presented
in this work have been demonstrated to be more
sensitive that other biosensors based on
impedimetric methods for the atrazine detection
(Helali, 2006; Hleli, 2006; Fredj, 2008), specially
taking into account that our immunosensor have
been tested using red wine samples.
10
-4
10
-3
10
-2
10
-1
10
0
10
1
10
2
10
3
10
4
400
800
1200
1600
2000
Impedimetric immunosensor
Red wine
[Atrazine, g L
-1
]

R
s

(

)

Figure 3: Impedimetric response when the immunosensor
are used to detect atrazine. Reprinted from Biosensors and
Bioelectronics, 23, J . Ramn-Azcn et al., An
impedimetric immunosensor based on interdigitated
microelectrodes (IDE) for the determination of atrazine
residues in food samples, 13671373, (2008), with
permission from Elsevier.
10
-4
10
-3
10
-2
10
-1
10
0
10
1
10
2
10
3
10
4
0
20
40
60
80
100
Impedimetric
Immunosensor
ELISA
[Atrazine, g L
-1
]
N
o
r
m
a
l
i
z
e
d

s
i
g
n
a
l
s

Figure 4: Normalized calibration curves of the optimized
atrazine immunoassay and the impedimetric
immunosensor presented. Measures were taken in diluted
PBS solution.
Table 1: Features of the atrazine assays in red wine
samples.
Impedimetric
Immunosensor
ELISA
IC
50
, g L
-1
1.8760.23 1.930.02
LOD, g L
-1
0.19 0.09
R
2
0.86 0.99
Reprinted from Biosensors and Bioelectronics, 23, J.
Ramn-Azcn et al., An impedimetric immunosensor
based on interdigitated microelectrodes (IDE) for the
determination of atrazine residues in food samples, 1367
1373, (2008), with permission from Elsevier.
IMMUNOSENSORS FOR ATRAZINE DETECTION IN RED WINE SAMPLES
229

4 CONCLUSIONS
Two immunosensors, one impedimetric and other
one conductimetric, for the atrazine detection in red
wine samples have been developed. Both devices are
based on an array of IDEs and in bioreagents
specifically developed. Impedimetric immunosensor
is able to detect atrazine in red wine at sub-ppb
concentrations, far below the Maximum Residue
Level (MRL, 50 g L
1
) required by EC. However,
this result could be improved using the
conductimetric device, which includes secondary
antibodies labelled with gold particles that increase
the conductive signal.
ACKNOWLEDGEMENTS
This work has been partially supported by the
Ministry of Science and Technology (Contract
number TEC2007-67081) and FEDER funds. The
MNT group is a consolidated Grup de Recerca de la
Generalitat de Catalunya since the year 2001
(expedient 00329). The AMR group is a
consolidated Grup de Recerca de la Generalitat de
Catalunya and has support from the Departament
dUniversitats, Recerca i Societat de la Informaci la
Generalitat de Catalunya (expedient 2005SGR
00207).
REFERENCES
Fredj, H. Ben., Helali, S., Esseghaier C., Vonna L., Vidal,
L., Abdelghani, A. Labeled magnetic nanoparticles
assembly on polypyrrole film for biosensor
applications. Talanta 75 (2008) 740-747.
Guan, J .G., Miao, Y.Q. Zhang, Q.J . Impedimetric
Biosensors. Journal of Bioscience and Bioengineering
97 (2004) 219226.
Helali, S., Martelet, C., Abdelghani, A., Maaref, M.A.,
J affrezic-Renault, N. A disposable immunomagnetic
electrochemical sensor based on functionalised
magnetic beads on gold surface for the detection of
atrazine. Electrochim. Acta 51 (2006) 51825186.
Hleli, S., Martelet, C., Abdelghani, A., Burais, N.,
J affrezic-Renault, N. Atrazine analysis using an
impedimetric immunosensor based on mixed
biotinylated self-assembled monolayer, Sensors and
Actuators B 113 (2006) 711717.
Katz, E., Willner, I. Probing Biomolecular Interactions at
Conductive and Semiconductive Surfaces by
Impedance Spectroscopy: Routes to Impedimetric
Immunosensors, DNA-Sensors, and Enzyme
Biosensors. Electroanalysis 15 (2003) 913947.
Luzinov, I., J ulthongpiput, D., Liebmann-Vinson, A.,
Cregger, T., Foster, M.D., Tsukruk, V.V., Epoxy-
Terminated Self-Assembled Monolayers: Molecular
Glues for Polymer Layers. Langmuir 16 (2000) 504-
516.
Maggio, E.T., 1981. Enzyme-Immunoassay, CRC Press,
Florida.
Murphy, L., Biosensors and bioelectrochemistry. Current
Opinion in Chemical Biology 10 (2006) 177184.
Pumera, M., Snchez, S., Ichinose, I., Tang, J .
Electrochemical nanobiosensors. Sensors and
Actuators B 123 (2007) 11951205.
Valera, E., Ramn-Azcn, J ., Rodrguez, A., Castaer, L.-
M., Sanchez-Baeza, F.-J., Marco, M.-P. Impedimetric
immunosensor for atrazine detection using
interdigitated -electrodes (IDEs). Sensors and
Actuators B 125 (2007) 526537.
Valera, E., Ramn-Azcn, J ., Sanchez-Baeza, F.-J .,
Marco, M.-P, Rodrguez, A.. Conductimetric
immunosensor for atrazine detection based on
antibodies labelled with gold nanoparticles. Sensors
and Actuators B (2008) 95-103.
Wang, J . Electrochemical biosensors: Towards point-of-
care cancer diagnostics. Biosensors and Bioelectronics
21 (2006), pp. 18871892.
Zhang, S.-B., Wu, Z.-S., Guo, M.-M., Shen, G.-L. Yu, R.-
Q. A novel immunoassay strategy based on
combination of chitosan and a gold nanoparticle label.
Talanta 71 (2007) 15301535.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
230
A LOW COST LED BASED BILIRUBIN METER
Description and Evaluation of a Low Cost Spectrophotometer Bilirubin Analyzer
L. A. L. Azeka and M. S. V. de Paiva
Department of Electrical Engineering, University of So Paulo
Avenida Trabalhador, So-carlense- 400, So Carlos, Brazil
[email protected], [email protected]
Keywords: J aundice, Bilirubin, Bilirubin meter, Spectrophotometry, Bilirubin Measurement.
Abstract: J aundice is a disease caused by excessive levels of bilirubin in the blood. It is very common in newborns
and mainly in premature ones. By using resources such as LEDs with high luminous intensity and light to
digital photoreceptors, the present article describes and evaluates a low cost bilirubin meter (Bilimed),
which measures the total bilirubin in the blood serum samples of neonates, using the technique known as
direct spectrophotometry. Samples of newborns were collected and measured in the equipment. The result
was compared with reference values of a commercial bilirubinometer and with laboratory analysis in order
to evaluate the equipments performance.
1 INTRODUCTION
During the first weeks of a newborns life it is
common the appearance of the illness called
jaundice or hyperbilirubinemia defined as a
yellowish discoloration of the skin and the mucous
membranes, caused by the accumulation of a yellow
pigment called bilirubin (Odell, 1980). It is
manifested in about 60% of the newborns, being this
rate much bigger in premature neonates (Knudsen,
1989, Taksande et al., 2005). The increase of the
bilirubin concentration in the blood is generally
caused by the immaturity of the liver in processing
it, among other external factors such as breast-
feeding, hepatic malfunctions and blood
incompatibility (Askin and Diehl-J ones, 2003). The
high value of bilirubin is toxic for the brain and may
cause an encephalopathy called kernicterus, leaving
irreversible sequels to the childs brain (Diamond,
1970).
The determination of the bilirubin levels is
generally carried out by three types of equipment
known as bilirubin meters or bilirubinometers: the
portable analyzer which measures the transcutaneous
bilirubin; the spectrophotometers which use the
photometric analysis in the blood serum to measure
the total bilirubin; and laboratory photometric
analyzers which analyze chemically reacted blood
samples by using scientific methods. The
transcutaneous bilirubinometers are the state-of-the-
art bilirubin measurement method and although
some authors have shown good correlation between
the transcutaneous bilirubin and the direct bilirubin
(Bhutani et al., 2000), not all the results show that
they are totally independent of factors such as race,
gestation and birth-weight (Maisels, 2006).
The bilirubinometers that use the
spectrophotometry as a principle - based on the light
absorption by the bilirubin molecule - are known as
direct spectrophotometers. This equipment was
firstly proposed in the 60s (J ackson, 1961) and
presented reliable results compared with the
transcutaneous method and other laboratory methods
(Grohmann et al., 2006).
The aim of this work is to describe the main
resources used by the direct spectrophotometer
Bilimed bilirubinometer, its technological
innovations, as well as to evaluate the equipment
performance through measurements in newborns
blood samples.
2 MATERIALS AND METHODS
The bilirubin molecule has a peak of light absorption
of about 455 nm. In this same wavelength there is
interference of the oxyhemoglobin molecule, being
necessary to measure its absorption (peaks of 545
nm or 575 nm) in order to make a compensation
(Roggan et al., 1999). The difference in absorption
231

in the bilirubin and oxyhemoglobin 455 and 575 nm
wavelengths is used to measure the concentration
value of the total bilirubin. The figure 1 shows the
absorption spectrum of bilirubin and
oxyhemoglobin.


Figure 1: Absorption spectrum of bilirubin and
oxyhemoglobin. (Du et al., 1998, Zijlstra et al., 1991).
The direct bilirubinometers operation principle
is the use of a high intensity light source in order to
generate the necessary light for the
spectrophotometric process. A white Philips
Lumiled Luxeon K2 LED is used as light source and
a circuit drives the LED through signals sent by the
microcontroller. The capillary tube containing the
serum to be analyzed is inserted in the optic set
through a holder. The light goes through the sample;
the light shaft is split and goes through 455 and 575
nm band-pass filters. The luminous signal is then
transformed in digital signal by the TAOS TCS 230
photosensors which are positioned just after the
filters. This signal is sent to the microcontroller and
a routine converts the frequency sent by the
photosensor into a value correspondent to the
luminous intensity. In order to make the calculation
of the bilirubin absorption, a measurement of a
capillary tube containing distilled water, which will
serve as a reference value, is carried out. The signal
is processed and is converted into a concentration
value of total bilirubin (mg/l or mol/dl).
The user interacts basically using two switches
and a LCD display. The figure 2 illustrates the
system.
Products which use high power LED have
recently revolutionized the medical area. Among
their application can be pointed out: lighting of
surgery rooms using surgical focus; source of
lighting which offers low level of photo biological
influence in closed environments, reducing
malfunctions in night workers and the J et-Lag
effect; dermatological treatment for acne, psoriasis

Figure 2: Chart of the electrical (continuous arrow),
luminous (dashed arrow) and digital signals (dotted arrow)
in the system of the bilirubinometer.
and eczemas; phototherapy for the treatment of
jaundice; ultraviolet LED based disinfection systems
(J ones and Barnett, 2006). The programmable
photosensors which convert the light into frequency
in a sole CI operate with photodiodes matrix in
series with a current to frequency converter circuit.
The frequency of the signal out is directly
proportional to the incident light, and may be
connected to a microprocessor without the need of
any additional electronic or analogical to digital
converter. There are many laboratory and medical
applications using these ICs, such as low cost LED
based spectrophotometer (Yeh and Tseng, 2006) and
glucose meters (King, 2006).
3 RESULTS
130 measurements were analyzed based on 26
samples from different bilirubin concentrations,
besides their respective zero measurements, which
formed a data base for the Bilimed analysis. The
samples were collected in different newborns
hospitalized in the pediatric sector of the Clinical
Hospital of the Faculty of Medicine of Ribeiro
Preto - Brazil.
The obtained values were compared with the
value which was measured by using a commercial
bilirubinometer (Photo Ictometer OHC Model IV-
OHara & Co LTD) and by laboratory analysis
carried out in the pediatric laboratory of the same
hospital. From the collected data, 41 measurements
(30% of the results) have concentrations which is
greater than 10 mg/dl (171 mol/dl). Figures 3 and 4
show the relationship of the concentration measured
by the Bilimed and the linearity of the results.
Electrical Sign
Optic Set
Microcontroler
Lenses and
Filters
Ultra
Bright LED
455 nm
Photosensor
Signal Processing
LCD
Capillary Tube
575 nm
Photosensor
Drive
LED
Circuit
Sensors
Reading
Switches
Regulated Source
Light Shaft Digital Data
Wavelenghts (nm)
420 460 500 540 580
1000
10000
100000
M
o
l
a
r

A
b
s
o
r
p
t
i
o
n

(
l
/

m
o
l
)

Oxyhemoglobin
Bilirubin
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
232



Figure 3: Graphic of the bilirubin concentration by the
concentration of the Bilimed, where the straight line is the
tendency found and the dots are the collected data.
0 5 10 15 20 25 30
0
5
10
15
20
25
30
Linearity
R
e
f
e
r
e
n
c
e

b
i
l
i
r
u
b
i
n

c
o
n
c
e
n
t
r
a
t
i
o
n

(
m
g
/
d
l
)
Bilimed bilirubin concentration (mg/dl)

Figure 4: Linearity of the Bilimed equipment (dots) by the
reference (straight line).

Figure 5: Error in the measurement of the Bilimed
bilirubin concentration (points). The mean is equal to 0
mg/dl (dotted line), the double of the standard deviation is
2,093 mg/dl (dashed line), and the tendency line is shown
by the full line.
The reading error graph, generated by the
difference between the concentrations measured and
the reference ones, is shown on figure 5. The
statistical results on table 1 (standard deviation,
variance and maximum error) of the Bilimed
readings were also calculated.
Table 1: Standard deviation, variance and maximum error
found in the Bilimed reading.
Standard
Deviation
1,048 mg/dl (17,864 mol/dl)
Variance 1,098 mg/dl (18,769 mol/dl)
Maximum Error 2,474 mg/dl (42,303 mol/dl)
3.1 Result Analysis
There is a large linearity in the measurements
carried out by the Bilimed. Through the analysis of
the graphs tendency curve of the bilirubin
concentration and of the signals sent by the
photosensors, the measurements are safe up to 30
mg/dl (513 mol/dl), as for this value the absorption
of the 455 nm light shaft is still within the limit of
the photosensor reading error. Above this value the
luminous intensity and the signal become very low,
significantly increasing the reading error. The
equipment presents standard deviation of 1,05
mg/dl (17,1 mol/dl), a result within the direct
bilirubin meters expectation. The maximal reading
error was 2,5 mg/dl (42,75 mol/dl) for a
concentration of 23,5 mg/dl (401,85 mol/dl). There
is a greater difficulty in obtaining samples with
values above this one, thus limiting the data base.
The main reasons for reading errors are related
to the temperature variations between the
measurements, the bad positioning of the capillary
tube in the holder, the bad separation of the serum in
the centrifugation process, and flaws and dirt in the
capillary tube.
In order to ensure the quality of the results it is
very important to follow the instructions correctly,
to carry out the measurement from zero before each
analysis, to avoid direct contact with the capillary
tube, and to centrifuge the sample so that a
homogeneous serum is obtained.
4 CONCLUSIONS
The measurement of the bilirubin concentration is a
very common procedure in hospitals and the
bilirubin meters are equipment which allow for a
quick evaluation of the patients condition. The
Bilirubin concentration (mg/dl)
Bilirubin concentration (mg/dl)
D
i
f
e
r
e
n
c
e

(
m
g
/
d
l
)

R
e
f
e
r
e
n
c
e

b
i
l
i
r
u
b
i
n

c
o
n
c
e
n
t
r
a
t
i
o
n

(
m
g
/
d
l
)

Bilirubin measurement error
Bilirubin concentration
A LOW COST LED BASED BILIRUBIN METER - Description and Evaluation of a Low Cost Spectrophotometer
Bilirubin Analyzer
233

direct spectrophotometers have been in the market
for a long time and Bilimed has the same reliability,
accuracy and simplicity of these equipments.
In spite of the limited data base, the tests carried
out with the Bilimed equipment show an optimal
linearity in the result, with accuracy of 1mg/dl,
upper range limit of 30 mg/dl and strong co-relation
with the results obtained with the commercial
bilirubinometer used as reference. In order to assure
the reliability of the data, some procedures must be
used.
Bilimed offers technological improvements
contrasting with available devices. There is a strong
tendency in the use of LEDs to substitute high
intensity luminous sources. Due to the use of the
white ultra bright LED and special photosensors, the
size and the number of electronic components
involved in the project were reduced. Another point
that must be emphasized is the reduced production
cost and, as consequence, the final client price.
With the use of filters, lenses, LEDs and
photosensors, the Bilimed equipment can get to 20%
of the price of a transcutaneous bilirubinometer and
50% of a conventional bilirubinometer.
ACKNOWLEDGEMENTS
The authors sincerely thank Olidef cz, PE Lucio
Kimura and Prof. Dr. Arthur Lopes Gonalves for
all the support.
REFERENCES
Askin, D. F. & Diehl-J ones, W. L. (2003) The Neonatal
Liver, Iii: Pathophysiology Of Liver Dysfunction.
Neonatal Network, 22, 5 - 15.
Bhutani, V. K., Gourley, G. R., Adler, S., Kreamer, B.,
Dalin, C. & Johnson, L. H. (2000) Noninvasive
Measurement Of Total Serum Bilirubin In A
Multiracial Predischarge Newborn Population To
Assess The Risk Of Severe Hyperbilirubinemia.
Pediatrics, 106.
Diamond, L. K. (1970) A History Of Jaundice, Baltimore,
Willians And Willians.
Du, H., Fuh, R. A., Li, J ., Corkan, A. & Lindsey, J . S.
(1998) Photochemcad: A Computer-Aided Design
And Research Tool In Photochemistry.
Photochemistry And Photobiology, 68, 141142.
Grohmann, K., Roser, M., Rolinski, B., Kadow, I., Mller,
C., Goerlach-Graw, A., Nauck, M. & Kster, H.
(2006) Bilirubin Measurement For Neonates:
Comparison Of 9 Frequently Used Methods.
Pediatrics, 117, 1174-1183
J ackson, S. H. (1961) A Simple Stable Instrument For
Determination Of Bilirubin By Direct
Spectrophotometry. Clinical Chemistry, 7.
J ones, G. & Barnett, G. (2006) High-Power Leds Provide
Illumination And Treatment In Medical Applications.
Leds Magazine.
King, R. (2006) Optoelectronic Sensors In Medical
Applications. In (Taos), T. A. O. S. (Ed.) Sensors.
Knudsen, A. (1989) Prediction Of The Development Of
Neonatal J aundice By Increased Umbilical Cord Blood
Bilirubin. Acta Paediatrica Scandinavica, 78, 217-21.
Maisels, M. J . (2006) Historical Perspectives:
Transcutaneous Bilirubinometry. Neoreviews, 7, 217-
225.
Odell, G. B. (1980) Neonatal Hyperbilirubinemia, Grune
& Stratton.
Roggan, A., Friebel, M., Drschel, K., Hahn, A. & Mller,
G. (1999) Optical Properties Of Circulating Human
Blood In The Wavelength Range 4002500 Nm.
Journal Of Biomedical Optics, 4, 36-46.
Taksande, A., Vilhekar, K., J ain, M., Zade, P., Atkari, S.
& Verkey, S. (2005) Prediction Of The Development
Of Neonatal Hyperbilirubinemia By Increased
Umbilical Cord Blood Bilirubin. Current Pediatric
Research, 9, 5-9.
Yeh, S. & Tseng, S. S. (2006) A Low Cost Led Based
Spectrometer. Journal Of The Chinese Chemical
Society, 53, 1067-1072.
Zijlstra, W. G., Buursma, A. & Roest, W. P. M.-V. D.
(1991) Absorption Spectra Of Human Fetal And Adult
Oxyhemoglobin, De-Oxyhemoglobin, Carboxy-
hemoglobin, And Methemoglobin. Clinical Chemistry,
37, 1633-1638.

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
234
CORRECTION OF ACOUSTIC LENS ERROR IN SPATIAL
COMPOUNDING OF ULTRASONIC DIAGNOSTIC IMAGES
Myoung H. Choi
Dept. of Electrical and Electronic Eng., Kangwon National University, 192-1 Hyoja- dong, Chuncheon, Korea
[email protected]
Keywords: Spatial compounding, Image registration, Lens error.
Abstract: Spatial compounding has been used in ultrasonic imaging for suppressing speckle noise. The technique
generally involves electronically steering the ultrasonic beams. The steering angle of the ultrasonic beam is
distorted by the acoustic lens structure of the probe that is used to focus the beam mechanically. These
errors introduced by the lens structure cause misalignment of the ultrasonic images received at different
steering angles, and consequently results in the blurred image after spatial compounding. In this paper, a
solution is proposed that corrects the lens error by using image registration. The lens error was compensated
by registering the wire target images before spatial compounding. An efficient registration algorithm was
developed to compute the transformation matrix required for the registration. The images were registered by
the transformation matrix before spatial compounding.
1 INTRODUCTION
Ultrasonic images show a characteristic granular
structure commonly known as speckle (Burckhardt,
1978, Wells and Halliwell, 1981). Speckle is one of
the fundamental problems of ultrasound imaging,
and it is a cause of major limitation on image
quality. Speckle in ultrasonic images arises from the
presence of closely spaced and randomly distributed
microscopic scatterers (Ahbott, 1979, Wagner,
1983). The coherence of the ultrasound sources and
the interference pattern caused by these tiny targets
result in fluctuations in the amplitude of the echo
called speckle. Although speckle noise carries some
information about the nature of the imaging object,
the speckle reduces the detection capability of
ultrasonic imaging systems, and makes it difficult to
identify specific target regions on the image.
Spatial compounding has been used to reduce
speckle brightness variations. The compounded
image is formed by, for example, averaging the
component images that have been acquired by
steering ultrasound beams in several different
directions. In the component images, the structural
targets show consistently strong echoes while
speckles show random variations. Consequently, the
structural targets in compounded images are
enhanced and variations in the soft tissues due to
speckle noise are averaged out (Shankar, 1986). As a
result, image contrast is improved, and electronic
noise is reduced and artifacts such as shadowing and
reverberation are suppressed. Examples in (Entrekin,
2001, Huber, 2002) show improvements in
visualization of breast lesions. These results indicate
that spatial compounding can enhance the
delineation of the boundaries and internal structure
of lesions. The improvement, however, is usually
gained at the price of spatial resolution. The
misalignment of the legions caused by aberration
and the loss of spatial resolution can degrade the
effectiveness of spatial compounding (Krcker,
2002, Meuwly, 2003).
Steering of electronic beam is accomplished
electronically by controlling the excitation of the
individual elements in the ultrasonic transducer
array. One of the errors introduced by the electronic
steering of the ultrasonic beam is caused by the
acoustic lens structure that is used to mechanically
focus the ultrasonic beam, such as the thickness and
acoustic speed of the lens. In Figure 1, ideal case is
compared with the real case. In an ideal case, the
ultrasound echo data reflected off the target is
received along a simple straight line. In a real
situation, the ultrasound echo data path is changed at
the interface between the lens and the tissue. The
steering angle is changed from
1
to
1
by the ratio
of the acoustic speed of the lens and the tissue.
These errors introduced by the lens structure cause
235

misalignment of the ultrasonic images received at
different steering angles, and consequently results in
the blurred image during the spatial compounding.










(a) Ideal
case






(b) Real
case
Figure 1: Acoustic lens error in the steering of ultrasound
beam.
The conventional solution to correct the errors
introduced by the lens structure involve manual fine
tuning of the lens thickness and the steering angles.
These parameters are used in the resampling process
which geometrically transforms the image data from
the probe coordinates to the patient coordinates as
shown in Figure 3(a). In ultrasound imaging systems,
this kind of manual fine tuning can be difficult and
time consuming as many different kinds of probes
are used interchangeably for different diagnostic
applications.
In this work, a systematic solution is proposed
that corrects the lens error by using image
registration. The lens error was compensated by
registering the wire target images before spatial
compounding. An efficient registration algorithm
was developed to compute the transformation matrix
required for the registration. The images were
registered by the transformation matrix before
spatial compounding.
2 SPATIAL COMPOUNDING
WITH LENS ERROR
CORRECTION
A typical spatial compounding scheme that uses 5
component image frames from a linear probe to
generate a compounded image is described in Figure
2. Each of the component images I
k
corresponds to a
steering angle of k, where is the incremental
angle and I
0x
is the image corresponding to the zero
rotation angle. Various methods were proposed to
compute the compounded image from the
consecutive frames which include linear averaging,
median, mean-excluding- minimum, root mean
square, etc (Wilhjelm,2004). In this work, a simple
linear averaging scheme was used.
Let I
i
, i= -N,0,N, denote the component
images that will be used in the spatial compounding.
The center image I
0
undergoes no rotation and thus
is free from the influence of the lens errors. Hence,
the center image I
0
is used as the reference frame
with respect to which all other images will be
transformed for registration. Let T
i
, i= -N,0,N,
denote the geometric transformation matrices that
transforms images I
i
such that J
i
=T
i
I
i
will be the
image registered with respect to the reference frame.
The alignment errors between the component images
are removed by the registration process. The
registered images J
i
can be used to produce the
compounded image by a spatial compounding
algorithm as shown in Figure 3.
Figure 2: Concept of Spatial Compounding.
The transformation matrix T
i
that registers the
component image I
i
with respect to the reference
image I
0
is computed from the images of a wire
phantom. The wire phantom images are converted to
binary images by using a suitable gray level
threshold. Threshold value is selected to generate
clear binary images of wire targets. For each image
I
k
, position vector of the wire targets u
kj
= [x
kj
, z
kj
]
T
,
j=1,M, are obtained by computing the center of
gravity of the wire targets, where M is the number of
wire targets, x
kj
and z
kj
are the x and z coordinates of
the position vector u
kj
.The position vector of the
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
236

same wire target in the reference image I
0
is denoted
as u
0j
= [x
0j
, z
0j
]
T
, j=1,M.
Let the position vectors be expressed in
homogeneous coordinates (Fu,1987). Let H
k
, k= -
N,0,N, denote the 4x4 homogeneous
transformation matrices that transform u
kj
to u
0j
.
Then we can write

1
0
1
0
0
0
kj
kj
k
j
j
z
x
H
z
x
, j=1,, M (1)

=
1 0 0 0
34 33 32 31
24 23 22 21
14 13 12 11
h h h h
h h h h
h h h h
H
k k k k
k k k k
k k k k
k


(a) Conventional spatial compounding.

(b) Spatial compounding with image registration.
Figure 3: Comparison of spatial compounding schemes.
We can remove the y component of the equation and
rewrite (1) as below.

1 1 0 0 1
34 33 31
14 13 11
0
0
kj
kj
k k k
k k k
j
j
z
x
h h h
h h h
z
x
, j=1,, M (2)
This equation can be expressed

1 1
0
0
kj
kj
k j
j
z
x
T z
x
, j=1,, M (3)

=
1 0 0
34 33 31
14 13 11
h h h
h h h
T
k k k
k k k
k

Here, T
k
is the transformation matrix that registers
the wire targets of I
k
with the reference image I
0
.
Since T
k
applies to all the wire targets in the image
I
k
,

1 1 1 1 1 1
2 1
2 1
0 02 01
0 02 01
L
L
L
L
L
L
kM k k
kM k k
k M
M
z z z
x x x
T z z z
x x x

(4)
Let the equation (4) be written in a simplified form
as
k k
U T U =
0
(5)
Then, the transformation matrix T
k
can be computed
by
1
0
) (

=
T
k k
T
k k
U U U U T
(6)
3 EXPERIMENTAL RESULTS
The images were denoted as I
i
, i=-3,,0,3, and
hence images corresponding to 7 different view
angles were used in the spatial compounding. Seven
consecutive images were used in the compounding
computation, each of 5232 x 256 RF sample data,
and compounded images are shown after scan
conversion into a 640 x 480 BW data.
The binary images of I
i
were obtained by selecting
a threshold. Six wire targets were used in the
computation. The wire target positions u
kj
= [x
kj
,
z
kj
]
T
, j=1,6, k= -3, , 0, , 3 were obtained by
computing the center of gravity of thebinary images
of wire targets. The transformation matrix T
k
were
computed using (6), and used to register the
component images I
i
to the reference image I
0
. The
registered images are denoted by that J
k
=T
k
I
k
. The
position error between the wire target of the k-th
image and that of the reference image were
computed before and after the registration and their
magnitudes are shown in Figure 4. Before the
registration, the position error increases with the
rotation angle, error increasing to over 110 pixels
(RF data) in I
-3
and I
3
. After the registration, the
position errors were reduced to less than or equal to
one pixel with the exception of I
3
where the
maximum error magnitude was three pixels.
CORRECTION OF ACOUSTIC LENS ERROR IN SPATIAL COMPOUNDING OF ULTRASONIC DIAGNOSTIC
IMAGES
237

The compounded image obtained by the
conventional spatial compounding scheme is shown
in Figure 5(a) and contain geometric errors caused
by the lens error and the result of compounding
these misaligned images is the blurry compounded
image. The proposed spatial compounding with
image registration was applied to the same image
and the result is shown in Figure 5(b).

Figure 4: Registration error of wire targets before and after
the registration.
(a) no lens error compensation (b) Proposed method
Figure 5: Spatial compounding results.
4 CONCLUSIONS
A lens error correction method for spatial
compounding is proposed that uses image
registration. The lens error was compensated by
registering the wire target images before spatial
compounding. An efficient registration algorithm
was developed to compute the transformation matrix
required for the registration. The images were
registered by the transformation matrix before
spatial compounding. It was shown that the
registration error that causes the blurring of the
spatially compounded images can be removed
effectively.
REFERENCES
C. B. Burckhardt, 1978. Speckle in ultrasound B scans,
IEEE Trans. Sonics Ultrason., SU-25, 1-6.
J . G. Ahbott and F. L. Thurstone, 1979. Acoustic speckle:
Theory and experimental analysis, Ultrason.
Imaging, 1, 303-324.
P. N. T. Wells and M. Halliwell, 1981. Speckle in
ultrasonic imaging, Ultrason., 19, 225-229,.
R. F. Wagner. S. W. Smith, J. M. Sandrik and H. Lopez,
1983. Statistics of speckle in ultrasound B scans,
IEEE Trans. Sonics Ultrason., SU-30, 156-163,.
D. P. Shattuck and O. T. von Ramm, 1982. Compound
scanning with a phased array, Ultrason. Imaging,
4(2), 93107.
M. ODonnell and S. D. Silverstein, 1988. Optimum
displacement for compound image generation in
medical ultrasound, IEEE Trans. Ultrason.,
Ferroelect., Freq. Contr., 35(4), 470476.
S. D. Silverstein and M. ODonnell, 1987. Speckle
reduction using correlated mixed-integration
techniques, in Proc. SPIE 768 Pattern Recognition
and Acoust. Imaging, 168172.
G. E. Trahey, S. W. Smith, and O. T. von Ramm, 1986.
Speckle pattern correlation with lateral aperture
translation: Experimental results and implications for
spatial compounding, IEEE Trans. Ultrason.,
Ferroelect., Freq. Contr., 33(3), 257264.
P. M. Shankar, 1986. "Speckle Reduction in Ultrasound
B-Scans Using Weighted Averaging in Spatial
Compounding", IEEE Trans. On Ultrasonics,
Ferroelectrics, and Frequency Control, Vol. UFFC-33(6).
R. R. Entrekin, B. A. Porter, H. H. Sillesen, A. D. Wong,
P. L. Cooperberg, and C. H. Fix, 2001. Real-time
spatial compound imaging: Application to breast,
vascular, and musculoskeletal ultrasound, Semin.
Ultrasound CT MR, 22(1), 5064.
S. Huber, M. Wagner, M. Medl, and H. Czembirek, 2002.
Real-time spatial compound imaging in breast
ultrasound, Ultrasound Med. Biol., 28(2), 155163.
J . F. Krcker, G. L. LeCarpentier, J . B. Fowlkes, and P. L.
Carson, 2002. Rapid elastic image registration for 3-
D ultrasound, IEEE Trans. Med. Imag., 21(11),
13841394.
J .-Y. Meuwly, J .-P. Thiran, and F. Gudinchet, 2003.
Application of adaptive image processing technique
to real-time spatial compound ultrasound imaging
improves image quality, Invest. Radiol., 38(5), 257262.
Wilhjelm, J . E., J ensen M. S., J espersen S.K., Sahl B.,
Falk E. 2004. "Visual and Quantitative Evaluation of
Selected Image Combination Schemes in Ultrasound
Spatial Compound Scanning," IEEE Trans. on
Medical Imaging, 23(2), 181-190.
Fu, Gonzales, and Lee, 1987. Robotics: Control, Sensing,
Vision and Intelligence, McGraw-Hill.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
238
AN AUTOMATED ATHLETE PERFORMANCE
EVALUATION SYSTEM
From Theory to Practice
Hugo Silva, Gonalo Martins, Susana Palma
PLUX, Biosensor Engineering., Av. 5 de Outubro n 70-8, Lisbon, Portugal
{hsilva, gmartins, spalma}@plux.info
Pedro Mil-Homens, Maria Valamatos
Faculty of Human Kinetics, Lisbon, Portugal
{pmil, mjvalamatos}@fmh.utl.pt
Keywords: Athletic Performance Evaluation, Wireless, Sensors, Real-Time, Automated signal processing.
Abstract: In order to obtain information on athletic performance, strength and power characteristics of the athlete are
generally evaluated. However, due to the large number of variables needed for the assessment, this kind of
evaluations is usually time consuming. Taking advantage of recent developments in the area of sensors and
acquisition systems and using signal processing algorithms reported in the literature, we developed a new
Athletic Performance Evaluation System. This system automatically determines evaluation parameters and
integrates them in ready-made reports, decreasing the time involved in the evaluation process. The system is
based on the installation of sensors and wireless acquisition systems at the assessment workstations of a
Sports Evaluation Laboratory. At present, J ump Platform, Leg Press and Multipower workstations are being
used. Strength and displacement data collected by the sensors at these workstations is automatically
processed in real time at the Central Base Station where standard force and power related evaluation
parameters are determined. Graphical representations of time evolution of the variables being measured by
the sensors are showed in real time on the screen. Each evaluation session is defined by a protocol that can
be specifically created by the coach for each athlete. The results of the evaluations are stored in an athletes'
database so that the historic performance of the athlete can be easily assessed. The resulting system presents
the deployment of sound theoretical evaluation metrics in a real time athlete performance evaluation system.
1 INTRODUCTION
Athletic performance can be assessed by analyzing
specific variables that provide information about the
physical condition of the athlete. Generally, strength
and power related variables are the gold standard for
athletic evaluation. For their assessment, specific
tests and procedures are used (Morrow et al, 2005).
Those tests are designed in accordance to
recommendations and guidelines that assist coaches
collecting valid and reliable data used to determine
the needed evaluation parameters (Brown et al,
2001).
Traditionally, and according to the assessment
guidelines, a large amount of evaluation variables
must be determined in order to obtain complete
information on the athlete's physical condition.
Usually, the data measured during the evaluation
tests is processed a posteriori and summarized in
reports that are analyzed a few hours or even days
after the tests are done, which also contributes to the
slowness of the evaluation. With the recent
technological advances, this situation can be
overcame. Besides introducing new levels of
objectivity, automation and usability into the
evaluation process, the use of new technological
tools may reduce markedly the time needed to
complete a performance evaluation session. The
training process itself is, nowadays, turning into a
process that uses the advantages of new technologies
(Liebermann et al, 2002): the use of sensors and
real-time presentation of the athlete's signals during
training provides athletes and coaches with
sophisticated objective information about the sport
performance evolution and it can also be used as a
real time feedback tool.
239
This paper describes a new Athletic Performance
Evaluation System based on real-time measurement
and processing of signals generated by the athlete
during the assessment tests. Using wireless sensors
and acquisition systems installed on different
evaluation workstations, an athlete-focused system
that includes automated algorithms for determining
strength and power-related variables was developed.
Standard evaluation methods (Hori et al, 2006,
Linthorne et al, 2001; Dowling et al, 1993) used for
manual analysis of data in the traditional evaluation
laboratories were automated in order to obtain a
faster, integrated evaluation system. In this way, the
time involved in testing each athlete is decreased by
a factor of three.
The computed variables are summarized in a
ready-made report and may be used as performance
indicators for purposes such as the quantification of
the relative contribution of strength and power to
athletics events, the identification of specific
weaknesses and prescription of suitable
training/rehabilitation programs, the follow-up of
training/rehabilitation programs or even the
identification of athletic talented individuals.
2 GENERAL DESCRIPTION
The setup of the system involves both hardware and
software modules. Signal acquisition and
transmission is done by wireless signal acquisition
hardware and measurement sensors installed on the
evaluation devices of the laboratory. Data analysis
and reporting as well as management of athletes
database and evaluation protocols are performed by
the different software modules installed on a Central
Base Station.
The system works on a workstation/protocol
basis.
Each workstation is composed of an evaluation
device instrumented with measurement sensors and
a wireless acquisition unit which collects the signals
measured by the sensors. The acquired data is
transmitted in real time to a Central Base Station via
Bluetooth, where it is automatically processed and
represented on a screen. For the moment, three
workstations are predefined: (a) Leg Press: for
evaluation of force production characteristics of the
lower members; (b) J ump Platform: for evaluation of
reactive force of the inferior members and (c)
Multipower: for evaluation of force production
characteristics of the superior members and dynamic
parameters such as power, velocity and muscular
resistance (Figure 1).
Figure 1: Schematic Representation of powerPlux athletic
performance evaluation system setup.
Each evaluation session is defined by a protocol
that is previously designed by the coach for the
athlete. To build a protocol, the coach chooses the
set and sequence of evaluations the athlete will
execute as well as the respective details. In the end
of each evaluation session a report is produced that
summarizes it and presents the best results, allowing
also the introduction of comments by the coach. If
several sessions of a same evaluation protocol are
performed, a comparative report may be produced.
Given the wireless connectivity between the
workstations and the Central Base Station, the coach
can easily follow the athlete's work, in three simple
steps: (a) connect: communicate wirelessly with the
workstation; (b) acquire: automatically access the
main performance indicators in real-time; and (c)
visualize: session and historic reporting and analysis,
with graphical curves representation and gain
indicators.
Simultaneously the athlete has real-time
feedback about his/her performance during the
evaluations, with automated visual and acoustic aids
to support the protocol execution. The fact that this
feedback information is given in real-time to the
athlete provide conditions for him/her to improve
significantly the process of skill acquisition and
sport performance (Schmidt et al, 1999; Liebermann
et al, 2002).
Besides the automated data acquisition,
processing and analysis module, powerPlux also has
a management module which includes an Athlete
Database to store relevant information for each
athlete, each evaluation session and corresponding
final reports.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
240
3 HARDWARE
The recent technological developments in the area of
hardware design and integration offer appropriate
tools for the development of compact, miniaturized
and versatile systems with a large range of
applications in the areas of real-time signal
acquisition and processing (Silva et al, 2005).
At a Sports Evaluation Laboratory, space needs
to be clear of obstacles and evaluation tools should
not interfere with athletes' performance, so that
parameters such as wireless connectivity,
miniaturization, versatility and usability are
desirable features when one thinks of a sports
assessment tool. Taking advantage of the new
acquisition hardware solutions (Silva et al, 2005),
we developed powerPlux Athletic Evaluation
System, which gathers the features referred above.
Each Workstation is instrumented with a 12bit,
1000Hz bioPlux8 wireless acquisition system and
suitable sensors for measuring the signals produced
by the athlete at those workstations (Table 1). In this
way, the analog data measured by the sensors are
converted to digital data by the bioPlux8 uint that
communicates via Bluetooth with the Central Base
Station, where the data is processed (Figure 2).
Table 1: Sensors used in the Workstations.
Workstation Sensors
Leg Press Load Cell
J ump Platform Force Platform
Multipower Load Cell; Displacement sensor

Figure 2: Schematic representation of powerPlux
hardware.
4 SOFTWARE
Signals measured by the sensors of each
workstations are managed by powerPlux Software
Application. It consists of five internal modules: (a)
device and sensor configuration; (b) acquisition and
display; (c) automatic signal processing; (d)
reporting; and (e) database. These modules receive
the signals collected at the different workstations in
order to determine evaluation parameters, present
them in real time and store them in the database.
The architecture of powerPlux software is
intentionally simple. A Home Screen allows to
choose between the Configuration and the Database
sections. The Configuration section is where the
coach manages the evaluation devices, protocols,
and evaluations. The database section is reserved for
accessing athletes profiles: viewing the historic of
evaluations results and performing new evaluation
sessions.
In order to obtain an objective quantification of
the athletes physical condition, reliable and valid
data needs to be collected. The design of evaluation
protocols may be guided by procedures
recommendations published by exercise science
experts (Brown et al, 2001). However, the ideal
evaluation protocol is not yet defined. Sports
scientists and coaches in evaluation laboratories
follow distinct recommendations and rules in order
to design their own evaluation protocols. Taking this
situation into account, we designed a versatile
software application that allows the coach to define
several parameters for his evaluation protocol and
sessions. In this way, a protocol may be costumized
according to the athletes' needs and according to the
different assessment guidelines. This is done in the
Protocol Configuration screen, where the coach may
choose the kind and sequence of the evaluations the
athlete needs to perform (Reactive Force, Isometric
Force or Velocity-Power) as well as a variety of
factors that need to be considered when testing: joint
angle at which to perform the testing, the rest
interval between consecutive repetitions, the number
of repetitions to perform (trials and definitives) or
the duration of each test. Other evaluation-specific
parameters may also be configured for each
evaluation protocol.
Once an evaluation protocol is designed and
chosen for an athlete, it is saved in the respective
page of the athletes database. The athlete may
execute as many evaluation sessions of that protocol
as needed.
The evaluation process is guided by visual
instructions. A graphic representation of the signal is
showed on the screen while the athlete is performing
the evaluations, which gives him real time visual
feedback of his performance.
In the end of an evaluation session a report os
produced with graphic representations of the best
execution with companion tables where the results
determined automatically by the signal processing
algorithms are shown. All the Session Reports are
saved to the Database and identified by date and
AN AUTOMATED ATHLETE PERFORMANCE EVALUATION SYSTEM - From Theory to Practice
241
time. For printing purposes, a printable format of the
Report may also be generated. In this way, the
historic performance progress of the athlete can be
easily accessed, allowing follow-up as well as
performance gains comparison throughout different
evaluation sessions.
4.1 Evaluations
From the analysis of the signals recorded at the
different workstations, powerPlux processing
algorithms determine standard performance
evaluation variables. The fact that the determination
of those variables is automated, introduces a new
level of objectivity into the evaluation process,
allowing a better characterization of the performance
of the athlete.
4.1.1 Reactive Force
J umping is frequently used as a method of
evaluation of reactive and explosive force in lower
members. The measurement and monitoring of
variables such as the jump height, contact time,
impulse or vertical velocity are used to study these
characteristics.
At the J umping Workstation, the athlete
performs vertical jumps while standing on a force
platform that collects the vertical force signal. The
processing of this data allows the computation of the
standard variables related with the evaluation of
lower members reactive force.
Before the evaluation takes place, the coach may
configure some specific Evaluation Parameters,
namely the kind of jumps the athlete will perform
(Squat-jump, Counter-movement J ump or Drop
J ump) (Linthorne, 2001), the number of trial and
definitive collections and the duration of each
collection.
Introduction of algorithm configuration values is
also possible. As an example, the force value above
which the algorithm will detect the beginning of a
jump or the force value for the detection of invalid
squat jumps due to the an initial downward
movement may be different in distinct evaluation
sessions.
powerPlux software automatically processes the
force signal measured during the jumps and
determinates the performance indicator variables in
real time: (a) contact time, (b) gravity center
elevation, (c) vertical velocity and (f) impulse
achieved in each of the repetitions. These
parameters, determined according to standard
methods in the literature (Dowling et al, 1993;
Linthorne, 2001), are presented while the athlete is
jumping (Figure 3) and are stored in the Evaluation
Session Report. The squat jump must always be
performed in a Reactive Force Evaluation Session in
order to assess the relative enhancements in jump
performance due to the effect of stretching the
muscle/tendon complex prior to contracting.
(Linthorne, 2001; Brown et al, 2001). The elevation
of the center of gravity achieved in this jump is
taken as base value with which the values achieved
in counter-movement and drop jump are compared.
Figure 3: Evaluation screen showing the results of a
counter movement jump evaluation.
4.1.2 Isometric Force
Isometric assessment of muscular function requires
the athlete to push maximally against a resistance,
where a measurement sensor is placed, without
movement taking place. This test is one of the oldest
methods used in sports science.
powerPlux allows the evaluation of isometric
force of both superior and inferior members. The test
is done at the Multi-Power and Leg-Press
workstations, respectively. The devices are
instrumented with load cells to measure force data.
The algorithms are applied in real time to this data
and determine, directly from the force-time curve,
the evaluation variables. Figure 4 shows a screen
shot of an evaluation of suprior members' isometric
force.
Maximum Strength (Maximum Voluntary
Contraction) and the speed with which force can be
developed (Rate of Force Production) are important
variables of the isometric force evaluation
(Abernethy et al, 1995; Wilson et al, 1996). These
variables are determined by powerPlux software for
both superior and inferior members by automatic
analysis of the force-time curve and respective
derivative. Other important variables are also
determined: the percentage of the Maximum Force
at the instant when the Maximum Rate of Force
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
242
Production is achieved, the Rate of Force Production
at 50, 150, 250 and 350ms and the Percentage of
Maximum Force at 50, 150, 250 and 350ms.
The evaluation may be unilateral or bilateral,
allowing the comparison between both left and right
members' performances.
Isometric assessments usually display high
test/retest reliabilities (Abernethy et al, 1995).
However, reliability varies between the muscle
groups, the parameter being assessed (Maximum
Force or Rate of Maximum Force Production) and
the posture at which the testing is performed (joint
angle). In order to facilitate the performance of a
comprehensive evaluation and follow-up of the
athlete, these parameters are recorded in the report
as well as the parameters configured before the
evaluation takes place.
Algorithm configuration parameters is also
possible for isometric force evaluation Workstations:
the Initial Force Value, the range of the derivative
and the Acquisition Frequency are configurable
parameters which the coach may adapt for different
evaluations.

Figure 4: Evaluation screen showing the results of a
Isometric Force Evaluation (unilateral).
4.1.3 Velocity-Power
The measurement of power output during exercise
gives useful information to evaluate athletes speed
strength (Hori et al, 2006), which can be used as an
indicator of performance in most athletic activities.
This evaluation can be done with an isotonic test that
consists of moving a sub-maximal load against
gravity as fast as possible.
Weightlifting exercises are effective training
methods to improve speed strength, which makes the
measurement of the power output in this kind of
exercises a helpful tool for coaches
powerPlux system integrates a Multipower
device instrumented with a sensor that measures the
time variation of the load displacement when the
athlete does the weightlifting. An algorithm
determines the velocity and acceleration of the
lifting exercise from the displacement data recorded
by the sensor mounted on the device (Hori et al,
2006). With this data, Force/Velocity and
Force/Power relations can be determined for
muscles under the isotonic situation, namely
regression equations for force and load as function
of velocity.
The maximum amount of weight that can be
lifted in one repetition, i.e., the One-Repetition
Maximum (1RM) is the most common measure of
isotonic strength. 1RM testing involves a trial and
error procedure in which progressively heavier
weights are lifted until the weight exceeds the
subject's ability. The standard procedure involves
starting from a percentage of a (estimated) reference
1 RM value and then lifting progressively heavier
loads until the heaviest successful lift is reached
(Brown, 2001). Before the evaluation takes place,
the coach indicates the estimated 1 RM. The lifting
evaluation starts with a load of 20% of the reference
1RM and continues with equally spaced increases of
loads until the maximum is reached. For each lifted
load, the following variables are automatically
computed and presented in the report: (a) Mean
Power; (b) Mean strength; (c) Mean and Maximum
Velocity of the lifted load; (d) Displacement of the
load; (f) Time to reach the Maximum Force and (g)
Strength Deficit.
Graphical representations of time variation of
displacement, force, force/velocity, force/power and
force deficit are also presented in real time, giving
the athlete feedback of his own work. At the end of
each evaluation session Full Power, ratio between 1
RM and Body Weight (1 RM/BW) and ratio
between the lifted load and Body Weight (W/BW)
are determined and presented in the report which can
be used for follow-up purposes.
Some evaluation parameters of the velocity-
power workstation may be configured, namely the
duration of each test (15s , 30 s, 45 s or 60 s), the
time until the start of the test and the type of
evaluation (unilateral or bilateral). The Acquisition
Frequency is also configurable.
4.2 Reports and Athletes' Database
At the end of the evaluation session, the results and
configuration parameters used are stored in the
Report. Results are organized in tables and
represented graphically for a better understanding
and analysis. When several sessions of the same
AN AUTOMATED ATHLETE PERFORMANCE EVALUATION SYSTEM - From Theory to Practice
243
evaluation protocol are performed, additional
comparative data, including gain indicators, is
generated and added to the summary tables. The
several evaluation session's reports are organized by
date and time, in order to allow a simple and easy
view of the historic evolution of each athlete for
follow-up purposes.
The Database is the key point of athletes'
management since all the data concerning
evaluations and results can be consulted or edited. It
includes the athletes' profiles, where personal data as
well as reports of previous evaluation sessions are
stored.
5 CONCLUSIONS
This paper describes an intergrated, athlete-focused
system for automatic athletic performance
monitoring and evaluation. Making use of the recent
advances in technology, we installed wireless
measurement sensors and data acquisition systems in
the devices traditionally used for physical condition
assessment and designed a specific software that
integrates and automatically processes the data
recorded by the sensors in real time. The signal
processing algorithms are based on standard
methods available in the literature and have some
configurable parameters that make them adaptable to
the specific characteristics of an athlete or group of
athletes.
Thanks to the automatic processing of data and
generation of reports the amount of time needed for
a coach to build and execute an evaluation session is
markedly reduced in contrast to the traditional
methods.
Each evaluation session is defined by a protocol
that is specifically created by the coach for the
athlete. In this way, a comprehensive athletic
evaluation can be made, showing the weaknesses
and strengths of the athelte and helping in the design
of a specific and optimized training program.
REFERENCES
Abernethy, P., Wilson G., Logan, P., 1995, Strength and
Power Assessment. In Sports Medicine, 19, 401-417.
Brown, L. E. and Weir, J. P., ASEP Procedures
Recommendation I: Accurate Assessment of Muscular
Strength and Power. In Journal of Exercise Physiology
online, 4 (3), August 2001.
Dowling, J . and Vamos, L., 1993, Identification of Kinetic
and Temporal Factors Related to Vertical J ump
Performance. In Journal of Applied Biomechanics, 9,
95-110.
Hori, N., Newton, R., Nosaka, K., McGuigan, M., 2006,
Comparison of Different Methods of Determining
Poer Output in Weightlifting Exercises. In Strength
and Conditioning Journal, 28 (2), 34-40.
Liebermann, D. G., Katz, L., Hughes, M. D., Barlett, R.
M., McClements, J ., Franks, I. M., 2002, Advances in
the Application of Information Technology to Sport
Performance. In Journal of Sports Sciences, 20 (10):
755-769.
Linthorne, N.P., 2001, Analysis of standing vertical jumps
using a force platform. In American Journal of
Physics, 69 (11), 1198-1204.
Morrow, J . M.J r., J ackson, A. W., Disch J . G., Mood, D.
P., 2005, Measurement and Evaluation in Human
Performance, 3
rd
edition.
Schmidt, R.A. and Lee, T., 1999, Motor Control and
Learning. IL: Human Kinetics.
Silva H., Gamboa H., Viegas V., and Fred A., 2005.
Wireless physiologic data acquisition platform. In
Proceedings of the 2005 Conference on
Telecommunications.
Wilson, G.J ., Murphy, A.J ., 1996, The use of Isometric
Tests of Muscular Function in Athletic Measurement.
In Sports Medicine, 22(1):19-37.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
244
SIMULATION AND EXPERIMENTAL DESIGN OF A
SYMMETRY CONTROLLER FOR FES CYCLING OPTIMISED
ON STROKE PATIENTS
Emilia Ambrosini, Simona Ferrante
NEARLab, Bioengineering Department, Politecnico di Milano, via Golgi 39, Milano, Italy
[email protected], [email protected]
Thomas Schauer
Technische Universitt Berlin, Control Systems Group, Einsteinufer 17, Berlin, Germany
[email protected]
Alessandra Pedrocchi, Giancarlo Ferrigno
NEARLab, Bioengineering Department, Politecnico di Milano, via Golgi 39, Milano, Italy
[email protected], [email protected]
Keywords: FES cycling, Control systems, Stroke, Rehabilitation.
Abstract: This study deals with the design of a controller for FES cycling able to assure a symmetrical pedalling. This
controller is exploitable in the rehabilitation of patients with unilateral neurological disorders which need to
recover a symmetrical use of the legs. The controller updates the values of pulse width used in the
stimulation of the two legs in order to nullify the unbalance between the torques produced at the cranks and
then to maintain a symmetrical pedalling. The controller was tested first in simulation by means of a neuro-
musculo skeletal model of a stroke patient which identifies the kinematic and dynamic of cycling induced
by FES. After a stability analysis and an optimization of the controller tuning performed in simulation, the
controller was tested experimentally on a healthy subject. The results of this trial show that the controller is
able to reach a symmetrical pedalling in about 18 s, starting from an unbalance of 0.73 Nm. Furthermore, it
is able to maintain a symmetrical task with small oscillations of the PWs. Thus, its employment in the
rehabilitation of stroke patients could be crucial in the recovery of motor functions, such as walking, where
a cyclical symmetrical motor task is required.
1 INTRODUCTION
Cycling induced by means of controlled functional
electrical stimulation (FES) of the large leg-
actuating muscles is an interesting method in the
rehabilitation of patients with neurological or
muscular disorders (Scheffler, 2007). We will refer
to this artificial movement with the term FES
cycling. In order to induce this task, superficial
electrodes are placed over the target muscle groups:
always over the quadriceps and the hamstrings,
sometimes over the gluteus maximum, the plantar-
flexors and dorsal-flexors of the ankle. Patients sit
on the ergometer and the computer controls the
appropriate sequence of stimulation of the muscles
in order to obtain the cycling movement. A shaft
encoder at the crank measures the crank angle; each
muscle is stimulated for a particular range of the
crank angle to turn the pedals. Usually, FES cycling
system use an active drive mechanism (for example,
an electric motor) to maintain a constant cadence.
Indeed, iso-kinetic training devices allow a larger
number of patients to undertake FES cycling, i.e.,
those unable to generate and maintain sufficient
muscular force to rotate a flywheel or those with low
tolerance to FES due to residual sensation.
FES cycling has become an established method
in the rehabilitation of patients with Spinal Cord
Injury (SCI) (Hunt, 2004; Faghri, 2005). Only
recently it was demonstrated the importance of FES
cycling on patients with stroke (Ferrante, 2008a).
When patients do not have a complete spinal lesion
245

the employment of FES cycling could become
crucial in re-learning the Central Nervous System
the proper sequence of activation of the muscles
involved in the task (Scheffler, 2007).
Designing a proper FES cycling controller
presents several challenges. First, the physiological
system to be controlled is strongly non-linear and
time variant. Further, there is an high degree of
uncertainty and variability in the response properties
of the system. In this context, it would be helpful to
design a neuro-musculo skeletal model able to
identify the response of the subject to FES in order
to increase the awareness about the system to control
and to test the implemented controllers before
performing experimental trials on patients. Another
crucial aspect in the development of control
strategies for FES cycling is the identification of the
rehabilitative objective and, thus, the choice of the
controller rationale; both these steps depend on the
target pathology. In this contest, another essential
step is the choice of the sensors needed on the
ergometer in order provide a real-time controlled
signal. Up to now only controllers for patients with
complete SCI aimed at the maximization of the
power output or at the minimization of muscular
fatigue were developed (Hunt 2004; Hunt 2006). In
those studies, the ergometer was equipped with a
torque sensor at the crank and thus the power output
could be chosen as the controlled signal (Hunt,
2008).
The aim of the study was the development of a
control system for FES cycling optimised for the
rehabilitation of patients with stroke. Because of the
laterality of the pathology, the recovery of the motor
symmetry is crucial in the rehabilitation of these
patients. Thus, the controller, starting from a real-
time measure of the unbalance between the torques
produced at the two cranks, modifies the stimulation
parameters of the two lower limbs independently in
order to achieve a symmetrical pedalling. The
implemented controller was first tested and validated
by means of a simulation model and then some trials
on healthy subjects were carried out.
2 THE DESIGN OF THE
SYMMETRY CONTROLLER
The structure of the closed-loop system developed is
shown in Figure 1. It includes two parallel branches
in order to control two systems contemporaneously.
Each system corresponds to a lower limb, whose
input (control signal) is the value of pulse width,
PW, used to stimulate the selected muscles of the leg
and whose output (controlled signal) is the torque, T,
produced at the crank. For each leg, the muscular
groups included in the stimulation strategy are the
quadriceps and the hamstrings, which are the
muscles which mostly contribute to the pedalling
(Ferrante, 2008b). The range of stimulation of each
muscle in respect to the crank angle is set following
(Ferrante, 2008a).

Figure 1: Block scheme of the symmetry controller; the
dotted line includes the whole controller.
The PW delivered to each leg is defined by a
pure integral controller (see Figure 1). The definition
of the reference signal of each controller is not
unique. Indeed, there are two possibilities to nullify
the difference between the T
R
and T
L
, which are the
T produced at the right and left crank, respectively:
- increasing the PW of the weaker leg;
- decreasing the PW of the stronger leg.
In the design of the controller, it is chosen to
stimulate as much as possible the weaker leg till the
maximum value (PW
max
). Then, if an unbalance is
still present, the PW of the stronger leg is decreased.
In order to take into account the unbalance only due
to the stimulation, the two active T, T
a,L
and T
a,R
, are
computed for each leg. T
a, L/R
is the difference
between the T produced when the stimulation is ON
and the one obtained during passive cycling, i.e,
when legs are driven by the motor at a constant
speed. Then,
,/

is computed averaging T
a,L/R

over the crank angle range during which the muscles
of the left/right leg are stimulated. Finally, the error
signals, e
L
and e
R
, are defined by comparing
,

and
,

, as shown by the flow diagram reported in

Figure 2.

Figure 2: The flow diagram which describes how the error
signals are computed.
PW
R

e
R

e
L

T
R

PW
L

T
L

Integral
Controller
Integral
Controller
Average &
Comparison
left leg
right leg
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
246

The two integral controllers, shown in Figure 1,
are parameterised only be the integral gain K
C
, as
described by equation (1)
), ( K ) 1 ( ) (
c
k e k PW k PW
i i i
+ = i=R, L
(1)
The values of PW
i
are updated every revolution.
K
C
was set at the same value for both the legs
and was defined according to the stability analysis
performed by means of the simulation model. The
integral controllers were developed with an integral
anti-windup design so that the PW
i
is constrained
between 0 and PW
max
. The controller was
implemented in Matlab.
3 STIMULATION PROTOCOL
A FES cycling protocol was defined both to perform
the stability analysis and to test the working of the
controller in simulation and in experimental trials.
Each trial lasted 3 minutes: the initial 60 s were
characterised by passive cycling while, during the
last 120 s, FES started with the symmetry controller
switched on. The patients was ask not to pedal
voluntary at all. An electric motor maintained the
angular velocity at a constant value of 20 rpm,
during the whole trial.
4 SIMULATION
4.1 The Neuro-muscolo Skeletal Model
for Cycling
To test the symmetry controller developed, a neuro-
muscolo skeletal model which simulates a stroke
patient pedalling by means of FES was designed.

Figure 3: Block structure of the simulation model.
The block structure of the simulator is reported in
Figure 3 and consists of three main parts:
1) Stimulation Pattern Generator, which defines
the PW, the frequency and the crank angular ranges
in which the stimulation has to be delivered to each
muscle involved in the stimulation strategy.
2) Muscular Model, which calculates the joint
moments, M
J
, produced by muscular contractions.
The model is inspired to a previous work (Riener,
1998). The muscle groups included in the model are:
mono-articular hip extensors, hamstrings, biceps
femoris-short head, rectus femoris and vasti. The
maximum isometric forces of all the muscles of the
right leg were set at the half of those of the muscular
groups of the left leg. This permits to reproduce the
muscular model of a stroke patient, with the right
side impaired. The fatigue occurrence shows a
decrease of the muscle activation to about 50% of its
nominal value over 100 s of stimulation with a PW
of 400 s, comparable to (Abbas, 1995).
3) Kinematics and Dynamics of Cycling, which
computes the crank angle,
C
, the cadence,
C
and
the T produced by each leg at the cranks, starting
from the M
J
. The mechanical structure consists of a
planar five-bar linkage (Figure 4). All the five links
(B1 to B5) are assumed to be rigid and are
connected by planar joints (J1 to J7), which
correspond to the hips, the knees, the ankles and the
crank shaft. The ankle joints coincide with the
pedals and any rotation around these joints is
forbidden; the positions of the two hip joints
coincide and they are fixed as well as the crank axis.
Thus, the entire system, has only one degree of
freedom and can be fully characterised by
C
. The
kinematics and dynamics were implemented using
the Open Dynamics Engine (ODE). The complete
model was developed in Matlab/Simulink.
More details on the simulator can be found in
(Ambrosini, 2008).

Figure 4: The five-bar linkage model. B1: right thigh, B2:
right shank, B3: left thigh, B4: left shank, B5: crank arms;
J 1: right hip, J 2: left hip, J 3: right knee, J 4: knee left, J 5:
right ankle, J 6: left ankle, J 7: crank shaft.
4.2 The Stability Analysis of the
Controller
The stability analysis of the controller was carried
out to tune K
C
. Only one branch of the block scheme
reported in Figure 1 was included in the analysis.
The simplified system analysed is shown in Figure
5; it refers to the left leg.


C


'C


Kinematicsand

dynamics

of cycling

Stimulation
patterngenerator

Muscular

model

T
PW MJ

J 1, J 2
x
y
J 3
J 5
J 4
J 6
J 7
B1
B2
B3
B4
B5



k
r
h
r
C

lh
ld
lmt lms
SIMULATION AND EXPERIMENTAL DESIGN OF A SYMMETRY CONTROLLER FOR FES CYCLING
OPTIMISED ON STROKE PATIENTS
247

In order to analyse only the region of linearity, it
was assumed that the controller output (PW
L
) was
not saturated and the system was approximated by a
linear transfer function P(q
-1
), computed as follows:
P(q
-1
) =K
P
q
-1
(2)
where the only parameter is the gain K
P
. Its value
was estimated by means of the simulation model,
setting the value of PW at 400 s, i.e., the value of
PW
max
, and calculating the consequent value of
,

,
produced. The value of K
P
was computed dividing
this value of
,

by the difference between 400 s
and the threshold value of PW over which the
stimulated muscles start to produce an increase in
the torque, i.e., 100 s. The value of K
P
obtained
was 0.0075 Nm/s. It was chosen to analyse the
stability of the system referred to the left leg
because, in the model, the left side was the healthy
one and, thus, its value of K
P
was the bigger.

Figure 5: Simplified version of the closed-loop system.
P(q
-1
) and C(q
-1
) represent the transfer function of the
system and of the controller, respectively.
The transfer function of the integral controller C(q
-1
)
was computed as follows:
1
C 1
- 1
K
) (

=
q
q C
(3)
From equations (2) and (3), it was possible to
calculate the closed-loop transfer function as:
1
C P
1
C P 1
) 1 K K ( 1
K K
) (

+
=
q
q
q H
(4)
The stability of a closed-loop system is
guaranteed if the poles of H(q
-1
) are inside the unit
circle. Therefore, our system is stable if:
| 1 K
P
K
C
| <1 (5)
From equation (5), it resulted that the maximum
value of K
C
to remain in the stability region is
260 s/Nm. Furthermore, to avoid oscillations of the
system output, the pole of the system should be real
positive. Thus, the maximum value of the controller
gain was fixed at 130 s/Nm.
To verify the results of the simplified stability
analysis, the controller was tested in simulation with
different values of K
C
. The results of the trials are
reported in Figure 6; only the first 60 s in which the
stimulation was on are shown. According to the
stability analysis, the system was stable if K
C
was
100 s/Nm (panels (a)-(d)), stable with some
oscillations if K
C
was 150 s/Nm (panels (b)-(e)),
and unstable with K
C
of 300 s/Nm (panels (c)-(f)).

Figure 6: Panels (a)-(b)-(c): values of PW
R
and PW
L
.
Panels (d)-(e)-(f): values of
,

and
,

. The trials were

carried out with different values of K
C
: 100 s/Nm (panels
(a)-(d)), 150 s/Nm (panels (b)-(e)), and 300 s/Nm
(panels (c)-(f)). Only the first 60 s in which the stimulation
was on are reported.
Finally, it would be better to fix K
C
at a value
lower than 130 s/Nm to be sure that the closed-
loop system is stable without oscillations. From
equation (5) follows that stability will not be lost for
any decrease in the value of K
P
during the cycling
session, e.g. caused by muscular fatigue.
4.3 Results
Figure 7 reports the results of a simulation trial
characterised by the protocol described in Section 3.
The initial values of PW
L
and PW
R
were the same
and fixed at 300 s; the value of K
C
of both the
integral controllers was set at 50 s/Nm to update
the PW gradually. As shown in panel (b), at the
beginning the value of
,

was lower than the one

of
,

, because the right was the impaired side in

the model. Thus, the PW
R
increased until the
difference between
,

and
,

became zero (at

about 90 s). Between the 90 s and the 130 s, the
controller maintained the symmetry. It is possible to
notice a slow decrease of both the
,/

due to the
occurrence of the muscular fatigue. Then (130 s-
140 s), in order to test the robustness of the
controller, a positive constant value of 4 Nm was
added to the
,

. Accordingly, because the value of

became higher than the one of

,

, the
controller tried to nullify the difference, first

P (q
-1
) C (q
- 1
)
Ta,R PWL Ta,L
-
e(k)
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
248

increasing thePW
L
till the maximum value and then
decreasing the PW
R
. When the disturbance ceased,
the
,

suddenly became lower than the

,

and,
therefore, the PW
R
increased again till the maximum
value and then the PW
L
decreased till the symmetry
was reached, at about 170 s.

Figure 7: Panel (a): values of PW
R
and PW
L
. Panel (b):
values of
,

and
,

. The black lines indicate the period

in which a positive constant value of 4 Nm was added to
the
,

. Only the phase in which the stimulation was on is

reported.
5 EXPERIMENTAL TRIALS
5.1 The Experimental Setup
The experimental setup developed includes an 8-
channel stimulator (Rehastim
TM
, Hasomed GmbH,
Germany) and a motorised cycle-ergometer
(THERA-live
TM
, Medica Medizintechnik GmbH,
Germany) equipped by resistance strain gauge
sensors able to measure the torques at the left and
right crank arm. These signals are transmitted from
the ergometer to a desktop PC via wireless,
providing a measure of the unbalance between the
two legs during cycling. More details on the setup
can be found in (Ferrante, 2008b).
In all the trials, an ON-OFF PW profile was
used. The stimulation currents were set at a value,
tolerated by the subject, which produces a good
muscular contraction. The stimulation frequency
was fixed at 20 Hz and all the signals were sampled
at 200 Hz.
5.2 Results
Figure 8 shows the results of the controller test
carried out on an able-bodied subject. The subject
was a female (24 years old, 166 cm and 52 kg). The
stimulation currents used were 30 mA for all the 4
muscles. The value of K
C
of both the integral
controllers was set at 50 s/Nm. Moreover, in this
trial, the values of PW were fixed at 300 s and
100 s for the left and the right leg, respectively, in
order to induce an unbalance between the two sides
in a healthy subject.
Figure 8 shows that, at the beginning, there was a
slight unbalance which the controller tried to
compensate increasing the PW
R
.

Figure 8: Panel (a): values of PW
R
and PW
L
. Panel (b):
values of
,

and
,

. The black lines indicate the period

in which the subject pedalled voluntary only with the right
leg. Only the phase in which the stimulation was on is
reported.
After 85 s, the controller achieved a symmetrical
movement. This symmetry was maintained till the
subject began to pedal voluntarily only with the right
leg (115 s-125 s) producing a high increment in the

(panel (b)). Thus, the controller increased the

PW
L
and then, when the saturation value was
reached, the controller reduced the value of PW
R

(panel (a)). This controller action was not sufficient
to gain symmetry because the subject was pedalling
voluntarily. When the subject stopped to pedal
voluntarily with the right leg,
,

suddenly
decreased to zero, which corresponds to the mean
value during passive cycling. Indeed, the leg was not
stimulated at all. Thus, the PW
R
started to increase
and the symmetry was re-gained in about 25 s and
maintained till the end of the trial.
SIMULATION AND EXPERIMENTAL DESIGN OF A SYMMETRY CONTROLLER FOR FES CYCLING
OPTIMISED ON STROKE PATIENTS
249

6 DISCUSSIONS
AND CONCLUSIONS
The present study deals with the design and the
testing of a novel closed-loop controller for FES
cycling, able to gain and then to maintain the
symmetry of the pedalling in stroke patients. This
controller could be useful in the rehabilitation of
these patients, who need to re-learn symmetrical
tasks in order to recover basic motor functions, such
as walking.
Furthermore, a neuro-muscolo skeletal model to
simulate cycling induced by FES in stroke patients
was developed. This simulator aided in the tuning of
the controller parameters and in the validation of the
controller before testing it experimentally.
Finally, first trials on healthy subjects were
carried out. Starting from a measurement in real-
time of the unbalance between the torques produced
by each leg at the cranks, the controller was able to
reach and then to maintain a symmetrical pedalling,
modifying the stimulation parameters of the two
lower limbs, independently. For example, the initial
unbalance of 0.38 Nm was nullified by the controller
in about 18 s as shown in the results reported in
Figure 8. The results of this trial showed also that
the controller maintained the symmetry of the
pedalling by means of small oscillations of the
values of PW, till an external contribution occurred.
When the subject started to pedal voluntarily only
with the right leg, an unbalance between the two
legs was introduced again and the controller
answered properly, without showing an unstable
behaviour. When the subject ceased to pedal
voluntarily, the unbalance of the pedalling was about
0.73 Nm and the controller re-gained the symmetry
in about 18 s. This trial showed clearly that the
system is not linear; indeed, even if the unbalance
doubles, the time needed to reach symmetry is the
same.
The automatic control system developed shows a
reliable behaviour. Thus, the next step will be the
testing of the controller on stroke patients to
demonstrate if this system could be actually useful
in the rehabilitation of these patients, accelerating
and improving the motor recovery of the lower
limbs.
ACKNOWLEDGEMENTS
This work was supported by the Italian Institute of
Technology (IIT). It was also partly funded through
grant by the German Federal Ministry of Education
and Research (BMBF) within the project
RehaRobES (FKZ 01EZ0766).
REFERENCES
Sheffler, L.R., Chae, J. (2007) Neuromuscular electrical
stimulation in neurorehabilitation, Muscle & Nerve,
vol. 35, no. 5, pp. 562-590.
Hunt, K.J ., Stone, B., Negard, N.O., Schauer, T., Fraser,
M.H., Cathcart, A.J ., Ferrario, C., Ward, S.A., Grant,
S. (2004) Control strategies for integration of electric
motor assist and functional electrical stimulation in
paraplegic cycling: utility for exercise testing and
mobile cycling, IEEE Transactions on Neural and
Rehabilitation Systems Engineering, vol. 12, no. 1, pp.
88-101.
Trumbower, R.D., Faghri, P.D. (2005) Kinematic
analyses of semireclined leg cycling in able-bodied
and spinal cord injured individuals, Spinal Cord, vol.
43, no. 9, pp. 543-549.
Ferrante,

S., Pedrocchi, A., Ferrigno, G., Molteni, F.
(2008a) Cycling induced by functional electrical
stimulation improves the muscular strength and the
motor controls of individuals with post-acute stroke,
European Journal of Physical and Rehabilitation
Medicine, vol. 44, no.2, pp. 159-167.
Hunt, K.J ., Ferrario, C., Grant, S., Stone, B., Mclean,
A.N., Fraser, M. H., Allan, D.B. (2006) Comparison
of stimulation patterns for FES-cycling using
measures of oxygen cost and stimulation cost,
Medical Engineering & Physics, vol. 28, no. 7, pp.
710718.
Ferrante,

S., Comolli, L., Pedrocchi, A., Bocciolone, M.,
Ferrigno, G., Molteni F. (2008b) Optimization of FES
cycling neuroprosthesis on stroke patients by means of
the left and right crank measurement, Biodevices
Conference Proceedings, International Conference on
Biomedical Electronics and Devices, Madeira,
Portugal, pp. 206-211.
Riener, R., Fuhr, T. (1998) Patient-driven control of FES-
supported standing up: a simulation study, IEEE
Transactions on Rehabilitation Engineering, vol. 6,
no. 2, pp. 113-124.
Abbas, J ., Chizack, H. (1995) Neural network control of
functional neuromuscular stimulation systems:
Computer simulation studies, IEEE Transactions on
Biomedical Engineering, vol. 42, no. 11, pp. 1117-
1127.
Ambrosini, E., Schauer, T., Ferrante, S., Pedrocchi, A.,
Ferrigno, G (2008) Neuro-muscolo skeletal model for
cycling induced by functional electrical stimulation in
individuals with stroke, accepted by 9
th
International
Workshop on Research and Education in
Mechatronics, Bergamo, Italy.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
250
EXPERIMENTAL DIGITAL BPSK MODULATOR DESIGN
WITH VHDL CODE FOR BIODEVICES APPLICATIONS
Gihad Elamary, Graeme Chester and J effery Neasham
School of Electrical,Electronics and Computer engineering, Newcastle University, Newcastle Upon Ttyne, U.K.
Gihad.elamary1, graeme.chester, [email protected]
Keywords: VHDL coding modulator, BPSK digital modulator, CPLD/FPGAs.
Abstract: In this paper, we propose a new simple design for a BPSK modulator applied to implant telemetry
applications as demonstrated. We used hardware description language VHDL (IEEE standard) to generate a
BPSK digital signal. The carrier and data are interfaced to the CPLD and FPGAs board for testing, we used
the local clock oscillator, which is operating at 25.175 MHz reduced into 12.5 MHz for the carrier and
2Mbps for data source. The modelled modulator has been designed simulated and performance was
evaluated by measurements, considering low power consumption and size for medical purpose. Moreover,
the advantages of this modulator are it can be reconfigured and upgraded to enhance the bit rate.
1 INTRODUCTION
Wireless infrastructure is fast growing and this
technology may also be used for medical
applications such as biotelemetry, telemedicine and
healthcare (R.Harrison, et al., 2006). One of these
applications is transcutaneous wireless implant
telemetry, in order to communicate through wireless
inductive coupling to transfer the power and data to
a battery less implant (K.D.Wiseet al, 2004; C.Sauer
et al, 2004). Demand for higher data rates is high as
a result of increasing the electrode numbers for
reading nerve signal information or controlling data.
In such applications BPSK has advantages over
ASK and FSK modulations. The advantage of
BPSK is having fixed carrier signal amplitude that
provides stable power transfer and independent data
modulation. This is suitable to provide a constant
RF signal into an implant device. The advantage is
high readability for DC voltage supply at different
distances between reader coils and implant part. We
propose in this work to develop VHDL code to
generate a digital BPSK signal for improving
modulator performance and increasing the data rate
(Bob Zeidman, 2002). Compared to the other
analogue modulators, this type of modulator
provides digital synthesis and the flexibility to
reconfigure and upgrade with the two most often
used languages VHDL-and Verilog- based
(P.Niktin, et al., 2005; J .S. Ruque, et al., 2005).
2 METHOD
The BPSK signal can be represented mathematically
in an analogue way as in equation (1).
As demonstrated in Figure 1, for this technique it is
essential to convert the binary data ) (t m into a
NRZ signal that maps a logic 0 to -1V(nominal)
and logic 1 to +1V. This data signal controls the
transition shift ) , 0 ( for the carrier signal. This
results in high power consumption for these types of
analogue modulators, reduces their efficiency and
limits their biomedical application. This also
increases the hardware complexity of the circuit and
produces a large physical device.
The proposed modulator is developed with
VHDL description code to generate carrier shifter

Figures 1: Illustrated the BPSK waveform with respect
NRZ) data transitions.
) (
2
) ( ) (
rf rf
b
b
BPSK
t w Cos
T
E
t m t S + =


(1)
251

frequencies (0, 180) which are controlled by the
input binary data to perform the transition of the
BPSK signal (I. Grout, et al, 2005). The modulator
consists of digital and analogue parts as depicted in
Figure 2. The output is selected by the multiplexer
then filtered with a passive Band Pass Filter (BPF)
to eliminate the higher frequencies and the
harmonics associated with the square wave signal in
order to provide the transmit analogue signal (Tx).


Figure 2: Illustration of the block diagram for the
proposed BPSK Modulator.
The simulated random data signal (Data_in) that is
generated by a PN sequence can be represented by
Fourier series analysis as in equation (2).
Where the input carrier signal is a periodic pulse
train signal and mathematically expressed by the
Fourier series as in equation (3).
The filter is essential for the modulator to complete
the process (off-chip). The output signal produced
by it is an analogue form. We investigated two types
of filters in this paper, Low Pass Filter (LPF) and
Band Pass Filter (BPF). The BPSK VHDL output
signal is fed into the test filters. These filter it to
pass the fundamental frequency ( data f ) and to
remove the harmonics and the DC component. Our
prototype analogue filter uses the Chebyshev filter
II as opposed to the other filters such as Chebyshev
I, Butterworth, Elliptic and Bessel. Our second
choice is Butterworth LPF as this was observed to
give better performance than the other types.
3 MODULE SIMULATION
A. Simulink/ MATLAB Simulation
The BPSK modulator was designed and simulated
with Simulink/MATLAB to verify and validate the
modulator specifications. The demodulator is also
constructed using the same tools to examine the
performance of the proposed modulator. The
architecture block diagram of Tx_Mod is shown in

Figure 3: The simulink BSPK modulator diagram.
) ( ) (
c
n
n
nT t p c t PN =

=
(2)

1
sin((2 1) ) 4
( )
(2 1)
c
k
k w t
Carrier t
k

=

(3)
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
252
Figure 3. The simulation result of the Simulink
modulator is presented in Figure 4. The signal in
Figure 4a presents the digital BPSK signal while the
signal in Figure 4b presents the filtered BPSK output
of the BPF. The Tx and the Rx data are presented in
Figure 4c and Figure 4d respectively. Figure 5 shows
the spectrum of the transmitted RF signal (CH1) and
the received RF signal (CH2) in the presence of noise
(AWGN). Finally Figure 6 shows the simulation
result for the Bit Error Rate (BER) of the designed
demodulator.


Figure 4: The simulated BPSK modulator output signals
results by Matlab/Simulink at frequency12.5 MHz.

Figure 5: The Spectrum of BPSK transmit and receive
signals at Data rate 2MHz, carrier 12.5MHz.

Figure 6: The BER simulation diagram of demodulator
signal from the proposed BPSK modulator.
B. VHDL-code Simulation
The modulator is implemented using VHDL code
and built by the Altera UP2 development kit board.
The carrier frequency 12.5 MHz was generated by
the local clock signal that operates at 25.175 MHz.
The signal data clock was reduced to 2MHz by a
frequency divider then used to generate a random
PN_sequence, which provided the transitions of the
carrier selected by the multiplexer (V.Pwdroni,
2007). The behavioural block diagram of the VHDL
code of the BPSK modulator is illustrated in
Figure_7, whilst the simulated result of this
modulator is presented in Figure 8 which
demonstrates the output signal waveforms indicating
the transitions (0,180) of the carrier signal (output
of multiplexer) due to the data source.


Figure 7: The VHDL Digital BPSK Modulator.
4 EXPERIMENTAL RESULTS
AND DISCUSSION
In this section we provide the measurements which
were conducted using the Altera UP2 Development
EXPERIMENTAL DIGITAL BPSK MODULATOR DESIGN WITH VHDL CODE FOR BIODIVECES
APPLICATIONS
253

Figure 8: The simulated waveform generated by Altera development tools for BPSK Modulator.
kit board for testing the VHDL code modulator and
comparing the performance with the simulated
BPSK modulation. The Agilent digital demodulator
(E8048A VXI) is used to receive the filtered RF
BPSK signal, and analyse the parameters of the
transmitted BPSK signal Tx as demonstrated in
Figure 9.

Figure 9: Illustrated the Lab measurement withUK2 Alter.
Aboard and Agilent digital demodulator (E8048A)The
desired carrier signal was generated from the master
clock on the circuit board that operates at 25.175
MHz, and the carrier phase acquires two discrete
states (0, ) at frequency 12.5 MHz. This
corresponds to the PN-code data source generated
with VHDL code inside the FPGAs at 2Mbps in
order to produce the BPSK modulation. The
behavioural simulation results produced BPSK
digital signal, passed through passive BPF for
harmonics separation. We investigated two
prototypes of filters in this paper, LPF and BPF.
Our choice was the Chebyshev filter II as this
has better performance than the other types. The
filtered output signals, captured with Tektronix
scope and Agilent spectrum analyser are
demonstrated in Figures 10 and 11 respectively.


Figure 10: The output of VHDL BPSK modulator (digital
signal) Top and filtered signal bottom at carrier frequency
12.5 MHz.

Figure 11: The Spectrum of BPSK transmitted signal at
carrier 12.5 MHz.
The Agilent digital demodulator was used to
measure the spectrum of the Rx BPSK signal as
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
254

shown in Figure 12 while the constellation diagram
of the demodulated signal is presented in Figure 13.


Figure 12: The Rx- BPSK signal spectrum at 12.5 MHz
carrier (data rate 2Mbps).

Figure 13: The constellation diagram of BPSK at
Demodulator using Agilent digital demodulator (E8048A).
5 CONCLUSIONS
We implemented a new simple BPSK digital
modulator model in the Simulink MATLAB
environment. This has been successfully
implemented using hardware description language
VHDL code by the Altera UP2 development board.
The modulator generates the BPSK signal directly
from the binary data in order to control the carrier
signal. The output producing a modulated digital
signal was filtered to transmit through a miniature
BPF. Experimentally measurements are presented at
carrier frequency 12.50 MHz, and data rate 2Mbps,
which present good performance with high data rate
and carrier suppression >35dB. The filter is a critical
part of the design. For this work we designed and
simulated different types of analogue filter and
compared them to choose the best filter
performance. The simulation results were that the
Chebyshev I & II, appeared optimum at BPFs
compared to the others, and optimum LPF
performance was form the Butterworth type. Digital
filters could be implemented to allow integration
with the digital modulator device. However the
disadvantage of digital filter is that they need
clocking at high multiples of the sampling frequency
which increases the power consumption and size. As
future work, it is still necessary to investigate other
methods to improve the harmonic rejection
performance of the analogue output filters by digital
synthesis of alternative waveforms at the modulator.
It is also an intention to up-convert the signal into an
ISM unlicensed frequency band (e.g. 402~405 MHz)
for biomedical telemetry purposes.
REFERENCES
R. Harrison, P. Watkins, R. Kier, R. Lovejoy, D.Black, R.
Normann, F. Solzbacher. (2006). Low power
Integrated Circuit for a Wireless 100 - Electrode
Neural Recording System IEEE International solid
state circuit conference.
K. D. Wise,D. J . Anderson, J. Fhetke, D. R. Kipke, K. N.
Ajafi. (2004). Wireless Implantable Microsystems:
High Density Electronic Interfaces to the Nervous
System. Proceeding of the IEEE, Vol. 92. No.1,
J anuary
C. Sauer, M. Stanacevic, G. Cauwenberghs, N. Thakor.
(2004). Power Harvesting and telemetry in CMOS
for Implanted devices IEEE.
Bob Zeidman.(2006) Designing with FPGAs and CPLD
P. Nikitin, E.Normark, C.Wakayama, R.Shi.( 2004).
VHDL-AMS Modelling and Simulation of BPSK
Transceiver system.
J . S. Ruque, D .I .Ruiz, C. E. Carrion. (2004). Simulation
and implementation of the BPSK Modulation on a
FPGA Xilinx Spartan 3 xcs200-4ftp256, using
simulink and System generator block set for
DSP/FPGA.
Chad A. Bossetti, J ose Marmena, Miguel A., L. Nicolelis,
Patrick D. Wolf. (2005). Transmission Latencies in
Telemetry Linked Brain Machine Interface
transactions on Biomedical Engineering IEEE.
Volnnei A. Pedroni Circuit Design with VHDL. (2007).
I. Grout, J .Ryan, T. O.Shea. (2005) Configuration and
debug of filed Programmable gate Arrays using
MATLAB/ SIMULINK. J ournal of Physics.
EXPERIMENTAL DIGITAL BPSK MODULATOR DESIGN WITH VHDL CODE FOR BIODIVECES
APPLICATIONS
255
MESOTHERAPY DEVICE FOR ESTHETIC APPLICATIONS
M. S. Martins
1
, V. M. G. Correia
2
, J . G. Rocha
3

Industrial Electronics Department, University of Minho, Campus de Azurm, 4800-058 Guimares, Portugal
1
[email protected],
2
[email protected],
3
[email protected]
J . M. Cabral
Industrial Electronics Department, University of Minho, Campus de Azurm, 4800-058 Guimares, Portugal
[email protected]
Keywords: Mesotherapy, Galvanic current, Step-Up circuit, Master-Slave Architecture, Microcontroller, H-Bridge,
PWM, CPU
Abstract: This article describes a complete system prototype to be used in aesthetic mesotherapy. The system is
composed by two main blocks: a Master block, whose chief component is a CPU, which provides the user
interface and a Slave block, implemented with a micro controller and a wave generator, which produces the
appropriated voltages and currents compatible with the mesotherapy treatments. The whole system is
powered by a 12V power supply and the output signal has a voltage that range between -54 V and 54 V. The
output signal is composed by the overlap of two frequencies: the first one is selected in the range from 1.2
kHz to 1.8 kHz and the second one is in the range from 0.07 Hz to 2 Hz. The system is being tested in
clinical environment with real patients showing very good promising results.
1 INTRODUCTION
Mesotherapy is a non-surgical cosmetic medicine
treatment that employs multiple injections of
pharmaceutical and homeopathic medications, plant
extracts, vitamins, and other ingredients into the
subcutaneous fat. Mesotherapy injections are
purported to target adipose fat cells, apparently by
inducing rupture and cell death among adipocytes.
(Tosti, 2007).
Mesotherapy has presented new opportunities in
aesthetic medicine because it is a non-evasive
treatment capable of reducing cellulite and weight
(Ward, 2002). Into these two capabilities are also
included protocols like Cellulite, Obesity, Facial and
Body Contouring. Besides these applications, the
mesotherapy can be applied in other treatments such
as acne, arthritis, sports injuries, ulcers, vein
treatments and rheumatology.
In this article we report some aspects of the
design and fabrication of a device prototype that
allows mesotherapic treatments. These treatments,
based on different layers of skin, are able to reach
9cm of depth (e.g. hypoderm or fat most affected
area).

The basic working principle of the device
consists on a set of electrical pulses that are applied
on the skin surface, through two treatment
electrodes: a roll-on, used as active electrode, and a
glove, used as return electrode, originating two
reactions:
Expansion of the cell pores and epidermis
alignment, as shown in figure 1;
The electric field attracts the cream molecules
into the skin. The cream can be positively or
negatively charged, as shown in figure 2.

Figure 1: Transcellular and intercellular routes of
treatment cream through the lipid matrix.
256

The main objective of this project is the
development of a device prototype capable to
perform the treatments referred above. In order to do
that, it is necessary to:
Develop a programmable signal generator;
Due to practical issues the system supply
voltage cannot exceed 12V, while the output
signal may contain peaks that reach 54V;
Develop the appropriate safeguards to avoid
patients injuries;
Develop a user-friendly interface to be used by
technical personal without expertise on
electronics and informatics;
Develop a flexible system able to introduce
new types of treatments without hardware
upgrades;
Develop a system capable to report any type of
anomaly;
Develop a system able to work 24h/7day.


Figure 2: Electric field generated by the electrodes It is
visible the response of cream particles along the
hypoderm.
2 BACKGROUND
This project is being funded by several companies
and its main goal is to develop an hardware
prototype capable of performing mesotherapy,
reducing the technical expertise needed in treatments
handling. On the other hand, it is possible to expand
the functionalities of this prototype in order to
achieve other treatments in the area of
electrotherapy, without the need of adding new
hardware components.
With the support provided by ISAVE University
(ISAVE, 2008) we have been able to incorporate all
the information required to produce the proper
signals required to generate other therapy types of
currents:
Galvanic Currents;
Russian Electrical Simulation (Ward and
Shkuratova, 2002);
Transcutaneous Electrical Simulation;
Iontophoresis Current (Carter, 2003);
Pain-Free Currents (J ohnson, 2003);
Electroforoporation Currents;

Thus, despite the development of this prototype
was targeted for aesthetic medicine, it can also
achieve other types of treatments since the platform
accepts other signals by changing only a few setup
parameters.
Electronic devices applied on esthetical medicine
must have a huge factor of accuracy and quality,
mostly the ones that are in direct contact with the
patient, leaving no space for errors.
In the past, all the options taken in the
development of these devices are mainly related
with the chosen materials and the methods that were
used to maximize perfection, quality, reliability and
performance.
In order to achieve the best results during a
treatment, the equipments must guaranty a clear and
precise signal output with a perfect repetition and
very low error tolerance.
In the esthetical medicine area there are different
types of treatments which can be held as a resource
of mesotherapy:
Anticellulite;
Located anticellulite;
Antistretch;
Antiflaccidity.

Each type of treatment needs a special signal
with a polarity, which may be positive or negative,
depending on the cream used. In all treatments the
applied signals obeyed to eq. 1:
SIGNAL =((HF * LF) +LF) (1)
where HF is the high frequency component,
whose value ranges between 1.2kHz and 1.8kHz,
and LF is the low frequency component, whose
frequency ranges from 0.07Hz to 2Hz (its period
ranges between 0.5 and 14 seconds).
Galvanic current is the base of this treatment. It
always circulates in the same direction (the voltage
is always positive or is always negative), once each
cream has its own polarity. Therefore the system
must guarantee this characteristic for the generated
voltage, that is, the system must follow the
algorithm:
Positive Polarity:
I f ( ( HF * LF) + LF) < 0
MESOTHERAPY DEVICE FOR ESTHETIC APPLICATIONS
257

t hen SI GNAL=0
El se
SI GNAL = ( HF * LF) + LF

Negative Polarity:
I f ( ( HF * LF) + LF) > 0
t hen SI GNAL=0
El se
SI GNAL = ( HF * LF) + LF
3 SYSTEM ARCHITECTURE
The general system architecture proposed in this
paper is a master-slave structure since there are two
control units: a high-level control unit (master) and a
low-level control unit (slave).
The high-level control unit or central processing
unit (CPU) is responsible for supporting the user
interface: exchanging and processing the necessary
information between the user and the low-level
control unit. The low-level control unit, whose main
component is a micro controller, is responsible to
convert digital data, received from the high-level, to
analogical signals that are applied directly to the
patient skin, through the connection devices.
Figure 3 shows the system architecture.

Mast er Sl ave

Figure 3: System architecture. The CPU operates as
master and the micro controller operates as slave. The
micro controller functionalities are controlled by the CPU.
Despite the system is of master-slave type, a
feedback from the output to the micro controller is
used in order to allow the detection of some
malfunctions.
The Master block is composed by a processor
unit, a touch-screen and an Ethernet interface, while
the Slave block is constituted by a micro controller,
a step-up converter, a driver, an H-bridge, a filter
and the patients protection circuit. Both blocks
communicate by means of an USB interface.
The following paragraphs describe, in detail, the
most important functions shown in figure 4.

Figure 4: Block Diagram of the device internal structure
and the links between the various sub-systems.
The CPU is the head of the system, responsible
for gather all the data, coming from the user-
interface console, process it and calculate the
appropriated coefficient vector that will be used by
the micro controller to produce the corresponding
output signal with a PWM (Pulse Width
Modulation).
If we consider that the technical personal do not
have expertise in informatics and electronics
technologies, it is important to provide an intuitive
graphical interface. For this purpose, it was used a
touch-screen monitor to operate the device in an
intuitive way. The programming language used was
C++, associated to an Open GL platform in order to
satisfy several requirements, namely the real time
operation, the support for micro controller systems
and the attainment of the maximum graphic
potentialities of the system.
It is important to refer that this platform is based
on a Linux operating system, due its great
potentialities for real time operation associated with
the advantage that it is a free software, making
easier and cheaper the system commercialization.
Another particular feature is the system
dynamism, which enables firmware upgrades and
interconnection with other devices in order to allow
a huge range of treatments.
All interconnections between CPU and other
functional blocks are bidirectional, allowing a
continuous feedback that enable instantaneous
detection of any anomaly.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
258

The micro controller is responsible for
assembling the information, sent by the CPU, in
order to drive the H-Bridge. It calculates the PWM
ON / OFF times, generating the desired signal to the
system output.
The micro controller uses an 8MHz crystal. In
order to generate a PWM signal that does not reach
frequencies higher than 30kHz, the sampling
frequency used by the micro controller is about 17
times higher than the frequency of the output signal
to ensure a low ripple at the filter output.
An AT90USB1287 low-power CMOS 8-bit
micro controller was used (J ohnson and Tabasam,
2003), based on the AVR with 64/128K bytes of ISP
Flash and an USB device controller with full speed
and low speed data transfer support, enhanced RISC
architecture. By executing powerful instructions in a
single clock cycle, the AT90USB1287 achieves
throughputs approaching 1MIPS per MHz allowing
the system designer to optimize power consumption
versus processing speed.
The interface between master and slave blocks is
achieved by means of a USB connection due to is
large bandwidth and Plug and Play interoperability.
A full-duplex low speed (1.5Mbit/s) data rate
connection is used that guarantees up to 512kbit/s of
bandwidth in each direction.
The Bus topology used is the Reduced Host
Topology showed in figure 5 (Atmel, 2007).


Figure 5: Reduced Host Topology. A reduced host
controller has a unique USB port and does not handle full
USB tree with hub. It means that a reduced host controller
is designed to handle a unique point-to-point connection
with an unique USB device.
Its main characteristics are:
USB Host:
There is only one host in any USB system, and
it operates as the master of the USB bus;
The USB interface to the host system is referred
to as the Host Controller.
USB Device:
An USB device operates as a slave node on the
USB bus;
Thanks to the USB hub (that also operates as an
USB device) up to 127 devices can be
connected on the USB bus. Each device is
uniquely identified by a device address.
It would be only necessary to implement a more
elaborated connection topology if the micro
controller interacts with more than one device
because it would be necessary to guarantee that it
communicates with the correct device. As this
system is composed by just one device, the use of
the simplified topology facilitates the
implementation of the connection between the CPU
and the micro controller.
The micro controller output signal produces a
peak voltage of 3.3V. To manage an effective
treatment it is necessary to amplify this signal to
higher levels, such as 54V. To carry out this task it
was implemented an H-bridge with two quadrants,
allowing the change of the signal output polarity
according to the corresponding treatment. The H-
Bridge was implemented by means of L6225 DMOS
Dual Full Bridge chip which combines isolated
DMOS Power Transistors with CMOS and bipolar
circuits on the same chip (STMicroelectronics,
2007). Figure 6 shows the schematic diagram of this
circuit.

Figure 6: H-Bridge circuit based on BCD technology.
Combines isolated DMOS Power Transistors with CMOS
and bipolar circuits on the same chip.
The signal output is a sum of two components: a
low frequency of 0.07Hz (Period =14s) and a high
frequency of 1.8kHz. A low-pass active filter, with a
Q factor of 2, and a cut-off frequency of 2kHz were
used. Thus the filter output no longer presents the
high frequency components, produced by the PWM
modulator, and all variations of the sign are
smoothed. Figure 7 shows the output waveform of
the PWM modulator. Figure 8 depicts a time
expansion between B1 and B2 bars of figure 7.
MESOTHERAPY DEVICE FOR ESTHETIC APPLICATIONS
259


Figure 7: Signal obtained at the output of the H-Bridge
circuit, with 2s in time axis and 5V in voltage axis.

Figure 8: Signal of figure 7, expanded with 20ms between
B1 and B2 bars.
Figure 9 shows the low pass filter response. Figure
10 depicts a time expansion between C1 and C2 bars
of figure 9.

Figure 9: Signal obtained at the system output. It
corresponds to the signal of Fig. 8 be low-pass filtered,
with 2s in time axis and 5V in voltage axis.

Figure 10: Signal Signal of figure 9, expanded with 20ms
between B1 and B2 bars.
Standard requirements (RSIUEE, 1974),
imposed on medical equipment, does not allow that
the power system, responsible for generating
treatment signals, have input voltages higher than
12V. Nevertheless, to obtain the expected results,
the signal output must produce peaks of 54V. This
task is achieved by means of a step-up circuit which
provides up to 54V linear voltage with small ripple
and 500mA of maximum current. In the present
application the current can not exceed 50mA. The
step-up circuit is implemented with the LT1680
(Linear Technology, 2007) whose schematic
diagram is shown in figure 11.

Figure 11: Step-Up circuit. High power, current mode
switching power supply controller optimized for boost
topologies. The IC drives an N-channel MOSFET switch
for DC/DC converter up to 60V output.
All equipment must obey to security rules
imposed by responsible entities. Therefore, in
addition to the main objective, that is, the
mesotherapic equipment development, we have also
to ensure the safety of the patient. In this way, it was
introduced a current limiter before the signal output.
When the current reaches higher values than 30mA,
the circuit output voltage is lowered, as it is shown
in figure 12.

Figure 12: Output voltage versus current. For currents
above 30mA the system protection responds by reducing
the output voltage.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
260

4 PROTOTYPE
To conduct the necessary system tests, a prototype
was built where the methodology described in this
article was implemented. This prototype is now
being used in the aesthetics mesotherapy treatments
in order to get a functionality and efficiency
diagnosis of the device. Therefore, it is necessary
that this prototype show the full potentialities of the
system but also show simple and functional user-
interface to simplify the use by the technical
personal. Figures 13 and 14 show some pictures of
the prototype.

Figure 13: Device internal circuits.

Figure 14: Device external appearance.
5 CONCLUSIONS
This article described a complete system prototype
for use in aesthetic mesotherapy. The system is
composed by a master-slave architecture in which
the master block is based on a CPU and the slave
block is based on a micro controller. The first
system prototype is already tested and fabricated in
laboratory conditions with performances that
correspond to the expected ones. It was also tested in
a clinical environment, with real patients. The results
of these tests are more or less subjective once consist
on the opinion of the technicians and patients, but
almost all consider the performance of the prototype
as good or very good. As a future work, the
prototype software will be adapted to perform
lymphatic draining or muscle stimulation treatments.
ACKNOWLEDGEMENTS
The authors would like to thank Doctor Cludio
J avier Martinez Leon, who helps the authors in
understanding aesthetics mesotherapy treatments
fundamentals.
REFERENCES
A. Tosti, M. Pia De Padova, "Atlas of Mesotherapy in
Skin Rejuvenation", Informa Healthcare, United
Kingdom 1 September 2007.
ISAVE Website, http://www.isave.edu.pt.
Carter R. Anderson, Russell L Morris, Stephen D Boeh,
Peter C Panus, Walter L Sembrowich, Effects of
Iontophoresis Current Magnitude and Duration on
Dexamethasone Deposition and Localized Drug
Retention, Physical Therapy. Volume 83. Number 2.
February 2003.
Charles Godbout, J rme Frenette, Periodic Direct
Current Does Not Promote Wound Closure in an In
Vitro Dynamic Model of Cell Migration, Physical
Therapy. Volume 86. Number 1. J anuary 2006.
Mark I. J ohnson, Ghazala Tabasam, An Investigation
Into the Analgesic Effects of Interferential Currents
and Transcutaneous Electrical Nerve Stimulation on
Experimentally Induced Ischemic Pain in Otherwise
Pain-Free Volunteers, Physical Therapy. Volume 83.
Number 3. March 2003.
Alex R. Ward, Nataliya Shkuratova, Russian Electrical
Stimulation, Physical Therapy. Volume 82. Number
10. October 2002.
S. Madhere, "Aesthetic Mesotherapy and Injection
Lipolysis in Clinical Practice", Taylor & Francis Ltd,
United Kingdom 15 J une 2006.
Atmel Corporation 2007 - Microcontroller at90USB1287,
http://www.atmel.com/dyn/resources/prod_documents
/doc7593.pdf.
STMicroelectronics 2007 - DMOS Dual Full Bridge
Driver L6225, http://www.st.com/stonline/products/
literature/ds/9451.pdf.
Decreto-Lei n 740/74 Art 598 Proteco contra
contactos indirectos, Regulamento de Segurana de
Instalaes de Utilizao de Energia Elctrica
(RSIUEE), December 1974, Portugal.
Linear Technology 2007 - High Power DC/DC Step-Up
Controller LT1680,
http://www.linear.com/pc/downloadDocument.do?navId=
H0,C1,C1003,C1042,C1031,C1115,P1597,D1509

MESOTHERAPY DEVICE FOR ESTHETIC APPLICATIONS
261
A RECONFIGURABLE ARRAY FOR BLIND SOURCE-SEPARATION
ON AN FPGA
Ricardo Escalona, Daniel Herrera and Miguel Figueroa
Department of Electrical Engineering, Universidad de Concepcion, Barrio Universitario S/N, Concepcion, Chile
[email protected], [email protected], [email protected]
Keywords:
Blind source-separation, Independent component analysis, InfoMax, Embedded signal processing, Field-
programmable gate arrays.
Abstract:
We present a recongurable array which performs blind source separation on a range of eld-programmable
gate array (FPGA) devices. Our array uses independent component analysis (ICA) with the InfoMax algorithm
to separate a mixture of signals without an external reference. We describe two congurations of the array,
representing distinct points in the design space. Our experimental results show a performance improvement of
more than one order of magnitude over an optimized software implementation of the algorithm on a computer,
with a power consumption of just 100mW. Our array successfully separates a fetal electrocardiogram (ECG)
mixture into the source signals of mother and fetus, enabling medical analysis on the resulting independent
components.
1 INTRODUCTION
Independent component analysis (ICA) is a signal
processing technique used to recover the original
sources from unknown mixtures captured by spatially
distributed sensors. Due to the weak assumptions im-
posed by ICAon the nature of the original signals, this
technique is widely used to perform blind source sep-
aration on medical applications such as electrocardio-
gram (ECG) and electroencephalogram (EEG) analy-
sis (Zeng et al., 2008; Potter et al., 2002), as well as
speech recognition, face classication, and data com-
munications (Bell and Sejnowski, 1997).
Despite these advantages, most ICA algorithms
require high computational throughput to operate in
real time. Typical software solutions on general pur-
pose computers are large and power hungry, and even
digital signal processors do not meet the power, per-
formance, and cost requirements of embedded and
portable applications. Custom VLSI implementations
usually feature the best power/performance tradeoff,
but they lack the exibility of software solutions, and
their design cycle is long and expensive.
We present a hardware implementation of the In-
foMax algorithm for ICA on eld-programmable gate
arrays (FPGAs). Unlike custom-VLSI circuits, FPGA
can be easily reprogrammed on-site, retaining the
exibility of software implementations while attain-
ing higher performance and lower power consump-
tion (Anguita et al., 2003). Unlike previous imple-
mentations (Yang et al., 2007; Li and Lin, 2005), we
can target our array at a wide range of devices, trad-
ing size and cost for performance depending on the
requirements of the application.
In this paper, we describe the architecture and tar-
get it at both an entry-level device and a high-end
platform FPGA. First, we describe the InfoMax al-
gorithm. Then, we describe the architecture of the
array and discuss the implementation of key func-
tions in xed-point arithmetic. We discuss and an-
alyze our design trade-offs for both versions of the
architecture and their impact on circuit area and per-
formance. Finally, we present experimental results on
the separation of mother and fetus EEG signals. The
performance of the array on the entry-level FPGA is
similar to that of an optimized software implementa-
tion on a desktop PC, while the array on the platform
FPGA improves the performance of software imple-
mentation by more than one order of magnitude. Both
arrays dissipate less than 100mW.
2 THE INFOMAX ALGORITHM
As stated in the previous section, Independent Com-
ponent Analysis performs blind-source separation of
an unknown mixture of independent signals. ICA
262
Orthogonalization
Weight
update
Forward
computation
Normalization
= u W x
1 1
2 tanh( )
T
k k k

= +

W W I u u W
1
1
,
,
j
k
k k k
j
proj
proj

=
=
=

z
z
w z
w z
z z
z w w
j
j
j
=
z
w
z
k
Figure 1: The InfoMax algorithm.
assumes only that the mixture is linear, the origi-
nal sources are statistically independent, and that at
most one of them exhibits a Gaussian distribution.
The technique applies a linear transformation on the
mixed signals and adapts its coefcients to maximize
the statistical independence of the outputs.
There are several algorithms that perform ICA,
but one of the most widely used is InfoMax (Car-
doso, 1997), which separates the sources by minimiz-
ing the mutual information between the outputs. Fig-
ure 1 outlines the data ow of the algorithm. Like
all ICA algorithms, the InfoMax computes the out-
put vector u(k) as the product of the n-input vector
x(k) = [x
1
(k). . . x
n
(k)]
T
and a weight matrix W(k):
u(k) = x(k)
T
W(k) (1)
where k is the time step. After each block of data (256
samples in our implementation), the algorithmapplies
a nonlinear learning rule to update the coefcients of
W. InfoMax maximizes the entropy of the output,
using the learning rule W(k +1) = W(k) +W(k),
where is the learning rate and
W(k) = (I (u(k))u
T
)W(k) (2)
is the weight increment at time k, where I is the iden-
tity matrix, (u) =
g(u)
u
and g() is an invertible and
nonlinear function.
Both the learning rate and the nonlinear function
g() are parameters chosen by the designer, and affect
the dynamic and stationary behavior of the algorithm.
One of the most widely used nonlinear functions is
g(u) = tanh(u) (u) = 2tanh(u).
To prevent all vectors of matrix W from converg-
ing in the same direction, the algorithm orthogonal-
izes and normalizes them after each iteration. We use
Gram-Schmidt orthogonalization:
pro j
z
w =
w, z
z, z
z
T
Figure 2: Array architecture.
z
1
= w
1
w
1
=
z
1
z
1

z
2
= w
2
proj
z
1
w
2
w
2
=
z
2
z
2

.
.
.
.
.
.
z
k
= w
k

k1
j=1
proj
w
j
z
k
w
k
=
z
k
z
k

(3)
where w
k
are the weight vectors, z
k
are their orthog-
onal projections before normalization, is the dot
product, and is the Euclidian norm.
3 THE ARRAY
3.1 Architecture
Figure 2 depicts the general architecture of our hard-
ware implementation of InfoMax. The algorithm ex-
ecutes in three main stages:
First, an array of hardware multipliers performs a
vector-matrix multiplication to compute the out-
puts following Equation 1. The weight matrix
W
nxn
is stored on on-chip RAM memory blocks.
Second, for every block of 256 samples, the
output values are used to compute the InfoMax
weight update according to Equation 2, which in-
volves computing the tanh function and a matrix
multiplication.
Finally, the array applies the updates to the stored
weight matrix, and then normalizes and orthog-
onalizes the weight vectors using Gram-Schmidt
(Equation 3), which require multiplications, divi-
sions, and square roots.
We exploit the available data parallelism both
within each stage, and also between stages using
pipelining. The scheduling of the computations dif-
fers between the two different congurations of the
array, and are explained in more detail in Sections 3.3
and 3.4. First, we discuss the implementation of basic
arithmetic functions that are common to both cong-
urations.
A RECONFIGURABLE ARRAY FOR BLIND SOURCE-SEPARATION ON AN FPGA
263
3.2 Basic Functions
The basic operations required by the InfoMax algo-
rithm are multiplication, addition, hyperbolic tangent,
square root, and division. Our current implementa-
tion uses Xilinx FPGAs, which feature 18-bit signed
integer multipliers. Therefore, the input signals, coef-
cients and most intermediate values are encoded as
18-bit xed-point quantities (5-bit integer and 12-bit
fractional, plus sign). Addition is performed using
18-bit ripple-carry adders supported directly by the
FPGA hardware with a fast carry chain. We use the
available 18-bit hardware multipliers to compute the
products, selecting the corresponding 18-bit slice out
of the 36-bit result, and performing arithmetic round-
ing to improve resolution.
Nonlinear functions such as tanh, division and
square root are not supported by the FPGA hardware,
so they must be implemented by hand. The algorithm
computes the tanh of every output value, therefore
its implementation must be fast. As depicted in Fig-
ure 3(a), we use on-chip block RAM to implement
lookup tables (LUTs) that contain selected values of
the tanh function, and use linear interpolation to ob-
tain intermediate values. Because the function is sym-
metrical, we only store values for positive arguments.
Each of the 512 table entries contains the slope and
offset of a linear segment of the approximation, and
the table is indexed using the 9 most-signicant bits
of the absolute value of the argument.
We also use LUTs to compute the square-root
function. The function is smooth for large arguments,
but changes rapidly for small ones. The resolution
of the square root function greatly affects the conver-
gence of the InfoMax algorithm, so the uniformly-
spaced table approach used for tanh is inadequate.
Our solution, depicted in Figure 3(b) is to use two
LUTs : the rst table provides an exact value for the
rst 512 values of the argument, while the second ta-
ble implements a linear approximation for the rest of
the arguments 18-bit dynamic range.
The convergence of InfoMax is very sensitive to
the resolution of the division operation. Therefore,
implementing it with LUTs would require extremely
large tables. Instead, we opted for an iterative algo-
rithm which computes the quotient with 18-bit reso-
lution in 24 clock cycles using a series of bit shifts,
comparisons, and subtractions. Figure 3(c) illustrates
the algorithm.
3.3 Fully-Parallel Conguration
The fully-parallel implementation of the algorithm
targets large FPGA devices to maximize performance
tanh( ) u
u
u
[
1
6
:
8
]
(a) Hyperbolic tangent.
u
u
[
1
6
:
9
]
u
[
7
:
0
]
(b) Square root.
Prenormalize
Partial dividend
Left shift register
Quotient
left shift register
Postnormalize
Initial
Load
A B
A - B
Recursive
load
Shift/load select
Divisor
B
B
A
A
Shift in
Quotient Dividend
(c) Division.
Figure 3: Implementation of basic functions.
at the expense of die area. In our current implementa-
tion, we target this conguration to a Xilinx Virtex-II
Pro XC2VP30 platform FPGA, which contains 136
multipliers, 30, 816 logic cells, and 2, 448 Kbits of
block RAM. The device also contains two PowerPC
processor cores, which we do not currently use. Our
current implementation reads 4 data streams and pro-
duces 4 output streams.
The fully-parallel array breaks up the algorithm
into four stages, each of them further decomposed in
substages which execute concurrently in a pipelined
fashion. The hardware multipliers ultimately con-
strain how many operations of the algorithm we can
perform in parallel on the chip. Table 1 summarizes
the actions performed by each stage and their resource
utilization. We briey describe their operation and re-
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
264
Table 1: Fully-parallel conguration.
Stage Operation Mult.
A1 Read input x
4x1
0
A2 u =W
4x4
x
4x1
16
A3a Read 4 tanh LUT 0
A3b 4 times: tanh(u
k
) = c
k
+m
k
u
k
4
A4 A
4x4
=A
4x4
+tanh(u
4x1
) u
T
4x1
16
B1 B
4x4
=I
4x4
2 A
4x4
0
B2 D
4x4
=B
4x4
W
4x4
16
B3 W
4x4
=W
4x4
+c D
4x4
4
C1 4 dot products a, b 16
C2 w, z/z, z 0
C3 3 times: z
k
=w
k

k1
j=1
pro j
w
j
z
k
12
D1 a =
4
j=1
z
jk
4
D2 b = sqrt(a) 1
D3 w
k
=z
k
/b 0
lationship to the equations in Section 2:
Stage A - Separation. Computes the output vector
and the tanh of each output element in a 256-
sample block. Stage A1 loads the input vec-
tor from memory, while A2 performs the ma-
trix - vector multiplication x(k)
T
W(k) depicted
in Equation 1. Stages A3a and A3b compute
tanh(u) using the LUTs, and stage A4 accumu-
lates (u) u
T
according to Equation 2.
Stage B - Update. Every 256 samples, the array up-
dates the weight matrix W according to the Info-
Max learning rule. Stage B1 computes I(u)u
T
in Equation 2, stage B2 completes the computa-
tion of W, and stage B3 applies the weight up-
date.
Stage C - Orthogonalization. Orthogonalizes the
new weights according to Equation 3. Stages C1
performs up to four dot products in parallel, C2
compute the projections, and stage C3 subtracts
them from the original weight vectors.
Stage D - Normalization. Normalizes the weight
vectors. Stages D1 and D2 compute the Euclidean
norm of each vector using multiplies and square-
root operations, and stage D3 divides each weight
vector by its norm.
Figure 4 depicts the schedule of execution of all
four stages in the fully-parallel architecture. The ar-
chitecture uses pipelining to overlap the execution
of different substages for consecutive input vectors.
Thus, in stage A up to ve substages are executed si-
multaneously on the array, and the array processes a
new input vector on each clock cycle.

.
.
.
.
(a) tanh(u) u
T
.
(b) Weight update.
(c) Orthogonalization and normalization.
Figure 4: Schedule of fully-parallel conguration.
3.4 Folded Conguration
In order to t the array into smaller (and thus less ex-
pensive) FPGA devices, we fold each stage of the ar-
chitecture onto itself to process fewer input elements
simultaneously. Thus, instead of performing every
operation of each substage in parallel, we use time-
multiplexing to share multiple hardware resources be-
tween different elements of an input vector. The cur-
rent target device for our folded conguration is a
Xilinx Spartan 3 XC3S1000, which features 24 mul-
tipliers, 17, 280 logic cells, and 432 Kbits of block
RAM. A software tool aids the designer in the pro-
cess of folding the architecture based on the available
resources in the target device, and generates the cor-
rect HDL code.
Table 2 shows the stages and resource utilization
of the folded conguration. Compared to Table 1, the
resource utilization has been now reduced to process
only one vector element at a time, while the fully-
parallel conguration processes a entire new vector at
each clock cycle. As a result, the folded conguration
uses approximately 25% of the resouces of the fully-
parallel version, at the cost of extended execution time
and a slightly more complex control structure.
Figure 5 shows the pipelined scheduling of the
operations within each stage in the folded congura-
tion. Stage A now processes a new input vector every
eleven clock cycles. Orthogonalization and normal-
ization take place sequentially, and the division oper-
A RECONFIGURABLE ARRAY FOR BLIND SOURCE-SEPARATION ON AN FPGA
265
Table 2: Folded conguration.
Stage Operation Mult.
A1 Read input x
4x1
0
A2 u
i4x1
=w
i1x4
x
4x1
4
A3a Read 1 tanh LUT 0
A3b 1 time: tanh(u
k
) = c
k
+m
k
u
k
1
A4 a
i1x4
=a
i1x4
+tanh(u
4x1
) u
T
4x1
4
B1 B
4x4
=I
4x4
2 A
4x4
0
B2 d
k1x4
=b
k1x4
w
k4x1
4
B3 w
k4x1
=w
k4x1
+c d
k4x1
1
C1 1 dot product a, b 4
C2 w, z/z, z 0
C3 1 time: z
k
=w
k

k1
j=1
pro j
w
j
z
k
4
D1 a =
4
j=1
z
jk
4
D2 b = sqrt(a) 1
D3 w
k
=z
k
/b 0
ation during normalization (D3) is computed simulta-
neously for all weights at the end of the update.
4 EXPERIMENTAL RESULTS
We mapped the fully-parallel conguration of the
array to a Xilinx Virtex-II Pro XC2VP30 platform
FPGA. The maximum clock rate achieved by our im-
plementation is 55.8MHz, corresponding to a critical
path of 18ns. The chip dissipates 103mW of power.
We tested the array on a 4-input experiment of blind
source-separation with InfoMax. The chip is capable
of processing a 256-sample block and update the In-
foMax weights in 7.8s, corresponding to 431 clock
cycles. The algorithm converges in approximately
7.5ms (less than 1000 block iterations).
The folded conguration mapped onto a Xilinx
Spartan 3 XC3S1000 FPGA exhibits a critical path of
22.3ns, achieving a clock rate of 49.9MHz, and dissi-
pating 93.4mW. On the same 4-input experiment, the
chip processes a 256-sample block in 113s, corre-
sponding to 5, 073 clock cycles. The algorithm con-
verges in less than 110ms in this conguration of the
architecture. Table 3 summarizes the resource utiliza-
tion (excluding multipliers and block RAM) of each
conguration of the array on their corresponding de-
vice.
For comparison purposes, we implemented the In-
foMax algorithm in software on a laptop PC featur-
ing oating-point arithmetic, a dual-core processor
running at 1.73GHz, and 2GBytes of DDR2 RAM.
An optimized C implementation of the InfoMax al-
gorithm running the same experiment used to test the
array, completes one iteration every 90s, and con-

.
.
. .
(a) tanh(u) u
T
.

.
.
.
.
.
.
.
.
.
(b) Weight update.

.
.

.
.
.
(c) Orthogonalization
(d) Normalization (1).

..

...
(e) Normalization (2)
Figure 5: Schedule of folded conguration.
verges in 85ms . Thus, the fully-parallel congura-
tion of the array reduces the execution time of the
software implementation by a factor of 11.3. On the
other hand, the execution time of the software version
is smaller than that of the folded array by a factor of
1.3. The power consumption of both hardware imple-
mentations is smaller than the computer by more than
two orders of magnitude.
We tested both arrays on fetal ECG signals. In
this application, spatially-distributed electrodes cap-
ture four mixtures of signals which contain the ECG
of the mother and the fetus. We recorded the results
obtained from both hardware implementations of the
algorithm, and from the software running on a PC.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
266
Table 3: Global resource utilization.
Fully parallel Folded
Resource Qty. % Qty. %
Slices 5,115 37% 6,680 86%
Flip Flops 4,010 14% 4,743 30%
Input LUTs 6,686 24% 11,009 71%
IOBs 44 7% 43 24%
Block RAM 1 24% 1 68%
Block ROM 8 2% 5 13%
Multipliers 91 67% 13 54%
Because they are based on the same architecture, both
arrays produce the same results. Figure 6 shows the
waveforms obtained from the hardware implementa-
tion of the algorithm. The chips successfully produce
the ECG signal of the fetus, two ECG signals of the
mother, and the noise. Figure 7 compares the software
and hardware results for one waveform. We measured
the error of the results produced by the array relative
to the oating-point software implementation. The
RMS value of the error, normalized to the amplitude
of the signal, varies between 0.07% and 0.09%.
0 0.2 0.4 0.6 0.8 1
1
0
1
(a) Time(Sec)
A
m
p
l
i
t
u
d
e
0 0.2 0.4 0.6 0.8 1
1
0
1
(b) Time(Sec)
A
m
p
l
i
t
u
d
e
0 0.2 0.4 0.6 0.8 1
1
0
1
(c) Time(Sec)
A
m
p
l
i
t
u
d
e
0 0.2 0.4 0.6 0.8 1
1
0
1
(d) Time(Sec)
A
m
p
l
i
t
u
d
e
0 0.2 0.4 0.6 0.8 1
0.1
0
0.1
(e) Time(Sec)
A
m
p
l
i
t
u
d
e
0 0.2 0.4 0.6 0.8 1
0.05
0
0.05
(f) Time(Sec)
A
m
p
l
i
t
u
d
e
0 0.2 0.4 0.6 0.8 1
1
0
1
(g) Time(Sec)
A
m
p
l
i
t
u
d
e
0 0.2 0.4 0.6 0.8 1
2
0
2
(h) Time(Sec)
A
m
p
l
i
t
u
d
e
Figure 6: (a)-(d): Measured mixtures, (e): ECG - fetus, (f):
Noise, (g)-(h): ECG - mother.
0 0.1 0.2 0.3 0.4 0.5
3
2
1
0
1
2
3
Time(sec)
N
o
r
m
a
l
i
z
e
d

A
m
p
l
i
t
u
d
e


Software
Hardware
Figure 7: ECG results - fetus.
5 CONCLUSIONS
We described an array architecture for ICA using the
InfoMax algorithm. The array can be recongured
to target a wide range of FPGA devices, represent-
ing different price/performance tradeoffs. We showed
two congurations: a fully-parallel version mapped to
a Xilinx Virtex-II Pro XC2VP30, and a folded imple-
mentation mapped to a Xilinx Spartan-3 XC3S1000.
The parallel array outperforms both the folded cong-
uration by a factor of 14.5, and a PC-based software
implementation by a factor of 11.3. Both hardware
arrays consume in the order of 100mW, use 18-bit
xed-point arithmetic, and achieve a resolution within
0.08% of a oating-point software implementation of
the algorithm. Future and ongoing work includes de-
veloping a software tool to automate the recongura-
tion process, and integrating the folded version of the
array on a portable ECG instrument.
ACKNOWLEDGEMENTS
This work was partially funded by grants Fondecyt
1070485 and Milenio ICMP06-67F.
REFERENCES
Anguita, D., Boni, A., and Ridella, S. (2003). A digital ar-
chitecture for support vector machines: Theory, algo-
rithm, and FPGA implementation. IEEE Transactions
on Neural Networks, 14:9931009.
Bell, A. J. and Sejnowski, T. J. (1997). The Independent
Components of Natural Scenes are Edge Filters. Vi-
sion Research, 37(23):33273338.
Cardoso, J. F. (1997). Infomax and maximum likelihood
for blind source separation. IEEE Signal Processing,
4:112114.
Li, Z. and Lin, Q. (2005). FPGA implementation of In-
fomax BSS algorithm with xed-point number repre-
sentation. Neural Networks and Brain, 2:889892.
Potter, M., Gadhok, N., and Kinsner, W. (2002). Separation
performance of ICA on simulated EEG and ECG sig-
nals contaminated by noise. IEEE Canadian Confer-
ence on Electrical & Computer Engineering, 2:1099
1104.
Yang, Y., Huang, X., and Yu, X. (2007). Real-time ECG
monitoring systembased on FPGA. 33rd Annual Con-
ference of IEE Industrial Electronics Society, pages
21362140.
Zeng, Y., Liu, S., and Zhang, J. (2008). Extraction of fetal
ECG signal via adaptive noise cancellation approach.
The 2nd International Conference on Bioinformatics
and Biomedical Engineering, pages 22702273.
A RECONFIGURABLE ARRAY FOR BLIND SOURCE-SEPARATION ON AN FPGA
267
IMPROVING SURFACE ENERGY AND HYDROPHILIZATION
OF POLY(ETHYLENE TEREPHTHALATE)
BY ENZYMATIC TREATMENTS
Isabel C. Gouveia
1,3*
, Laura C. Antunes
3
and J oo A. Queiroz
2,3

1
Departamento C.T. Txteis,
2
Departamento Qumica,
3
R&D Materiais Txteis e Papeleiros
Universidade da Beira Interior, 6201-001 Covilh, Portugal
*[email protected]
Keywords: Poly(ethylene terephtalate), Enzymatic treatment, Contact angle, Surface energy, Hydrophilicity, Esterases,
Lipases, Adhesion.
Abstract: In order to increase the hydrophilicity and adhesion of poly (ethylene terephthalate) (PET) fabrics it was
studied the action of three types of enzymes (Texazym PES sp5, Aspergillus niger and Aspergillus oryzae)
applied at different incubation times and concentrations. This processes aims to modify morphologically
and chemically the superficial structure of the polymeric materials (PET), forming new carboxyl, hydroxyl
and other polar groups at the surface, in order to increase adhesion and hydrophilicity. The increase in the
hydrophilicity of the fabric was evaluated by measuring the contact angle being the best results obtained for
the Texazym PES (87.45), much smaller than the non-treated fabric (122.95); and by the wicking height,
which revealed an important improvement in the hydrophilicity. The formation of carboxyl and hydroxyl
groups was evaluated by a staining procedure with a cationic and reactive dye, respectively. It was also
confirmed by the increasing in the polar component of the surface energy, determined by the Qwens-Wendt
method. The higher surface energy and thus, the higher adhesion properties, were obtained for the esterase
Texazym, using 0.12U during 90 minutes. The surface morphology of the non-enzymatic-treated and
enzymatic-treated samples was analyzed by scanning electron microscopy (SEM) showing no degradation
of fibers treated under the selected optimum conditions. In contrary, this method showed an important
surface cleaning action by removing some undesirable polyester oligomers.
1 INTRODUCTION
Advances of biotechnology in textile industry have
brought new products and processes for specialty
applications as for instance in biomedical materials.
Polymers and textiles are usually used as films
and foils for packaging, protective coating, material
for biomedical and sealing applications because of
their superior bulk properties, such as transparency,
high resistance, strength, good thermal resistance,
etc. But these excellent characteristics are often
unsuitable for biomedical applications due to their
low surface energies. Therefore, surface treatments
are usually necessary to improve surface wetting and
adhesion properties (Inagaki et al, 2001), (Yang and
Gupta, 2004), (Guebitz and Cavaco-Paulo, 2008),
(Huemann et al, 2006).
The synthetic fibers, in particular, polyester
made from poly(ethylene terephthalate), (PET) have
a reduced number of polar groups (hydroxyl and
carboxyl groups) capable to establish hydrogen
bonds with water, reflecting in its weak capacity to
absorb water also related to its high degree of
cristalinity. This property can be changed appealing
to chemical methods as, for example, the alkaline
treatment. This method can, however, damage not
only to the fibers but can also be harmful to the
environment. In this way, alternative processes,
simultaneously ecological, efficient and safe, have
been studied.
Earlier studies demonstrated that the application
of lipases, cutinases and esterases to synthetic fibers,
help increasing hydrophilicity through the hydrolysis
of ester bonds, under moderate conditions (low
concentration and low reaction time at room
temperature), accompanied by a slight reduction of
the resistance to rupture and weight loss
(Vertommen et al, 2005), (Heumann et al, 2006).
Several studies of enzymatic treatments have
been proposed in order to modify the surface
properties of polymers such as adhesivity,
268

hydrophobicity, oleophobicity, wettability/
hydrophilicity, pilling and static charges, by
hydrolysing polymers without affecting the bulk
properties, having the advantage of being eco-
friendly compared with conventional chemical
treatments (Guebitz and Cavaco-Paulo, 2008),
(Heumann et al, 2006), (Vertommen et al, 2005).
Other authors (Heish and Cram, 1998) confirmed
that the increasing of the hydrophilicity after
modification of polyester with lipases was superior
to the one achieved with conventional chemical
treatments (alkaline treatment: 3N of NaOH during 2
hours).
Our new approach focus on the surface
modification of PET fabrics by enzymatic treatments
using three different enzymes (Texazym PES and
Aspergillus niger, Aspergillus oryzae, respectively),
in order to form new polar groups (carboxyl and
hydroxyl) at the surface, capable to establish
hydrogen bonds with water and capable to improve
surface adhesion. The study was undertaken to
analyze and compare the effect of enzymatic
treatments applied in a textile material (100% PET
fabric), by studying the morphological and chemical
changes at the surface, the mechanical properties
and surface energy, in order to establish whether or
not the material can be functionalized and its surface
adhesion properties can be improved.
2 EXPERIMENTAL
The enzymatic treatment aims to improve the
hydrophilicity without harming the mechanical
properties of the material. For that purpose, it was
investigated a new approach by studying the effect
of the three types of enzymes, esterases and lipases
(Texazym PES sp5, Aspergillus niger and
Aspergillus oryzae), varying the incubation time and
the enzyme concentration.
The chemical modifications were investigated by
measuring the contact angle and the wicking height.
The indirect determination of the formed carboxyl
and hydroxyl groups was measured by staining with
a cationic dye (Methylene Blue) and a reactive dye
(Reactive Black 5), respectively, and by measuring
the surface energy by the Owens-Wendt method.
The surface morphological changes were analyzed
by scanning electron microscopy (SEM).
2.1 Materials
The substrate (textile material) used in this work was
a 100% poly (ethylene terephthalate) fabric, Batavia
Twill, with the characteristics indicated in Table 1.
The substrate was pre-washed with 1 g/L Plurafac
LF 400, at 50C during 60 minutes, with mild
mechanical agitation (25 rpm). Subsequently, the
substrate was rinsed and washed under running
water, followed by a thermofixation at 170C during
15 minutes.
2.2 The Enzymes
The enzymes selected for this study, were an
esterase (Texazym PES sp 5 from inoTEX Ltd.) and
two lipases (Aspergillus niger and Aspergillus
oryzae from Sigma). These enzymes were applied
according to the literature reviewed under the
conditions of pH and temperature indicated by the
manufacturer. In Table 2 are presented the principal
characteristics of the enzymes used.
It was studied the action of Texazym PES,
Aspergillus oryzae and Aspergillus niger, applied at
different concentrations (0.06, 0.09, 0.12, 0.15 and
0.18 U) and different incubation times (30, 60, 90
minutes and 24 hours) with a liquor ratio of 1:25.
The enzymatic treatments with Texazym were
performed at 30C, using 50 mM of sodium acetate
buffer solution (pH 5.5). The enzymatic treatments
with Aspergillus niger and Aspergillus oryzae were
performed by incubating 2 g of polyester fabric at
45C and 40C, using 50 mM phosphate buffer (pH
7.0). Immediately before and after treatments all
samples were placed in a standard atmosphere
(20 2 C, 65% HR) during 242 hours.
Table 1: Fabric characterization.
Characterization Test Method Value
Warp direction
Linear Mass (Tex)
Density
(Yarns/cm)
Diameter (Den)

NP 4105
NP EN
1049-2
NP 3160

48.47
21.4
3.3
Weft direction
Linear Mass (Tex)
Density
(Yarns/cm)
Diameter (Den)

NP 4105
NP EN
1049-2
NP 3160

36.67
32
3.3
Weaving
construction

Mass per area
(g/m
2
)

Composition


NP EN 1700

NP EN 1701

NP EN 1808.
2247 and
2248
Batavia twill

211.65

100%
Polyester
IMPROVING SURFACE ENERGY AND HYDROPHILIZATION OF POLY(ETHYLENE TEREPHTHALATE) BY
ENZYMATIC TREATMENTS
269

2.3 The Dyes
The dyes were selected to fulfill the objectives
focalized on assessing the number of hydroxyl and
carboxyl groups.
Therefore, a reactive dye Remazol Black B (C.I.
Reactive black 5, from DyStar) and a cationic dye
(Methylene blue, from Sigma-Aldrivh) were
selected. The reason for this choice is essentially the
ability to establish bonds with hydroxyl and
carboxyl end groups in the polyester fabric,
respectively.
Table 2: Characterization of the enzymes.
TexazymPES
sp 5
Aspergillus
o.
Aspergillus
n.
pH Optimal 5.5 7.0 7.0
Temp. () 30 40 45
Origin Thermobifida
fusca
Aspergillus
oryzae
Aspergillus
niger
Activity 0.6 U/ml 50 U/mg 4 U/g
2.4 Evaluation of the Effectiveness
of the Enzymatic Treatments
2.4.1 Determination of the Contact Angle
and Surface Energy Estimation
The Dataphysics is composed by a camera, a
computer and a monitor which are used to measure
the contact angle on the samples. Liquid drops were
dispersed on each fabric sample using a micrometer
pipette. The image of each drop was captured by the
camera connected to a computer, and the captured
images were viewed at the monitor. The standard
testing methods were according to Tappi T 558 pm-
95. The liquid used in the experiment was glycerol
and a drop of 5 l was deposited on the fabric
surface and the contact angles were measured. The
measurements were performed, after one week since
enzymatic application, were repeated eight times and
the average was calculated. Afterwards, surface
energies were determined according to the Owens-
Wendt approach (Owens and Wendt. 1969). This
method takes into account the dispersive and polar
components of the surface energy. Using different
test liquids, in this case water and glycerol, it is
possible to calculate the solid surface energy as the
sum of the polar and dispersive contributions.
Constant values for the test liquids used for contact
angle measurements are as follows:
Water: =72.8 mJ /m
2
;
s
d
=21.8 mJ /m
2
;
s
p
=51.0
mJ /m
2
;
Glycerol: =64.0 mJ /m
2
;
s
d
=34.0 mJ /m
2
;
s
p
=
30.0 mJ /m
2
.
2.4.2 Wicking Rate
The determination of the wicking rate by measuring
the rising height was according to DIN 53924
vertical wicking tests and performed after one week
since the enzymatic application. Samples of 3cm
10 cm were prepared and were suspended in a
standard atmosphere (20 2 C. 65% HR) for 242
hours. The samples were then placed in a solution
0.05 % w.o.f. of dye (Methylene Blue), and
immersed at a height of 1 cm. After 10 minutes the
samples were removed and the rinsing height was
measured up. It was used a solution of dye instead of
water to facilitate the reading and the measurement
of the rinsing height.
2.4.3 Determination of Carboxyl End
Groups
The carboxyl end groups were determinate after
dyeing the samples at 50 C with a cationic dye
(Methylene Blue. 0.5 % (w.o.f.)) during 20 minutes.
Subsequently, the samples were washed in hot and
cold water and then dried in an oven at 40 C during
24 hours. The dyed samples were analyzed in a
reflectance measuring apparatus (Spectraflash 300
Datacolor, LAV/Spec. Incl., d/8. D65/10). This
procedure aims to evaluate the increase or decrease
in the intensity of color (by determining the value of
K/S), in order to evaluate the formation of carboxyl
groups. The relative color strength (K/S values)
were established according to the Kubelka-Munk
equation (1), where K and S stands for the
absorption and scattering coefficients and R stands
for the reflectance value, respectively (Shah and
Gandhi, 1990; Pandiyaraj and Selvaranjan, 2008):
( )
R
R
S
K
2
1
2

=

(1)
An increase in the value of K/S, when compared
to the non-treated sample, indicates an increase of
carboxyl groups known to react with this type of
dye.
2.4.4 Determination of Hydroxyl End
Groups
The hydroxyl end groups in the non-treated and
treated samples were determinate by a dyeing
procedure performed at 60 C with a reactive dye
(Remazol Black B) during 90 minutes. The dye bath
contains 2% w.o.f. of dye, 20 mg.ml
-1
of Na
2
CO
3

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
270

and 60 mg.ml
-1
of Na
2
SO
4
, pH 11, with a liquor ratio
1:100 and a mechanical agitation of 40 rpm.
Subsequently, the samples were washed in hot and
cold water and then dried in an oven at 40 C during
24 hours. After dyeing, the samples were analyzed
in the reflectance measuring apparatus above
described to evaluate the increase or decrease in the
intensity of color, in order to evaluate the formation
of hydroxyl groups. An increase in the K/S values
indicates an increase in the formation of hydroxyl
groups which react covalently with this type of dye.
2.5 Quality Control Test
2.5.1 SEM Analysis
The surface morphology of the treated polyester
fabric was observed using a scanning electron
microscope. SEM analysis was performed in all
samples after one week since the application of the
enzymatic treatments, using a HITACHI S2700
Electron Microscope and an EMITECH-K550 gold
evaporator.
2.5.2 Determination of the Resistance to
Abrasion
The Martindale evaluation system was used for
measuring the resistance to abrasion (mechanical
properties) of the non-treated and treated samples.
The standard testing method was according to IWS
TM 112. In this method, the samples are tested
under a weight of 9 KPa and run until the rupture of
two yarns.
3 RESULTS AND DISCUSSION
3.1 Evaluation of the Effectiveness of
the Enzymatic Treatments
The results for the effect of the enzymatic
treatments, over the contact angle and the
hydrophilicity are indicated in Figures 1 and 2. The
results for the contact angle (Figure 1) showed a
slight decrease for all enzyme concentration, but the
most significant decrease was observed with
Texazym PES, when compared to the non-enzymatic
treated sample (Control). For the others enzymes
(Aspergillus niger and Aspergillus oryzae) it can be
also observed a slight decrease. However, for higher
concentrations of enzyme and incubation times a
significant decrease of contact angle in relation to
the control can be observed, being in accordance
with the results obtained by other authors (Hsieh e
Cram. 1998).
After a comparative analysis, we can conclude
that either the use of Aspergillus oryzae or an
Aspergillus niger made possible to achieve a lower
contact angle, hence greater hydrophilicity for
polyester fabric. However the Texazym PES led to
the better results. Definitely, by applying the
Texazym PES with a concentration of 0.12 U during
90 minutes it can be achieved a contact angle of
87.45; by applying the Aspergillus niger with a
concentration of 0.15 U during 60 minutes a contact
angle of 101.50, is obtained.
The use of Aspergillus oryzae, (0.12 U during 60
min) also allows to achieve lower values for the
contact angle (109.03). The more favourable
conditions to decrease the contact angle seens to be
using the Texazym PES with a concentration of 0.12
U during 90 minutes. For lower incubation times (30
to 60 minutes) there are no important changes in the
contact angle values.
80
85
90
95
100
105
110
115
120
125
130
0,00 0,05 0,10 0,15 0,20
C
o
n
t
a
c
t

A
n
g
l
e

(

)
Enzyme concentration (U)
Control Texazyme - 30 min
Texazym- 60 min Texazym- 90 min
Texazym- 1440 min Oryzae - 30 min
Oryzae - 60 min Oryzae - 90 min
Oryzae - 1440 min Niger - 30 min
Niger - 60 min Niger - 90 min
Niger 1440 min

Figure 1: Values of contact angle for all enzymes.
The results of hydrophilicity obtained by the
effect of capillary are illustrated in Figure 2. Their
analysis indicates that the best results are achieved
using the Texazym PES, followed by the use of the
Aspergillus niger. In those circumstances its
obtained a wicking height of 5.40 cm and 4.60 cm,
respectively. Other studies revealed that the
application of other enzymes promoted similar
results, however with a lower wicking height ( 4
cm) (Alisch-Mark et al, 2006). In agreement with
the results of the contact angle and hydrophilicity,
the use of Aspergillus oryzae leads to lower values
IMPROVING SURFACE ENERGY AND HYDROPHILIZATION OF POLY(ETHYLENE TEREPHTHALATE) BY
ENZYMATIC TREATMENTS
271

in the capacity of absorption of water by capillarity
effect (Figure 2).
The results obtained by the analysis of the
carboxyl end groups (Figure 3), formed by the
application of Tezaxym PES for any concentration
and incubation time, showed a significant increase
detected by the reaction with the cationic dye, when
compared to the other enzymes and the control.

Figure 2: Values of hydrophilicity (expressed by the
wicking height).

Figure 3: Values of K/S (620 nm), after dyeing with
cationic dye.
The more favourable conditions to the grafting of
these groups are by applying a concentration of 0.12
to 0.15 U for any incubation time. These results are
in accordance with the results obtained before
(contact angle and hydrophilicity), showing that the
presence of carboxyl groups favours the
establishment of hydrogen bonds with water. An
application with Aspergillus oryzae seems to be less
desirable with regards with the formation of these
functional groups, reflecting up in the lower capacity
of the treated fabrics to absorb water.
The formation or not of the hydroxyl groups,
using different enzymes, can be indirectly measured
by the higher or lower intensity of color that results
from the reaction of a reactive dye with the hydroxyl
groups (-OH) by the formation of covalent bonds.
After analyzing the results presented in Figure 4, it
can be concluded that a higher formation of the
hydroxyl groups is achieved by using the Texazym
PES, whichever the incubation time, comparing with
the others enzymes and with the control. This feature
is extremely important with respect to future
applications of this enzyme in the textile industry,
where the incubation time is a cruel factor in textile
treatments.


Figure 4: Values of K/S (600 nm), after dyeing with
reactive dye.
To better elucidate which were the best
conditions for each enzyme, considering the best
results of the contact angle and wicking height, the
selected optimum conditions for each enzyme are
presented in Table 3. The best results regarding the
lower contact angle and higher wicking height are
obtained for the Texazym PES us1ing a
concentration of 0.12 U during 90 minutes, followed
by the Aspergillus niger using 0.15 U, 60 minutes
and finally, by the Aspergillus oryzae using 0.12U
during 60 minutes. In those conditions, an
application with Texazym PES yielded a rinsing
height of 5.40 cm, which is a very good value when
compared with the ones obtained by other authors
(Alisch-Mark et al, 2006). In this study they
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
272

achieved a rinsing height of 4.2 cm (approximately)
after 48 hours of incubation with T.fusca (16 U). Our
results revealed an important achievement since
much lower incubation time and enzyme
concentration was used.
However, the other results regarding the quality
control parameters are important to better define the
optimal conditions overall.
Table 3: Optimal conditions for different operating
parameters.

Contact
Angle
(glycerol)
()
Rising
Height
(cm)
Cationic
Dye
K/S
Reactive
Dye
K/S
Control 122.95 2.7 0.60 0.34
Texazym
PES
(0.12 U,
90min)
87.45 5.40 0.81 0.62
Asperg.
niger
(0.15 U,
60 min)
101.50 4.60 0.50 0.28
Asperg.
oryzae
(0.12U,
60 min)
109.03 0.40 0.60 0.34

The total surface energy, the dispersion
component and polar component of the fabrics were
calculated according to the Owens-Wendt approach
(
s
p
.
s
d
.
s
are the polar component, the dispersion
component and the total surface energy, of fabric,
respectively) and are indicated in Tables 4 to 6.
It is clear that the total surface energy increases
with incubation time and enzyme concentration. The
values of surface energy obtained for the control was
0.60 mJ /m
2
for the polar component, 3.28 mJ /m
2
for
the dispersion component and 3.88 mJ /m
2
for the
total surface energy. Analyzing the results, an
important increase of the polar component for the
Texazym PES is verified when compared with the
other enzymes. The higher value (187.31 mJ /m
2
)
was obtained to the incubation time of 1440 minutes
with a concentration of 0.15U. However, good
values for the polar component can be obtained for
lower incubation times (183.85 mJ /m
2
, 153.52
mJ /m
2
), when a concentration of enzyme between
0.12U to 0.18 U is used during an incubation time of
90 minutes. Considering the other enzymes, much
lower values for the polar component are obtained:
30.36 mJ /m
2
(using 0.18 U during 60 min for the
Aspergillus niger), and 11.47 mJ /m
2
and 10.04
mJ /m
2
using 0.12 U during 30 minutes and 0.15 U
during 60 minutes, respectively, for the Aspergillus
oryzae.
Table 4: Surface energy determination for different
operating parameters of Texazym PES.
Time
(min)
Concentration of
enzyme (U)

s
p

(mJ /m
2
)

s
d

(mJ /m
2
)

s

(mJ /m
2
)
30
0.06 22.49 1.9 29.4
0.09 67.88 15.9 83.78
0.12 99.57 29.15 128.72
0.15 97.6 29.76 127.36
0.18 82.13 20.29 102.41
60
0.06 31.02 1.78 32.8
0.09 81.07 21.81 102.88
0.12 98.54 24.12 122.66
0.15 110.76 32.74 143.5
0.18 102.32 32.31 134.62
90
0.06 61.00 6.32 67.32
0.09 122.07 29.78 151.85
0.12 183.85 63.43 247.38
0.15 153.52 39.21 192.73
0.18 141.68 29.69 171.36
1440
0.06 45.98 5.70 51.68
0.09 117.12 42.4 159.53
0.12 147.85 54.21 202.06
0.15 187.31 91.66 278.97
0.18 145.94 57.2 203.14
Table 5: Surface energy determination for different
operating parameters of Aspergillus niger.
Time
(min)
Concentration of
enzyme (U)

s
p

(mJ /m
2
)

s
d

(mJ /m
2
)

s
(mJ /m
2
)
30
0.06 2.67 2.76 5.43
0.09 3.9 1.94 5.84
0.12 0.1 11.34 11.43
0.15 0.00 11.55 11.55
0.18 14.95 0.02 14.97
60
0.06 0.52 6.49 7.01
0.09 0.88 5.84 6.71
0.12 0.04 15.78 15.82
0.15 0.45 25.27 25.72
0.18 30.36 1.77 32.13
90
0.06 2.44 2.94 5.37
0.09 3.11 3.82 6.93
0.12 0.28 10.81 11.09
0.15 1.79 3.62 5.41
0.18 5.82 1.39 7.21
1440
0.06 5.31 1.01 6.38
0.09 9.47 0.42 9.89
0.12 21.49 1.47 22.97
0.15 3.03 2.19 5.22
0.18 0.42 8.99 9.42
It is mainly due to the incorporation of polar
groups like carboxyl and hydroxyl on the fabric
IMPROVING SURFACE ENERGY AND HYDROPHILIZATION OF POLY(ETHYLENE TEREPHTHALATE) BY
ENZYMATIC TREATMENTS
273

surface, verified by other authors (Pandiyaraj and
Selvarajan, 2008) and in our present study.
Table 6: Surface energy determination for different
operating parameters of Aspergillus oryzae.
Time
(min)
Concentration
of enzyme (U)

s
p

(mJ /m
2
)

s
d

(mJ /m
2
)

s

(mJ /m
2
)
30
0.06 18.48 2.11 20.59
0.09 4.85 0.07 4.92
0.12 3.24 2.66 5.09
0.15 11.47 0.17 11.64
0.18 0.92 6.98 7.89
60
0.06 2.66 1.97 4.63
0.09 6.74 0.29 7.03
0.12 0.05 12.22 12.27
0.15 10.04 0.22 10.26
0.18 1.91 4.07 5.98
90
0.06 6.68 0.06 6.74
0.09 0.00 10.31 10.31
0.12 8.77 0.01 8.78
0.15 5.49 0.72 6.21
0.18 0.95 5.31 6.26
1440
0.06 1.18 2.21 3.39
0.09 3.16 1.48 4.64
0.12 5.69 0.07 5.76
0.15 2.05 2.39 4.43
0.18 1.54 3.61 5.15

The change in the polar component verified for
all enzymatic treatments, when applying different
enzyme concentrations and incubation times, affects
the total surface energy as a function of enzymatic
operating parameters. These results show that the
main contribution to the increase in surface energy is
due to the polar components which can incorporate
with moisture through hydrogen bonds. Thus, good
wettability is obtained when the values of polar
component is high (Pandiyaraj and Selvarajan,
2008), being the most favorable conditions the ones
for the enzymatic treatment with Texazym PES.
3.2 Quality Control Tests
With the test of resistance to abrasion was intended
to assess whether the mechanical properties of
polyester fabrics would have suffered any significant
change after the treatment with enzymes (Texazym
PES, Aspergillus niger and Aspergillus oryzae).
These results were referenced in Table 7.
After their evaluation it can be seen that there
werent verify important losses in the resistances to
abrasion using any enzymatic treatment when
compared with the control. The loss of around 4% in
the resistance is perfectively safe for the textile.
Table 7: Values of the resistance to the abrasion in the
optimal conditions for each enzymatic treatment.

Resistance to
abrasion (cycles)
Control 24000
Texazym PES (0.12 U. 90min) 23000
Aspergillus niger (0.15 U. 60 min) 23000
Aspergillus oryzae (0.12U. 60 min) 23000

The SEM analysis aims to observe any surface
changes on fibers after the enzymatic treatment. The
images presented are referred to the control and to
the optimum conditions of application of the
Texazym PES. The others enzymes didnt revealed
any changes when compared to the control. Looking
at images of Figures 5 and 6 it can be seen that there
is no degradation of the surface of the fibers under
study. In contrary, the use of the Texazym PES
applied under the optimum conditions seems to be
the cause of a greater effect of cleaning since it can
be seen the presence of a smaller number of particles
deposited on the fibers (oligomers). This effect of
total or partial elimination of them is of utmost
importance, especially in what concerns to adhesion
and dyeing properties.

A B
Figure 5: SEM micrographs of Control samples (A-
magnification: 2500x, B-magnification: 500x).
A B
Figure 6: SEM micrographs (Texazym PES, using 0.12 U
during 90 min).
4 CONCLUSIONS
The effect of the different enzymatic treatments,
incubation time and enzyme concentration on the
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
274

hydrophilization of PET fabrics and surface energy
increasing was analyzed. Depending on the
enzymatic process used, the wettability of PET
fabrics can be significantly improved. It was found
the formation of carboxyl and hydroxyl polar groups
by the Texazym PES action. The enhancement of
polar groups on the fabric surface was confirmed
with a cationic dye (Methylene blue) and a reactive
dye (Remazol black B). Thus, improvement in
adhesion properties can be expected.
Our results revealed an important achievement
since much lower incubation time and enzyme
concentration was used, comparing to previous
investigations.
The increasing in the hydrophilicity and surface
energy of PET fabrics are also known to have
importance in the increasing in adhesion of cells and
tissues, which is a very important property for
permanent biomedical implants.
REFERENCES
Alisch-Mark, M., Herrmann, A., Zimmermenn, W., 2006.
Increase of the hidrophilicity of polyethylene
terephthalate fibres by hidrolases from T.fusca and
F.solani f. sp. pisi. Biotechnology Letters, 28, 681-685.
Guebitz, G. M., Cavaco- Paulo, A., 2008. Enzymes go big:
surface hydrolysis and functionalisation of synthetic
polymers. Trends in Biotechnology, 26 (1), 32-38.
Heumann, S., Eberl, A., Pobeheim, H., Liebminger, S.,
Fischer-Colbrie, G., Almansa, E., Cavaco-Paulo, A.,
Guebitz, G.M., 2006. New model substrates for
enzymes hydrolysing polyethyleneterephthalate and
polyamide fibres. J . Biochem. Biophys. Methods, 39,
89-99.
Hseih, Y. and Cram, L. A., 1998. Enzymatic Hydrolysis to
Improve Wetting and Absorbency of Polyester
Fabrics. Textile Research Journal, 68 (5), 311-319
Inagaki, N., Tasaka, S., Narushima, K., Kobayashi, H.,
2001. Surface Modification of PET Films by Pulse
Argon Plasma. Journal of Applied Polymer Science,
85 (14), 2845-2852.
Owens, D.K., Wendt, R.C., 1969. Estimation of the
surface free energy of polymers. J. Appl. Polym. Sci.,
13, 174.
Pandiyaraj, K.N., Selvarajan, V., 2008. Non-Thermal
plasma treatment for hydrophilicity improvement of
grey cotton fabrics. J ournal of Materials Processing
Technology, 199,130-139.
Shah, H.S., Gandhi, R.S., 1990. Instrumental Colour
Measurements and Computer aided Colour matching
for Textile, (Eds) Mahajan Book Distributors.
Vertommen, M.A.M.E., Nierstrasz*, V.A., van der Veer,
M., Warmoeskerken, M.M.C.G., 2005. Enzymatic
surface modification of poly(ethylene terephthalate).
Journal of Biotechnology, 120, 376-386.
Yang, S., Gupta, M.C., 2004. Surface modification of
polyethyleneterephthalate by an atmospheric-pressure
plasma surce. Surface & Coatings Technology, 187
(2/3), 172-176.
IMPROVING SURFACE ENERGY AND HYDROPHILIZATION OF POLY(ETHYLENE TEREPHTHALATE) BY
ENZYMATIC TREATMENTS
275
USING MULTI-AGENT SYSTEMS TO STUDY PARACRINIENNE
CELLS INTERACTION
Lynda Dib
University Badji Mokhtar, Computer Science Department, BP12 Annaba, Algeria
[email protected]
Keywords: Multi-agent model, System of multi-agent simulation, Agent interaction, Paracrinien communication.
Abstract: This paper presents our multi-agent framework for modelling and predicting the emergent behaviour
resulting from the presence of distinct environmental conditions that lead to bad interaction of cells in their
tissue. As the cellular interaction is an important behaviour permitting the survive of cells in their tissue, the
objective of our simulator is to be a virtual world of cellular biology while analyzing and simulating the
control mechanisms during the paracrinien communication between cells in order to help its specialists to
better understand, to good interpret and to warn changes of cell states according to its actual internal state
and to the state of its environment.
1 INTRODUCTION
Biological, especially the study of the human body is
a complex field. In biological phenomena numerous
parameters intervene but their exact influence is
often difficult to determine. If it is easy to find a
mathematical model describing the evolution of an
illness for example, but it is difficult in contrast to
model and to understand what happens at the cell
level.
The modelling of the biological and medical
systems by the multi-agent approach is in its early
stages (Giuliano, Denzinger, Merelli, Miles,
Tianfield, &Unland, 2005). Several recent works are
interested in the modelling of cellular behaviour, for
example, the modelling of intracellular signals
(Boss, J onker, & Treur, 2005). Other works are
interested on the intercellular modelling, for
example, the multi-agent simulation of cellular
migration (Dib, Guessoum, Bonnet, & Laskri, 2005)
(Dib, Guessoum, Laskri, Fartas, & Guettar, 2006)
(Dib, Guessoum, & Laskri, 2006). The multi-agent
system models the neurons functioning (Colloc,
2005), and the control mechanisms modelization, for
the formation of granulomes during the tuberculosis
infection (Segovia-J uarez, Ganguli, & Kirshner,
2004). Using multi-agent systems to study cells
interaction (Dib, &Guessoum, 2007) (Dib, 2008).
Our system falls into in this category. The aim of our
work is about offering to biologists the possibility to
model and to simulate the complex systems
containing cells, molecules and their interactions in
their environment. Agent paradigm provides a very
good solution to model and simulate cells and their
interactions, especially the paracrinien
communication.
Paracrinien signaling, acting on cells
immediately adjacent to the sending cell (to a
proximity <1mm), is a complex phenomenon.
Paracrinien signals (mediators) are chemical
substances with local diffusion, in an extracellular
middle (Berridge, 1985). They are recognized by
every cell using the membranes receptors (Smith,
Hill, Lekowitz, Handler, & White, 1983).
That the cell is considered as a structural and
functional unit of the organization cannot live in
isolation, so it is necessary that there is a highly
developed communication network between it
(Masliah, & Housset, 2008). The big vital functions
(such as the breathing, digestion, movements, etc)
are not possible because cells communicate
between each other in a harmonious way. Every
cell receives and sends signals permanently toward
the neighbouring cells (Berridge, 1985) (Smith,
1997). These multiple signals are messages that cells
interpret. In response to the whole received message,
the cell chooses an action: to divide, to specialize or
to die (Chauffert, 2004).
This article presents our system simulating the
paracrinien communication. It is organized as
follows: In the second section, we describe the
proposed model for the simulator realization. The
third section will be devoted to the simulation and
276

the results obtained by the system. The conclusion is
made in section fourth.
2 MULTI-AGENT MODEL
Our system is composed of three categories of
agents: Cells, Molecules and Mediators. We also
distinguish a set of objects: the environment, the
receptors and the extracellular matrix.
2.1 The Agents
2.1.1 The Cells
The AgentCell is principally defined by a set of
characteristics represented by the following
parameters:
CE: Cellular Energy,
NR: Number of Receptors bounded to the cell,
LinkR: state of the liaison between the cellular
membrane plasmic and the Receptors. This last,
takes the value 1 if the receptors are attached to
the AgentCell membrane plasmic, the value 0 if
this junctions is not established,
NRBM: Number of Receptors Bounded to the
Mediators,
NFR: Number of Free Receptors,
NAR: Number of Active Receptors,
NNAR: Number of No Active Receptors,
StatC: internal Stat of AgentCell,
VN: Vector of AgentCell Neighbouring,
VR: Vector of Receptors belonging to the
AgentCell.
In a paracrinien interaction the communicating
AgentCell realizes the following actions:
definition of the achieved goal (secretion or
reception of a paracrinien signals) according to
the internal state of the signalling AgentCell and
to the environmental information;
sending (secretion) of the mediator by the
signalling AgentCell;
activation of the target AgentCells receptor;
reception of the mediator by the target
AgentCell;
interpretation of the mediator (signal captured by
the target AgentCells receptor);
response appropriate to the received mediator.
From the biological real characteristics of the
communicating cell the automaton concerning its
behaviour is realized (Figure 1).
S
ecretion
of chem
ical
m
essenger
AgentCell in the initial state
Signalling AgentCell
R
e
c
e
p
t
io
n
o
f
t
h
e
c
h
e
m
ic
a
l
m
e
s
s
e
n
g
e
r
I
n
t
e
r
p
r
e
t
a
t
i
o
n
o
f
t
h
e
m
e
s
s
e
n
g
e
r
t
o

d
i
e
I
n
t
e
r
p
r
e
t
a
t
io
n
o
f
t
h
e
m
e
s
s
e
n
g
e
r
f
o
r

it
s
d
if
f
e
r
e
n
c
in
g
Interpretation
of
the
m
essenger
to survive
In
te
r
p
r
e
ta
tio
n
o
f
th
e
m
e
s
s
e
n
g
e
r
to
p
r
o
life
r
a
te
Survival targets AgentCell
Proliferatedtargets AgentCell
differential targets AgentCell
Died targets AgentCell
Targets AgentCell

Figure 1: Behaviour Automaton of a communicate cell.
2.1.2 The Molecules
The cells consist of a molecule assembly. All the
activities of the cell, including the different cellular
structure formations, depend on the interaction of a
particular group of molecules. We distinguish three
important groups: water, the inorganic ions and the
organic molecules.
The AgentMol is principally defined by a set of
characteristics represented by the following
parameters:
numMol: number of the Molecule,
RC: Rate of Concentration of the actual quantity
of molecules in the environment. The RC may be
increased RCA or decreased RCD.
LimitMin, LimitMax, express a minimal and
maximal limit of resource or molecule
concentration, in the environment, that will not
be clear.
2.1.3 The Chemicals Mediators (CM)
A Chemical Mediator (CM) is a molecule that can
be attached to a cellular receptor. In our system a
CM is generated from a molecule.
An AgentCM has the same characteristics as an
AgentMol. It is described in our model by a set of
characteristics represented by the following
parameters:
State: describes the current state of the
AgentCM,
LCMR: Link between Chemical Mediator and
Receptor. This parameter takes the value 1 when
the link between AgentCM and receptor is
established and 0 in the contrary case,
CMInterpreted: this parameter takes the value 1
when the AgentCM is interpreted by the target
AgentCell and 0 in the contrary case,
CMA: CM Active or no.
USING MULTI-AGENT SYSTEMS TO STUDY PARACRINIENNE CELLS INTERACTION
277

Also, an AgentCM possesses certain behaviour
during the communication between two AgentCells.
At every behaviour the AgentCM passes from its
current state to another that can be the inactivate
state, the destroyed or the captured state.
From the biological real characteristics of the
chemical molecule (mediator) the automaton
concerning its behaviour is realized (Figure 2).
Secretes the chemical
messenger
Ties to one
AgentCell targets
Targets AgentCell
interpreter the messenger
Targets AgentCell
answers by an
apoptose
Targets AgentCell answers
by a proliferation or a
differentiation or a survival
Signalling AgentCell
AgentCM
Created & Activated
AgentCM
Interpreted
AgentMC
Inactived
AgentCM
Captured
AgentCM
Destroyed

Figure 2: Behaviour Automaton of a communicate
mediator.
2.1.4 The Objects
In addition to these agents, we find other very
important entities in the simulated system: the
environment and the receptors.
The environment is represented by a particular
class Environment, and evolves dynamically to
every change of the cellular and the mediator
state. It contains a cell population, represented
by a vector Vcell, a population of molecules
and resources for the cellular survival (sugar,
k+...) represented by another vector Vmol,
The receptors are protein molecules situated on
the membrane or in the cell. The receptor is
principally defined by a set of characteristics
represented by the following parameters:
TR: Type of Receptor,
SR: Stat of Receptor. This parameter takes the
value 1 where the junction between CM and
receptor is established and the value 0 in the
contrary case,
RA: Receptors are activated or not. This
parameter takes the value 1 where the
Receptor is activated, the value 0 in the
contrary case.
3 SIMULATION
Our model has been implemented in the DIMA
environment (Development and Implementation of
Multi-Agent system) (Guessoum, Meurisse, & Briot,
2002). DIMA groups classes that can be re-used
and/or adapted to easily construct agents.
The simulator, witch is the scheduler can be
activated, suspended, resumed or stopped.
In our multi-agent simulator the paracrinienne
communication is realized by an interaction between
the AgentCells and an interaction between the
AgentCell and AgentMC.






Figure 3: Simulation of the communication.
After a time T, the communicating AgentCell
executes the following signalling stages:
If the AgentCell is in the sending state (it is a
signalling AgentCell), so it secretes (activates) a
messenger in the extracellular environment. This
messenger is an AgentCM sent from the
signalling AgentCell to its nearer AgentCell,
(Figure 3 (a&b)).
To identify the neighbour AgentCell that has the
smallest distance, target AgentCell, we applied
the equation used in (Dib, 2008).
Once the target AgentCell and AgentCM are
identified the communication, the attraction
between specific receptors of targets AgentCell
and AgentCM, will be achieved as follows:
the AgentCM looks for identifying a specific
receptor to this mediator in its membrane. In
an affirmative case, it sends an attraction
a)- Send of the mediator at the instant t0.
Target
AgentCell
Secreted
Mediator
Signalling
AgentCell
b)- Send of the mediator at the instant t1.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
278

signal to the AgentCM but in the contrary case
it ignores it.
When the research is positive, the AgentCM-
Receptor link will be established, and the
AgentCM passes from its active state to the
captured state (Figure 4). In the contrary case
the AgentCM will seek among other
AgentCells neighbors the opportunity to
communicate. If no neighbor has a specific
receptor permitting the establishing of the
AgentCM-Receptor link so this AgentCM will
be ignored by any neighbor and it will passes
from its current status (active) to the new one
"destroyed".



Figure 4: Simulation of the attraction of the chemical
mediator by the target cell.
Once the attraction between target AgentCell
and AgentCM is realized then the target
AgentCell responds to this action by treating
the received signal and generating an
appropriate response;
Once the cellular response is generated so the
target AgentCell passes to a new state
corresponding to the produced response
(Figure 5);
The AgentCell finish the communication by the
inactivation of the AgentCM (mediator). This

Figure 5: Simulation of the communication: Target
AgentCell responses to this message by a proliferation and
the system generate a daughter AgentCell.
stage must be fast to permit other mediators to
express themselves. An AgentCell can react
simultaneously to origin signals. The
evolution of the simulation is illustrated by the
following figures.

Figure 6: Simulation of a new communication: Send of the
mediator.

Figure 7: Simulation of a new communication: Target
AgentCell responses to this message by a proliferation and
the system generate a daughter AgentCell.

Figure 8: Simulation of communication between two pairs
of AgentCell.
Cellular Answer Generated by the Signalling
AgentCell. The link of the AgentCM to the target
AgentCells receptor provokes an appropriate
cellular response that can be the cellular survival, the
cellular proliferation, the cellular differentiation or
the cellular death (Figure 9).
Attraction of the mediator by
the target AgentCells receptor
USING MULTI-AGENT SYSTEMS TO STUDY PARACRINIENNE CELLS INTERACTION
279

AgentCell signal ant
Signaling AgentCel l
AgentMC
AgentCM
AgentCell cible
Target AgentCell
Proliferation
Proliferation
Differentiation
Differentiation
Apoptose
Apoptosis
survi val
survi val
InformativeMolecule
Cell
Cellular answer

Figure 9: Interaction paracrinien: communication between
two AgentCells and target AgentCells response after its
interaction with the AgentCM
In our framework, this response, which is the
result of the interaction between two different types
of agent (AgentCell and AgentCM), moves the
AgentCell and the AgentCM from their current state
to another state appropriate for the generated
response.
4 CONCLUSIONS
In this article, we presented our system that we have
realized under the multi-agent platform DIMA. It
allows to model and to simulate the biological
cellular environment specifically the cellular
interaction process via chemical mediators
(paracrinien communication). This modelling is
achieved through interactions between the different
agents of the system.
The simulation achieved by our system reflects
the reality of the biologic nature. The objective of
our system is to be a virtual world of this cellular
biology helping its specialists to better understand,
to good to interpret and warn changes of cell states
according to its actual internal state and to the state
of its environment.
In this system, the cells (AgentCell)
communicate with each other to live, to control their
growth as well as for regulating their functions. The
cell is either normal (in an initial state), or signalling
cell (secrets mediator) and whether a target cell
(receipts mediator and generates an appropriate
response). At the same time, the chemical mediator
(AgentCM) is either active (identified and attracted
by the target AgentCells receptor) and whether an
ignored mediator (not identified by the target
AgentCell).
From our simulator, all these biologic
phenomenons are studied and simulated as well as
the evolution of a cellular population in the time is
calculated and is presented to the user by a sequence
of animated images.
REFERENCES
Dib, L. (2008). Multi-agent systems simulating the
physiological role of plasmic membrane. Elsevier
Journal CBM: Computers in Biology and Medicine.
Volume 38, Issue 6, 676-683.
Dib, L., & Guessoum, Z. (2007): Using Multi-agent
Systems to Study Cells Interaction. SWIN:The
Systemics and Informatics World Network. ITSSA
Journal International Transactions on Systems
Science and Applications. Vol 3, Num 3, 269278.
Dib, L., Guessoum, Z., Laskri, M.T., Fartas, H., &
Guettar, M. (2006). Systme multi-agent simulant
lendocytose et lexocytose. Troisime International
Workshop AMINA: Applications Mdicales de
lInformatique Nouvelles Approches. Facult de
Mdecine de Monastir Tunisie, Dc.
Dib, L., Guessoum, Z., Bonnet, N., & Laskri, M.T. (2005).
Multi-agent system simulating tumoral cells
migration. Published in LNCS, AI05: Advances in
Artificial Intelligence. (pp. 624 632). Australia.
Dib, L., Guessoum, Z. (2006). CellMigration system.
Published by IOS Press book: "Advances in
Intelligent IT: Active Media Technology. Fourth IEEE
International Conference on Active Multi-media
AML'06. (pp. 400 403).
Boss, T., J onker, C-M., & Treur, J . (2005). Modeling the
dynamics of intracellular processes as an organization
of multiple agents. In MAS- BIOMED 05 First
Intrnational Workshop on Multi-agent systems for
Medicine, Computational Biology and Bioinformatics.
(pp. 107-121).
Colloc. J . (2005). Un Systme multi-agent neuronale : vers
des systmes dinformation Epigntiques. SIM05. 5(4).
Giuliano, A., Denzinger, J ., Merelli, A., Miles, E.,
Tianfield, S., & Unland, H. (2005). MAS- BIOMED 05
First International Workshop on Multi-agent systems
for Medicine, Computational biology and
bioinformatics; Ultrecht, Hollande 2005
Guessoum, Z., Meurisse, T., & Briot, J . P. (2002).
Modular construction of agents and adaptive multi-
agent systems in DIMA. TSI, thematic number:
Environment of development of multi-agent systems.
Chauffert, B. (2004). Cours sur la biologie cellulaire
PCEM 1-PC K,
Masliah, J., & Housset, C. (2008). PCEM2- Biochemistry
- cellular Biology. Cellular communication and
physiopathology: example of the tumorigeness
Berridge, M. (1985). The molecular basis of
communication within the cell. 253(4):142-150,.
Smith, A. D. (1997). Oxford Dictionary of Biochemistry
and Molecular Biology. Oxford University Press,.
Smith, E. L., Hill, R. L., Lekowitz, I. R. J ., Handler, P., &
White, A. (1983). Principles of Biochemistry:
Mammalian Biochemistry, 6
th
ed. McGraw-Hill.
Chapters 11 through 20 describe in detail the
Biochemistry of the endocrine systems.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
280
A RF TRANSCEIVER FOR WIRELESS MONITORING
SYSTEMS OF THE VERTEBRAL COLUMN BEHAVIOUR
J . P. Carmo and J . H. Correia
University of Minho, Dept. Industrial Electronics, Campus Azurem, 4800-058 Guimaraes, Portugal
[email protected]
Keywords: Wireless acquisition system, RF transceiver, Spine deceases, e-health systems.
Abstract: This paper presents a radio-frequency (RF) transceiver designed, using a standard 0.18 m CMOS process,
for operation in the 2.4 GHz ISM band. The receiver has a sensivity of -64 dBm and consumes 3.5 mW
from a 1.5 V supply. The RF transmitter delivers an output power of +11 dBm (15 mW) with a power
consumption of 45 mW. These features make the RF transceiver suitable to be integrated in microsystems,
where the low-power is a major requirement. The application of these microsystems is to monitoring the
influence of heavy loads, in the behaviour of the vertebral column.
1 INTRODUCTION
The human posture has been an object of studies in
biomechanics, once some deviations of structural
and functional positions induce an unbalanced body.
These deviations usually, affects the vertebral
column and are caused by physical efforts, bad
postures in work, deficiency in sustentation muscles,
infections and congenital causes. The main
pathologies of vertebral column caused by the
referred deviations, are the scoliosis and lordosis.
Sometimes these pathologies appears in children
when they carry the heavy backpacks on the backs,
in this case, its very important monitoring the
influence of loads (backpack weight) in vertebral
column behavior. Figure 1 shows an adolescent
female with scoliosis in the vertebral column. Her
rib proeminence is most obvious upon her bending
forward. The radiograph demonstrates a right
thoracic scoliosis. The study of influence of
backpack weighs on the vertebral column of children
is an important issue, that has been worked by many
researchers for years (Palastranga et al, 2002). The
application cited in this work uses the following
approach: applying indirect information, using the
electrical potential generated by the muscles, when
they contract and when they are rest. The technique
which could measure this electrical potential is the
electromyography (EMG). Based on the results of
EMG, combined with the movements of the body,
measured by the accelerometers, it is possible to
know by numerical simulation, the displacement
occurred on the insertions points between the
muscles and the vertebral column (Pato et at, 2007).
Using these values of displacements in a finite
element code, like ANSYS, it is possible to compute
the value of stress field in the vertebral column,
especially it is possible to observe where are the
points more affected and the respective stress value.

Figure 1: An adolescent female with scoliosis.
The radio-frequency (RF) transceiver proposed in
this paper, was designed using a standard 0.18 m
CMOS process. This process allows to have the
power supply of 1.5 V. The proposed low-
power/low-voltage transceiver, is intended for use in
wireless sensor networks, more specifically, for the
monitoring the influence of heavy loads, in the
behavior of the vertebral column.
281

~
Local oscillator (LO)
@ 2.4/2.5 GHz
Envelope
detector
IF
(100 MHz)
Filter
Switch
RXD
TXD
LNA
PA
A
2.5 GHz
2.4 GHz

Figure 2: The block schematic of the RF transceiver.
2 TRANSCEIVER'S DESIGN
The transceiver has a receiver, a transmitter, an
antenna-switch and a phase-locked loop (PLL) as
frequency synthesiser. The Figure 2 shows the
architecture of the proposed transceiver, where the
reception is made by means of direct demodulation,
using the technique of heterodyne detection. The
final demodulation step is made with the use of a
envelope detector, applied for an intermediate
frequency of 100 MHz.
The quality requirement for the proposed RF
transceiver is a transmission with a bit error
probability less than 10
-6
(one error for each one
million bits transmitted) with a sensitivity of
-64 dBm, in a transmitted power of +11 dBm using
Amplitude Shift Keying (ASK)
modulation (Carlson et al, 2002). All of these
specifications are useful to make this transceiver
suitable for short range applications (e.g., between
fifty five and sixty meters - #55 / #60 meters), and
obviously, to the target biomedical application,
which will be further explained in the section 4.
2.1 Receiver
The Figure 3 shows the receivers front-end
schematic. This circuit has a low-noise amplifier
(LNA) that provides a 50 input impedance, using
a tuned load to provide high selectivity. The
amplified RF signal is directly converted to an
intermediate frequency (IF) with a mixer, followed
by a low-pass filtering and a post-amplifier. The
final downconversion to the base-band is made with
envelope detection.
The low-noise amplifier (LNA) is the first gain
stage in the receiver path. In a LNA, the signal must
be amplified as much as possible, keeping the
signal-to-noise ratio (SNR) as low as possible. This
is achieved with the best noise figure (NF). The
LNA is an inductively degenerated common source
amplifier. The cascoding transistor M
2
is used to
increase the gain, to better isolate the output from
input and to reduce the effect of M
1
s C
gs

(Yao et al, 2007). The LNA enters in the sleeping
mode, when the current in the polarization stage is
switched off. The same principle applies to the all
subsystems.

LO
Mixer circuit
Bias4
LO
L
s
L
g
C
b
Bias1
Input
M
1
M
2
LNA
= 50
in
Z
L
d
C
b
Bias3
C
f
R
f
IF filter
Bias2
Bias5
M
1 M
2a
M
2b
M
2c M
2d
M
3a
M
3b
M
3c M
3d C
b
C
b
C
b
C
b
R
p
R
p R
p
R
p
Post-amplifier
Bias6
Output
stream
Output buffer
C
1
C
2
C
3
R
2
R
1
Envelope
detector
M
3
M
4
M
5

Figure 3: The schematic of the receiver.
M
1
M
2
L
d1
M
1
L
d2
Class E
PA filter
A
n
t
e
n
n
a
TDX
LO
Class E Power amplifier PA driver

Figure 4: The schematic of the transmitter.
The downconversion to the IF uses a four-quadrant
multiplier Gilbert cell as a mixer. This mixer
performs well with local signals (LO) at 2.5 GHz
and a square shape. The main advantage is the
possibility to use ring-oscillators, rather than tuned
LC, thus, a big on-chip area saving is possible to be
achieved, without a severe degradation in the
phase-noise. Moreover, this don't poses special
cares, because the down-conversion from IF to the
base-base, is made by way of an envelope detection,
whose input signal comes from the output of a low-
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
282

pass filter detector. The IF frequency is produced
from the 2.4 GHz RF and from the 2.5 GHz local
frequency. A minimum IF level at the envelope
detector defines the receivers sensitivity. This
envelope detector is of active type, e.g., it provides
gain, compared with the conventional topology
using a simple clamping circuit. Thus, the sensitivity
of the receiver is bigger, compared when a passive
detector is used.
2.2 Transmitter
Figure 4 shows the schematic of the transmitter. The
upconversion to the pass-band at 2.4 GHz, is made
by a cascade circuit, comprising transistors M
1
and
M
2
. This circuit combines the transmitted
bitstream (TXD) with local carrier generated in the
frequency synthesizer described further. This circuit
produces an AKS digital signal compatible with the
usage of a switched power amplifier of class E
(Sokal et al, 1975). The external filter that follows
the power amplifier is a typical class E network, and
removes the spectral components around the
2.4-GHz carrier frequency.
2.3 Frequency Synthesiser
The local generation of the frequency synthesiser is
a phase-locked loop (PLL) with integer division in
the feedback path. The Figure 5 shows the block
diagram of the PLL, which has a reference generator
circuit with a crystal based oscillator at 20 MHz,
followed by a phase-frequency difference
circuit (PFD) without dead zone, a current steering
charge pump (CP) and a third order passive filter.
The passive section output is connected to the VCO,
which generates the desired frequencies of 2.4 GHz
or 2.5 GHz. These frequencies must be divided by
120 or by 125 and connected to the PFD again,
closing the loop. For the TSPC logic, it is required a
rail-to-rail input to work properly. At these
frequencies, the power consumptions are lowest,
when compared with the SCL logic
(Pellerano et al, 2004). In real PFDs there is an
offset around the zero phase difference, and a gain
inversion region takes place for phase differences
higher than 2- rad. In this gain inversion region,
the PFD outputs the wrong control signals increasing
the phase and frequency differences between the
inputs, and the lock time takes a sudden turn for the
worse (Lee et al, 2003). The PFD has a linear gain in
the range [-, +] and a constant gain in the remain
interval. This constant gains increases the bandwidth
of the PLL, making it faster to lock, compared with
those containing other structures (Kim et al, 2005).

20 MHz XTAL
reference
generator
PFD
VCO
F
ref
F
div
Up
Down
I
cp
I
cp
C1 C2
R2
C3
R3 V
control
High-frequency buffers
F
out
2
Static logic
30/32 2
TSPC
a)
Up
Down
Fdiv
Fref
PFD
IUp
IDown
Down
Up
CP output
Control
V
Charge-pump
b)
Control
V
Bias
VCO
Bias
Bias
Tuning
V
VCO circuit
Control
V
c)
Figure 5: a) The PLL structure, b) the schematic of the
PFD-CP, and c) the schematic of the VCO.
The charge pump (CP) is of current steering type
(Chih-Ming et al, 2002). This circuit avoids the
problems in conventional CPs. In spite of being
switched, the currents are routed from the load to an
alternative path, and from that path to the load. To
finish, the voltage controlled oscillator (VCO) is of
ring type, in order to save on-chip area and because
the phase and frequency errors are not critical in the
used modulation.
3 EXPERIMENTAL RESULTS
For the receiver, simulations shown a sensivity of
-64 dBm and a consumption of 3.5 mW from a 1.5 V
A RF TRANSCEIVER FOR WIRELESS MONITORING SYSTEMS OF THE VERTEBRAL COLUMN BEHAVIOUR
283

supply. The RF transmitter delivers an output power
of +11 dBm (15 mW) with a power consumption of
45 mW. The receiver has a total power consumption
of 3.5 mW for the receiver (1.5 mW for the LNA,
0.6 mW for the down-conversion mixer and 1.5 mW
to the post-amplifier and envelope detector). The
transmitter has the power consumptions in the
following blocks: 2.7 mW in the driver and
41.5 mW in the power amplifier.
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.0 2.8
Frequency [GHz]
f
s
(
V
a
n
t
e
n
n
a
)

[
V
]
m3
0.5
1.0
0.0
1.5
m3
f
MHz
= 2.4 GHz
fs(V
antenna
) = 1.229 e
+j59.9
[V]

Figure 6: Amplitude spectrum of the amplified signal at
the antenna terminals.
The Figure 6 shows the amplitude spectrum of the
amplified signal at the terminals of an antenna, with
an input impedance of 50 .
4 APPLICATION
The application for microsystems using the
transceiver proposed in this paper, is to make
wireless acquisition modules, in order to allow the
monitoring of heavy loads influence on vertebral
columns behaviour. Each module makes the
electromyography (EMG), to measure the electric
potentials on the iliocostalis and longissimus
thoracis muscles, and use a dual-axis accelerometer
to get the movements of the body, in order to obtain
the complete behaviour of the vertebral column. The
acquired information is to be transmitted with the
maximum rate of 250 kbps, however, the
simulations shown, that the baud-rate can be
extended without jitter problems, for frequencies up
to 10 Mbps. An analog channel with differential
input connected to the electrodes, is required to
measure the EMG signal, while the remained
channel is to measurements of the patients
movements.
The analysis of the EMG signal must be made in the
amplitude domain, thus, before proceeding to the
ADC conversion, it is required a peak detection of
the amplified EMG signal, followed by an
integration (Robertson et al, 2004). This mandatory
process eliminates the fluctuations that characterize
the EMG signals.
The measurements of the motion and the
positioning of the patients body is made with the
use of a commercial dual-axis accelerometer of
MEMS type. This chip connects to a commercial
microController by way of an integrated Serial Port
Interface (SPI).
The Figure 7 shows the block diagram of a first
possible prototype, which contain the sensor
interface read-out, the electronics for data
acquisition (amplifications and analog-to-digital
conversion) a micro-controller, and the proposed RF
transceiver. A coin-sized 1.5 V battery will provide
the supply and a commercial DC/DC step-up
converter makes possible to supply the remaining
components of the prototype with different voltage
levels. As it can be seen in this Figure, the anti-static
(ESD) protections are provided by way of power-rail
connected diodes (Ker et al, 2005).
FromEMG
electrode
ADC
RF transceiver microControler
From
antenna
Interface
(ESD protections)
A
1
.
5

V
>
1
.
5

V
DC/DC
step-up
2-axis
acelerometer
SPI
interface
Out-supply

Figure 7: Block diagram of the microsystems for use in
wireless modules.
5 CONCLUSIONS
This paper presented a low-power/low-voltage
radio-frequency (RF) transceiver at 2.4 GHz, for
working with a single 1.5 V coin-sized battery.
Simulations shown for this transceiver a
consumption of 3.5 mW in the receive mode and
15 mW of transmitted power, with a power
consumption of 45 mW in the transmitting mode.
These characteristics fulfil the requirements for
short-range communications.
The target application for this transceiver is in
wireless acquisition modules for monitoring of
heavy loads influence on vertebral columns
behaviour, in order to understand the influence of
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
284

heavy loads as a risk factors in the vertebral column,
such as the scoliosis and lordosis. These factors
normally associated to appears in children when they
carry the heavy backpacks on the backs. Thus, its of
extremely importance to characterise the influence
of heavy loads (backpack weight) in the vertebral
column behaviour. This solution fits the medical
doctors requirements for an easy placement and
removal of wireless modules. The main advantage of
this solution, is the maintenance of the mobility and
lifestyle of patients during the diagnosis.
The possibility to control the receiver and the
transmitter subsystems, allowing them to be
switched on and off is another advantage of this RF
transceiver. This is a specially important topic of
design, in applications where power efficient
algorithms are mandatory, e.g., in wireless sensors
networks (Enz et al, 2004).
REFERENCES
Carlson, B., et al, 2002, Communication systems: An
introduction to signals and noise in electrical
communications, 4th edition, McGraw-Hill, 2002.
Chih-Ming, H., et al, 2002, A fully integrated 1.5 V
5.5-GHz CMOS phase-locked loop, IEEE J ournal of
Solid-State Circuits, Vol. 37, No. 4, pp. 521-525.
Enz, C., et al, 2004, WiseNET: An ultralow-power
wireless sensor network solution, IEEE Computer,
Vol. 37, Nr. 8, pp. 62-70.
Ker, M., et al., 2005, ESD implantations for on-chip ESD
protection with layout consideration in 0.18-m
salicided CMOS technology, IEEE Transactions on
Semiconductor Manufacturing, Vol. 18, Nr. 2,
pp. 328-337.
Kim, B., et al, 1005, A 250-MHz2-GHz wide-range
delay-locked loop, IEEE J ournal of Solid-State
Circuits, Vol. 40, Nr. 6, pp. 1310-1321.
Lee, K., et al, 2003, Phase-frequency detectors for fast
frequency acquisition in zero-dead-zone CPPLLs for
mobile communication systems, Proc. of the 29th
ESSCIRC, 16-18, Estoril, Portugal.
Palastanga, N., et al, 2002, Anatomy and human
movement, 4th edition, Butterworth Heinemann,
pp. 445-537.
Pato, M., et al, 2007, A finit element model for squeletical
muscles, Proc. of the CMNE/CILAMCE, 2007.
Pellerano, S., et al, 2004, A 13.5 mW 5-GHz frequency
sinthesizer with dynamic logic frequency divider,
IEEE J ournal of Solid-State Circuits, Vol. 39, Nr. 2,
pp. 378-383.
Robertson, D., et al, 2004, Research methods in
biomechanics, Human Kinetics.
Sokal, N., et al, 1975, Class E-A new class of
high-efficiency tuned single-ended switching power
amplifiers, IEEE J ournal of Solid-State Circuits,
Vol. 10, Nr. 3, pp. 168-176.
Yao, T., et al, 2007, Algorithmic design of CMOS LNAs
and PAs for 60-GHz radio, IEEE J ournal of
Solid-State Circuits, Vol. 42, No. 5, pp. 1044-1057.
A RF TRANSCEIVER FOR WIRELESS MONITORING SYSTEMS OF THE VERTEBRAL COLUMN BEHAVIOUR
285
A MULTI-LAYERED MICROFLUIDIC DEVICE
FOR MAGNETOPHORETIC CELL SEPARATION
Hye-Lyn Lee, Suk-Heung Song, Hee-Taek Lim, Hyung-J oon Kim, Min-Suk Park and Hyo-Il J ung
School of Mechanical Engineering, Yonsei University, Seoul, Republic of Korea
{hyelyn, shsong}@yonsei.ac.kr, [email protected], [email protected]
[email protected], [email protected]
Keywords: Multi-layered microfluidic channel, Magnetophoretic cell separation, Microelectromagnet, Microbeads,
Magnetic field.
Abstract: In this paper, we present the design and experimental results of a multi-layered microfluidic electromagnetic
cell separation device. Our channel consists of top and bottom layers in order to separate magnetically
labeled cells in the vertical direction. Rapid separation of magnetic beads in top and bottom channel can be
used in high throughput screening to monitor the efficacy and drug compounds. The experiments using the
device were carried out with 4.5m magnetic bead and magnetic labeled J urkat cell under electromagnetic
field of 1.55mT. Without the magnetic field, the magnetic labeled cells started to flow from the bottom inlet
and exit out of the bottom channel outlet. In the presence of the magnetic field, the cells started in bottom
channel are attracted upward by the electromagnetic field and flow through the top layered. Finally, the
labeled cells flow out the top channel outlet. The separation efficiencies of the multi-layer structured
microfluidic channel showed more than 95%. We found that the multi-layer structured microfluidic channel
was very effective in enhancing the separation. This microfluidic channel can be potentially applied to Lab-
on-a-chip system because of its attractive features such as high throughput, continuous sorting, simple and
rapid fabrication.
1 INTRODUCTION
There is a growing interest of microfluidic cell
separation systems as they are useful for high
throughput drug screening and medical diagnosis
(Inglis at al. 2004) (Pamme and Wilhelm, 2006).
The Fluorescent Activated Cell Sorter (FACS) is one
of the most common methods to detect and separate
cells but the bench-top volume of the device is a
barrier to its miniaturization. A microfluidic device
is a proven way of minimization and there have been
several reports regarding the separation of specific
cells in an optical microfluidic system (Wolbers et al.
2004). However, optical systems have the
disadvantage of requiring external observation and
are not suitable for opaque samples like blood.
A Magnetic Activated Cell Sorter (MACS) can
overcome all the defects of cell sorters. It is simple
to operate and is generally not affected by the
electrical properties of a solution, pH, temperature or
impurities (Pamme 2006). Separation of human
peripheral T lymphocytes has been reported using
MACS with permanent magnets and quadropole
fields (Sun et al. 1998). Microfluidic MACS for
HeLa cell and macrophage sorting have recently
been developed (Pamme and Wilhelm 2006).
It is a generally, thought that conventional
magnetophoretic separating devices produce small
magnetic fields in the microfluidic channel. Also,
most use mono-layered channel that separate in the
horizontal direction (Kim at al. 2007).
In this research, we demonstrate a new
microfluidic channel consisted of top and bottom
layers in order to separate in the vertical direction.
This device can get easily high degree of separation
efficiency although in the small magnetic fields.
2 MATERIALS AND METHOD
2.1 Theory
There are three forces acting on a magnetically
labeled cell surface: magnetic force, drag and
gravity as shown in Fig.1.
Because the effect of gravity is negligible owing
286



Figure 1: Three forces acting on the surface of a cell.
to the small size of the magnetic bead labeled cell
(Qasem et al. 2004), the forces responsible for
deflection of cells coated with magnetic beads are
magnetic and drag forces. The magnetic force
exerted on a magnetic bead in magnetic field can be
calculated as follows (Williams et al. 1999):


Hydrodynamic drag force is estimated according to
the Stokes drag equation:

Finally, considering the force balance between
magnetic and hydrodynamic drag (neglecting the
inertial term of Newtons law), the deflection of the
cell is given by (Qasem et al. 2004):

In Eq. (3), V
m
is a volume of magnetic beads.
was calculated from the literature (Lagae et al. 2005).
Finally, as determined by measuring magnetic
fields with a gauss meter (LakeShore 475 DSP) and
calculating based on the inverse square law. The
deflection of the cell, z, generated by one magnetic
bead attached on the cell surface was 4.5 m. In
most cases, the number of magnetic beads bound
with cell was more than 1, so the vertical deflection
should be more than 17.5 m. We fabricated a
microfluidic device to separate cells assuming that
z is more than 17.5 m.
2.2 Fabrication
2.2.1 Electromagnet
The microelectromagnet was fabricated using
MEMS technology. The fabrication process for on-
chip microelectromagnet is in Fig. 2: (a) SiO2 was
deposited 1m on a double-side-polished Si wafer
by furnace and the microconductor was patterned
using UV-lithography; (b) a copper microcoil, as a
conductor for the microelectromagnet, was
manufactured by 25m thick electroplating with a
photoresistor mold. For the electrical insulation, the
dielectric layer based on polymer material (AZ 4620,
Clariant, Korea) was deposited between the
microcoil and the magnetic plate; (c) The polymer,
as a dielectric layer, was encapsulated on the
microcoil and was hard-baked; (d) The seed layer,
Ti/Ni 500/3000, was deposited onto the dielectric
layer for electroplating of the nickel plate; (e) the
nickel, as a magnetic plate, was electroplated 25m
thick and (f) the PDMS microfluidic channel system
was integrated.
The size of the electromagnet is 4 x 4 mm
2
and
the height of the magnetic plate is 25 m.


Figure 2: Fabrication processes of the microelectromagnet
and microfluidic system.
2.2.2 Microfluidic Channel
A microfluidic channel was fabricated according to
standard softlithography and replica molding process.
The silicon wafer was washed first by methanol,
followed by acetone, and de-ionized water, then a
SU-8 negative epoxy-based photoresist (SU-8 2100,
MicroChem Corp.) was spin-coated on the wafer.
The spin-coated wafer was baked using a hot plate
(95C, 35 min) to remove unwanted area from the
photoresist. The wafer was then exposed to UV light
( =365 nm, 60 s), baked again in two steps (65C,
1 min and 95C, 15 min), and developed by the SU-
8 developer (Sigma Aldrich) for 15 min. The result
was a 130 m high photoresist mold. After
A MULTI-LAYERED MICROFLUIDIC DEVICE FOR MAGNETOPHORETIC CELL SEPARATION
287

preparing the SU-8 mould, a PDMS gel mixture (DC
184-A:B =9:1, Dow Corning) was poured on the
wafer, the gel mixture was baked in an oven (80C,
45 min) and detached from the mold. The PDMS
microfuidic channel was finally treated with O2
plasma and bonded with a glass substrate. The size
of this device is 25mm x 14mm x 5mm and the
microfluidic channel length is 17mm, the width is
150m and the depth is 100m.


Figure 3: Illustration of the multi-layered microfluidic
channel with on-chip electromagnet.
2.3 Cell Culture
J urkat clone E6-1 cells (Organ: acute T cell
leukemia/ human blood) were cultured under
standard conditions (37C, 5%CO2) in RPMI-1640
medium comprised of 10% (v/v) fetal bovine serum
(FBS) and 1% antimycotic antibiotic (all purchased
from Cambrex, USA).
2.4 Sample Preparation
Dynabeads

CD3 are superparamagnetic, micro

sized particles with a characteristic polymer surface
for coupling with CD3 T-Cells, thus making it
possible to sort out the human T cells in this
experiment. Dynabeads

CD3 (4x10
8
bead/ml, 25 l)
and J urkat clone E6-1 cells (1x10
7
cell/ml, 1ml)
were incubated for 10 min at 2-8C with gentle
tilting and rotation. Finally, J urkat cells (13 m
diameter) are bound with magnetic beads CD3 (4.5
m diameter).
2.5 Experimental Set-up
A schematic view of the multi-layered microfluidic
channel using on-chip electromagnet is illustrated in
figure 4. The multi-layered microfluidic channel is a

Figure 4: Schematic diagram of the multi-layered
microfluidic channel.
straight type with a square cross-section. The fluid is
assumed incompressible the flow will be laminar
and boundary condition is the no-slip on the channel
walls. The magnetic beads and cell mixture started
in bottom channel at flow rate of 5L/min are
attracted upward by the electromagnetic field and
flow through the top layered. Finally, magnetic
beads flow out the top channel outlet.
A syringe pump (KDS scientific, CMA
instruments) is connected to the microfluidic
channel to supply different types of fluid through
two syringes (sample solution syringe and buffer
solution syringe). The syringes and channel inlets
were linked by Teflon tubes (500 m, Nano Port).
The microfluidic channel was placed on an optical
microscope and monitored by a CCD camera. We
used a microelectromagnet of 1.55mT (Tesla) placed
on the top of the main channel.
3 RESULTS AND DISCUSSION
The sample solution was flowed through a
microfluidic channel and the J urkat cells labeled
magnetic beads were separated by a 1.55mT
microelectromagnet at 5L/min of flow rate.
Magnetic beads are introduced to magnetic fields
and then experience a magnetic field while flowing
in the microfluidic channel.
Figure 6 shows comparison of the efficiencies
without and with magnetic field by electromagnet.
In the absence of magnetic fields, the flow of cells
labeled magnetic beads is shown negligible in
outflow 1 of top channel with an efficiency of 2.8%.
However, in the presence of magnetic fields, the
magnetically labeled cells are dominant in outflow 1
with a high efficiency of 95.3%. It demonstrates
magnetic beads were deflected and separated
through outlet 1 when a magnetic field was
introduced.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
288



Figure 5: (a) Photographic image of microfluidic channel
(b) Photographic image of J urkat cells and magnetic beads
from outlet of the channel.

Figure 6: Experimental results of the separation
efficiencies.
4 CONCLUSIONS
A new multi-layered channel for separating cells has
been introduced using microelectromagnet. Our
experiments demonstrate that specific cells can be
separated simply using a multi-layered microfluidic
channel with high efficiencies.
The efficiency of the separation by our approach
was comparable with that of conventional
magnetophoretic cell sorters (Bu at al. 2008)
(Smistrup at al. 2005). Our results identify a new
multi-layered microfluidic channel to isolate cells
for drug discovery and Lab-on-a-chip system
because of its attractive features such as high
throughput, continuous sorting, simply and rapidly
fabricated system.
ACKNOWLEDGEMENTS
This work was supported by National Core Research
Center (NCRC) for Nanomedical Technology of the
Korea Science & Engineering Foundation (Grant no.
R15-2004-024-01001-0), Seoul Research &
Business Development (R&BD Program, 11128)
and Korea Research Foundation Grant funded by the
Korean Government (MOEHRD) (KRF-2007-313-
D00073).
REFERENCES
Inglis, D., Riehn, R., Austin, R., & Sturm, J . 2004.
Continuous microfluidic immunomagnetic cell
separation. Applied Physics Letters, 85, 5093.
Pamme, N., & Wilhelm, C. 2006. Continuous sorting of
magnetic cells via on-chip free-flow magnetophoresis.
[Article]. Lab on a Chip, 6(8), 974-980.
Wolbers F, Andersson H, van den Berg A, Vermes I. 2004.
Apoptosis induced kinetic changes in autofluorescence
of HL60 cells-possible application for single cell
analysis on chip. Apoptosis 9:749755
Pamme N. 2006. Magnetism and microfluidics. Lab Chip
6:2438
Sun L, Zborowski M, Moore LR, Chalmers J J . 1998.
Continuous, flow-through immunomagnetic cell
sorting in a quadrupole field. Cytometry 33:469475
Kim, H., Son, O., Kim, K., Kim, S., Maeng, S., & J ung, H.
2007. Separation of apoptotic cells using a
microfluidic device. Biotechnology Letters, 29(11),
1659-1663.
Qasem R, Victor S, Daniel P, Chen Y. 2004. On-chip
microelectromagnets for magnetic-based bio-
molecules separation. J Magnetism and Magnetic Mat
281:150172
Williams PS, Zborowski M, Chalmers J J . 1999. Flowrate
optimization for the quadrupole magnetic cell sorter. Anal
Chem 71:37993807
Lagae L, Wirix-Speetjens R, Liu CX, Laureyn W, Borghs
G, Harvey S, Galvin P, Ferreira HA, Graham DL,
Freitas PP, Clarke LA, Amaral MD. 2005. Magnetic
biosensors for genetic screening of cystic fibrosis.
IEEE Proc Circuit Device Syst 152:393400
Bu, M., Christensen, T. B., Smistrup, K., Wolff, A., &
Hansen, M. F. 2008. Characterization of a
microfluidic magnetic bead separator for high-
throughput applications. Sensors and Actuators, A:
Physical, 145-146(1-2), 430-436.
Smistrup, K., Hansen, O., Bruus, H., & Hansen, M. 2005.
Magnetic separation in microfluidic systems using
microfabricated electromagnets-experiments and
simulations. J ournal of Magnetism and Magnetic
Materials, 293(1), 597-604.
A MULTI-LAYERED MICROFLUIDIC DEVICE FOR MAGNETOPHORETIC CELL SEPARATION
289
PAIN AND EFFICIENCY IN NEONATAL BLOOD SAMPLE
SCREENINGS
New Devices for Reducing Pain and Improving Blood Sample Quality
Bruno Wacogne, Christian Pieralli
Institut FEMTO-ST, Dpartement d'Optique P.M. Duffieux, UFR Sciences et Techniques
Route de Gray, 25030 Besanon cedex, France
[email protected], [email protected]
Gonzalo Cabodevila, Nolwenn Baron
Institut FEMTO-ST, Dpartement MN2S, Avenue de l'Observatoire, 25030 Besanon cedex, France
[email protected], [email protected]
Sandrine Marioli
Service de Pdiatrie 1, CHU Saint Jacques, 2 place Saint Jacques, 25030 Besanon cedex, France
[email protected]
Lionel Pazart
Centre d'Investigation Clinique en Innovation Technologique,CHU Saint Jacques
2 place Saint Jacques, 25030 Besanon cedex, France
[email protected]
Keywords: Neonatal screening, pain evaluation, micro-needles arrays, image processing.
Abstract: Neonatal blood sample screening is recognised as a difficult gesture and painful to the newborns. The
number of detected diseases is still relatively low and depends on the country where it is performed. There
is a real need for new techniques that reduce pain, facilitate the blood sampling, increase the quantity of
sampled blood and improve the collection of blood of the cardboard blotter actually used. In this paper, we
present systems that are currently developed in Besanon (France) in collaboration between the FEMTO-ST
Institute and the University Hospital. They mainly concern micro-needles arrays and pressure free blood
sampling devices. The choice of these systems has been dictated by a study of the pain that newborns feel
during the gesture. The ulterior motive of this work is to improve neonatal blood sample screenings and
therefore, to increase the number of screened diseases and try to generalise this technique to places where it
is not yet done.
1 INTRODUCTION
The blood sample screening of several congenital
diseases (phenylketonuria, hypothyroidism, adrenal
hyperplasia, ...) is performed routinely at birth in
many countries. The number of detected diseases
depends on the screening policy of the country and
the technical limitations of the methods. For
neonates, techniques are mainly limited by the
difficulties of realisation of the sampling gesture, the
small volume of punctionable blood and the pain
caused by the gesture. Currently, the French
Association for the Detection and Prevention of
Handicaps for Children (AFDPHE), responsible for
organising the screening in France, favors (at the
third day of life) a collection of capillary blood after
a bite by a retractable lancet at the heel of the
newborn. This gesture is followed by successive
pressures on the heel in order to collect blood
droplets on a cardboard blotter. This gesture is
recognised as painful (Facchini, 2005), (Owens,
1984). It should be noted that in some places,
screenings are performed with venous blood instead
of capillary blood. In this case, the gesture is more
technical and it presents risks for both the nurse and
the newborn. In this paper, we restrict the discussion
to the case of capillary blood.
290

The orientation of the present works is intended
to develop medical devices less painful and that
offer a higher potential screening of a larger number
of diseases in newborns, replacing technology
currently used.
The context of the study, together with the
evaluation of pain during the blood collection
gesture is presented in section 2. Section 3 deals
with the study of a painless micro-needles array
meant to replace the classical lancet. In section 3, we
present a device developed to obtain blood without
successive pressures on the newborn's heel.
Furthermore, this system can be used to efficiently
deposit the collected blood on the cardboard blotter.
Then, a conclusion and some perspectives will be
proposed at the end of this manuscript.
2 CONTEXT AND PAIN
EVALUATION
Screenings are achieved by nurses on the third day
of life. They require the use of an automatic, sterile
and disposable lancet that pierces the skin of the
heel. The heel is then pressed to obtain droplets of
blood that are collected on circles drawn on a
cardboard blotter (as seen on figure 1). Since the
blood flow is not sufficient, successive manual
pressures to the heel are required. Circles on the card
must be completely filled and the blood must cross
the card (this is not the case for all the circles in
figure 1). If the first attempt does not give enough
blood, a second is made on the same heel or on the
other one. In the same way, if the droplets do not
cross the screening card it is necessary to add blood.
This tracking is recognised as painful.
Newborns feel more or less pain during the
blood sampling gesture. Pain can be estimated using
various behavioural scales (Destuynder, 1991),
(Uyan, 2008). Both D.A.N. (Douleur Aigu du
Nouveau-N, Newborn Intense Pain ) and E.D.I.N.
(Echelle de Douleur et dInconfort du Nouveau-N,
Newborn Pain and Discomfort Scale) scales are
adapted to the evaluation of newborn's pain. We
used the D.A.N. scale because it is more specific to
intense pain. It allows quoting three criteria: the
facial answer, the movements of the members and
the vocal answer. Each criterion is quoted either
between 0 and 3 or between 0 and 4. We evaluated
the pain at five different moments: before the
gesture, when the nurse takes the baby's heel, during
the lancet sting, when the nurse presses the heel and
after the gesture.
For a first analysis, we observed 55 children for
which 58 stings were needed. Indeed, no blood was
obtained in 3 cases at the first attempt.


Figure 1: Actual blood sampling technique (top) and
cardboard blotter used to collect blood.
The gesture proving to be painful for a majority
of children, we refined our study at three moments:
the taking of the heel, the sting, the pressing of the
heel. The results obtained showed that the taking of
the heel is not painful (only 1 case). The sting is
painful in 10 cases, what nearly represents 25 % of
the cases. The pressing of the heel is the most
painful moment (26 cases representing 68%).
Therefore, we oriented our works toward two
new systems of blood sampling. The first one is a
micro-needles array meant to reduce the pain during
the sting (in replacement of the lancet). The second
one is used after the heel has been stung with the
lancet. It is designed to collect enough blood without
any pressure on the heel. Moreover, this system
leads to a perfect impregnation of the cardboard
blotter.
3 MICRO-NEEDLES ARRAY
The goal of this work is to estimate the geometry a
micro-needles array used to collect blood at the
newborn's heel. This array should replace the lancet.
The goal is to penetrate the newborn's heel and to let
blood flow in the holes of the micro-needles by
capillarity. In a final version, a tank should be
designed to store the sampled blood. Various studies
concerning micro-needles have been published, but
PAIN AND EFFICIENCY IN NEONATAL BLOOD SAMPLE SCREENINGS - New Devices for Reducing Pain and
Improving Blood Sample Quality
291

very little on blood collection. For example, we can
mention the work presented by (Mukerjee, 2004). In
this case, a liquid is obtained after a relatively long
time but this liquid does not contain only blood.
Other systems, inspired by mosquitos morphology,
are presented but some complementary studies are
still required (Oka, 2001), (Sharaf, 2003). Figure 2
shows an example of a micro-needles array
fabricated in our laboratory. This particular
architecture was designed for drug delivery and not
for blood sampling.


Figure 2: Example of micro-needles array fabricated in our
laboratory.
For our specific application, the characteristics
of the micro-needles have to be rethought according
to the newborn's heel anatomy. The characteristics
we studied are the shape, the dimensions and the
number of micro-needles in a array. This work is
based on a capillaroscopy study of the newborn's
heels that helped to define the depth of the children's
capillary network (350 m) and its density (70
capillaries by mm
2
). In what follows, we present the
conclusion drawn from our studies. The details of
numerical simulations can be obtained on demand.
Some parameters used in this study come from
the work conducted for several years in our
laboratory. Micro-needles are fabricated with
silicon. The silicon surface is oxidized in order to
make the device bio-compatible. The needles are
cylindrical. In order to improve their strength the
basis of the needles is conical. The tip of the needles
is bevelled in order to facilitate the penetration of the
stratum corneum. Finally, the needles exhibit a
longitudinal slit. It is used to maximize the contact
surface between the capillaries and the hole of the
needle. Figure 3 shows a schematic diagram of one
micro-needle.

Figure 3: Schematic representation of a micro-needle
dedicated for newborns blood sampling.
3.1 Dimensions of the Needles
The height L of the needles is fixed to 1200 m to
ensure a penetration until the dermis of skin. The
inner and outer diameters (D
i
and D
o
) are 60 and 150
m respectively. These diameters increase the
probability to meet a capillary.
The bevel exhibits an angle = 41. This value
is calculated considering that a third of the needle
penetrates the skin (400 m) and that the hole of the
needle must be at the level of the capillaries (350
m).
The slit must not extend outside the area where
the capillaries are located. In this way, we reduce the
risk to collect something else than blood. Its length
is therefore limited to L
s
=200 m. The width of the
slit doesn't have an influence on the needle strength.
Numerical simulations showed that the Von Mises
constraints are almost not influenced by this
parameter. For the moment, we consider a slit width
of 60 m.
The basis of the needle is conical. This cone is
defined by its height H and its angular aperture .
Numerical simulations showed that the Von Mises
constraints are minimized for a height H =600 m
and an angular aperture <30.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
292

3.2 Probability to Collect Blood with
the Needles
To summarize, the micro-needles are cylindrical
with a conical basis. They exhibit an opening used to
collect blood. This opening consists of the hole of
the needle and the longitudinal slit. When the micro-
needle has penetrated the skin, two surfaces can be
defined: the total surface of the needle in skin and
the surface of the opening that is susceptible to be in
contact with the capillaries. The ratio between these
two surfaces allows estimating the probability to
collect blood.
Studies showed that an array of 8 needles gives a
probability to collect blood equal to 67%. It is
necessary to increase the number of needles up to 24
to reach a probability of 96%. We consider that the
distance between two needles must be at least equal
to 1 mm. Below this size, the needles may not
penetrate the skin (fakir effect). Therefore, a 24
needles array is about 4*6 =24 millimetre square,
which is quite large. Another possibility is to use a 8
needles array (8 millimetre square) three times on
three different parts of the heel.
3.3 Perspectives
To conclude this section, we have to mention that
we are actually working on capillarity studies that
should answer the question: how long will it take to
collect enough blood ? It is indeed possible that a
pumping system will be required in order to reduce
the gesture duration.
Another aspect that must be addressed concerns
the penetration of the needles into the skin.
Techniques developed in the frame of other projects
should be transposable to the specific anatomy of
newborn's heel.
4 PRESSURE FREE
COLLECTION SYSTEM
We recall that most of the children feel pain when
their heel is pressed. The pressure on the heel is
necessary to get the required quantity of blood in
order to correctly fill the circles of the cardboard
blotter. There are two main difficulties:
1. the quantity of the sampled blood with respect
of the felt pain
2. the impregnation of the card circles; blood must
cover the whole surface of the circles and the
rear face of the card must be correctly
impregnated.
To bring some solutions to these two constraints,
a particular tip is studied (figure 4). This tip is fixed
to the extremity of a reversed syringe that is not
represented on the figure. The tip consists of a
transparent bell onto which a micro-grid is stuck. A
flexible interface is fixed on the micro-grid. The
latter must be removable. It consists of thick disc
with an aperture in its centre. We proceed as
follows.


Figure 4: Principle of the pressure free system.
1. We pierce the newborn's heel with the
conventional lancet.
2. We apply the device onto the cut. The flexible
interface ensures the bloodproofness of the
whole system and prevents the skin to be in
contact with the micro-grid. Without this
interface, the probability that the cut coincides
with a few micro-apertures of the grid would be
very weak and the quantity of sampled blood
likely insufficient.
3. We start the aspiration by means of the reversed
syringe. The pressure on the heel is no longer
required.
4. Blood fills the transparent bell. The volume of
the bell is fixed to 1 ml. We stop the aspiration
when the bell is filled.
5. We withdraw the device from the heel.
6. We remove the interface. The dimension of the
grid apertures is calculated so that blood
remains in the bell during these operations.
PAIN AND EFFICIENCY IN NEONATAL BLOOD SAMPLE SCREENINGS - New Devices for Reducing Pain and
Improving Blood Sample Quality
293

7. We apply the tip on the card. The apertures of
the grid fill the whole surface of the circles
printed on the card. By stamping, the circles are
correctly filled. The capillarity forces are
sufficient for blood to distribute efficiently.
Grids are micro-machined by Deep Reactive Ion
Etching. Three characteristic dimensions of the
apertures have been considered: 300 m, 200 m
and 75 m. Two grid thicknesses have been tested:
300 m and 525 m.
The results presented here only concern the tests
of homogeneity of the blood deposited on the cards.
The blood aspiration through the grid will be tested
subsequently. An example of grid attached to the
transparent bell is shown in figure 5.


Figure 5: View of a micro-machined grid fixed on a
transparent bell.
We analyzed the uniformity of the blood
collection with the help of a light source equipped
with a diffuser that permits to illuminate the
cardboard blotter in a homogeneous manner. A CCD
camera is used to acquire pictures of the different
blood impregnated circles. Finally, a suitable image
processing is used to measure the uniformity of
blood. An example of impregnated card is presented
on figure 6 (top). In this case, only 800 l was
necessary to completely fill the circles. Besides, no
pressure on the syringe was required, the capillarity
forces being sufficient. The homogeneity of the
blood is clearly observed. Also, we can note that
blood completely crosses the card as it can be seen
in figure 6 (bottom). In all, six cards have been
impregnated with grids of different thicknesses and
apertures of various dimensions.
Two steps are required to evaluate the
homogeneity by image processing. The first one
consists in semi-automatically detecting the circle
that contains blood. The second one consists in
evaluating the homogeneity of the blood. To do this,
the pictures are convolved with a test window of 5x5
then of 25x25 pixels. We calculate the standard
deviation on the considered window; the lower the
standard deviation, the better the homogeneity. A
typical result is given in figure 7 where the
homogeneities obtained with apertures of 300 and 75
m are shown. In abscissa we have the number of
the card and in ordinates the value of the
homogeneity indicator.


Figure 6: Picture of both front and rear side of a cardboard
blotter impregnating with the pressure free device.

Figure 7: Experimental estimation of the homogeneity of
the cardboard blotter impregnation for two aperture sizes.
The conclusion seems obvious. For equal micro-
grid thicknesses, homogeneity is better for large
apertures than for small ones. The grid thickness has
no influence. 300 m apertures seem interesting for
two reasons. First, the apertures are small enough so
that the blood doesn't escape from the grid before the
latter is in contact with the cardboard blotter.
Secondly, such a dimension is compatible with the
industrial machining. The manufacture of such tips
will therefore be cost-effective.
5 CONCLUSIONS
In this paper, we have presented new medical
devices that reduce pain and should allow screening
a larger number of diseases in newborns. They are
meant to replace the technology currently used in
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
294

neonatal blood sample screening. The origin of this
work is a study of the pain felt by the newborns.
This has highlighted that both, the sting and the
pressure required for the blood collection where
painful. We therefore studied a micro-needles array
as well as a system that suppresses pressures on the
children's heel. The ulterior motive of this work is to
improve neonatal blood sample screenings and
therefore to increase the number of screened
diseases and try to generalise this technique to a
places where it is not yet done.
ACKNOWLEDGEMENTS
This work is supported by the French Health
Ministry through a Hospital Protocol for Clinical
Research.
REFERENCES
Destruynder, R., Lassauge, F., Menget, et al., 1991, Pain
in the newborn in an intensive care unit, Pediatrie,
Vol. 46, pp. 535-539.
Facchini, A., Bellieni, CV., 2005, Relating pain intensity
of newborns to onset of nonlinear phenomena in cry
recordings, Physics Letters A, Vol. 338, pp. 332-337.
Mukerjee, E., Collins, S., Isserof, et al., 2004,
Microneedles array for transdermal biological fluid
extraction and in situ analysis, Sensors and Actuators
A, Vol. 114, pp. 267-275.
Oka, K., Aoyagi, S., Isono, et al., 2001, Fabrication of a
micro-needle for a trace blood test, Transducer'01
Digest of Technical Papers, pp. 412-415.
Owens, ME., Todt, EH., Pain in infancy : neonatal
reaction to a heel lance, 1984, Pain, Vol. 20, pp. 77-
86.
Sharaf, R., Aggarwal, P., Kaler, K., et al., 2003, On the
design of an electronic mosquito : design and analysis
of the micro-needle, Proceedings of the International
Conference on MEMS, NANO and Smart Systems,
pp. 32-35.
Uyan, ZS., Bilgen, H., Topuzoglu, et al., Comparison of
three neonatal pain scales during minor painful
procedures, 2008, J ournal of Maternal-Fetal and
Neonatal Medicine, Vol. 21, pp. 305-308.
PAIN AND EFFICIENCY IN NEONATAL BLOOD SAMPLE SCREENINGS - New Devices for Reducing Pain and
Improving Blood Sample Quality
295
THE DESIGN AND FABRICATION OF
IMPLANTED INTRACRANIAL PRESSURE SENSOR
Tian Bian, Zhao Yulong and J iang Zhuangde
State Key Laboratory for Manufacturing Systems, Xian Jiaotong University, Xian 710049, China
[email protected], [email protected]
Keywords: Piezoresistive, Pressure sensor, Implantable intracranial pressure, Biocompatible.
Abstract: For the purpose of intracranial pressure measurement, implantable intracranial pressure monitoring sensor
applying to the long-term and real-time monitoring to the intracranial pressure of the brain patients, a
pressure sensor was designed based on the piezoresistive principle. The fabrication of pressure sensor
adopted the technology of bulk micromachining to form the structure, and used the ion implanted technique
to form resistances. The packaging was successfully fabricated by using biocompatible material, such as
titanium alloy and polyurethane. The output characteristic of the sensor is measured. It was demonstrated
that this pressure sensor has good performance, include linearity, accuracy and sensitivity for medical
applications.
1 INTRODUCTION
The intracranial pressure (ICP) is extremely valuable
in many cases in order to monitor and control the
clinical condition of a patient. Presently there are
essentially three types of intracranial-pressure
sensors: (a) the sensors requiring handling of the
cephalorachitic intraventricular liquid or cisternal
liquid, (b) the so called sub dural sensors to be
implanted in the subdural space between the dura
mater and the arachnoid, (c) the extra-dural sensors
to be implanted on the dura mater between this dura
mater and the skull.
The sensors of the first type (a) measure directly
the pressure of the cephalorachitic liquid which is
transmitted by a catheter to a transducer. The sensors
of the second type (b) measure the intracranial
pressure by using the arachnoid as an interface, the
arachnoid being a very fine and very flexible
membrane capable of integrally transmitting the
pressures. In both cases the pressure to be measured
is directly accessible, without there being
distortions, whereby the measurement provides
significant information without need for special
precautions and while making use of conventional
pressure sensors which are merely selected to have
the required sensitivity.
However implanting the sensors (a) or (b) entails
effraction of the dura mater necessitating a far more
complex surgical intervention than that required by
an extradural implant (c) and carries risks well
known to the practitioners; in particular, only the
extradural implant is suitable for danger-free, long-
term surveillance.
In this study, compared with conventional
monitoring, adopted the type of extradural implant, a
pressure sensor based on MEMS was designed for
ICP motoring. Considering minimized size and low
range of sensor, the corresponding finite element
analysis modules of the pressure sensor was set up.
The material and structure design of packaging were
also involved for compliant and stable implantation
purpose.
2 STRUCTURE DESIGN
AND FABRICATION
The ICP monitoring was evaluated at the standard of
2.00KPa, normally, the ICP under 2.00KPa, but the
lightly increased value of ICP is between 2.00KPa to
2.70KPa. Compared with other kind of pressure
sensors, piezoresistive pressure sensors have more
advantages which could satisfy the demand of ICP
monitoring, such as micro size, simplicity
fabrication process and high sensitivity. To fulfill
the requirement of pressure measurement for high
sensitivity and low range applications, this paper
presents a piezoresistive pressure sensor developed
on Si wafers.
296

In this research, it was utilized the anisotropy
characteristic along different orientation of single
crystal silicon. The piezoresisitance characteristic is
used to produce the pressure sensor. There is the
largest piezoresistance coefficient along the crystal
direction [110] or [110]. However there is almost no
piezoresistance coefficient along the crystal
direction [100] and [010] in (100) silicon. The
simulation was done for the pressure range 0~10kPa
by finite element method (FEM) software ANSYS.
Since the pressure sensor device is a quartered
symmetry, the quarter FEM of pressure sensor was
established, and the stress distribution of structure is
showed in Fig 1. The stress concentration zone is
located at the edge of the membrane, where the
resistors are implanted. The mechanical stresses
obtained by FEM should be transformed into output
voltage in such a way that the simulation stress value
can be applied to predict the equivalent output
electrical signal. All the four piezoresistances of the
pressure sensor are formed the Wheatstone bridge
circuit. Eq. (1) indicates the output voltage,
resistance and stress variation relation
l
is the
longitudinal piezoresistance coefficient and
t
is the
transverse piezoresistance coefficient.
l
is the
uniaxial stress, and a transverse stress
t
:
1 1 t t
in
V R
V R

= = = +

(1)
The pressure sensor chip is processed from a 4
inch (100) orientation silicon wafer using
conventional lithographic technology. The thickness
of silicon wafer is 400 m . The single crystal
silicon is n-type. The fabrication and the packaging
processes comprise several steps, a schematic view
of the device layout is showed in Fig 2.
First, a silicon oxide layer with the thickness of
120 nm is deposited on the substrate silicon by high
thermal way, and patterned by the mask for the
piezoresistance of pressure sensor. Then the boron
ion is implanted to the substrate through the pattern
of the mask to a depth of 2 ~2.5m at the dose of
2.0
15
10 cm-2 with the energy of 80keV. This forms
resistors patterns of pressure sensor. The purpose of
resistance is about 25 /. A Si3N4 layer with
depth of 120 20nm is formed using Low Pressure
Chemical Vapor Deposit (LPCVD) to protect the
circuit of the sensors. And then etch the backside of
the substrate to form a 25m silicon diaphragm. An
aluminum film which thickness is 1.5 microns is
splashed to form the interdigitated electrode and
connecting wire in the chip. The fabricated sensor
was showed in Fig 3. The size of pressure sensor
is5500 5500 400 m m m , and the membrane
is4500 4500 m m .





Figure 1: Stress distribution of structure.





Figure 2: Schematic view of the device.





Figure 3: Photo of the fabricated sensor.
3 PACKAGING OF SENSOR
3.1 Titanium Alloy Packaging
Light, strong and totally biocompatible, titanium is
one of few materials that naturally match the
requirements for implantation in the human body.
The high strength, low weight, outstanding corrosion
resistance possessed by titanium and titanium alloys
have led to a wide and diversified range of
successful applications which demand high levels of
reliable performance in surgery and medicine. The
THE DESIGN AND FABRICATION OF IMPLANTED INTRACRANIAL PRESSURE SENSOR
297

natural selection of titanium for implantation is
determined by a combination of most favourable
characteristics including immunity to corrosion, bio-
compatibility, strength, low modulus and density
and the capacity for joining with bone and other
tissue - osseointegration. For the advantages above,
the titanium alloy TC4 type was used for the
material of packaging. As showed in Fig 4, the shell
of titanium alloy packaging has two parts, the cover
and foundation. The extended wire was connected
with outer through the hatch of the cover. The liquid
was introduced to the pressure senor through the
hole of foundation for sensing.

3.2 Extended Wire
The extended wire introduced the signal of pressure
sensor to peripheral part for monitoring. For the
requirement of biocompatible and possibility of
long-term implants, the wire is covered with
polyurethane for medical application. Polyurethane
film's performance and characteristics make it a
perfect fit for use in the medical industry. Polyether's
unique combination of strength, biocompatibility,
and innate anti-microbial qualities make it an ideal
material for extensive use in the medical field. The
diameter of extended wire is 1.45mm, and the photo
of wire is showed in Fig 5.

3.3 Process of Packaging
Considering minimized size of hatch which was
punctured on skull, the diameter of packaged sensor
was set to 11mm and 3mm high. The diameter of
transfer circuit is 10mm and the height is 0.2mm.
The structure of packaged sensor was showed in Fig
6, and the transfer circuit is showed in Fig 7. The
process of packaging has several steps as follows:
First, the pressure sensor chip was covered by an
insulated macromolecule material parylene film
which transmited the pressure. The insulated film
could conduct intracranial pressure indirectly and
biocompatible with body. Through the technology of
ultrasonic cleaning make the titanium alloy shell
pure to achieve the standard for medical implanted.
The inside packaging cover was spread with
insulated material. The insulated material could
descend the possibility of creepage which is due to
the contact between wire and titanium alloy shell
accidently. Second, pressure sensor chip and signal
transfer circuit were glued on the foundation of shell
by cyanoacrylate, and connected by spun gold
welding. The extended wire was also weld to circuit.
Finally, the cover of shell was glued on foundation
part.





Figure 6: Structure of packaged sensor.




Figure 7: Transfer circuit.
4 RESULT AND ANALYSIS
The output of the pressure sensor is showed in Fig 8
and the performances of pressure sensor are showed
in Table 1, respectively. The inc represents the
Figure 5: Photo of extended wire.
a. Cover b. Foundation
Figure 4: Shell of titanium alloy packaging.
Titaniumalloy shell Insulated layer Extended wire Pressuresenor chip
Parylenefilm Circuit board Cyanoacrylate
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
298

pressure inputs from lower to higher, contrarily, the
dec shows the outputs when pressure lower. The
sensitivity is 5.66mv/KPa. The non-linearity and
hysteresis of the sensor are less than 0.1%FS and
0.05%FS, respectively.

Figure 8: The output characteristic of the sensor.
Table 1: The performances of pressure sensor.
5 CONCLUSIONS
It has been shown that the design and fabrication of
pressure sensor including packaging. The pressure
sensor was fabricated for intracranial pressure
monitoring based on MEMS. The sensor chip
possesses the better characteristics including size,
linearity, and accuracy. The packaging of the sensor
was designed for well biocompatible implanted in
skull. The pressure sensor chip is able to measure
the parameter for the demands of less volume and
less pressure range conditions for medical
applications.
ACKNOWLEDGEMENTS
This work was supported by the National Natural
Science Fund (Item NO.: 50535053) and
international cooperation item (Item
No.:2006DFA73620). The authors appreciate Dr
J ingbo Xus help from institute of Precision
Engineering, China, for providing pressure sensors
and relative testing, and thank Mr Gaofeng Zhou
from our lab for their help with fabrication.
REFERENCES
N. Bruder, P. N'Zoghe, N. Graziani, D. Pelissier, F. Grisoli
and G. Franois. A comparison of extradural and
intraparenchymatous intracranial pressures in head
injured patients [J ]. Intensive Care Medicine, 1995,
21(10):850-852.
U. Kawoos, G.K.Mugalodi, M.R.Tofighi, S.Neff, A.Rosen.
A Permanently Implantable Intracranial Pressure
Monitor[C]// Proceedings of the 2005 IEEE 31st
Annual Northeast Bioengineering Conference,
Hoboken, NJ , United States, 2005:17-19.
Procaccio F, Stocchetti N, Citerio G, et al. Guidelines for
the Treatment of Adults with Severe Head
Trauma(part ) [J].J Neurosurg Sci, 2000, 44(1)1~10.
Gregory T A 2003 Micromachined Transducers
Soucebook(Beijing: Science Press) pp 4750
Kroetz G H, Eickhoff M H and Moeller H 1999 Silicon
compatible materials for harsh environment
sensors,Sensors Actuators 74 182-9
Parameter Value
Pressure range (KPa) 10
Repeatability (%FS ) 0.03
Non-linearity (%FS) 0.01
Accuracy(%FS) 0.04
Hysteresis (%FS) 0.02
System error (%FS ) 0.01
Diameter of sensor (mm) 11
THE DESIGN AND FABRICATION OF IMPLANTED INTRACRANIAL PRESSURE SENSOR
299
INERTIAL SENSOR BASED IDENTIFICATION OF
HUMAN MOVEMENTS
Ivo Stancic
1
, J osip Music
1
, Ana Kuzmanic Skelin
1
, Tea Marasovic
1
, Norberto Salgado
2

Tamara Supuk
1
and Vlasta Zanchi
1
1
Faculty of Electrical engineering, Mechanical Engineering and Naval Architecture FESB
University of Split, Rudjera Boskovica bb, Split, Croatia
2
Department of Electronic, Telecommunications and Informatics, University of Aveiro, Aveiro, Portugal
{ivo.stancic, josip.music, ana.kuzmanic, tea.marasovic}@fesb.hr, [email protected]
{tamara.supuk, vlasta.zanchi}@fesb.hr
Keywords: Inertial sensors, Head movement, Standing up movement, Kalman filtering, Spine load during sitting and
standing.
Abstract: The scope of this paper is the presentation of experiments which involve measurements and identification of
human movements by using the inertial sensors. We describe the purpose, design and obtained results of
two experiments, as well as our future plans which include the exploration of the forces acting at spine
segments by measurements with inertial sensors. The first experiment implemented the method for
measuring the range of motion (RoM) of head in transverse plane (Kuzmanic, 2007). It was done in the
Laboratory of Biomechanics and Automatic Control LaBACS, University of Split. In the second
experiment we analyzed the standing up movement and we used the robot assistive device for the support
of human while performing the standing up task. Measurements for purposes of this experiment were done
in the Laboratory of Biomedical Engineering and Robotics, University of Ljubljana. We have proposed the
new method which uses the Extended Kalman filtering for combining the data acquired from inertial sensor
measurements of standing up movement with data from the dynamic human body model (Music, 2008).
Our plans regarding the next experiment are focused on the identification of the spinal load during sitting
and standing, by using the inertial sensors measurement system.
1 MEASUREMENT OF THE
HEADS RANGE OF MOTION
The measurement of range of motion (RoM) and
static posture of the head gives important physical
parameters for clinical assessment and diagnosis
related to cervical spine functions. In literature, this
movement is referred as a cervical range of motion.
The detection of an abnormal RoM or asymmetrical
patterns is an essential for preventing cervical
dysfunction (McAviney, 2005; Wu, 2007).
One of our aims was to investigate the feasibility of
the use of inertial sensors in routine clinical
assessments. Therefore, our goal was to design the
system based on inertial sensors and to propose the
method for measuring the range of motion (RoM) of
head in transverse plane. The measurement was
performed using single inertial measurement unit
MTx XSens sensor (XSens Motion Technologies,
Netherlands), Fig. 1. Specialized software for sensor
data acquisition, with high visualization abilities has
been developed in LaBACAS. MTx XSens sensor
can provide useful, noninvasive measurement of
head motion in three cardinal planes for the fast
evaluation of disturbances related to head/neck
problems and cervical dysfunctions. The advantage
of the proposed method over standard methods is the
ability to measure unilateral RoM of the head. This
technique overcomes the limits of gold standard
measurement devices by estimating the neutral
position, which is assumed to be a nontrivial problem
in standard RoM measurement. In addition, a
proposal for use of sensor for visual feedback RoM
assessment is presented. LaBACS MTx Software
was developed by in-house, to control the operation
of the inertial measurement unit (IMU), acquire the
data and display them in the real time. Program was
developed under Microsoft Visual Studio 2005, using
MFC (Microsoft Foundation Class). Fig. 2. shows a
frame of running software. Measurement was done
on 6 subjects without any known symptoms of
300

cervical dysfunction. Five repetitions of movements
were analyzed and averaged for each subject in order
to eliminate the variability during movement
recording. In accordance with standard, total RoM is
calculated by subtraction of maximal and minimal
angle, or by summation of left and right RoM,
assuming that the neutral position angle is known.
During the measurement neutral position is identified
statistically, over time interval of five repetitions of
cyclic RoM movement, Fig. 3 b).

















Figure 1: Measurement setup: Subject with sensor
mounted on a cap.

Figure 2: User interface of LaBACS MTx Software.
1.1 Results of the First Experiment
The results of the measurement on 6 asymptomatic
subjects are given in Table 1. Resulting angles of
each group are described in terms of mean RoM
standard deviation [
o
], for the movement on the
left (LRoM) and right side (RRoM). The results of
the present study demonstrate similar ranges of
motion as found in literature (Dvir, 2000), although
the existing results are obtained with different
instrumentation. Measurement of individual neutral
position has a standard deviation ranging from
minimally 1.12
o
to maximally 3.36
o
. These results
imply that the subjects are able to return the head to a
self-defined neutral position. Therefore, the
measurement method of head motion based on
inertial sensors is valid for current application of
RoM in transverse plane and is suitable for
measurement of head neutral position, as well.
Table 1: RoM results for 6 asymptomatic subjects.
left side:
LRoM
right side:
RRoM
RoM (total) LRoM/
RRoM 1) 2)
73.02
0
7.61
0

74.34
0
9.44
0

147.28
0
15.51
o

147.36
0
15.54
o

1.022
0

0.096
0

1)
L

R
where
L
is maximal and
R
is minimal head
angle; 2) LRoM +RRoM

a)

b)
Figure 3: a) Recorded angles of cyclic movement; b)
Histogram computation of head neutral position and
endpoint angles.
2 KINEMATIC MEASUREMENTS
OF STANDING UP MOTION
Number of aiding systems has been developed for
the purpose of standingup support. Recently, robot
assistive devices have been introduced and their
benefits demonstrated. In acquisition systems, the
kinetic and kinematic parameters of the subject are
required for operation of the robot control algorithm.
Kinematic measurements are usually performed with
INERTIAL SENSOR BASED IDENTIFICATION OF HUMAN MOVEMENTS
301

optical motion analysis systems that are unsuitable
for clinical applications. Therefore, introduction of
miniature, low cost inertial sensors (accelerometers
and gyroscopes) as a body mounted sensors, has
shown to be promising.
We propose a new approach in which the
Extended Kalman filtering (EKF) technique is used
to fuse data acquired from inertial sensor
measurements with data from the dynamic human
body model (Music, 2008). In this way we believe
that better kinematic measurements in ambulatory
settings are possible. We named the approach Model
Based Inertial Sensing - MoBIS.
The proposed human body model consists of
shank, thigh and HAT (Head-Arms-Trunk)
segments, Fig. 4.
Figure 4: Measurement setup: (1) linear infrared cameras,
(2) HAT inertial sensing unit, (3) thigh inertial sensing
unit, (4) shank inertial sensing unit, (5) AMTI OR6-6-1
force plate, (6) seat, (7) J R3 40E15 force sensor, (8)
standing-up robot assistive device.
The segments are assumed to be rigid bodies
with their masses contained at center of mass
(CoM). Segment masses, lengths, moments of inertia
and CoM positions are defined using anthropometric
data. Three joints (ankle, knee and hip) are assumed
to be ideal pin joints with no added friction during
rotation. The model is in contact with its
environment only by the distal end of the shank
segment i.e. by subjects feet. The assumption of
symmetry of sit-to-stand motion in respect to sagittal
plane was adopted in modeling phase. This
assumption enables the measurements to be carried
out only on one side of the body and results
projected on the other side. The symmetry
assumption does introduce certain error.
2.1 Results of the Second Experiment
The method is validated on both simulated and
measured data. The presented results (Figures 5, 6
and Table 2) show that Model Based Inertial
Sensing (MoBIS) in robot assisted standing-up is a
reliable alternative to optical measurements systems
for motion kinematics assessment. To improve

method performance in terms of accuracy and
reliability, further development (e.g. extensive
testing on a group of healthy and impaired subjects,
introduction of adaptive EKF) is suggested (Music,
2008).

Figure 5: Comparison of actual and measured angles.

Figure 6: Comparison of Optotrak and EKF data.
Table 2: Measurement error.
Shank
RMSE
[deg]
Thigh
RMSE
[deg]
HAT
RMSE
[deg]
Normal/self-selected standing up speed
Meas. 01 6.1 4.1 3.8
Meas. 02 2.1 7.8 6.8
Meas. 03 2.4 5.2 6.8
Meas. 04 3.7 3.5 5.9
Average 1 3.6 5.2 5.8
Fast standing up
Meas. 05 1.4 2.5 3.2
Meas. 06 4.8 2.6 4.7
Meas. 07 4.9 2.4 5
Meas. 08 3.9 2.1 5
Average 2 3.8 2.4 4.5
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
302

3 DISCUSSION
AND CONCLUSIONS
Results obtained in described experiments show that
inertial sensors can be implemented into different
measurement systems and bio devices as reliable
and yet inexpensive tool for identification of
versatile human movements. Our future work
regarding the implementation of inertial sensors
includes the identification of the spinal load during
sitting and standing.
3.1 Segmental Spine Load: Model
and Force Analysis
The main idea is to explore forces acting at single
spine segments. All the measurement procedures,
used in the research, will be noninvasive. The
identification of the single spine segment
coordinates will be done using the inertial sensors.
The later stages will also include the analysis and
calculations of corresponding forces, and therefore,
to that end, ground reaction forces will be measured
(Supuk, 2002). Research will be performed in static
and dynamic conditions, on sitting and standing
subject, Figures 7 and 8. Partial differential
equations will provide mathematical support during
the modeling process, keeping in mind that we are
dealing with compartmental system. Configuration
of the spine, that will be identified based on the
results of measurements obtained by five different
sensor outputs, along with the seat reaction forces
will serve as input parameters for the calculations of
forces acting at the diverse points of the spine (33
vertebrae including the five that are fused to form
the sacrum (the others being separated by
intervertebral discs) and the four bones which form
the tailbone.).


Figure 7: Standing subject spine model.

Figure 8: Musculoskeletal model used to identify spine
configuration of the subject in the seated position.
REFERENCES
Dvir, Z., Prushansky, T., 2000. Reproducibility and
instrument validity of a new ultrasonography-based
system for measuring cervical spine kinematics,
Clinical Biomechanics, Vol. 15, Issue 9, pp. 658-664.
Kuzmanic Skelin, A., Vlak, T., Stancic, I., 2007.
Inertial sensor measurement of head-cervical range of
motion in transverse plane. Proc. of 3rd WSEAS Intl.
Conference on Remote Sensing, Venezia, Italy,
WSEAS Press, pp. 47-51.
McAviney, J ., Schulz, D., Bock, R., Harrison, DE.,
Holland, B., 2005. Determining the relationship
between cervical lordosis and neck complaints, J
Manipulative Phisyiol Ther, Vol.28, 2005, pp. 187-
193.
Music, J ., Kamnik, R., Munih, M., 2008. Model based
inertial sensing of human body motion kinematics in
sit-to-stand movement, Simulation Modelling Practice
and Theory, Vol. 16, No. 8, pp. 933-944.
Supuk, T., Kuzmanic Skelin, A., Kuzelicki, J ., Zanchi, V.,
2002. Estimation of vertical component of ground
reaction force during sit to stand movement: Direct
dynamics approach. In Proceeding of Int.
Electrotehnical and Computer Science Conference
ERK2002, Portoroz, Slovenija, pp. 23-25
Wu, Shyi-Kuen et al, 2007. The feasibility of a video-
based motion analysis system in measuring the
segmental movements between upper and lower
cervical spine, Gait and Posture, Vol.26, 2007, pp.
161-166.
INERTIAL SENSOR BASED IDENTIFICATION OF HUMAN MOVEMENTS
303
ADAPTIVE AURICULAR ELECTRICAL STIMULATION
CONTROLLED BY VITAL BIOSIGNALS
Transition from Fixed to Adaptive and Synchronized Electrical Stimulation
Controlled by Heart Rate Variability and Blood Perfusion
Eugenijus Kaniusas
Institute of Electrical Measurements and Circuit Design, Vienna University of Technology
Gusshausstrasse 27-29/E354, Vienna, Austria
[email protected]
J ozsef Constantin Szeles, Tilo Materna
Department of Surgery, University of Vienna, Vienna, Austria
[email protected]
Giedrius Varoneckas
Sleep Medicine Centre, Klaipeda University Hospital, Klaipeda, Lithuania
[email protected]
Keywords: Electrical stimulation, heart rate variability, physiological sensors, adaptive stimulation, ear.
Abstract: The auricular electrical punctual stimulation is usually applied for pain relief. The common application
involves fixed stimulation parameters, which makes the simulation insensitive to prevailing pain or stress
level and may lead to a disadvantageous over-stimulation. In order to address this issue, the given position
paper presents an experimental background leading to a conceptual design of an adaptive and synchronized
stimulation technique. Here parameters of the heart rate variability are used as stimulation biofeedback,
while the stimulating signal is synchronized with cardiac or respiratory activity to boost stimulation effects.
1 INTRODUCTION
The auricular electrical punctual stimulation (P-
Stim) is an electrical nerve stimulation technique,
newly introduced by Dr. Szeles (Szeles, 2001a). The
P-Stim is usually applied for acute and chronic pain
relief. A reduction of pain perception and pain-
relieving medications is attained (Szeles, 2001b;
Sator-Katzenschlager, 2006; Likar, 2007), even with
an induction of anaesthesia state (Litscher, 2007).
Furthermore, reduction of body mass index (BMI) in
obese patients (Szeles, 2001b), increase of blood
flow velocity and oxygenation (Szeles, 2004) were
reported during the P-Stim application. The
advantages of the electrical stimulation over
conventional (manual) acupuncture with respect to
pain relief, well-being and sleep quality were
documented in (Sator-Katzenschlager, 2004) for
extended periods of time up to 3 months.
Figure 1: a) Ear with indicated approximate auricular
branch of vagus nerve according to (Peuker, 2002; Gao,
2008). b) Electrical punctual stimulation of the auricular
vagus nerve (P-Stim).
Stimulating device
(=reference electrode)
Vagal stimulating point
Vagal nerve
branch
a) b)
304

The particular beneficial effects of the P-Stim are
still under discussion, whereas a number of the
following mechanisms seem to be involved. The
electrical stimulation of the afferent nerve receptors
may influence gate mechanisms in the central
nervous system (CNS), preventing pain-related
action impulses from reaching the CNS and avoiding
the persons perception of pain. Furthermore, an
indirect stimulation of pain receptors and activation
of inhibitory pain control systems may be involved,
as well as a stimulated release of neurotransmitters,
e.g., endorphins and other endogenous opioids.
Though the efficiency of the P-Stim was subject-
ively proved in many cases and the P-Stim is already
in clinical use, only recently some objective and
statistical evidence was established on the
stimulation effects. Given that an auricular branch of
vagal nerve (Fig. 1a) is electrically stimulated by the
P-Stim device (Fig. 1b), effects on the heart rate
variability (HRV) were assessed in the time and
spectral domain (Kaniusas, 2008; Gbaoui, 2008a)
and in the state space (Gbaoui, 2008b) by our group.
In addition, blood perfusion (BP) changes during
stimulation were investigated (Kaniusas, 2008). In
the latter studies optical plethysmography (OPG)
served as biofeedback to derive the HRV and BP.
Here the suitability of the HRV and BP analysis
is given by the fact that the stimulated afferent vagal
nerve goes to the nucleus solitarius in the CNS,
whereas the sinus node of the heart is controlled by
the efferent vagus nerve from the nucleus ambiguous
in the CNS. The node initiates heart contractions
with particular rate dynamic and ejection strength,
thus the HRV and BP being the appropriate
parameters to register the stimulation effects.
The given position paper is intended to introduce
a novel technology for an adaptive and synchronous
P-Stim controlled by the HRV and BP. As a starting
point, technical data and new experimental results
concerning parasympathetic/sympathetic power in
the HRV from the standard P-Stim are presented,
which yield a substantial basis and arguments for the
introduction of the adaptive stimulation.
2 ESTABLISHED STIMULATION
2.1 Methodology
The P-Stim was applied in supine position of three
healthy volunteers: two men aged 41/29 with BMI
25/23 kg/m
2
and one female aged 19 with BMI of
20 kg/m
2
. A precise positioning of the needle in the
vicinity of the vagal nerve (Fig. 1) was facilitated by
local conductivity measurements, for the local
conductivity increases in the region of the nerve and
its supporting blood vessels.
As demonstrated in Fig. 2a, the voltage U of the
electrical stimulation comprises monophasic
impulses with changing polarity, stimulation
(=repetition) rate f
S
of 1 Hz, amplitude A of 4 V and
impulse duration of about 1 ms.
The duration of the recordings was about 15 min
before, during, and after the stimulation,
respectively. At least two recordings were performed
per volunteer with a time-lag in-between of more
Figure 2: Stimulation waveforms of a fixed (a) and (b) adaptive electrical punctual stimulation.
a)
b)
U
t
A
1 /f
S

Pause Stimulation Stimulation
1 /f
A

U, I
t
A
1 /f
S
1

Pause
1 /f
S
2
( 1 /f
S
1
)
ADAPTIVE AURICULAR ELECTRICAL STIMULATION CONTROLLED BY VITAL BIOSIGNALS - Transition from
Fixed to Adaptive and Synchronized Electrical Stimulation Controlled by Heart Rate Variability and Blood Perfusion
305

than 10 days. It should be noted that the needles for
stimulation were inserted about 5 min before the
recording to avoid needles positioning effects, i.e.,
to avoid temporal effects of manual acupuncture.
In parallel, the OPG signal s
OPG
from the finger
was assessed as biofeedback. Here the relatively
high sampling rate of 2 kHz is needed for an
accurate HRV analysis (Guidelines, 1996). A typical
course of s
OPG
is depicted in Fig. 3a.
The instantaneous heart rate f
C
for the HRV
analysis was estimated from s
OPG
, as demonstrated
in Fig. 3a, with artefacts and noisy segments being
manually removed. The prominent minima in s
OPG
,
which correspond to the onset of the systole or blood
ejection, were detected as fiducial points for the
calculation of the instantaneous f
C
.
The investigation of the resulting f
C
sequence in
the spectral domain comprised power in the
established frequency ranges (Guidelines, 1996):
low frequency range 0.04-0.15 Hz corresponding to
sympathetic power P
SYM
and high frequency range
0.15-0.4 Hz corresponding to parasympathetic
power P
PAR
. Both P
SYM
and P
PAR
were estimated for
sequence windows of 300 s with 50 % overlap. It
should be noted that there are controversial
indications that P
PAR
is also present in the low
frequency range.
The BP is given by the course of s
OPG
(Fig. 3a).
In particular, the amplitude deflection of s
OPG
within
a single heart cycle corresponds approximately to
both amount of blood ejected (=left ventricular
stroke volume) and vesicular compliance.
The respiration reference s
R
(Fig. 3b) was
established by a skin curvature sensor on the chest,
as described in (Pftzner, 2006; Kaniusas, 2004).
2.2 Results
2.2.1 Heart Rate Variability
Fig. 4b and Fig. 5b demonstrate a temporal increase
of P
PAR
during stimulation, which temporal
activation is given in Fig. 4a and Fig. 5a. The
relative increase of P
PAR
among volunteers was
about 20 %, which was observed in all sessions but
one, probably because of a relatively high initial
value of P
PAR
. A temporal dip of P
PAR
was often
observed during the stimulation.
No unique tendencies were registered in the
behaviour of P
SYM
, as demonstrated in Fig. 4c and
Fig. 5c. However, stress relaxation effects could be
observed in some cases even in healthy volunteers.
In Fig. 4b,c and Fig. 5b,c dashed ellipses mark the
corresponding time intervals, where P
PAR
increases
and P
SYM
concurrently decreases. In general, such
changes of P
PAR
and P
SYM
tend to indicate ongoing
restorative effects.
The stimulation effects on P
PAR
were discussed
in a wider context in (Kaniusas, 2008; Gbaoui,
2008a), considering additionally parameters in the
Figure 3: a) Optical plethysmography signal s
OPG
with an estimated cardiac rate f
C
from indicated systolic onset points (*)
and an estimated respiratory rate f
R
from the envelope. b) The corresponding respiration signal s
R
from the chest skin
curvature sensor.
a)
b)
s
OPG
(rel.units)
s
R
(rel.units)
t (s)
Cardiac components
Envelope
1/ f
R
1/ f
R
1/ f
C
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
306


Figure 4: Effects on heart rate variability in the female
subject. a) Temporal activation of the electrical
stimulation (P-Stim OFF or P-Stim ON). b) The
corresponding parasympathetic power P
PAR
. c) The
corresponding sympathetic power P
SYM
.
time domain and state space. Aforementioned
tendencies of P
PAR
were also found in (Haker, 2000),
even during non-electrical auricular stimulation by
acupuncture needle.
In contrast to P
PAR
, none of the mentioned
studies indicate clear tendencies of P
SYM
. This is
likely to be attributed to the study enrolment of only
healthy unstressed pain-free individuals in resting
state, where potential changes or improvements of
P
SYM
are strongly restricted.
2.2.2 Blood Perfusion
The BP is given by the course of s
OPG
, as show in
Fig. 3a. It is important to observe that not only the
instantaneous cardiac activity but also the respiration
can be derived from s
OPG
.
In particular, the systolic onset points, as marked
by asterisks in Figure 3a, give a useful reference to
heart excitation. These points are delayed by about
200 ms from the actual excitation of the heart
ventricles (=R peaks in electrocardiography (ECG))
with the delay being nearly constant.

Figure 5: Effects on heart rate variability in a male subject.
a) Temporal activation of the electrical stimulation (P-
Stim OFF or P-Stim ON). b) The corresponding
parasympathetic power P
PAR
. c) The corresponding
sympathetic power P
SYM
.
The respiratory cycle can be derived from s
OPG
,
as indicated by the envelope in Fig. 3a. Here the
amplitude modulation of s
OPG
results from the
respiratory induced modulation of the left ventricu-
lar stroke volume which temporally increases during
expiration. The simultaneously recorded respiration
reference s
R
(Fig. 3b) proves the respiratory related
modulation of the s
OPG
deflection.
3 PROPOSED STIMULATION
3.1 Rationale
Since the spectral HRV parameters are specifically
influenced by the standard P-Stim application and
the instant cardio-respiratory data can be derived
from the BP, as shown above, a novel adaptive and
synchronized P-Stim could be established.
A targeted control of the stimulation waveform
(compare Fig. 2) is highly reasonable for avoiding
over-stimulation and realising stimulation on-
ADAPTIVE AURICULAR ELECTRICAL STIMULATION CONTROLLED BY VITAL BIOSIGNALS - Transition from
Fixed to Adaptive and Synchronized Electrical Stimulation Controlled by Heart Rate Variability and Blood Perfusion
307

Figure 7: Establishment of biofeedback for controlling and synchronization purposes with IG as the impulse generator.
demand controlled by HRV parameters. In other
words, if pain perception is already reduced, as
detected by e.g., reduced stress and diminished
P
SYM
, then A, f
S
(Fig. 2b) could be reduced as well.
In addition, efficient energy use in the stimulation
would be facilitated.
The synchronization of the stimulation waveform
with the cardio-respiratory activity would allow a
constructive interference of the stimulated pain-
relieving effects and the residual body attempts. In
particular, the cardiac synchronization would allow a
timely activation of the gate mechanisms in the CNS
or a timely and indirect stimulation of receptors
(e.g., blood pressure), regulating vital body
functions. The respiratory synchronization would
help to interfere with body phenomena like
respiratory sinus arrhythmia, yielding a forced
increase of P
PAR
in the expiration phase.
3.2 Realization
The proposed set-up is shown in Fig. 6. The input
parameters A, f
S
and the activation rate f
A
of the
impulse generator are adaptively adjusted according
to the HRV parameters via a control loop. The
cardio-respiratory synchronization signal for the
impulse generator is also derived from s
OPG
.
In particular, Fig. 7 suggests the difference P
PAR

- P
SYM
as a possible realization of the stimulation
feedback, while the targeted value could be the pain
intensity to be reduced. That is, the higher P
PAR
and
the lower P
SYM
get in the course of the stimulation,
the more strongly the pain has already been reduced.
Similar behaviour of P
PAR
and P
SYM
during
stimulation was already observed in Fig. 4 and Fig.
5. Obviously the ratio P
PAR
/ P
SYM
could be used
instead of the difference.
According to Fig. 7 an adaptive control of A and
f
S
is established, assuming that these parameters are
directly interrelated with the stimulation strength. In
an analogous way, a composition of bursts by
controlling of f
A
could be attained (compare Fig. 2).
Here a proportional-integral controller or integral
controller could be applied, for the human (Fig. 6)
can be roughly approximated as a proportional
control process with a single time constant (compare
Fig. 4b). The time delay in Fig. 7 may be needed for
synchronizing the stimulation pulses with a
particular time instant in the heart cycle.
Fig. 2b exemplifies a possible adaptive control-
ling of the stimulation curve, while more efficient
biphasic impulses are used (compare Fig. 2a). In
addition, constant current stimulation would be
preferred over voltage application, for the skin
impedance is relatively low with electrode needles
inserted and thus the risk of local tissue damage
Figure 6: Control loop of the adaptive auricular stimulation with OPG as the optical plethysmography.
Impulse generator Ear Body
Stimulation electrode
OPG
Human (=control process)
Parameters
of OPG
Stimulation
parameters: A, f
S
, f
A

Synchronization
(heart or respiration)
Controller
Targeted
values
Feedback (=actual
HRV parameters)
s
OPG

Finger
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
308

though locally increased current density is low.
4 DISCUSSION
It is worth to note that the HRV is usually derived
from the ECG (Guidelines, 1996). However, the P-
Stim induced very strong artefacts in the ECG since
the stimulation and the ECG have the same electrical
origin. In contrast, the OPG with optical origin
serves as a reliable biosignal, being independent of
the P-Stim activation. However, the OPG conveys
mechanical information on the systole-diastole cycle
rather than electrical on the heart excitation (=origin
for the HRV). In addition, the OPG exhibits
relatively slow changes if compared to the ECG, for
the pulse waves are much more inert than electrical
heart excitation. The use of the OPG may have
reduced an effective time resolution of f
C
.
The time delay of about 200 ms between the
systolic onset in the OPG and the R peak in the ECG
depends on the speed of the heart excitation and
mechanical vessel properties. Nevertheless, the
delay can be assumed to be constant, if the respirato-
ry induced blood pressure changes and thus arterial
distension and stiffness changes can be neglected.
Lastly, the limitations of the presented
experimental results should be mentioned. The
observed effects, especially concerning P
SYM
, are
restricted by the fact that all volunteers were young
pain-free healthy persons. Furthermore, the
stimulation duration was relatively short: 15 min
versus 4 hours (with 4 hours pause in-between) over
at least seven days, as clinically applied and
subjectively verified for being effective. The initial
state of the volunteers, as their possible excitation at
the beginning of the recording, and their mental
activity changes during the investigation - both
influencing the HRV - may have limited the range of
potential changes or improvements of HRV
parameters during the stimulation.
However, the provided experimental background
leads to a comprehensible design of an adaptive and
synchronized stimulation technique. This would
allow a pain sensitive adjustment of the stimulating
parameters avoiding over-stimulation and
comforting the patients.
REFERENCES
Gao, X.Y., Zhang S.P., Zhu, B., Zhang H.Q., 2008.
Investigation of specificity of auricular acupuncture
points in regulation of autonomic function in
anesthetized rats. Autonomic Neuroscience: Basic and
Clinical, 138, 50-56.
Gbaoui, L., Kaniusas, E., Szeles, J.C., Materna, T.,
Varoneckas, G., 2008a. Heart rate variability during
electrostimulation on ear: spectral domain versus state
space (in german). Proceedings of XXII Symposium on
Measuring Technique, 230-238.
Gbaoui, L., Kaniusas, E., Szeles, J.C., Materna, T.,
Varoneckas, G., 2008b. Effects of the auricular
electrical stimulation on heart rate variability assessed
in phase space: pilot study. Accepted for IEEE Sensors
2008 in Lecce, Italy.
Haker, E., Egekvist, H., Bjerring, P., 2000. Effect of
sensory stimulation (acupuncture) on sympathetic and
parasympathetic activities in healthy subjects. Journal
of the Autonomic Nervous System, 79, 52-59.
Kaniusas, E., Gbaoui, L., Szeles, J.C., Materna, T.,
Varoneckas, G., 2008. Validation of auricular
electrostimulation by heart rate variability and blood
perfusion: possibilities and restrictions. Proceedings of
Microelectronics Conference 2008, 180-184.
Kaniusas, E., Pftzner, H. et al., 2004. Magnetoelastic skin
curvature sensor for biomedical applications.
Proceedings of IEEE Sensors, 1484-1487.
Likar, R., J abarzadeh, H. et al., 2007. Auricular electrical
punctual stimulation (P-STIM): a randomized, double-
blind, controlled pilot study in laparoscopic
nephrectomy (in german). Schmerz, 21(2), 154-159.
Litscher, G., Wang, L., Gaischek, I., 2007. Electroence-
phalographic responses to laserneedle and punctual
stimulation quantified by bispectral (BIS) monitoring:
a pilot study to evaluate methods and instrumentation.
Internet Journal of Laserneedle Medicine, 1(1).
Peuker, E.T., Filler, T.J ., 2002. The nerve supply of the
human auricle. Clinical Anatomy, 15, 35-37.
Pftzner, H., Kaniusas, E. et al., 2006. Magnetostrictive
bilayers for multi-functional sensor families. Sensors
and Actuators A: Physical, 129, 154-158.
Sator-Katzenschlager, S.M., Scharbert, G. et al., 2004. The
short and long term benefit in chronic low back pain
through adjuvant electrical versus manual auricular
acupuncture. Anesthesia & Analgesia, 98, 1359-1364.
Sator-Katzenschlager, S.M., Wlfler, M.M. et al., 2006.
Auricular electro-acupuncture as an additional
perioperative analgesic method during oocyte
aspiration in IVF treatment. Human reproduction,
21(8), 2114-2120.
Szeles, J .C., 2001a. Therapy appliance for punctual
stimulation. Patents WO 01/35897, US7336993.
Szeles, J .C., Hoda, M.R., Polterauer, P, 2001b. Appli-
cation of electrostimulation acupuncture (P-Stim) in
clinical practice. Schmerznachrichten from Austrian
Pain Association, 1.
Szeles, J .C., Litscher, G., 2004. Objectivation of cerebral
effects with a new continuous electrical auricular
stimulation technique for pain management.
Neurological Research, 26(7), 797-800.
Guidelines, 1996. Heart rate variability: standards of
measurement, physiological interpretation, and clinical
use. Circulation, 93(5), 1043-1065.
ADAPTIVE AURICULAR ELECTRICAL STIMULATION CONTROLLED BY VITAL BIOSIGNALS - Transition from
Fixed to Adaptive and Synchronized Electrical Stimulation Controlled by Heart Rate Variability and Blood Perfusion
309
DEVELOPMENT OF A MECHANICAL INSTRUMENT
TO EVALUATE BIOMECHANICALLY THE SPINAL COLUMN
IN PREGNANT WOMEN
Cludia Quaresma, Mrio Forjaz Secca
CEFITEC, Dep. of Physics, Faculdade de Cincias e Tecnologia, UNL
Quinta da Torre, P-2829-516, Caparica, Portugal
[email protected], [email protected]
J oo ONeill
Faculdade de Cincias Mdicas, UNL , Quinta da Torre, P-2829-516, Caparica, Portugal
[email protected]
J orge Branco
Maternidade Dr. Alfredo da Costa, Lisboa, Portugal
[email protected]
Keywords: Biomechanics, Vertebral column, Non-invasive instrument, Evaluation, Biomechanical, Standing position.
Abstract: The incidence of problems related to rachialgiae is so frequent and usual that it must be studied as if it were
an epidemic and social disease (Knoplich, 2003). Instruments that evaluate the spinal column in a standing
position, in a global way, are needed to attain a better insight into this problem. This work presents a
completely safe instrument to assess the evolution of all the vertebra locations throughout time, in order to
study the biomechanical changes in women during pregnancy. A mechanical system, registered as Vertebral
Metrics, with the objective of evaluating the curvatures and lateral deviations of the spinal column in the
standing position was built. A measuring part, which is the body of the instrument, and a supporting part,
where the previous part is mounted, constitute the new non-invasive instrument. The measuring part
consists of 17 identical adjustable mechanical blocks that allow us to reproduce the position of each vertebra
of the spinal column, from the first cervical vertebra to the first sacral vertebra. Vertebral Metrics was
originally planned and built to be applied to pregnant women. However, after redefining dimensions of the
different parts it can be applied to any type of population, in the future.
1 INTRODUCTION
Rachialgiae constitute a relevant problem in modern
society (Alexandre & Moraes, 2001). In many
women this problem appears for the first time during
pregnancy and remains for the rest of their lives,
causing serious problems of absenteeism and
consequently a great loss under the point of view of
the economy of the country. It is necessary to have
instruments that evaluate, in a global way, the spinal
column in standing position, in order to understand
better the behaviour of the spinal column during the
gestational period.
There are many instruments to evaluate the
spinal column. However, most use ionizing radiation
and very few allow us to analyse the spinal column
in a standing position.
The instrument that is most used to evaluate the
spinal column in the standing position is the
radiograph. However, since it uses ionizing
radiation, it should not be used on pregnant women
(Harlick e al, 2007; Pinel-Giroux e tal, 2006) and,
besides, it would only give a collapsed two-
dimensional view.
The study presented here is part of a broader
analysis where we intend to identify and describe the
biomechanical alterations of the spinal column that
occur throughout pregnancy. We found it was
necessary to build a non-invasive instrument that
would allow the reproduction of the position of each
of the vertebrae, through the identification of each of
the vertebral apophyses of the spinal column, from
310

the first cervical vertebra to the first sacral vertebra.
In a global way Vertebral Metrics evaluates the
curvatures and lateral deviations of the spinal
column in the standing position.
After an exhaustive search of existing
instruments we found the inexistence of an apparatus
with the required characteristics, which led us to
prepare and elaborate an equipment to measure the
lateral deviations and curvatures of the spinal
column.
The elaboration started with the conceptual
design of the apparatus, based on some existing
instruments. Measurements on 134 women were
carried out in order to define its dimensions,
including the height of the body, the support, the
size of horizontal pieces and the stand of Vertebral
Metric. The experimental procedure of the
previously mentioned study was performed as
follows:
1- The length between various points of reference
was measured in each woman.
2 - The Normality test for each of the variables was
applied.
3 - Significance>0.05 was found.
4 - The confidence intervals for 99% were calculated
The details of the mechanisms of each of the pieces
and the choice of the convenient materials to use
(polycarbonate, brass, steel and duralumin) to build
the instrument were defined afterwards.
Finally, Vertebral Metrics was built (Figure 1)
and the necessary adjustments were performed.
2 VERTEBRAL METRICS:
DESCRIPTION
AND FUNCTIONING
The objective of Vertebral Metrics, a non-invasive
mechanical instrument, is to identify the x, y and z
positions of each vertebra, from the first cervical
vertebra to the first sacral vertebra. After entering
these data into a mathematical model of the spine,
the curvatures and lateral deviations of that segment
in the standing position can be calculated.
The components of the apparatus (Figure 2) are
the body (1) and the support (2).
The body has a vertical piece (5), mounted in
the support (2), with 18 horizontal pieces (4) called
2D Positioner. Two pieces, one vertical, where
the body of the instrument fits, and a stand where the
person to be evaluated stands up, constitute the
support.


Figure 1: Image of Vertebral Metrics.

Figure 2: Diagram of Vertebral Metrics.
Figure 2 presents two views of the whole
instrument, with (a) and (b) showing the frontal view
and the left lateral view, respectively.
The Vertical piece of the body (Figure 3)
consists of a rectangular profile with two plates at
the ends, which provide rigidity and the fixation to
the system.

Figure 3: Vertical piece of the body of Vertebral Metrics.
A detail of the instrument, is presented in Figure
4. The 2D Positioner (4) slides along the scaled
ruler attached on the vertical piece (3).
DEVELOPMENT OF A MECHANICAL INSTRUMENT TO EVALUATE BIOMECHANICALLY THE SPINAL
COLUMN IN PREGNANT WOMEN
311



Figure 4: One of the 2D Positioner and the vertical
piece of the body of the instrument.
Figure 5 shows components, 6, 7, 8, 9 and 10 of
the 2D Positioner. The component 7 is a square
sectioned rod, with a cone shaped tip, which is
where it makes contact with each vertebra apophysis
contact point. This rod is manoeuvred by two
pinions, (6), placed horizontally and touching the
rod via a neoprene O-ring surrounding the pinions. It
fixes the y position of the vertebral apophysis. This
coordinate is calculated through the distance from
the conical end, (5), to each 2D Positioner, (4), of
the body of the instrument.
The pieces 6 and 7, moved by the horizontal
threaded rod (8), are fixed at the top by a plate and
two bolts. These bolts are inserted in two vertical
threaded holes, passing through the rotational axis of
the pinions 6.
At the fixed end of the rod 8 there is a small handle
which rotation enabling rotational movement of the
rod, which, in turn, allows movement of piece 7.
Measurements of the x position (see Figure 4) of the
vertebral apophysis of each of the vertebrae are then
possible.













Figure 5: Top view of 2D Positioner.
The component 9 is a piece that fixes the 2D
Positioner which moves upwards/downwards on
the vertical piece (3), as component 10 rotates.
Finally, coordinate z is obtained from the
measurement of the position of the horizontal part in
piece 4 (see Figure 1).
The support of Vertebral Metrics has two pieces:
one horizontal plate supported on four feet, where
the person to be evaluated stands, and one vertical
piece, shown in Figure 6, where the vertical part of
the body of the instrument (see Figure 2) fits.

Figure 6: The support of Vertebral Metrics.
After marking the vertebral apophyses (Figure 7),
with a washable pen, each pregnant woman stands
up on the stand of the support of the instrument with
the posterior face of the trunk facing the body of
Vertebral Metrics.

Figure 7: Marking the vertebral apophyses with a
washable pen.
All the 18 2D Positioners of the body of
Vertebral Metrics are identical and adjustable,
allowing the identification of the x, y and z positions
of the vertebral apophyses, from the first cervical
vertebra to the first sacral vertebra.
Three of the 2D Positioner pieces are used
distinctively. The first is placed in the occipital
region and is used, during the data collection, as a
reference point. The second piece collects the data of
the cervical vertebra and the piece number fifteen
collects the data from the first, second and third
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
312

lumbar vertebra. The remaining horizontal pieces
will identify the position of all other vertebral
apophyses of the spinal column (Fig 8).
The x, y and z positions of each 2D Positioner are
then obtained.

Figure 8: Example of application of Vertebral Metrics.
Each data collection lasts seven minutes. Figure 9
shows the position of the 2D Positioner after
application of Vertebral Metrics.


Figure 9: After the application of Vertebral Metrics.
The collected data are then recorded and
transferred to a specific data basis with correction
factors associated with the instrument included. The
final data will then be inserted into the previously
mentioned mathematical model.
3 CONCLUSIONS
Vertebral Metrics is a non-invasive mechanical
instrument, which assesses the curvatures and lateral
deviations of the spine in a standing position. The
patent was registered and the study with pregnant
women was accepted by the Ethics Committees of
the Maternidade Dr Alfredo da Costa and Regional
Health Administration of Lisbon and Vale do Tejo.
The validation process for the instrument,
including the calculation of the correction factors
and uncertainties associated with it was
accomplished.
This instrument allows a global assessment of the
spine. Thus, identification of dysfunctions and / or
diseases of the spinal column in pregnant women,
will be shown on a thorough diagnosis. Intervention
programs, directly connected to specific problems of
each person, may then be elaborated and
implemented.
Vertebral Metrics was originally planned and
built to be applied to pregnant women. However, it
can be applied to any type of population after
redefining the dimensions of the different parts.
REFERENCES
Alexandre, N.; Moraes, M. (2001) Modelo de avaliao
fisio-funcional da coluna vertebral. Rev. Latino-am
Enfermagem Maro, 9 (2); 67-75.
Knoplich, J . (2003) Enfermidades da coluna vertebral..,
Robe Editorial. So Paulo, 3ed
Harrison et al (2005) Sagittal skin contour of the cervical
spine: interexaminer and intraexaminer reability of the
Flexicurve instrument. Journal of Manipulative and
Physiological Therapeutics. Vol 28, 7; 516-519
Harlick, J . Milosavljevic, S. (2007) Palpation
identification os spinus processes in the lumbar spine.
Manual Therapy 12, 56-62
Hinman, M. (2004) Comparison of thoracic kiphosis and
postural stiffness in younger and older women. Spine
Journal 4; 413-417.
Lee, Y.; Chen, Y. (2000) Regressionally determined
vertebral inclination angles of the lumbar spine in
static lifts. Clinical Biomechanics 15; 672-677.
Norton, BJ ; Ellison, J B. (1993) Realiability and concurrent
validity os the Metrecom for length measurements on
inanimate. Phys. Ther. 73; 266-274.
Teixeira, FA; Caravalho, GA( 2007) Confiabilidade e
validade das medidas da cifose torcica atravs do
Mtodo Flexicurva. Revista Brasileira de Fisioterapia.
Vol.11, 3; 199-204.
Pinel-Giroux, F.; Mac-Thiong, J .; Guise, J .; Labelle, H.
(2006) Computerized assessment of sagittal curvatures
of spine Comparison between Cobb and tangent
circles techiques. J. Spinal Disord. Tech, vol.
19,7;507-512
Walsh e Breen (1995) Reability and validity of the
Metrecom Skeletal Analysis in the assessment os
sagittal plane lumbar angles. Clinical Biomechanics
vol. 10 n4; 222-223
DEVELOPMENT OF A MECHANICAL INSTRUMENT TO EVALUATE BIOMECHANICALLY THE SPINAL
COLUMN IN PREGNANT WOMEN
313
DEVELOPING A PUPILLOMETER
Gonalo Leal
1
, Pedro Vieira
2
and Carlos Neves
3
1
Department of Physics, Science and Technology Faculty, New University of Lisbon, Portugal
2
Institute of Biophysics and Biomedical Engineering, Science Faculty, University of Lisbon, Portugal
[email protected], [email protected]
3
Department of Physiology, Institute of Molecular Medicine, School of Medicine, Lisbon University, Portugal
[email protected]
Keywords: Pupil, Pupillometer, Medical Instrumentation, Ophthalmology.
Abstract: This project presents stable and robust optical equipment for detecting the area of the pupil and its variation
on a temporal scale. An algorithm was developed to detect the pupil contour, implemented in a simple,
intuitive and user friendly interface programme. Using the equipment specifically developed, measurements
were taken of the area, perimeter, horizontal diameter and vertical diameter. After a statistical and
comparative study, it was possible to reach conclusions regarding the general dimensions of the pupil, its
variation prompted by a given stimulus and the clinical viability of the equipment concerned.
1 INTRODUCTION
The area of the pupil changes in response to the
variation in light intensity in the retina, with a view
to assisting the optimizing of visual perception. In
dim light, pupil dilation (midriasis) is an effective
way to maximize the number of photons reaching the
retina, which in turn activates adaptive mechanisms
to low light intensity. When exposed to bright light,
miosis causes an adequate reduction in the intensity
of light in the retina, acting as immediate response to
the mechanisms of adapting to intense light (Kardon,
2003). The study of the way the pupil changes its
size is of relevant clinical interest, for it acts as an
objective indicator of the retinas sensitivity to light
and, as a result, of the optical nerve functionality.
The state of a persons pupil allows for several
diseases to be diagnosed, among which sleep
disturbances (narcolepsy), photophobia,
schizophrenia (pharmacological reaction), Adies
Syndrome, Alzheimers, narcotic addiction, among
others.
2 STATE OF THE ART
Nowadays, Pupillometry is mostly based in
computers. This technology, based on image
processing, numerically analyses the pupil features.
The sensitivity of existing systems varies much and
depends to a major degree on the spatial and
temporal resolution of the acquisition devices and
also on the algorithms used. Pupillometers differ also
in portability and applicability. The majority of
current Pupillometry devices use algorithms that are
based on physical models and that use a circle or an
ellipsoid as an approach to the pupil contour (Kim et
al, 2004).
Recent studies have been made using Fourier
series to determine the pupil contour (Rakshit and
Monro, 2007).
3 MATERIALS
3.1 Material Used
The current system is based on a LE175C
LUMENERA camera that is connected, via Ethernet,
to a personal computer (figure 1). A floodlight of
500W was used as external illumination and as a
stimulator of the subjects pupil. The camera is
placed in a mechanical arm attached to a table. The
arm can be moved vertically and horizontally in a
small area (figure 2a). The subjects head is
positioned in a mechanical support that can be
moved vertically (figure 2b).
314
3.2 System Concept

Figure 1: First version of the Pupillometer System

designed in AutoCAD.

Figure 2: (a) Mechanical arm that moves the camera

horizontally and vertically. (b) Mechanical device where
the subject places his/her head. This device allows vertical
movement.
4 METHODS
4.1 Algorithm
The Matlab programme was chosen to develop all
the software required for the system, since it is a
very resourceful software and has very good imaging
tools and functions.
An algorithm was developed based on the
concept of Intensity Threshold. It starts with an
initial point (calculated by a secondary algorithm
shown in figures 3 and 4) and traces lines to every
direction separated by a 5 degree angle (LI,
Dongheng and Derrik J . Parkhurst, 2006). It then
compares the intensity of gray levels of consecutive
points along the traced line. If the variation of values
is bigger than the Threshold value, the algorithm
stops and defines a point that characterizes the pupil
contour. Then a second line is traced using the
previously calculated point as the new start point,
but in the opposite direction, in order to calculate the
opposite contour point and so speed up the time of
measurement. Once all points are defined, there are
specified algorithms that calculate the area,
perimeter, horizontal diameter and vertical diameter
of the image. The value of Threshold can be
manually inserted or can be calculated by two
algorithms developed for that purpose (Square
Threshold and Circle Threshold).

Figure 3: Analysis of gray intensity in the image. We can

see the pupil zone featured by low levels of intensity. This
is how the programme calculates the initial point for the
beginning of the cycle.

Figure 4: In order to improve the analysis shown in figure

3 we can plot the intensity of an image in a 3D mesh plot.
The pupil is shown in dark blue, since it has the lowest
intensity gray levels. Both images are the same, but image
b) is the result of a) with a Low Pass filter applied. This
feature reduces the error because the eyelashes are also
dark and could be misunderstood as an area of the pupil by
the algorithm.
4.2 Graphic User Interface (GUI)
To put all the algorithms together, a user friendly
(a) (b)
(a)
(b)
DEVELOPING A PUPILLOMETER
315

interface was developed in Matlab, which acquires

all data by filming the eye of the subject. This
interface also measures pupil dynamics by
processing large amounts of images and determines
the statistical data of the results.
There are three main panels: data acquisition
panel, pupil detection panel and statistics panel.
The first panel (figure 5) shows the images that
are being stored in real time. We can also edit the
subjects demographic data and the calibration data.
There are event buttons that store the time when a
stimulus is applied to the subject (the better to
analyse the statistical data).

Figure 5: Data Acquisition Panel.
The second panel (figure 6) is used to measure the
pupil dynamics during a period of time. There is a
menu where the user can easily calculate the
threshold value for automatic detection (figure 7).

Figure 6: Pupil Detection Panel.

Figure 7: a) Square threshold (compares the intensity gray
levels of 5 pupil points to 5 iris points). b) Circle
Threshold (compares a torus of points of the pupil to a
torus of points of the iris, mainly in the pupil/iris border.
The function of the third panel (figure 8) is to view
all the data statistics in two ways: by plots and by
tables. The user can also view the standard deviation
of each variable and save all data to a .txt file that
can easily be exported to other programmes used in
Fourier and Wavelet analysis.

Figure 8: Statistics Panel.

Outside the three panels, the interface has three
buttons, one for exiting the programme, a second one
to view the help file (a .pdf document with the user
guide of the software) and a third to consult the IDs
of all subjects in the database.
4.3 Data Acquisition
For this preliminary study, the right eye of nine
individuals was measured in a dark room where the
only light source was a 500W floodlight. Each
subject was illuminated with a bright light, so that
the pupils accommodation over time could be
detected (first the pupil contracted and then the pupil
dilated over time). Each subjects head was
positioned at a distance of about 25cm from the
camera. The light stimulus was applied to the subject
and didnt change during the measurements.
In order to remove the reflexes from the subjects
eye, no direct illumination was used.
A hundred images were taken for each subject
(for about 10-15 seconds, depending on the speed of
image storing, which is independent of the amount
of data in the camera and also with the processing
speed of the computer used).
Table 1: Acquired data statistical information.
Subject Data Mean
Standard
Deviation
Test 1
Area 14444.80 1509.91
Perimeter 493.35 35.20
Horizontal
Diameter
143.10 7.00
Vertical
Diameter
140.57 7.08
Test 2
Area 11062.80 1508.11
Perimeter 482.12 49.79
Horizontal
Diameter
127.84 6.93
Vertical
Diameter
135.20 10.01
(a) (b)
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
316
Table 1: Acquired data statistical information (Cont).
Test 3
Area 18489.40 1861.11
Perimeter 574.14 32.25
Horizontal
Diameter
159.24 6.30
Vertical
Diameter
162.39 6.38
Test 4
Area 19133.60 1497.48
Perimeter 698.84 68.99
Horizontal
Diameter
163.64 6.70
Vertical
Diameter
163.32 6.27
Test 5
Area 16065.20 1281.32
Perimeter 529.56 47.55
Horizontal
Diameter
153.01 3.09
Vertical
Diameter
150.80 4.58808
Test 6
Area 16112.10 2789.19
Perimeter 571.28 86.97
Horizontal
Diameter
147.74 11.03
Vertical
Diameter
153.64 12.29
Test 7
Area 33239.40 1991.74
Perimeter 1168.96 267.96
Horizontal
Diameter
211.41 3.93
Vertical
Diameter
239.35 25.56
Test 8
Area 10985.10 1161.59
Perimeter 533.17 82.11
Horizontal
Diameter
128.34 4.06
Vertical
Diameter
125.25 5.35
Test 9
Area 29139.60 1974.06
Perimeter 1989.11 916.88
Horizontal
Diameter
197.40 4.75
Vertical
Diameter
248.56 14.49
Note 1: Subjects 7 and 8 used contact lenses on purpose so that
the influence of contact lenses in the measurement could be
studied.
Note 2: All results are in number of pixels, since the zoomlens
changes the size of each pixel.




An example of the output data for one subject is
shown below:

Figure 9: Example of the pupil detection of an amount of
16 consecutive frames. All frames look alike because this
measurement was made at high speed.
Eye color: Brown.
Number of images: 100.
DEVELOPING A PUPILLOMETER
317

Figure 10: Plot of the Area of each frame along all 100
images.

Figure 11: Plot of the Perimeter of each frame along all

100 images.

Figure 12: Plot of the Horizontal diameter of the pupil
along all the thirty images.

Figure 13: Plot of the Vertical diameter of the pupil along
all the 100 images.
Output .txt file
14: 24
03- Oct - 2008
I d: Test 1
Thr eshol d: 3
Ar ea Per i m Di st _h Di st _v
12282. 0 441. 2 133 131
13289. 5 458. 2 134 131
12339. 0 458. 2 132 130
( )
Elapsed processing time: 00:07:14 (hms).
5 RESULTS AND DISCUSSION
The acquired results displayed in table 1 show that
the subjects pupil dilated during the period of
accommodation. It is clear that the pupil size
changed during the acquisition period and, since this
is a preliminary study, the results fit the objectives.
However there is some error caused by the reflex of
visible light in the images taken.
The pupils of subjects 7 and 9 showed more
dilatation, since they suffer from myopia. These
people tend to have a larger pupil, as this study
successfully shows. A careful analysis of the pupils
diameters plots (figures 12 and 13) show that the
pupils fluctuations behave like a sinusoidal curve.
This behaviour is very interesting, for we can predict
the period of the sinusoidal curve.
A detailed analysis of the plots shows that we
can relate the pupils fluctuations to the sympathetic
and parasympathetic nervous flows in the subjects
brain. This feature can be used in the future to
compare the pupils fluctuations in a normal subject
those of the pupil of a subject suffering from a
neurological disorder (Alzheimers or Narcolepsy,
for example).
By analyzing the statistical values of the
processed data we understand that the area and
perimeter algorithms must be optimized, since they
are expressing an error of about 7% and 5% (Matlab
error analysis), respectively.
Because grey scale (256 levels) images were
used, the system has the ability of detecting smaller
variations whenby comparing to some other studies
made in this area, such as IACOVIELLOs (2006).
This feature allows the detection of small variations
in the digital image, so that the error can be reduced
and more information gathered.
6 CONCLUSIONS
This paper presents a complete system that can
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
318
accurately measure the dynamics of pupillary
movements given various stimuli. This is commonly
considered one of the most important parameters to
make non-invasive diagnosis of many neurological
disorders.
This system has a very user friendly interface that
will allow doing clinical trials by people not
specialised in this specific technique.
The acquired results showed some error in the
contour detection. In order to surpass this problem
the detection algorithm must be optimized and a
different type of camera must be used. Since the tests
were made using a camera that works in the visible
wavelengths, there was some light reflex in the
subjects eye. For better results an infrared camera
should have been used, but no such camera was
available at the time of this study.
The results also showed some pupil noise, that is
chiefly due to the sympathetic and parasympathetic
neural flow. The main algorithm has been improved
to a good ratio between statistical error and
processing speed.
It is relevant to say that the detection algorithm
does not use physical models and does not
approximate the pupil contour to any geometrical
figure. The speed of the algorithm is compromised
but the results are more precise.
This study sets out to design a medical
instrument that can be used by any technician or
physician to measure pupil dynamics. For example,
it could be attached to a hospital bed, to monitor the
pupil activity of patients. The interface is compiled
in a .exe file that can easily be installed in every
computer even if the computer does not have Matlab
installed.
6.1 Future Perspectives
To improve the existing system it is clear that it must
evolve into an optical device that works in infrared
light. With this feature the light reflexes in the eye
will not influence the acquired data and the system
will also work in the dark. Using a CCD camera and
an infrared light system, an image of the anterior
surface of the eye can be obtained, even when
external lighting is not present (without interference
from non-controlled stimuli).
We intend to process the acquired data using
Fourier and Wavelet analysis, to work in frequency
and time domains.
The main algorithm will adapt to every image in
such a way that the threshold value will be
calculated for each frame, so that the digital image
features may vary (brightness, contrast and gamma).
In the developed GUI all results are expressed in
pixels, but, to facilitate the physicians work, an
algorithm must be created to convert pixels into
millimeters. However, it will not be easy, since the
system uses a zoom lens and so the pixel size is
dependent of the zoom setting.
We believe that, in the near future this
methodology can be of assistance to Ophthalmology
diagnosis by quantifying the sympathetic and
parasympathetic pupillary dilatation components.
ACKNOWLEDGEMENTS
We thank the Department of Physiology of the
Institute of Molecular Medicine for technical advice
and for sponsoring the project, and the colleagues of
Hangar 4 of FCT-UNL for laboratory assistance, and
the colleagues of the Institute of Biophysics and
Biomedical Engineering for technical advice and
helpful discussions.
REFERENCES
Enderle, J ohn, Susan Blanchard and J oseph Bronzino.
Introduction to Biomedical Engeneering. 2
th
Edition.
San Diego: Academic Press, 2005.
Filipe, J oo A. Capo, F. Falco Reis and J . Castro
Correia. Assessement of autonomic function in high
level athletes by pupillometry. Autonomic Neuro-
science: Basic and Clinical. Vol.104. 2003, pp. 66-72.
Gonzalez, Rafael C., Richard E. Woods and Steven
L.Eddins. Digital Image Processing using MATLAB.
New J ersey: Prentice Hall, 2004.
Hachol, A., et al. Measurement of pupil reactivity using
fast pupilometry. Physiol. Meas. Vol.28. 2007, pp. 61-72.
Howarth, Peter A., Gordon Heron and Louise Wittaker.
The Measurement of Pupil cycling time. Graete's
Arch Clin. Exp. Ophthalmol. Vol.238, 2000, pp. 826-832.
Iacoviello, Daniela. Analysis of pupil fluctuations after a
light stimulus by image processing and neural
network. Computers and Mathematics with
Applications. Vol.53. 2006.
Kardon, Randy. 2003. Pupil. In Adler's Physiology of
the Eye. Edited by Paul L. Kaufman and Albert Alm.
10th Edition. St Louis : Mosby, 713-743.
Kim, J ieun, Kyungmo Park and Gon Khan. 2004. A
Method for Size Estimation of Amorphous Pupil in 3-
Dimensional Geometry. Paper presented at the 26th
Annual International Conference of the IEEE EMBS,
in San Francisco.
Li, Dongheng and Derrik J . Parkhurst. Starburst: A robust
algorithm for video-based eye tracking. openEyes.
http://thirtysixthspan.com/openEyes/publications.html.
Rakshit, Soumyadip and Donald M. Monro. Pupil Shape
Description using Fourrier Series. 2007.
Smith, Shirley A. and S. E. Smith. Pupil function: tests
and disorders, in MATHIAS, Christopher J . and Sir
Roger Bannister, Autonomic Failiure: A Textbook of
Clinical Disorders of the Autonomic Nervous System.
4
th
Edition. Oxford : Oxford University Press, 2001.
DEVELOPING A PUPILLOMETER
319
QUALITY ASSESSMENT IN COLONOSCOPY
New Challenges Through Computer Vision-based Systems
Fernando Vilari no
Computer Vision Center and Computer Science Dep. Universitat Aut` onoma de Barcelona, Spain
[email protected]
Gerard Lacey
Graphics, Vision and Visualisation Group (GV2), School of Computer Science and Statistics, Trinity College Dublin, Ireland
[email protected]
Keywords:
Colonoscopy, Quality assessment, Eye-tracker, Computer vision, Polyps, Colon cancer.
Abstract:
The assessment of the quality of the colonoscopic interventions arises as a most relevant issue once the number
and the availability of these clinical procedures are increased day by day. The use of the latest computer vision-
based techniques can provide the physician with both qualitative and, most important, objectively veriable
quantitative indicators of performance. In this paper we present a study in which we propose the automatic
analysis of colonoscopy video for the quality assessment of the intervention from different points of view:
1) We propose the characterization of the different parts of the colon in order to obtain metrics of the time
used for navigation, portion of gut analyzed, etc. 2) We analyze the image contents in order to automatically
characterize the presence of polyps. 3) We use the information obtained by and eye-tracker in order to assess
the physicians skills.
1 COLON CANCER AND
INTESTINAL SCREENING
1.1 Colon Cancer in Numbers
The main lesions associated to the intestine are:
bleeding, lump, ulcer, Crohn disease, and cancer.
During the last 20 years, colon cancer has been
the second leading cause of cancer deaths in the
United States, behind lung cancer, with approxi-
mately 60, 000 deaths per year as shown in OBriens
report (OBrien et al., 1990), and also analyzed in
other studies (U.S. Department of Health and Human
Services, 2003). America Cancer Societys 2007 re-
port (American Cancer Society, 2007) provides an ex-
tended summary of facts and statistics about colon
cancer prevalence and impact in the population. Col-
orectal cancer is the second leading cause of cancer-
related deaths in Singapore and Europe (Ministry of
Health Singapore, 1990). Although colon screening
has become the main alternative for prevention of col-
orectal cancer, recent data suggest that there is a sig-
nicant miss-rate for the detection of even relatively
large polyps and cancer (Pabby et al., 2005). For this
reason, special efforts have been focused on the devel-
opment of computer-aided systems for the detection
of this type of pathologies. Nowadays, novel lines
of research are oriented to widen this perspective by
means of the implementation of objective indicators
for the assessment of the procedures used in colon
screening, since the miss rate of polyps is highly cor-
related to elements such as the quality of the prepa-
ration, the time consumed for the intervention, the
amount of intestinal surface screened, the kills of the
physician in the manipulation of the endoscope, etc.
Moreover, this indicators present a potential value for
the training of future endoscopists as reference values
to measure abilities in and objective way.
1.2 Screening Technics
Intestinal endoscopy is referred to as the technique
for screening the intestinal lumen. For the case
of large intestine, this technique receives the name
of colonoscopy. Fiberoptic colonoscopy (FOC) is
widely accepted as the denitive method for diagno-
sis of colonic polyps. FOC allows direct visualiza-
tion of the intestinal surface and affords the possibil-
ity of obtaining a in-situ biopsy as well as cauteriza-
320
tion and clinical intervention such as polyp removal
(Winawer et al., 1997). FOC is a minimal invasive
surgery (Hunter and Sackier, 1993), (Hulka and Re-
ich, 1994) consisting of the introduction through the
anus of a exible probe which a camera and an il-
lumination device on its tip. The probe consists of
a exible cable which can be controlled by the ex-
pert in order to reach every part on the intestinal wall.
There exist several reference books regarding clini-
cal colonoscopy for the reader interested in deepen-
ing in the specicities of colonoscopy, amongst which
Katos textbook (Kato and Baron, 2003) can provide
an insightful introductory view. Several authors have
assessed that endoscopic images possess rich infor-
mation (Nagasako et al., 1998), which facilitates the
abnormality detection by multiple techniques (Zheng
et al., 2005). The main drawbacks related with FOC
can be enumerated as follows (Winawer et al., 1993),
(Eddy, 1990): risks of perforation; costs of the inter-
vention; difculty in visualizing the 100% of the in-
testinal surface; high number of patients for a reduced
number of specialists provide a stress and shrink in
the intervention time; difcult visualization due to
the intestinal content; preparation needed; imprecise
localization of events for a subsequent interventions,
and high rate of negative results in inspections look-
ing for polyps (Gokturk and Tomasi, 2001).
In the last years, other modalities for the visual-
ization of the colon have arisen. One of the most
relevant is virtual colonoscopy, a technique consist-
ing of the construction of a virtual 3D model of the
colon from computed tomography (CT) data (Liang
et al., 2004), and for this reason several authors have
proposed automatic techniques for the detection of le-
sions in this modality of images. On the other hand,
Wireless Capsule Video Endoscopy (WCVE) (Iddan
and Meron, 2000) has devoted particular attention, re-
cently. WCVE (Fireman et al., 2002) consists of a
capsule with a camera, a battery and a set of led lamps
for illumination attached to it, which is swallowed by
the patient, emitting a radio frequency signal which is
received and stored in an external device. The result
is a video movie which records the trip of the capsule
along the intestinal tract with a rate of two frames
per second, and that can be easily downloaded into
a PC with the camera software installed. It is much
less invasive, since the patient simply has to swallow
the pill, which will be expelled in the normal cycle
through defecation. Moreover, there is no need of
hospitalization nor expert support through the process
and the patient can lead an ordinary life, since the at-
tached device is recording the video movie emitted
by the camera in the capsule (Vilari no, 2006; Vilari no
et al., 2009). However, although these new modalities
provide new ways of clinical analysis, colonoscopy is
the reference technique for clinical intervention in the
case of colon screening and colon cancer.
2 COMPUTER VISION FOR
QUALITY ASSESSMENT
We state that computer vision-based techniques can
be used in order to obtain objective indicators of the
quality of the interventions in an automatic way. Our
approach provides a framework for the quality assess-
ment which is implemented in the following 3 areas
in which we are developing our research, namely: 1)
Computer-aided intervention: On-line detection and
characterization of potential targets in intervention-
time, 2) Post-interventional quality metrics: Auto-
matic computation of the quality measures related
to the intervention such as quality of preparation,
amount of bowel surface visualized, time measures,
etc. 3) Evaluation of skills: Analysis of 3D trajec-
tories of the endoscopes and the screening behavior.
For this nal point, we propose to use the trajecto-
ries together with the information obtained from the
tracking of the gaze position of trainees in order to
assess their skills. The perspective presented in this
paragraph is graphically depicted in Figure 1.
Figure 1: Our framework for automatic quality assessment
in colonoscopy.
3 COLON CANCER: CLINICAL
CHARACTERIZATION
Colon cancer has devoted wide attention in many
studies in order to nd proper descriptions and cat-
egorization (Rembacken et al., 2000), (Saitoh et al.,
2001), (Paris Workshop Participants, 2003). Adeno-
matous polyps, particularly those larger than 1 cm in
QUALITY ASSESSMENT IN COLONOSCOPY - New Challenges Through Computer Vision-based Systems
321
diameter, are the most likely precursors of colorectal
carcinoma (Gokturk and Tomasi, 2001), (Thoeni and
Laufer, 1994). The main features used for cancer and
general abnormality characterization are: color, shape
and texture.
Color. Color colonoscopic images tend to ex-
hibit the same color features for the same colon
status (Kato and Baron, 2003). Malignant tu-
mors are usually inated and inamed and this in-
ammation is usually reddish and more severe in
color than the surrounding tissues. Benign tumors
exhibit less intense hues. Redness may specify
bleeding and black may be treated as deposits due
to laxatives. Green may be the presence of fae-
cal materials, which are not clear during the pre-
operative preparation, and yellow relates to pus
formation (Tjoa and Krishnan, 2003).
Shape. Shape is a relevant cue since polyps are
associated to rounded or peduncular shapes. Pe-
duncular polyps are relatively easy to visualize
during a screening session. Flat polyps present
a higher difculty, and in addition, they are more
likely to develop into malignant polyps.
Texture. Texture is known to be an important
cue to be evaluated for the discrimination between
malignant and benign lesions (Kudo and Kashida,
2000), (Nagata et al., 2000).
The high-level characterization of cancer ex-
plained above is usually translated into an image-
based feature extraction stage, focused on color, tex-
ture or shape cues. Following the feature extraction,
a discrimination procedure, based on simple compar-
isons or more sophisticated machine learning tech-
niques must be applied. In our approach, efcient
methods for color, texture and shape characteriza-
tion must be oriented towards the high speed require-
ments of on-line procedures. The use of histogram
quantization (HQ) (Swain and Ballard, 1991) and the
use of different color spaces (Paschos, 2001) in a
whole-image level (Hai et al., 2006) or a multi-scale
framework (Li et al., 2005) for color; Gray level
co-occurrence matrices (GLCM) (Srivastava et al.,
2005), fractal dimension (Chaudhuri and Sarkar,
1995), histograms of oriented gradients (HOG) (Won
et al., 2002) and wavelets (Karkanis et al., 2001) for
texture; and MPEG-7 descriptors and others for shape
(Coimbra and Cunha, 2006) should be adapted to run
online in order to provide a efcient characteriza-
tion of our system. This orientation towards the real
time performance of discriminant features is one of
the most relevant challenges of our current line of re-
search.
Figure 2: Three polyps: at, peduncular and mixed. Pedun-
cular polyps are prone to their development into malignant
cancers.
In recent works we presented preliminary results
of discriminative features and classication systems
for colon cancer detection and its a posteriori char-
acterization in different types of polyps. Figure 2
shows examples of a) peduncular, b) at and c) mixed
polyps, which have a different degree of clinical rel-
evance, as appearing in our latest contribution (Vi-
lari no et al., 2007). Our next step is, hence, to
put these techniques in the horizon of real-time per-
formance. In addition, we argue that the use of
mixed strategies (color/texture/shape) for characteri-
zation would potentially provide a reacher informa-
tion than each of them alone, and the study of the
selection of appropriate features represents an open
eld of research.
4 POST-INTERVENTIONAL
QUALITY METRICS
We state that computer vision-based techniques can
be used in order to provide post-interventional metrics
in an automatic way. We dene four main lines or
research in this area:
1. Automatic estimation of presence of intestinal
content for the assessment of the quality of the
preparation.
2. Percentage of the intestine visualized.
3. Measure of the intervention time.
4. Automatic detection of the different moments
of the intervention (introduction of the endo-
scope/withdrawal).
5. Characterization of the motion of the endoscope.
All these points can be gathered into a general
framework which could be stated in the following
way: The denition of a 2D map of the patients gut
in order to have patient-oriented representation of the
colon, together with their singular features, lesions,
etc. In order to get this wide target, color, textural and
motion features must be used together. Color appears
as a main cue for intestinal content characterization
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
322
Figure 3: Eye-tracker conguration system.
(Vilari no et al., 2006). Texture can provide a descrip-
tion of the intestinal folds, and motion characteriza-
tion is essential for the measure of transition times. In
a addition, deeper studies must be performed in order
to build up a large database of cases which allows us
to carry out statistical test with the aim of unveiling
the most suitable features and indicators for quality
measures.
5 EYE-TRACING INFORMATION
FOR SKILL ASSESSMENT
The last line of research we introduce in this paper is
related to the assessment of the skills of the trainees
in endoscopy training programs. We hold that using
both the information of the trajectories of the endo-
scopes -which can be potentially obtained by means
of the analysis of the camera movement- and the
tracking of the gaze position -which can be obtained
by an eye-tracker device-, we can provide objective
indicators for the evaluation of the skills. Moreover,
it will be possible to characterize those skills sepa-
rately: ability in endoscope manipulation vs. visual
search.
An eye tracker consists of an adjustable device
with a set of cameras pointing towards the users eyes.
The eye tracking procedure involves the recording of
the gaze position and the translation of this values into
the location on the image where the user is focusing
the attention. Figure 3 shows a scheme of the EyeLink
II eye-tracker which was used for our experiments.
The video signal from the cameras is sent to a host
PC in which the eye tracker software is installed, and
which communicates to the experiment PC in which
the visual stimuli are shown. The pupil position is
then translated into gaze position on the stimuli screen
at a very high speed (up to one record each four mil-
liseconds).
6 RESULTS
In this position paper, we would like to highlight pre-
liminary results which were obtained in the colon
cancer characterization and eye-tracker based analy-
sis. The computation of automatic quality assessment
measures is a key part of our ongoing research, and
it will be analyzed in more detail in the Discussion
section.
The eye-tracker information can be used to la-
bel massive data by using the gaze position of the
specialist while screening the colonoscopy video off-
line (in a post-interventional session). Those areas
the specialist is steering the sight towards are asso-
ciated to visual salient features which are of inter-
est for the colonoscopists. This has two main con-
sequences: On the one hand, we can build up an an-
notated database of clinical cases in which the posi-
tion of the observers gaze determines the position of
the cancer screened. On the other hand, we can use
the information of the the gaze trajectory to charac-
terize the abilities of the experts in the visual search.
In our case, we built up a database consisting of 6
cases, which is freely accessible for the public and
from the following link: hhtp://www.cs.tcd.ie/
colon/colon_et_database.zip.
We used this database to for the experiments as-
sociated to the colon cancer detection and character-
ization. We applied support vector machine classi-
ers (SVM) (Vapnik, 1995) in order to distinguish
between polyp images and random images of 6 dif-
ferent videos. Each image was 500x700 pixels. We
manually selected those frames were the polyps were
present, and then we trained the classier with exam-
ples of polyps and non-polyps patches of 128x128
pixels by using the gray-level image, achieving a
working point of around 80% of both sensitivity and
specicity. The visual analysis of these data pro-
vided promising results in the clustering of the dif-
ferent types of polyps into three basic types (pedun-
cular, at and mixed). For a further analysis of these
results we refer to the IbPria conference proceedings
(Vilari no et al., 2007).
Regarding the analysis of the visual trajectories
obtained from the eye-tracker, we carried out a new
set of experiments in which 20 full colonoscopy stud-
ies were annotated in a two-stage strategy: 1) First,
the expert manually selected those sections of the
video study in which the expert detected the presence
of cancer -this action was performed by clicking and
holding a mouse during the visualization of the polyp-
. 2) Then, only those parts of the video selected by the
specialist in the previous stage were screened with the
eye-tracker. We repeated this experiment with differ-
QUALITY ASSESSMENT IN COLONOSCOPY - New Challenges Through Computer Vision-based Systems
323
ent experts and novices in order to look for differences
in annotation. The results, which are to be deeply ana-
lyzed in a current study to be submitted in a forthcom-
ing publication, point out that experts and trainees
show different behaviors in the visual search in terms
of reaction time, frequency of saccades, geometry of
the 3D trajectories of the gaze position, etc. These
preliminary results show statistical signicance, and
they must be contrasted in terms of inter- and intra-
observer variability. This is one of the most relevant
drawbacks of this kind of studies since, the time avail-
ability of the physicians is a major constraint for large
validation studies.
7 DISCUSSION
The preliminary results shown above pave the way
to the computation of quality assessment measures.
For the case of polyps detection, the main aim is to
provide the physician with candidates of polyps to
be analyzed during the intervention. Different statis-
tics, such as the number of polyp-candidates analyzed
and not analyzed, total time consumed in this analy-
sis, etc., can be pulled out in order to obtain objective
indicators. These statistics and indicators must be de-
ned together with the physicians in order to get clin-
ical signicance and the appropriate tolerance levels.
For the case of eye-tracker data, statistically sig-
nicant indicators of the physicians visualization
skills are of relevant importance in a two-fold way:
First, we would be able to provide the physician with
objective metrics that measure high level skills, such
as reaction time, search activity, robustness of the
search pattern, etc., together with general indicators
such as miss rate of polyps. In addition to the for-
mer, we make it possible to decouple the manoeuver-
ing ability from the visual search skills in an objective
way. This information, together with the motion anal-
ysis, can provide indicators regarding the smoothness
of the trajectories that the endoscope performs.
Finally, the automatic estimation of presence of
intestinal content, the quality of the preparation, the
percentage of the intestine visualized, the measure of
the intervention time, and the automatic detection of
the introduction and withdrawal stages must be put
into a clinical framework of quality assessment. This
provides both a study-based assessment and, into a
historical archive, a log of the physicians indicators
along different interventions. In order to get this done,
a full study is being performed currently together with
the Royal College of Surgeons of Ireland, funded by
an Enterprise Ireland project of the Irish government
for the acquisition and annotation of a full database
of 100 cases of colonoscopy showing colon cancer in
high denition videos. This database comprises sev-
eral tera bytes of video data and its corresponding an-
notation both in manual and eye-tracked versions by
several specialists with different levels of expertise.
8 CONCLUSIONS
Quality assessment of colonoscopy videos is a rele-
vant issue, since the evaluation of different aspects
of the intervention, by providing objective indicators,
sets up the foundations for a control and reduction of
miss rates in colon cancer detection. We proposed an
approach of quality assessment by means of computer
vision-based techniques which is underpinned by: 1)
The automatic suggestion of potential candidates of
colon cancer during the intervention time, 2) The au-
tomatic computation of objective quality metrics after
the intervention, and 3) The use of eye-tracking in-
formation in order to provide metrics to evaluate the
skills of the physician both in the visual search and
in the endoscope manoeuvering. Preliminary results
showed the suitability of such techniques for polyp
detection and experts vs. trainees discrimination. The
deeper analysis of quality metrics and their correla-
tion is devoted to a further piece of research whose
study our team is carrying out currently.
ACKNOWLEDGEMENTS
We would like to acknowledge the collaboration of
M.D Steven Patchett and M.D Hugh Mulcahy, from
the Royal College of Physicians of Ireland, for their
collaboration in the clinical study of this work. This
research is supported by Enterprise Ireland contracts
PC-2006-038 and CFTD-2006-216.
REFERENCES
American Cancer Society (2007). Cancer Facts and Figure.
American CancerSociety.
Chaudhuri, B. and Sarkar, N. (1995). Texture segmentation
using fractal dimension. IEEE Trans. Pattern Anal.
and Mach. Intel., 17:7277.
Coimbra, M. and Cunha, J. (2006). MPEG-7 visual descrip-
tors: Contributions for automated feature extraction
in capsule endoscopy. IEEE Transactions on Circuits
ans Systems for Video Technology, 16(5):628637.
Eddy, D. (1990). Screening for colorectal cancer. Annals of
Internal Medicine, 113:373384.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
324
Fireman, Z., Glukhovsky, A., et al. (2002). Wireless cap-
sule endoscopy. Israel Medical Association Journal,
4:717719.
Gokturk, S. and Tomasi, C. o. (2001). A statistical 3-d pat-
tern processing method for computer-aided detection
of polyps in ct colonography. IEEE Transactions on
Medical Imaging, 20(12):12511260.
Hai, V., Echigo, T., et al. (2006). Adaptive control of video
display for diagnostic assistance by analysis of cap-
sule endoscopic images. In ICPR, volume III, pages
980983.
Hulka, J. and Reich, H. (1994). Textbook of Laparoscopy.
W.B. Saunders, Phipladelphia, PA, 2 edition.
Hunter, J. and Sackier, J. (1993). Minimally invasive
surgery. McGraw-Hill, New York.
Iddan, G. and Meron, G. e. a. (2000). Wireless capsule
endoscopy. Nature, 405:47.
Karkanis, S., Iakovidis, D., et al. (2001). Detection of le-
sions in endoscopic video using textural descriptors on
wavelet domain supported by articial neural network
architectures. In ICIP, volume II, pages 833836.
Kato, H. and Baron, J. (2003). Electronic Videoendoscopy.
Jpapan: Harwood Academic Publisher.
Kudo, S. and Kashida, H. e. a. (2000). Colonoscopic diag-
nosisi and management of nonpolypod and early col-
orectal cancer. World Journal of Surgery, 24(9):1081
1090.
Li, P., Chan, K., et al. (2005). Learning a multi-size patch-
based hybrid kernel machine ensemble for abnormal
region detection in colonoscopic images. In CVPR,
volume II, pages 670675.
Liang, J., Higgins, W., et al. (2004). Introduction to the spe-
cial section on virtual endoscopy. IEEE Transactions
on Medical Imaging, 23(11):13331334.
Ministry of Health Singapore (1990). MOH Annual Report.
Ministry of Health Singapore.
Nagasako, K., Fijimori, Y., et al. (1998). Atlas of gas-
troenterologic endoscopy by High-Resolution Video-
endoscopy. Igaku-Shoin Ltd. Tokyo.
Nagata, S., Tanaka, K., et al. (2000). Pit pattern diag-
nosis of early colorectal carcinoma by magnifying
colonoscopy: clinical and histological implications.
International Journal of Oncology, 16(5):927934.
OBrien, M., Winawer, S., et al. (1990). The national
polyp study: Patient and polyp characteristics asso-
ciated with high-grade dysplasia in colorectoral ade-
nomas. Gastroenterology, 98:371379.
Pabby, A., Schoen, R., et al. (2005). Analysis of colorec-
tal cancer occurence during surveillance colonoscopy
in the dietary prevention trial. Gastrointestinal en-
doscopy, 61(3):385391.
Paris Workshop Participants (2003). The Paris endospcopic
classication of supercial neoplastic lesions. Gas-
trointestinal Endoscopy (Supplement), 58(6):323.
Paschos, G. (2001). Perceptually uniform color spaces for
color texture analysis: empirical evaluation. IEEE
Trans. Image Processing, 10:932937.
Rembacken, B., Fuji, T., et al. (2000). Flat and de-
pressed colonic neoplasms: A pospective study of
1000 colonoscopies in the uk. Lancet, 17:11571167.
Saitoh, Y., Waxman, A., et al. (2001). Prevalence and dis-
tinctive biologic features of at colorectal adenomas
in a north american population. Gastroenterology,
120:16571665.
Srivastava, S., Rodriguez, J., et al. (2005). Analysis of con-
focal microendoscope images for automatic detection
of ovarian cancer. In ICIP, volume I, pages 1113
1116.
Swain, M. and Ballard, D. (1991). Color indexing. Journal
of computer vision, 7(1):1132.
Thoeni, R. and Laufer, I. (1994). Textbook of gastrointesti-
nal radiology, chapter Polyps and Cancer, page 1160.
Philadelphia, PA: Sanders.
Tjoa, M. and Krishnan, S. (2003). Feature extraction for the
analysis of colon status from the endoscopic images.
BioMedical Engineering OnLine, 2(9).
U.S. Department of Health and Human Services (2003).
Cdc, the importance of prevention and early detection
cdc. http://www.cdc.gov/cancer/colorectal/.
Vapnik, V. (1995). The nature of statistical learning theory.
Springer.
Vilari no, F. (2006). A machine learning approach for in-
testinal motility assessment with capsule endoscopy.
PhD thesis, Dept. Computer Science, Universitat Au-
tonoma de Barcelona, Spain.
Vilari no, F., Lacey, G., et al. (2007). Automatic labeling
of colonoscopy video for cancer detection. LNCS,
4477:290297.
Vilari no, F., Spyridonos, P., et al. (2006). Automatic de-
tection of intestinal juices in wireless capsule video
endoscopy. In Pattern Recognition, ICPR, volume 4,
pages 719 722.
Vilari no, F., Spyridonos, P., et al. (2009). Intestinal motil-
ity assessment with video capsule endoscopy: Au-
tomatic annotation of phasic intestinal contractions.
IEEE Trans. on Medical Imaging (in press).
Winawer, S., Fletcher, R., et al. (1997). Colorectal cancer
screening: Clinical guideliness and rationale. Gas-
troenterology, 112:594642.
Winawer, S., Zauber, A., et al. (1993). Prevention of col-
orectal cancer by colonoscopy polypectomy. New
England Journal of Medicine, 329:977981.
Won, C., Park, D., et al. (2002). Efcient use of MPEG-7
edge histogram descriptor. ETRI, 24(1):2330.
Zheng, M., Krishnan, S., et al. (2005). A fusion-based clin-
ical decision support for disease diagnosis from endo-
scopic images. Computers in Biology and Medicine,
35(3):259274.
QUALITY ASSESSMENT IN COLONOSCOPY - New Challenges Through Computer Vision-based Systems
325
DEVELOPMENT OF A SLEEP MONITORING SYSTEM WITH
WEARABLE VITAL SENSOR FOR HOME USE
Takuji Suzuki
1
, Kazushige Ouchi
1
, Ken-Ichi Kameyama
1
and Masaya Takahashi
1,2
1
Humancentric Laboratory, Corporate Research & Development Center, Toshiba Corporation, Japan
2
National Institute of Occupational Safety and Health, Japan
{takuji1.suzuki, kazushige.ouchi, kenichi.kameyama}@toshiba.co.jp, [email protected]
Keywords: Daily sleep care, Sleep monitoring, Wearable sensor and Data viewer.
Abstract: This paper describes a new sleep monitoring system for home use. The basic system consists of a wearable
physiological sensor and PC software for analyzing sleep quality from users wrist motion and heart rate
variability. Different from a conventional sleep monitoring device used in a hospital, the sensor is so small
and easy-to-use that a normal person can use it at home. This means that the system is useful for a sleep
specialist who wants to check a patient's daily sleep pattern. The system can also be used for self-care. We
have developed a wrist-watch-shaped physiological sensor that monitors users wrist motion and pulse wave
interval. We have also developed the algorithm for computing the quality of sleep from these physiological
data on PC. Although sleep is a kind of brain activity and our sensor can not directly measure it, the output
of our algorithm is close to medically evaluated sleep quality. We performed dozens of comparison
experiments and found that its accuracy was about 73.5% on average. The value of the accuracy is enough
for assessing a normal persons sleep quality.
1 INTRODUCTION
In recent years, many people have been suffering
from sleep disorder caused by mental stress,
irregular lifestyle or shift work. However, it is not
easy to determine the quality of sleep because deep
sleep is not always good sleep and shallow sleep is
not always bad sleep. For example, it is natural that
a person cannot sleep well because of jet lag.
However, a person who is always sleepy in the
daytime for a period exceeding one month might
have a health problem. Therefore, it is important for
a doctor to check a patients sleep habits for several
days in order to diagnose and cure his/her sleep
disorder properly. Moreover, it is necessary for a
person to check his/her own sleep habits and to
change his/her lifestyle (self-care).
However, there is no good system to record and
analyze daily sleep. For example, most medical
sleep sensors, such as those employed for
polysomnography (PSG), are for recording many
kinds of physiological data (EEG, EMG, EOG and
so on) for only one or two nights, not for recording
sleep habits with natural state in daily life. It is also
too difficult for a normal person to handle PSG at
home because it involves the use of many electrodes
for measuring the physiological data. A doctor can
attempt to learn a patient's sleep habits by
interviewing him or her, but this is an inherently
unreliable approach. A simple and easy-to-use sleep
monitoring system that can be used in the home is
strongly desired in order to get objective data on
sleep habits.
In order to develop such a system, we have
created a wrist-watch-shaped wearable physiological
sensor that monitors users wrist motion and pulse
wave intervals (Pulse-to-Pulse Intervals: PPIs). The
sensor can be made small and simple because wrist
motion and pulse wave can be easily measured
compared to the case of using PSG. We have also
developed the algorithm for computing the quality of
sleep from these physiological data. The algorithm
can distinguish sleep stage (wake /REM /NREM
[shallow /deep]) using the relationship between
autonomic nervous activity and sleep stages.
Although sleep is a kind of brain activity and our
sensor cannot directly measure it, the output of our
algorithm is close to medically evaluated sleep
quality.
In the following sections, the way of expressing
sleep data, related works, our systems hardware and
software, and the validation result of the sleep
estimation are discussed.
326

2 SLEEP DATA
Generally speaking, sleep is a kind of brain activity
and its purpose is recovery from brain fatigue.
Therefore, sleep state is measured mainly by EEG,
and is classified into several stages. Sleep state is
roughly divided into REM (rapid-eye movement)
sleep and NREM (Non-REM) sleep. NREM sleep is
divided into 4 stages. Stages 3 and 4 of NREM sleep
are so called deep sleep, and stages 1 and 2 are
shallow sleep. These stages are decided by a sleep
specialist using PSG data (Rechtschaffen, 1968), and
their change is shown in a graph called a hypnogram
(shown in Figure 1).
A doctor mainly uses a hypnogram for
evaluating a persons sleep quality. For example, the
doctor checks the quantity of deep sleep if a patient
complains about oppressive drowsiness in the
daytime. If the patient frequently wakes up in the
night and experiences difficulty in breathing, he/she
might be suffering from sleep apnea syndrome. If
REM sleep always occurs soon after falling asleep,
there might be a problem concerning the patients
nervous system. From the viewpoint of healthcare, it
is important to check the balance of deep sleep,
REM sleep or sleep cycle. Therefore, a sleep
monitoring system for home use can also show the
result of one nights data in a graph similar to a
hypnogram.







Figure 1: Sleep hypnogram.
There are many studies on the relationship between
physiological parameters and sleep stages. For
example, Baharav et al. stated that autonomic
nervous activity level derived from heart rate
variability (HRV) during sleep changes in response
to the sleep stages (Baharav, 1995). A value of
LF/HF shows the activity of the sympathetic nerve.
During a REM sleep, a value of LF/HF and the
variability of that are large, and the value of LF/HF
decreases during a NREM sleep, particularly in the
case of deep sleep (Slow Wave Sleep). Since the
brain stem controls both the cerebrum and the
autonomic nervous system, it may be possible to
estimate the sleep stage using HRV.
3 RELATED WORKS
A number of trials have been conducted with a view
to developing sleep monitors for home use. For
example, body/wrist motion has been used for
wake/sleep identification. The amount of activity
(number of subtle wrist motions per minute) measured
from acceleration sensors is often used for
monitoring wake/sleep rhythms (Sadeh, 1989)
although the sleep stages (ex. REM sleep / NREM
sleep) cannot be determined from the data.
More recently, researchers have focused on
measuring heart/pulse rate and analyzing its
variability: HRV (Watanabe, 2004, Michimori, 2003
and Wakuda, 2007). The sleep stages can be
calculated from HRV if the indices of HRV are
properly mapped for the sleep stages.
However, there are two problems in this
approach. One is that body/wrist motion often
disturbs heart/pulse sensing and the HRV value can
not be calculate correctly. The other is that the level
of autonomic nervous activity differs according to
age, sex and body/mental condition. For example,
the autonomic nervous system of the young is
generally more active than that of the old. Sleep
stages cannot be classified using static thresholds.
Our sensor measures both pulse wave interval
and wrist motion. The wrist motion data are used not
only for counting the amount of activity, but also for
detecting errors in HRV data. This solves the first
problem mentioned above.
For the second problem, we employ a statistical
method for deciding sleep stages (Suzuki, 2007). We
assume that there are several stages in a certain
period of sleeping time since the sleep stage
cyclically repeats about every 90 minutes. It means
that the data of autonomic nervous activity can be
classified into several groups if we have any 90-120
minute dataset. In this way, the thresholds for
dividing sleep stages are changed flexibly along with
the dataset.
4 THE OVERVIEW OF THE
SYSTEM
4.1 Wearable Physiological Sensor
Figure 2 shows our wearable physiological sensor.
The size of the sensor is 50mm*60mm*13mm and
the weight is only 35g. A rechargeable battery is
used as an electrical power source. It is possible to
measure physiological data for over 40 hours after
Wake
REM
Stage1
Stage2
Stage3
Stage4
DEVELOPMENT OF A SLEEP MONITORING SYSTEM WITH WEARABLE VITAL SENSOR FOR HOME USE
327

full charge. The sensor incorporates a photoelectric
pulse wave sensor and a 3-axis accelerometer.
Besides, it has an external pulse wave sensor.
Therefore, pulse waves can be measured on the
users wrist or on his/her finger, depending on
his/her preference. The front panel serves as a wrist-
watch displaying date and time, and as a sensor
displaying time and the amount of activity. The
sensor has only two buttons; namely, one is a light
switch, and the other is a switch to start/end sensing.

Figure 2: Wearable physiological sensor.
The sensor measures pulse waves and accelerations
on a users wrist and stores the computed pulse-to-
pulse intervals (PPIs) and the amount of activity in a
flash memory (4MiB). Both analog and digital filters
are used to remove the fluctuations of the amplitude
and the basal line of pulse waveform, which makes
PPIs more precise. As the size of the data measured
in one night (7 hours) is 256 KiB, the sensor can
store almost 2 weeks data in the flash memory.
The sampling rate of the pulse wave and 3-axis
accelerations is 64Hz. However, the resolution of the
PPI is 0.1 ms by using linear interpolation to detect
pulse peak.
The amount of activity is calculated as the
number that the scalar of the 3-axis acceleration is
larger than 0.01 G, which is the same as Actigram
(Cole, 1992).
The stored PPIs and amount of activity data are
sent to PC via USB.
We evaluated the performance. Firstly, the
correlation coefficient between the amount of
activity counted by the sensor worn on the left
forearm and that measured by an actigraph
(Micromini-Motionlogger Actigraph, Ambulatory
Monitoring Inc.) worn on the right forearm during
sleep was 0.95 (average of 3 healthy subjects).
Figure 3 shows an example of the result.

Figure 3: Actigram during sleep. Upper graph shows the
result measured by Actigraph and lower graph shows that
measured by our sensor.
Besides, the correlation coefficient between the PPIs
computed by the pulse wave measured by the sensor
and the R-R intervals computed by a simultaneously
measured electrocardiogram during sleep was
evaluated. Single-channel ECG was measured by
CM5 lead using PSG (Polymate AP1124, TEAC
Corporation, sampling rate: 1 kHz) simultaneously
with the PPI measured by our sensor. R-R intervals
were computed using commercially available R-R
interval analysis software for the PSG (NoruPro
Light Systems, J apan). The correlation coefficient is
0.96 (average of 3 healthy subjects). Figure 4(a)
shows the correlation plot, and (b) shows the Bland
& Altman plot between R-R intervals of ECG and
PPIs.

These values are accurate enough to use the sensor
as a medical device.
4.2 PC Software
Figure 5 shows the flow of the algorithm for
computing sleep stages from the data of PPIs and the
amount of activity.
We employ Coles algorithm for wake/sleep
identification from the amount of activity data (Cole,
1992). This algorithm cannot determine wake/sleep
in real time, but its accuracy is about 90%. At the
same time, the indices of autonomic nervous activity
are derived from frequency analysis of the
variability of PPIs. Firstly, sampled PPIs dataset in
a minute is interpolated at even intervals by cubic
spline interpolation by the minute. Next, Fast
Fourier Transformation (FFT) is executed for the
even-interval PPIs to get the frequency spectrum. In
the frequency domain, the integral value of the
power from 0.04Hz to 0.15Hz is called LF (low
frequency), which shows both sympathetic and
parasympathetic nervous activities. The integral
value of the power from 0.15Hz to 0.4Hz is called
HF (high frequency), which shows parasympathetic
nervous activity. Therefore, we can get the balance
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
328

R -R intervals of EC G (s)
P
u
l
s
e
-
t
o
-
P
u
l
s
e

I
n
t
e
r
v
a
l
s

(
s
)
1.50 1.25 1.00 0.75 0.50
1.50
1.25
1.00
0.75
0.50

(a)
A verage of P -P interval and R -R interval of EC G (s)
D
i
f
f
e
r
e
n
c
e

i
n

P
-
P

i
n
t
e
r
v
a
l

a
n
d

R
-
R

i
n
t
e
r
v
a
l

o
f

E
C
G

(
s
)
1.5 1.3 1.1 0.9 0.7 0.5
0.5
0.4
0.3
0.2
0.1
0.0
-0.1
-0.2
-0.3

(b)
Figure 4: Correlation plot (a) and Bland & Altman plot (b)
between R-R intervals of ECG and PPIs.
of sympathetic and parasympathetic nervous activity,
which is related to sleep stages as we mentioned
above. In order to classify the sleep stages from the
dataset of LF and HF values, the k-means clustering
method is adopted. Firstly, REM/NREM sleep is
divided from 2-hour data set, and then, shallow/deep
sleep is divided from its NREM dataset.
We developed sleep analysis software on
Windows XP/Vista using this algorithm. The
program was coded and compiled by Visual Basic
Ver.6 (Microsoft Corp.).


Figure 5: Algorithm for computing sleep stages.
Figure 6 and 7 are the picture image of the
software. Figure 6 is one nights data, which shows
pulse rate, variability of pulse rate, LF and HF
trends, amount of activity and the simplified
hypnogram.


Figure 6: Result of one nights data.
Figure 7 is a summary of a 2-week hypnogram
showing sleep habits. This is the most useful
function for sleep care, which cannot be
implemented in the conventional sleep monitoring
systems.
DEVELOPMENT OF A SLEEP MONITORING SYSTEM WITH WEARABLE VITAL SENSOR FOR HOME USE
329


Figure 7: Result of summary of a 2-week hypnogram.
5 VALIDATION OF THE SLEEP
ESTIMATION
Correlation between the sleep stage estimated by the
proposed method using our wearable physiological
sensor and the sleep stage estimated using PSG by
sleep specialists was evaluated. EEG, EOG, chin
EMG, ECG, respiration and SpO2 by PSG
(Polymate AP1000, TEAC Corporation, Sampling
rate: 250Hz), the pulse wave and acceleration by our
sensor was recorded simultaneously in a night (8
hours). The test was held in two cites (Showa
University East Hospital, Tokyo, J apan and The
Institute for Science of Labour, Kawasaki, J apan).
45 normal healthy subjects (30 males and 15
females, 19-72 years old) are measured. All subjects
had informed consent.
The sleep stages of PSG were distinguished
manually by sleep specialists (doctor, clinical
laboratory technologist, or sleep researcher) who
belong to those cite based on Rechtschaffen & Kales
method (Rechtschaffen, 1968) by the minute. Our
sensor also estimated the sleep stages also by the
minute.
We defined coincidence ratio as an evaluation
function to compare the estimation result by our
sensor with the result by PSG.
The coincidence ratio is defined as a correlation
coefficient of moving average of sleep stages (20-
minutes time window) between the stages estimated
by this method and those from PSG. Table 1 shows
the result of the comparison.
Table 1: Result of the comparison.

Figure 8 shows an example of the estimation results.

Figure 8: An example of the estimation results (upper:
sleep stage distinguished by medical professionals using
PSG, lower: sleep stage distinguished automatically by the
sensor).
6 DISCUSSIONS
Beat-to-beat pulse interval detection is necessary
to obtain autonomic nervous activity. Our algorithm
has enough ability to get beat-to-beat pulse
intervals. However, this algorithm is applicable for
healthy subjects, except for cardiac disease,
peripheral blood circulation disorder.
The coincidence ratio of our sleep stage
estimating algorithm is 0.735. Although it is a rather
low value, PSG results also varied depending on the
examiner (variance is about 20%), and therefore it
seems to be acceptable for home healthcare use.
However, it is also applicable for healthy subjects,
except for autonomic nerve disorder, cardiac disease.
7 CONCLUSIONS
Measurement of sleep habits is a promising new
medical field. However, there are no systems
suitable for it. This is because current sleep
monitoring systems cannot satisfy the needs for
accurate analysis of sleep and convenience in use. In
order to provide a solution, we developed a small
and easy-to-use sensor device. We also developed an
algorithm for analyzing sleep stages. We confirmed
sufficient accuracy in the detection of PPIs by
comparison with R-R Intervals by ECG, and that in
the estimation of the sleep stage by comparison with
the result of PSG. The software can display the sleep
data of one night and the summary of a 2-week
hypnogram. This function is useful not only for a
doctor analyzing a patient's sleep habits, but also for
a user analyzing his/her sleep.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
330

REFERENCES
Rechtschaffen A., and Kales A., A Manual of standardized
terminology, techniques and scoring system for sleep
stages of human subjects. Public Health Service, U.S.
Goverment Priting Office, 1968.
Sadeh A., Alster J ., Urbach D. and Lavie P.,
Actigraphically based automatic bedtime sleep-wake
monitor scoring: validity and clinical applications. J
Ambul Monit 2, pp. 209216, 1989.
Watanabe T. and Watanabe K., Noncontact Method for
Sleep Stage Estimation, IEEE Transactions on
Biomedical Engineering, Vol.51, No.10, pp.1735-
1748, 2004.
Michimori A., Fukushima K. and Hagiwara H., Sleep
Monitoring System by Using Heart Rate Variability
Analysis, Matsusita Technical Report, No.82, 29-33,
2003 (in J apanese).
Wakuda Y., Noda A., Hasegawa Y., Arai F., Fukuda T.,
and Kawaguchi M., Biological Rhythm Based
Wearable Sleep State Observer, Journal of Advanced
Computational Intelligence and Intelligent
Informatics, Vol.11, No.2, pp.232-241, 2007
Baharav A., Kotagal S., Gibbons V., Rubin B.K., Pratt G.,
J . Karin and S. Askelrod, Fluctuations in. autonomic
nervous activity during sleep displayed by. power
spectrum analysis of heart rate variability, Neurology
45:66, 1183-1187, 1995.
Suzuki T, Ouchi K, Moriya A, Kameyama K and
Takahashi M. Development of a Sleep-Stage
Estimation Method using Heart RateVariability and
Actigraphy measured by Wearable Sensor, Sleep and
Biological Rhythms; 5 Suppl.1:A38, 2007.
Cole RJ , et al., Automatic sleep/wake identification from
wrist actigraphy, Sleep, 15(5), 461-469, 1992.
Heart rate variability: standards of measurement,
Physiological interpretation and clinical use. Task
Force of the European Society of Cardiology and the
North American Society of Pacing and
Electrophysiology. Circulation. 93, pp.10431065,
1996.
DEVELOPMENT OF A SLEEP MONITORING SYSTEM WITH WEARABLE VITAL SENSOR FOR HOME USE
331
POSTERS
A WAY FOR PREDICTING AND MANAGING THE GLYCAEMIC
INSTABILITY OF THE DIABETIC PATIENT
Farida Benmakrouha, Christiane Hespel
IRISA-INSA,20 Avenue des Buttes de Coesmes, 35043 Rennes cedex, France
[email protected], [email protected]
Mikhail V. Foursov
IRISA-Universite de Rennes-1, 35042 Rennes cedex, France
[email protected]
Jean-Pierre Hespel
Centre Hospitalier Universitaire de Rennes, 35000 Rennes, France
[email protected]
Keywords:
Dynamical system, Bilinear model, Regulation, Stability, Insulin perfusion.
Abstract:
We study the Bounded-Input-Bounded-Output (BIBO) stability of the system modeling the behavior insulin
delivery/glycaemia of the diabetic patient, under continuous insulin infusion, continuous glucose monitoring,
in order to point out that the patient is entering in a period of stable/unstable equilibrium.
The model is a bilinear dynamical system predicting for an interval of 15 minutes, with an average error of
15%. In case of stable equilibrium, the prediction will be valid for a longer time interval, when in case of
unstable equilibrium, it will leads one to reduce the time intervals.
The BIBO stability is studied by computing the generating series G of the model. This series, generalization
of the transfer fuction, is a tool for analyzing the stability of bilinear systems. It is a rational power series in
noncommutative variables and by evaluating it, a formal expression of the output in form of iterated integrals
is provided. Three cases arise: rstly, the output can be explicitly computed; secondly, the output can be
bounded/unbounded if the input is bounded; thirdly, no conclusion seems available about the BIBO stability
by using G. We propose a stabilizing constant input by studying the univariate series G

.
1 INTRODUCTION
Among the different medical possibilities to adminis-
ter insulin, the sub-cutaneous route is most secure and
easy to implement, but it lacks reliability. The intra-
venous route is the most rapidly responding method,
but it may cause vascular complications. The in-
traperitoneal route seems to be the most physiological
one.
In order to carry out a glycaemic regulation by an in-
traperitonal infusion of insulin, it is necessary to pre-
dict the glycaemia as a function of the insulin infu-
sion rate, for a given patient and a given insulin. The
rst closed-loop regulation method was developed by
A. Albisser back in 1974 (A. M. Albisser, ). Among
other methods, we can mention (J. L. Selam, 1992).
But, in spite of many positive aspects of these meth-
ods, none of them was unanimously accepted by the
medical community. This is partly due to insufcient
frequency of glycaemic sampling and the difculty to
vary rapidly the insulin infusion rates.
Recent technical progress made it possible to over-
come these difculties. In 2000 appeared the rst
holter glycaemic device: the CGMS (Continuous
Glucose Monitoring System) which allows one to
measure the glycaemia every 3 minutes. A rst reg-
ulation system based on the CGMS was developed in
2001 by E. Renard of CHU of Montpellier in collab-
oration with Medtronic Minimed (Renard, 2003). In
2006, E. Renard (E. Renard, 2006), analyzes the im-
plantation of three components: a pump for peritoneal
insulin delivery, a central intravenous glucose sensor
and a controller. Meals are preceded by a handheld
programmed bolus calculed for pre-meals. A prob-
lem remains in case of unstability of the glycaemia
close to a meal, a stress, a physical effort.
335
Then we proposed a bilinear modeling giving a
good approximation of the behavior insulin deliv-
ery/glycaemia on an interval of 15 minutes, in stan-
dard conditions (C. Hespel, 2000).
Once the model is known, the regulation consists
in inverting the input/output behavior of the system
(M. V. Foursov, 2003; M. V. Foursov, 2004). One has
to calculate the input in terms of the output function
one wishes to obtain (Fig.1). This regulation is par-
Figure 1: Regulation.
tially closed-loop because the glycaemic values are
only used every 15 minutes in order to compute the in-
sulin delivery. On constant intervals of time [t
i
, t
i+1
]
i
,
we compute some model M
i
and function insulin de-
livery u
i
(t) in order to follow an ideal trajectory y
i
(t)
for the glycaemia. On every time interval, the trajec-
tory is recalculed because of the variation between the
ideal trajectory and the true trajectory.
The crucial point consists in determining the size of
the time intervals, i.e. the frequency of the changes of
the insulin delivery. The study of the stability leads
us to reduce the size of the interval when the system
is unstable.
2 PRELIMINARIES: MODELING,
REGULATION
Two techniques are used to achieve this goal:
Phenomenological modeling requires a knowledge of
the equations governing the evolution of the process.
Such models of the glycaemic behavior of diabetics
were developed (Albisser, ).
In the behavioral modeling, the system is regarded as
a black box (J. Sjoberg, 1995). The goal is to con-
struct a model that approximates the system with a
desired precision (Chen and Chen, ). The parame-
ters involved in the obtained system of equations have
no practical signicance, but their number depends on
the required precision.
A commonly-used class of models is formed by linear
dynamical systems. Linear-model-based regulation is
simple to implement and it gives quite satisfactory re-
sults in many cases. It did not seem to be sufcient
for regulating the glycaemia of diabetics.
Another class of models consists of bilinear systems.
A bilinear system is quite similar to a linear one: it is
additionally linear as a function of the input. One has
thus more leeway to approximate the real system with
a better precision. We choose to model the system by
a bilinear system whose dimension is not xed.
Since a diabetic does not respond in the same way to
equal doses of insulin at different times of the day, we
suppose that she is described by different systems at
different time instants. So we construct a collection
of models that describe the behavior of the glycaemia
under certain conditions. Each model is thus valid
only for a certain period of time (at least 15 minutes).
Mathematically, the problem is to identify locally
(near t
0
) up to a given order k, a single input sys-
tem with drift considered as a black box, when only a
sample of the input/output data is known. Our method
involves the identication up to order k of the gener-
ating series G of the unknown system and the con-
struction of a bilinear system (B
k
) approximating the
unknown system up to k.
An advantage of this method is the possibility to pro-
vide, in terms of the order k, a system (B
k
) approxi-
mating the unknown system. (B
k
) is chosen so that its
output and the output of the unknown system coincide
up to k. The problem of identication of dynamical
systems in a neighborhood of t
0
is the following: to
determine the generating series of the unknown sys-
tem, up to a given order k, given the Taylor series of
the input and corresponding output functions.
The identication involves the following 3 steps. Dur-
ing the rst step, one obtains a system of linear equa-
tions that express the relationship between the deriva-
tives of the input and the output. The unknown pa-
rameters are certain linear combinations of the coef-
cients of the generating series. On the next step, these
linear combinations of the coefcients are identied
from the available data, by choosing appropriate in-
put/output sets. Finally, the coefcients are identied
by solving another system of linear equations.
A bilinear system corresponding to a rational series of
minimal rank is constructed providing a local model.
2.1 The Bilinear Model
A bilinear system (B) with a single input u
1
(t) and a
drift u
0
(t) 1 is given by its state equations
(B)

x
(1)
(t) = (M
0
+u
1
(t)M
1
)x(t)
y(t) = .x(t)
(1)
x(t) an Rvector space Q, M
0
, M
1
, are Rlinear.
We consider the alphabet Z ={z
0
, z
1
}, where z
0
codes
the drift and z
1
codes the input. The expansion of the
generating series G built on Z, by noting w a word
Z

, is G =
wZ
G|ww.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
336
G is a rational series dened from (1) by:
G = .x(0) +

0
1

j
0
, , j

=0
.M
j
0
M
j

x(0)z
j
0
z
j

(2)
We compute the rational expression associated with
(2), by generalizing the Schutzenbergers method
(Schutzenberger, 1961).
By evaluating the expression of G, we obtain a for-
mal expression of the output (Fliess, 1981)
y(t) =

wZ

G|w

t
0
(w) =

t
0
(G) = (G) (3)
where the iterated integrals are recursively dened by:

t
0
(w) =

1 i f w = 1
Z

t
0
(

0
v)u
i
()d i f w = vz
i
(4)
And we compute the iterated integral

t
0
(G).
2.2 The Regulation Method
Once the model is known, the regulation consists in
inverting the input/output behavior. One has to cal-
culate the input in terms of the output function one
wishes to obtain. This regulation is partially closed-
loop, since the glycaemic values are used to recalcu-
late the insulin infusion rates every 15 minutes.
In a previous paper (M. V. Foursov, 2002) we have
shown that we are capable of nding the Taylor se-
ries expansion of the command from the Taylor series
expansion of the desired output trajectory, using gen-
erating series techniques similar to those used during
the identication and the modeling. The algorithm
consists in sequential solving a system of polynomial
equations. If the model of a diabetic were an ex-
act one, this would be sufcient to regulate the gly-
caemia. But since our bilinear model is an approxi-
mation of the actual one, the glycaemic behavior will
eventually deviate from the chosen trajectory. Then,
the trajectory has to be recalculated in order to com-
pensate for these deviations and the insulin infusion
device has to be reprogrammed accordingly.
An important point is to determine the frequency of
change of the insulin infusion rates. The rst tests of
our modeling method (C. Hespel, 2000) showed that
we can predict the glycaemia over 15-minute inter-
vals with an error of about 10% or 15%. We need to
provide a new data to the pump every 15 minutes.
3 STUDY OF THE BIBO
STABILITY
A dynamical system is Bounded-Input-Bounded-
Output (BIBO) stable if its output y(t) is dened and
bounded for every bounded input u(t).
The output of a bilinear dynamical system can be
computed in evaluating its generating series. This
evaluation consists in integrating every term of this
series and in summing. We use the theorem of Hoang
(Hoang, 1990) :
Theorem 1. k, let us suppose that G
k
is exchange-
able and let us denote (G
k
) by g
k
((t))
g
k
((t)) = g
k
(t,
1
(t), ,
m
(t)) (5)
where
j
(t) is the primitive of the input u
j
(t) can-
celling for t = 0. Then, k, the series
S
k
= G
0
z
i
1
G
1
z
i
k
G
k
(6)
where z
i
1
, , z
i
k
Z, has the following evaluation:
(S
k
) = y(t) =

t
0

k
0

2
0
g
0
((
1
))g
1
((
2
) (
1
))
g
k
((t) (
k
))d
z
i
1
(
1
) d
z
i
k
(
k
)
(7)
Three cases occur: In some cases, this process is easy
and y(t) can be explicitly computed. In other cases, if
we assume that u(t) is bounded by 2 values Min, Max,
then we can know if so is y(t), without computing
explicitly y(t). Lastly, in some difcult cases, we only
try to nd some stabilizant constant inputs u(t) =
such that the output remains bounded, if it is possible.
We prove that the output of the bilinear system for the
input u(t) = consists in evaluating some univariate
series G

. This series being rational, can be written

as a quotient of 2 polynomials. We can then use 2
propositions (F. Benmakrouha, 2007) dealing with the
poles of G

in order to decide that a stability exists for

u(t) =
Proposition 1. A necessary condition for the BIBO
stability of (B), is that, for every R, the real part
of the poles of G

is 0 and the imaginary poles of

G

are single.
Proposition 2. If there exists such that every pole
of G

has a negative real part and if every imaginary

pole is single, then u(t) = is a stabilizing input
3.1 Example 1
For an insulin delivery u(t), a glycaemia y(t), the state
equations of the system (B
2
) are

x
(1)
(t) = (

0 0
a b

+u(t)

0 0
1 0

)x(t)
y(t) = (1.5 1) x(t)
(8)
The generating series is :
G
2
= (z
1
+az
0
)(bz
0
)

+1.5 (9)
G
2
= G
21
+G
22
+1.5 and y(t) = (G
2
)
with G
21
= z
1
(bz
0
)

, G
22
= az
0
(bz
0
)

A WAY FOR PREDICTING AND MANAGING THE GLYCAEMIC INSTABILITY OF THE DIABETIC PATIENT
337
We obtain an explicit expression of y(t):
(G
21
) = e
bt

t
0
e
b
1
d
1
(
1
)
(G
22
) = ae
bt

t
0
e
b
1
d
1
(10)
For u(t) = then
1
(
1
) =
1
, we get :
y
2,
(t) =
+a
b
(e
bt
1) +x(0)
This system is not BIBO for b > 0.
This system is BIBO for b < 0 (if M
1
u(t) M
2
then y(t) is bounded)
For instance, for a > 0, b < 0, 0 u(t) M, then
y(t) x(0) +
M+a
b
3.2 Example 2
The state equations of the bilinear system (B
3
) are

x
(1)
(t) = (

0 0 0
a b c
0 a 0

+u(t)

0 0 0
1 0 0
0 1 0

)x(t)
y(t) = (1.5 1 0) x(t)
(11)
The generating series is
G
3
= (z
1
+az
0
)(bz
0
+(z
1
+az
0
)cz
0
)

+1.5 (12)
We compute G
3,
by substituting z
0
to z
1
in G
3
:
G
3,
= 1.5+
(a+)z
0
1bz
0
(a+)cz
2
0
.
If =a then we decompose G
3,
in partial fractions
for studying the constant stabilizing inputs
u(t) = depending on the parameters a, b, c.
4 CONCLUSIONS AND FUTURE
WORKS
The BIBO stability of a bilinear system cannot be
generally studied by considering its state equation. In
this paper, we use the evaluation of its generating
series G. If the rational expression of G is simple or
obtained by concatenating some simple rational ex-
pressions, then the use of the generating series of the
system provides an answer about the stability and a
bound for the output. Otherwise, we can look for a
stabilizant constant input u(t) = by using the uni-
variate series G

.
By applying this method to the bilinear model approx-
imating the behavior insulin delivery/glycaemia,
we expect an information about the stability of the
system describing really this behavior.
A specic surveillance depending on whether the sys-
tem is stable/unstable will be set. Rather than take
constant interval of 15 minutes for recalculate the
ideal trajectory of the glycaemia, we propose that the
time intervals depend on this information about the
stability. In case of unstability, the varying size of the
intervals of time would be dened in order to keep the
glycaemia between some moderate bounds.
REFERENCES
A. M. Albisser, B. S. Leibel, T. G. E. e. a. An articial
endocrine pancreas. Diabetes, 23:389396.
Albisser, A. M. Intelligent instrumentation in diabetes man-
agement. CRC Crit. Rev. Biomed. Eng., 17:124.
C. Hespel, J. P. Hespel, E. M. G. J. G. F. B. (2000). Al-
gebraic identication: Application to insulin infusion.
ISGIID2000 Evian.
Chen, T. and Chen, H. IEEE T.Neur.Net.
E. Renard, G. Costalat, H. C. J. B. (2006). Arti-
cial beta-cell: clinical experience toward an im-
plantable closed-loop insulin delivery system. Dia-
betes Metabolism, pages 497502.
F. Benmakrouha, C. H. (2007). Generating formal power
series and stability of bilinear systems. 8th Hellenic
European Conference on Computer Mathematics and
its Applications (HERCMA 2007).
Fliess, M. (1981). Fonctionnelles causales non lin eaires et
ind etermin ees non commutatives. Bull . Soc . Math.
France, 109:340.
Hoang, N. M. (1990). Contribution au d eveloppement
doutils informatiques pour r esoudre des probl` emes
dautomatique non lin eaire. Masters thesis, Univer-
sit e de LilleI.
J. L. Selam, P. Micossi, F. L. D. e. a. (1992). Clinical trial
of programmable implantable insulin pump for type i
diabetes. Diabetes Care, 15:877885.
J. Sjoberg, Q. Zhang, L. L. e. a. (1995). Nonlinear
black-box modeling in system identication: a unied
overview. Automatica, 31:16911724.
M. V. Foursov, C. H. (2002). On formal invertibility of the
input/output behavior of dynamical systems. preprint
IRISA, 1439.
M. V. Foursov, C. H. (2003). On algebraic modeling and
regulation of the behavior of diabetics. Innovative
Technologies for Insulin Delivery and Glucose Sens-
ing.
M. V. Foursov, C. Hespel, D. B. J. H. (2004). Nouvelle
m ethode de mod elisation alg ebrique et de r egulation
de la glyc emie. Infu-Syst` emes 21, 3:2224.
Renard, E. (2003). Closed-loop system with implantable
pump and iv sensor. proceedings of the symposium
Looking to the future, pages 7879.
Schutzenberger, M. P. (1961). On the denition of a family
of automata. Inform. Contr., 4:245270.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
338
DEVICE FOR SYNCHRONIZED ROTATION
Shuh J ing Ying, Rufael Berhane and Rajiv Dubey
Department of Mechanical Engineering
University of South Florida
Tampa, Florida, U.S.A.
[email protected], [email protected], [email protected]
Keywords: Synchronization, Synchronized rotation.
Abstract: This device makes two shafts rotating in synchronized angular displacement. The sensors are potentiometers.
The device is designed for use in a driving simulator. The first shaft is connected to a steering wheel and
controlled by a driver so this shaft is the master, the second shaft is connected to the computer software that
displays the surrounding sceneries, therefore this shaft is the follower. The two shafts must be rotated in
synchronized mode and accurately. Major components in this device are operational amplifiers, NAND
gates, power MOFET and relays. Details in design and manufacturing are reported in this paper. This device
may be used for any place that requires two shafts rotating synchronically.
1 INTRODUCTION
Electronic Mobility Control Co.(EMC) developed an
Advanced Electronic Vehicle Interface Technology
(AVEIT) that converted the steering wheel, gas
pedal and brake pedal in a car into one joystick.
Based on the EMC system, we are building a driving
simulator for training disabled person to drive by
using a joystick for gas, brake and steering. The
forward movement of joystick is for gas pedal,
backward for brake, and right and left for steering.
However it is really difficulty for a person to drive a
real car without going through a simulator. A
simulator to be built is similar to an airplane
simulator for training pilot to drive an airplane. But
it is not so simple to build a driving simulator, many
man powers are required to work on the project from
design, computer software, and manufacturing.
Because of the limited man power and financial
resources, we bought a regular car simulator from
Simulator Systems International (SSI). It has three
screens to show the surrounding sceneries for the
driver to see. We hope to build a device that can
make EMC and SSI systems working
simultaneously. That is the whole purpose of this
project.
There are two possibilities for reaching the
synchronized rotation, mechanical and electrical
devices. After a length consideration the final
decision was on electrical approach.
Once electrical device is chosen we looked into
existing literature but to our surprise we could not
find any in our library or any nearby libraries.
Although we were pretty certain that synchronized
motors have been used in large airplanes but no
information is found. Anyway the existing device
may not fit what we need.
So we start to design a circuit for our purpose
and we build the device exactly as we want. The
details are given in this paper. After the device is
completed certainly we had quite extensive test to
check out every part. However we have not reached
to the point to train a disabled person actually to
drive.
2 DESIGN OF THE CIRCUIT
The circuit required is to operate a motor which has
enough power to rotate the steering mechanism of
the SSI system. The directions of rotation can be
clockwise and counterclockwise. The size of motor
is chosen to be similar to the servomotor used in the
EMC system. The input power of the motor is 93
watts with torque of 8.7 n-m. The motor is a geared
DC motor. Because there is no precise position
required in the operation analogue circuit is chosen
in the design. Potentiometers are used for the
position indicators and relays are used for the
control of the motor.
339
2.1 Differential Amplifier
The difference of the voltage signals from the
potentiometers are to be amplified. Positive signal
will make the motor to rotate in one direction and
negative signal then will rotate in the other direction.
A bipolar op amp, AD706, is chosen to build a
differential amplifier. The schematic circuit diagram
is shown below.
It will be idea if the amplifier is very sensitive to
the input voltage and the dissipation power is not
very large as compared to the motor power. The
output voltage is given by
( )
4
#1 #2
3
1 4 2 3
1
out in in
R
V V V
R
for R R and R R

= +


= =


Figure 1: Differential Amplifier.
See Ref. 3 for details. The specification of AD706 is
given in Appendix.
2.2 NAND Gate
Because the output voltage of AD706 is not high
enough to turn on the power MOSFET to operate the
relay, a NAND gate is used to further amplify the
positive signal. NTC4011B is chosen for achieving
this purpose. NTC 4011B is a 2 input positive logic
NAND gate. There are four NAND gates in one IC.
The specification of NTE4011B is given in
Appendix. One input of the first NAND gate is
connected to a fixed positive voltage, hence only
positive signal will go through the first NAND gate.
Two of the 4 NAND gates are used for amplifying
the positive signals when the output of AD706 is
positive. However when the output of AD706 is
negative we must lead the signal to go to other
branch and to make the motor to turn in another
direction. A negative latch is used for that purpose.
2.3 Negative Latch
A negative latch is a resetable memory block. The
output goes to high as soon as the input signal goes
to negative. This is exactly we need for the output of
AD706 goes to negative.
Two NAND gates are needed. The circuit
diagram is given as follows

Figure 2: Negative Latch.
2.4 Power MOSFET
A power MOSFET is a specific type of Metal Oxide
Semiconductor Field-Effect Transistor. It is
designed to handle large power. IRFP250N is
chosen for this circuit. When the voltage applied to
G reaches 12 v, the switch is turned on from S to D.
The specification of IRFP250N is given in appendix.
2.5 The Circuit Diagram
With some details given above, the whole circuit can
be presented as shown in Fig 3. The input signals are
from the potentiometers, one is connected to the
EMC system and one is with SSI system. These
signals go through the differential amplifier AD706.
When the difference is positive, the signal goes
through the upper branch in the circuit diagram
because one of the inputs in the NAND gate TC4011
is set to +5Vdc. This positive signal then turn on the
power MOSFET and make the motor to rotate. As
the output of AD706 is negative, the signal triggers
the negative latch and the power MOSFET in the
lower branch in the diagram is turned on
consequently the motor is turning in the another
direction.
However, when the circuit was tested, the motor
will rotate back and forth as the input difference of
AD706 is nearly zero. This will severely shorten the
life of the motor. The power supplies in the circuit
are using voltage dividers. This is only for the
saving of space in the device.
Also we experienced the lives of ICs are very
sensitive to the voltages applied. Follow the
specifications carefully, is very important. That is
why all the specification are attached in appendix.




BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
340

Figure 3: Circuit Diagram for Synchronized Rotation.

Figure 4: Modified Input Circuit.
2.6 Modified Input Circuit
To stop the rotation of the motor as the input of
AD706 is nearly zero. The outputs of NAND gates
and the inputs of the power MOSFETs are
disconnected The connection of these wires are
controlled by this modified input circuit that is
shown in Fig 4. Note that in this circuit the input to
IC AD706 is through a full wave rectifier. When the
signal is nearly zero, the voltage is not strong
enough to trigger the relay in this circuit so that the
relay in the previous diagram will not be turned on.
Consequently the motor will not be turned on. Now
as the EMC steering wheel rotates approximately
45
o
, the motor will start to rotate in the same
direction. And the motor will stop as the difference
between the potentiometers is within 45
o
. This
operation is considered as perfect.
3 CONCLUSIONS
Although this device is aimed originally for the use
of the synchronized rotation in a driving simulator,
actually it can be used in many other places
requiring a synchronized rotation. Because of this,
the information presented here could be invaluable
to many research workers in this community. On the
other hand this circuit is rather simple and easy to
follow, mechanical engineers can use it. The device
has been tested extensively in the simulator and it is
working perfectly well. However, the simulator has
not been used for training disabled person to drive.
This is considered as the next phase of our work.
J ust to make the information completely presented.
The specifications of ICs are given in appendix.
DEVICE FOR SYNCHRONIZED ROTATION
341
REFERENCES
Narsingh Deo, 1983, System Simulation with Digital
Computer, Prentice-Hall Englewood Cliffs, N. J .
Geert Van der Plas, Georges Gielen and Willy Sansen,
2002, A computer-aided design and synthesis for
Analog Integrated Circuits, Kluwer Academic
Publishers, Boston, MA.
Amplifier Reference Manual, 1992, Analog Devices
Norwood, MA.
The ARRL Handbook for radio 2006, 2005 83
rd
Edition A.
R. R. L.
Ying, Shuh J ing, Patti Barrett, and Stephen Sundarrao,
2008, A Device Turning on Strobe Light from Horn
Signal, BMES Annual Fall Meeting, St.Louis, MO
Ying, Shuh J ing and Stephen Sundarrao, 2007 Design and
Manufacturing of A Wheel Chair Treadmill, RESNA
2007
APPENDIX
AD706 SPECIFICATIONS
Features:
High DC precision.
100V max offset voltage
1.5 V/
o
C max offset drift
200 pA max input bias current
0.5 V p-p voltage noise, 0.1 Hz to 10 Hz
750 A supply current
Maximum Ratings:
Supply Voltage.....18V
Internal Power Dissipation
(Total: Both Amplifiers)....650mW
Input Voltage...........V
s

Differential Input Voltage.......0.7V
Output Short Circuit Duration.........Indefinite
Storage Temperature Range....-65
o
to +125
o
C
Operating Temperature Range
AD706J..........0 to +70
o
C
AD706A.....-40 to +85
o
C
Lead Temperature(Soldering 10sec)........300
o
C
Notes:
Stresses above the maximum ratings may cause
permanent damage to the device. This is a stress
rating only. Exposure to max rating conditions for
extended periods may affect device reliability. The
input pins of this amplifier are protected by back
toback diodes. If the differential voltage exceeds 0.7
V, external series protection resistors should be added to
limit the input current to less than 25 mA.
TC4011-SPECIFICATIONS
The TC4011B is a quad 2 input NAND gate.
Maximum Ratings:
DC Supply voltage............18 V
Input voltage..18.5 V
Output Voltage...18.5 V
DC input current.....10 mA
Power dissipation..... 300 mW
Operation Temperature range...... -40 to 85
o
C
Storage Temperature range -65 to 150
o
C
IRFP250N-SPECIFICATIONS
MaximumRatings:
I
D,
continuous drain current......... 30A
I
DM,
pulsed drain current.....120A
Power dissipation.......214W
Gate to source voltage......20V
Single pulse avalanche energy..315mJ
Avalanche current........30A
Repetitive avalanche energy......21mJ
Peak diode recovery........8.6V/ns
Operating temperature range......-55 to 175
o
C
Soldering Temperature (10 sec)300
o
C
Mounting Torque .... 1.1 N-m (10 lbf-in)
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
342
HAND-HELD LUMINOMETER WITH ECL-BASED BIOSENSOR
FOR LACTATE DETERMINATION
A. Martinez-Olmos, A. J . Palma
Department of Electronics and Computer Technology, University of Granada, Granada 18071 Spain
[email protected], [email protected]
J . Ballesta-Claver, M. C. Valencia-Miron and L. F. Capitan-Vallvey
Department of Analytical Chemistry, University of Granada, Granada 18071, Spain
[email protected], [email protected], [email protected]
Keywords: Electrochemiluminescence, Portable instrumentation, Potentiostat, Biosensor, Screen-printed device,
Lactate determination, Saliva.
Abstract: A new hand-held luminometer for electrochemiluminescence (ECL)-based one-shot biosensor for lactate is
described. The lactate recognition system is based on lactate oxidase and the transduction system consists of
luminol, all reagents immobilized in a Methocel membrane. The measurement of ECL from a screen-
printed electrode by a portable instrument designed and developed by the authors makes it possible to
determine lactate concentration. The compositions of the membrane and reaction conditions have been
optimized to obtain adequate sensitivity. The one-shot biosensor responds to lactate rapidly, with the typical
ECL acquisition time being 3 min, with a linearized dependence whose dynamic range was from 910
-6
to
210
-3
M, a detection limit of 2.410
-6
M and a sensor-to-sensor reproducibility (relative standard deviation
RSD) around 10 % at the medium level of the range. The performance of the ECL one-shot biosensor and
portable instrument was tested for the non invasive analysis of lactate in saliva, validating the results against
a reference procedure.
1 INTRODUCTION
L(+)-Lactate is produced in the anaerobic
metabolism of glucose and its determination is of
interest in clinical analysis, sports medicine and food
analysis. The measurement of lactate is routinely
performed with liquid chromatography (Ewaschuk,
2002), spectrophotometry (Benthin, 1991) and
amperometry, mainly with enzymatic electrodes
(Compagnone, 1998). Lactate analysis is needed in
different fields such as food, sports medicine and
health. In foodstuffs, lactate is produced by bacterial
fermentation and is an essential component related
to the manufacture of cheese, yoghurt, milk, etc.,
thus monitoring lactate being an important quality
control parameter.
Rapid evaluation of lactate levels can be
performed with one-shot sensors, that mainly are of
electrochemical type (Klonoff, 2003).
Chemiluminescence measurement could be of
interest for one-shot sensor design due to its good
sensitivity and selectivity, although the use of
electrochemiluminescence (ECL) could offer clear
advantages for controlling the chemical system
(Richter, 2004).
The use of screen printing technologies, with
benefits for low cost and mass production, appears
to be interesting to develop ECL one-shot biosensors
next to portable instrumentation.
The presented lactate biosensor is based on its
enzymatically catalyzed oxidation and back ECL
transduction using luminol (L) according to:
Lactate +O
2

LOD
Piruvate +H
2
O
2
L e L* (electrode)
H
2
O
2
+L* 3-aminophthalate +N
2
+H
2
+h
2 MATERIALS
The one-shot biosensor is formed by a screen printed
electrode where the working electrode contains all
343

needed reagents immobilized in a Methocel
membrane. The sensing layer was spotted as
solutions of luminol, lactate oxidase (LOD), BSA,
sodium chloride and 8.8 pH phosphate buffer in
aqueous solution of Methocel. The screen-printed
electrode was covered by a thick overlapping plastic
layer with a 40-l volume hole in the electrode area
to place the sample.
The characterization of the screen-printed
electrodes was investigated through cyclic
voltammetry. The ECL measurements were
performed measuring the light intensity emitted
while triggering the ECL reaction.
3 INSTRUMENTATION
After describing the biosensor, a portable instrument
based in ECL detection and designed and fabricated
for this sensor will be detailed. The prototype has
been applied to lactate concentration determination
in saliva.
The system is based on a solid-state photodiode
detector, which generates an electric current
proportional to the ECL being measured. In Figure 1
the general scheme of the system is presented.

Figure 1: Block diagram of the instrument.
The light resulting from the ECL reaction on the
sensor excites the photodiode detector (PD) (S1227-
66BR, Hamamatsu Photonics) which generates an
electric current in response. The analog circuit for
measuring this current is shown in Figure 3. ECL is
produced when a voltage difference of 0.5V is
applied between the reference and the working
electrodes in the biosensor. This polarization of the
sensor is carried out using a programmable built-in
potentiostat, which is designed to apply variable
voltage steps between the sensor electrodes.
A detailed schematic of the potentiostat is
presented in Figure 2. In this circuit, a serial digital-
to-analog converter (DAC) (DAC8574, from Texas
Instruments) generates an analog voltage from a 16-
bit digital word sent by the microcontroller, which is
the input value to the potentiostat. If the
electrochemical cell is full of a conductive liquid,
the operational amplifiers A1 and A2 form a non-
inverting amplifier stage with gain 2. This
establishes a voltage at the working electrode that is
double than the input voltage value. The voltage at
the reference electrode is forced to virtual ground
because of the negative feedback of the operational
amplifier A3. Thus, the voltage difference between
the working and the reference electrodes is simply
twice the analog value generated by the DAC.

Figure 2: Potentiostat circuit.
The voltage at the working electrode is
monitored directly by the microcontroller, since its
function is to detect when the test drop is deposited
on the biosensor. This event causes the start of a
time count, thus allowing a precise determination of
the time elapsed between the drop deposition and the
beginning of the measurements. Therefore, a perfect
timing control of the measurement procedure can be
achieved.

Figure 3: Measurement circuit.
The current is converted into voltage trough the
current-to-voltage converter formed by the
operational amplifier A1 (TLC277, from Texas
Instruments). This device has a feedback network
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
344

compounded by the resistors R
f
, R
1
and R
2
, which
results in a high gain conversion, being the output
voltage of A1:
( )
1 2 1 2 2
2
( ) ( )
pd f f
o
V R R I R R R R R
V
R
+
+ + +
=

(1)
where V
+
is the voltage at the non-inverting input of
the operational amplifier A1, and I
pd
is the current
generated by the photodiode. The voltage V
+
takes a
value of few mV and can be changed through the
variable resistor R
var
. The effect of this parameter is
to compensate the input offset voltage of the
operational amplifier, which can have a high
influence on the output because of the great gain of
this stage.
Under ideal conditions (V
+
=0) and assuming
R
f
>>R
2
, Equation (1) can be rewritten as:
1
2
1
o pd f
R
V I R
R

= +



(2)
Selecting a high value for R
f
and making R
1
>R
2
,
a gain factor of 10
11
10
13
V/A can be achieved.
The output voltage of A1 is conditioned using
two parallel stages , formed by the operational
amplifiers A2 and A3. In each stage, the signal is
firstly filtered through a RC low-pass filter. The
operational amplifier A2 acts as a buffer, whereas
A3 amplifies the output voltage of the I/V converter
before sending it to the microcontroller. In this way,
the C receives two signals, one corresponding to
the filtered output of the first stage, and another that
is an amplification of this last one. The purpose of
having two different channels for measuring the
same signal is to expand the range of lactate that can
be analysed.
The outputs of A2 and A3 are connected directly
to the microcontroller (C) (PIC18F2550,from
Microchip Inc.), which uses an internal 10-bit
analog-to-digital converter to alternatively sample
these signals at high frequency. A serial EEPROM
module (24LC512, from Microchip Inc.) of 512kbit
is used to store the sampled data. Finally, once the
calibration function (see next section) programmed
in the microcontroller is applied, results are sent to
the LCD display (Figure 1). All electronic circuitry
is included in an enclosure with optical, magnetic
and electrical shielding.
Moreover, control software written in Visual
Basic allows the user to optionally communicate the
instrument with a computer via an USB port to
receive the data for further analysis.
Main advantages of our design lie on portability,
low cost because of the use of a photodiode instead
of a costly or bulky photomultiplier, and the use of
non invasive samples. Most commercial portable
lactate meters use blood or serum for lactate analysis
(www.lactate.com, Poscia, 2005).
4 BIOSENSOR COMPOSITION
AND MEASUREMENT
CONDITIONS
Composition of sensing membrane was optimized in
terms of type and concentration of membrane
polymer, supporting electrolyte, pH and buffer,
luminol concentration, enzyme units, and BSA
concentration.
0
200
400
600
0 5 10 15 20
t / s
E
C
L

r
e
l
a
t
i
v
e

s
i
g
n
a

Figure 4: Relative ECL lactate signals.
Different types of ECL analytical signals were
studied using the instrument described in the
previous section in order to obtain an analytical
parameter for lactate concentration. The intensity of
the collected light, resulting from the reaction on the
sensor, did not show a direct relationship with the
lactate concentration, as can be seen in Figure 4,
where three steps at fixed potential were applied to
the same problem drop. The intensity of the light is
increased with successive potential steps. From these
current pulses, a kinetic signal derived from the
relative increase of the signal was chosen for the
measurement of the lactate concentration, since it
remains stable for different excitation pulses. The
measurement conditions studied were: a) applied
potential (0.5 V); b) waiting time before the first
pulse (3 min); c) time between pulses, being 10 s for
better sensitivity; d) pulse time with 1 s as best for
sensitivity and time of analysis.
The sample volume in the screen-printed device
was spotted with a micropipette. From the influence
of sample volume, studied between 20 and 40 L.
Low volumes have high ECL signals but poor
repeatability. The signal and the standard deviation
decrease when the volume increases. The reason of
this behaviour is that low volumes dont cover the
HAND-HELD LUMINOMETER WITH ECL-BASED BIOSENSOR FOR LACTATE DETERMINATION
345

three electrodes totally, specially the reference
electrode, making then oscillating potentials.
Therefore a volume of 35 uL for the test drop has
been selected, wich provides a good precision (5-8
% RSD).
5 ANALYTICAL
CHARACTERIZATION
The dependence of ECL signal with lactate
concentration was studied between 10
-7
and 10
-3
M
obtaining a linear relationship between 10
-6
M and
210
-4
M (Figure 6).
0
200
400
600
800
1000
0.00 0.50 1.00 1.50 2.00 2.50
lactate10
4
M
C
L

r
e
l
a
t
i
v
e

s
i
g
n
a
l

Figure 6: Linear calibration of lactate biosensor.
Table 1 shows some analytical parameters of
biosensor for lactate.
Table 1: Analytical characteristics.
Parameter Value
Linear range (M) 910
-6
210
-4
Intercept 1.87
Slope 786150
r
2
0.9956
Detection limit (M) 2.410
-6
RSD blank (%) 8.2 %
RSD lactate (%) 810
-4
M 10.3 %
This biosensor was applied to lactate
determination in saliva obtaining good preliminary
results.
6 CONCLUSIONS
A new hand-held luminometer for
electrochemiluminescence (ECL)-based one-shot
biosensor for lactate is described. Exciting the
sample volume with consecutive steps of 0.5V and
measuring the light resulting from the reaction on
the sensor provides a method for the evaluation of
the lactate concentration. A good linear calibration
in the range of 910
-6
to 210
-4
M has been achieved,
what indicates that lactate in saliva, rather than
lactate in blood can be measured. This fact results in
a better behaviour of the prototype than the existing
commercial instruments, because of its minimal
invasive requirements for the measurement of lactate
in humans. The use of a solid-state photodiode as
optical detector, instead of a photomultiplier, which
is the usual technique in available commercial ECL
systems, as well as the integration of the potentiostat
and the measurement electronics in the same design
has allowed a low cost and compact instrument.
ACKNOWLEDGEMENTS
This work is supported by the Ministerio de
Educacin y Ciencia, Direccin General de
Investigacin (Spain), under projects CTQ2005-
09060-C02-01 and CTQ2005-09060-C02-02, and by
the J unta de Andaluca under project P06-FQM-
01467.
REFERENCES
J . B. Ewaschuk, G. A. Zello, J . M. Naylor, and D. R.
Brocks, J . Chromatog. B, 781 (2002) 39-56.
S. Benthin, J . Nielsen, and J . Villadsen, Anal. Chim. Acta,
247 (1991) 45-50.
D. Compagnone, D. Moscone, and G. Palleschi, Recent
Res. Devel. in Pure & Applied Anal. Chem., 1 (1998)
73-86.
D. C. Klonoff, Diabetes Tech. Therapeut., 5 (2003) 929-
931.
M. M. Richter, Chemical Rev., 104 (2004) 3003-3036.
www.lactate.com
A. Poscia, D. Messeri, D. Moscone, F. Ricci, and F.
Valgimigli, Biosensors and Bioelectronics 20 (2005)
2244-2250
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
346
SELECTIVE OSTEOBLASTIC CELL MICRO-ARRAYS ON
DIAMOND FILMS
Bohuslav Rezek, Lenka Michalkov, Egor Ukraintsev, Alexander Kromka
Institute of Physics, Academy of Sciences of the Czech Republic, Cukrovarnicka 10, Prague 6, Czech Republic
[email protected]
Marie Kalbacova
Institute of Inherited Metabolic Disorders, 1st Faculty of Medicine, UK, Ke Karlovu 2, Prague 2, Czech Republic
[email protected]
Keywords: Cell adhesion, Proteins, Diamond, Biotechnology, Biosensors, Osteoblasts, Atomic force microscopy.
Abstract: Unique combination of chemical and biocompatible properties with semiconducting properties makes
diamond an attractive material for merging solid state and biological systems. Microscopic chemical
patterning of diamond films by hydrogen and oxygen surface atoms is used for self-assembly of human
osteoblastic cells into micro-arrays. The cell adhesion and assembly on the diamond is further controlled
and optimized by cell and protein (fetal bovine serum - FBS) concentration. The cells are characterized by
fluorescence microscopy of actin fibers and nuclei. The protein adsorption is studied by atomic force
microscopy (AFM). The cells are arranged into arrays on O-terminated patterns. The best cell selectivity is
achieved for the lowest cell concentrations of 2500 cells/cm
2
. Higher cell concentrations enable to colonize
unfavorable H-terminated regions due to mutual cell communication. Based on AFM, the proteins are
present on both H- and O-terminated surfaces, however, pronounced differences in the thickness, surface
roughness, morphology, and phase images indicate different conformation of the proteins and hence the cell
selectivity. There is no cell selectivity when no protein is supplemented in the medium. These results may
be applicable in tissue engineering, implants, bio-electronics, and biotechnology in general.
1 INTRODUCTION
Diamond is not only a gem but also a promising
technological material. Its properties include high
hardness, fracture toughness, low friction
coefficient, high Young modulus, increased wear
resistance and a variety of substrates onto which it
can be prepared (Potocky, 2007). Although diamond
is considered inert, its surface can be functionalized
by various atoms or molecules (Rezek, 2007a). This
gives rise to striking properties (Nebel, 2003).
For instance, electrical conductivity and electron
affinity of diamond are strongly influenced by the
O- or H-termination of the diamond surface
(Kawarada, 1996; Maier, 2001). The differences are
mainly caused by the surface dipole of C-H and C-O
bonds (Tachiki, 2003). O-terminated diamond is
highly resistive, whereas H-terminated surface
induces p-type surface conductivity even on an
undoped diamond (Maier, 2001). These features can
be applied for field-effect transistor (FET) devices
(Rezek, 2007b; Garrido, 2003).
Furthermore, O-terminated surfaces are
hydrophilic while H-terminated surfaces are
hydrophobic. H-terminated surfaces were thus found
less favorable for osteoblastic cell adhesion,
spreading and viability compared to O-terminated
surfaces (Kalbacova, 2007a). On the other hand, H-
terminated diamond surface is an ideal starting point
for covalent attachment of biomolecules (Yang,
2002). Chemical functionalization can also lead to
bio-passivation or bio-active properties (Bajaj,
2007).
Unique combination of the mechanical,
chemical, and biocompatible properties (Tang, 1995,
Kalbacova, 2007a) with semiconducting properties
makes diamond an attractive material for merging
solid state and biological systems (Yang, 2004;
Rezek, 2007b). Hence the hydrogen and oxygen
surface patterns are highly relevant for bio-
electronics as well as for tissue engineering.
347

Characterization of the interaction between cells and
surfaces is essential for cell-based biosensors,
engineered tissue therapies and the optimization of
implant biomaterials. Cells recognize their
environment and consequently start to change it by a
production of appropriate extracellular matrix
(ECM) proteins to form the basis for cell spreading,
increased adhesion and expression of differentiated
phenotypes (Schakenraad, 1989). This complex and
flexible process is dependent on culture conditions,
including the underlying substrate and the pre-
adsorbed protein layer.
Until now, the research on the cell-diamond
interfaces has been focused on overall homogeneous
surface terminations (Yang, 2004; Kalbacova,
2007a; Rezek, 2007c; Song, 2007). In this work we
show selective adhesion and arrangement of
osteoblast-like cells on NCD thin films that are
microscopically patterned with H- and O-terminated
regions (Michalikova, 2008). We control initial cell
density and serum concentration in medium
influencing cellular colonization of patterned
susbtrate. Furthermore, we employ atomic force
microscopy (AFM) to characterize the structural
properties of mediating proteins (fetal bovine serum,
a crucial component for the cell growth) adsorbed
onto the diamond micro-patterns. The data are used
to discuss the selectivity of the cell adsorption on the
patterns, i.e. to what degree the cell adhesion and its
selectivity is driven by serum adsorption and
conformation on H- and O-terminated surfaces or by
a direct effect of diamond surface dipoles on the
cells. We also provide perspectives for potential bio-
electronic applications.
2 MATERIALS AND METHODS
Diamond films are grown on (100) oriented silicon
substrates (13 mm in diameter, 500 m thickness,
RMS roughness of <0.6 nm) by microwave plasma
process using total gas pressure 50 mbar, substrate
temperature 800C, 1% CH
4
in H
2
and total power
2.5 kW. This process results in a growth of
continuous, smooth and high quality nanocrystalline
diamond film (Potocky, 2007; Kromka, 2008). X-ray
photocurrent spectroscopy (XPS) detects that the
films are 95% pure diamond (Zemek, 2006). The
diamond film thickness is 300-400 nm. Average
crystal size is 50 nm, RMS roughness at 1x1 m
2

area is 15-20 nm as measured by AFM using
standard silicon tips of nominal radius <10 nm. The
silicon substrates are coated with NCD film on both
sides, silicon is thus hermetically encapsulated in the
diamond.
The diamond films were further chemically
cleaned in acids (97.5% H
2
SO
4
+ 99% powder
KNO
3
) at 200C for 30 minutes. The surface was
then hydrogenated at 800C for 10 minutes. NCD
films were lithographically processed to generate
alternating H- and O-terminated patterns of 30 to
200 m widths. A positive photoresist ma-P 1215
(micro resist technology GmbH, Germany) was
applied. NCD films with lithography mask were
treated in oxygen radio-frequency plasma (300W
power, 3 minutes process time) to oxidize the
surface and hence to generate the hydrophilic
patterns. Then the sample was rinsed in a stripper,
de-ionized water and dried. This process removed
possible surface contamination (Rezek, 2006a). The
H-/O-termination quality was proved by a scanning
electron microscope (SEM; J EOL Superprobe 733).
Electronic measurements detected a surface
conductivity of 10
-5
S/sq on the H-terminated
surfaces (Kozak, 2008). Surfaces with O-termination
were highly resistive. The NCD samples were
sterilized in 70% ethanol for 10 minutes prior to cell
plating. The device concept is schematically shown
in Figure 1.

Figure 1: Schematic picture of silicon substrate
hermetically coated with NCD layer with stripe-like
patterns having hydrogen or oxygen surface termination.
Cell adhesion on the O-terminated region is also
schematically indicated.
SAOS-2 cells (human osteoblast-like cell line)
(DSMZ GmbH), were grown in McCoys 5A
medium (BioConcept) supplemented with heat
inactivated fetal bovine serum (FBS; Biowest) of
various concentrations (0-15%), penicillin (20 U/ml)
and streptomycin (20 g/ml). Note that the SAOS-2
is a standard and well-defined cell line. Thus the
results can be compared between various series of
experiments as well as with reports in the literature.
Cells were plated at the densities of 2,500 and
10,000 cells/cm
2
using a droplet technique: substrate
surface was covered by 100 l droplet of cell
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
348

Figure 2: Nanocrystalline diamond film with 200m wide H-/O-terminated patterns: (a) optical (bright field) image prior to
cell plating showing optically transparent and featureless surface, (b) scanning electron microscopy image prior to cell
plating where bright stripes correspond H-termination and dark stripes O-termination of the diamond surface due to their
opposite electron affinity, (c) fluorescent microscopy image of osteoblastic cells cultivated on the substrate. The alignment
cross is used for correlation of the surface termination micro-patterns with the cells.

Figure 3: Fluorescent microscopy images of osteoblastic cells (SAOS-2) cultivated in McCoy's medium supplemented with
15% FBS for 2 days on H-/O-terminated stripes of different widths (30m, 60m, 100m and 200m) on diamond films.
Initial cell concentration was 2,500 cells/cm
2
. The fluorescence shows actin stress fibers (green) and nuclei (blue). Scale bar
is 100m.
suspension in the appropriately supplemented
medium, let to incubate for 2 h (adhesion time), and
then 1.4 ml of the medium was added. In case of 0%
FBS, the cells were plated and incubated for 2 h in
the medium without the serum. Then the 15% FBS-
supplemented medium was added to facilitate
further cell cultivation. After plating, the cells were
cultivated for 48 hours in 5% CO
2
at 37C.
An advantage of the applied droplet technique is
a precise control of the applied number of cells on
the sample. A disadvantage is slightly non-
homogenous distribution of cells over the sample
with lower concentration on the edge and higher
concentration in the middle of the sample.
Therefore, the microscopic images were taken from
comparable areas on the samples.
Adhesion and morphology of SAOS-2 cells were
characterized by fluorescent staining of actin stress
fibers (phalloidin-Alexa 488 - 1:100, Molecular
Probes) and nuclei (DAPI - 1:1000, Sigma)
according to the protocol in Ref. (Kalbacova,
2007b). The staining was visualized using the E-400
epifluorescence microscope (Nikon); digital images
were acquired with a DS-5M-U1 Color Digital
Camera (Nikon).
As the adhesion and growth of osteoblastic cells
is mediated by proteins, the adsorption, adhesion,
and conformation of FBS itself on the H- and O-
terminated diamond was also investigated. Polished
IIa (100) mono-crystalline diamonds were used as
substrates to minimize the contribution from surface
morphology of NCD films. The mono-crystalline
diamond surface was H- or O-terminated using the
same procedures as for NCD films. A droplet of
15% FBS in the McCoys 5A medium was applied
on the diamond substrates for 10 min. Then the
whole samples were immersed in the fluid cell
containing the same FBS/McCoys medium and
characterized by AFM (Ntegra, NTMDT).
AFM measurements were performed in the
medium using doped silicon cantilevers
(BSMulti75Al) with the typical force constant of 3
N/m, resonance frequency 75 kHz in air (30 kHz in
the medium), and nominal tip radius <10 nm.
Surface morphologies were investigated in
oscillating-mode AFM (OM-AFM), where the tip-
surface interaction is controlled by adjusting the
AFM amplitude set-point ratio. Free oscillation
amplitude of 60 nm and the set-point ratio of 50%
were typically used. The parameters were optimized
not to influence the soft FBS layer yet to provide
optimal resolution and contrast. A nanoshaving
procedure (Rezek, 2006b; Rezek, 2007a) was
applied to evaluate the protein layer thickness. First,
SELECTIVE OSTEOBLASTIC CELL MICRO-ARRAYS ON DIAMOND FILMS
349

Figure 4: Fluorescent microscopy images of osteoblastic cells (SAOS-2) cultivated for 2 days on 100m H-/O-terminated
stripes on diamond films: (a) low initial cell seeding concentration (2,500 cells/cm
2
), (b) high initial cell seeding
concentration (10,000 cells/cm
2
), and (c) cells bridging of unfavorable H-terminated regions. The fluorescence shows actin
stress fibers (green) and nuclei (blue). Scale bar is 100m.

Figure 5: Fluorescent microscopy images of osteoblastic cells (SAOS-2) cultivated for 2 days on 200m H-/O-terminated
stripes on diamond films in McCoy's medium supplemented with different fetal bovine serum (FBS) concentrations (0, 5,
10, and 15%). Scale bar is 100m.
a region of 2x2 m
2
was scanned in contact AFM
(C-AFM) and then re-measured across somewhat
larger area by OM-AFM. The force applied during
C-AFM was approx. 200 nN. The interaction forces
in OM-AFM are orders of magnitude lower. The
FBS layer thickness was then determined as the
difference between average height values across 1
um
2
of the FBS layer surface and 1 m
2
of the
nanoshaved area where FBS was removed. Several
regions were probed on each sample to determine
the error bar from root-mean-square (RMS)
roughness values and statistical errors.
Autocorrelation function of the images was
calculated to determine typical lateral feature size
(Lx).
3 RESULTS
Correlation of oxygen- and hydrogen-terminated
micro-patterns on the diamond films with patterns of
cell adhesion on such structures is illustrated in
Figure 2. Figure 2(a) shows a bright field image of
the micro-structured sample in optical microscope
before cell seeding. The surface is featureless as the
patterns are optically invisible. Figure 2(b) presents
SEM image of the sample, where H- and O-
terminated patterns (width of 200 m) are clearly
identified due to their different electronic properties.
The bright stripes correspond to the H-terminated
NCD surface, having negative electron affinity
(Maier, 2001). The dark stripes represent the O-
terminated NCD surface. Fluorescently stained
human osteoblasts adherent on 200 m wide
patterned surface are presented in Figure 2(c). By
correlating a position of the alignment mark in SEM
and fluorescent microscopy pictures, it is evident
that the osteoblastic cells preferentially colonize the
O-terminated (hydrophilic) patterns.
Figure 3 shows that the cells adhere
preferentially onto O-terminated stripes
independently of the stripe width in the range of 30-
200m. Two types of cell adhesion patterns are
detectable. Cells on the narrow stripes (30m
smaller than the cell size) are elongated and form
cell-by-cell arrays. On wider stripes (60, 100, and
200m bigger than the cell size) the cells spread
and fill the entire width of the stripe. At the micro-
pattern borders they form a sharp boundary.
Osteoblast adhesion onto the NCD surface is
affected by the initial cell seeding concentration.
Figure 4 illustrates higher selectivity for cell
adhesion on the O-terminated surface at lower initial
cell seeding density (2,500 cells/cm
2
). There is still
some free space for cell spreading and expansion
within the hydrophilic region. On the other hand,
cells plated at the higher density (10,000 cells/cm
2
)
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
350

colonize not only hydrophilic areas but also
unfavorable hydrophobic regions (Figure 4(b)).
Figure 4(c) presents an abnormally long single cell
(left image side) as well as clusters of several cells
(right image side) that can bridge and colonize the
hydrophobic area.
Figure 5 demonstrates the influence of different
initial FBS concentrations (0, 5, 10, and 15%) in the
culture medium on the cell attachment onto the H-
/O-patterned surface. The range of serum
concentrations 5-15% does not significantly affect
the cell adhesion pattern. The cells follow the H-/O-
terminated micro-patterns in the same way as shown
in the previous figures. In a sharp contrast, cells
plated in FBS-free medium colonize the surface
independently of the micro-patterns. The cell
selectivity is obviously determined by the FBS
proteins.
Figure 6 shows OM-AFM topography image of
the FBS layer on diamond with stripe-like patterns
of hydrogen and oxygen surface terminations. The
diagonal lines in the background are due to polishing
of the diamond substrate. The roughness of diamond
substrate is about 0.6 nm. On this background one
can see clear stripes on the O-terminated surface.
There are also some small scattered islands of
similar thickness on the H-terminated stripes, most
likely due to certain degree of non-specific
adsorption. When the height of stripes is probed by
the nanoshaving method, we find that the layer
thickness of the layer adsorbed on O-terminated
diamond is 4 2 nm. Even on H-terminated surface
(outside of the islands) there is a thin layer of 1.5 2
nm. Hence the FBS layer is present on both types of
diamond surfaces, although in the different
thickness.

Figure 6: AFM topography image of a fetal bovine serum
(FBS) layer on the diamond with stripe-like patterns of H
and O surface terminations.

Figure 7: AFM force curves on H- and O-terminated
diamond with the adsorbed FBS layer.

Figure 8: AFM measurements in FBS/McCoys medium
on hydrogen- and oxygen-terminated diamond surfaces
with adsorbed FBS layers: topography and phase image on
(a-b) FBS/H-terminated diamond (c-d) FBS/O-terminated
diamond.
Figure 7 shows force curves obtained by force
spectroscopy on H- and O-terminated diamond. The
force curves exhibit 500 100 pN interaction
between tip and surface on both H- and O-
terminated diamond. Similar forces and shapes were
found between cantilevers functionalized by bovine
serum albumin and glass surfaces after deposition of
proteins (Popov, 2007). Hence also the protein
molecules from FBS are present on both H- and O-
terminated diamond.
Figure 8 shows the detailed topography and
phase images on both types of surfaces. Values of
RMS roughness and lateral feature size (Lx) are also
given. The roughness of FBS layer on O-terminated
diamond (1.7 nm) is about two times higher
SELECTIVE OSTEOBLASTIC CELL MICRO-ARRAYS ON DIAMOND FILMS
351

compared to H-terminated diamond (0.6 nm). The
topographic features are different, with a kind of
ridge-like shapes around valleys on H-terminated
diamond and hillock-like shape on O-terminated
diamond. Correspondingly, the feature size is also
different, about 10 nm on H-terminated diamond and
20 nm on O-terminated diamond. A pronounced
difference is detected also in the AFM phase images.
AFM phase image of the adsorbed layer on H-
terminated diamond is dominated by dark dots
correlated with protrusions in morphology. On O-
terminated diamond, brighter spots having darker
boundaries are correlated with the hillocks.
4 DISCUSSION
In correlation with previous reports on homogeneous
surface termination of diamond (Kalbacova, 2007a,
Bacakova, 2007) we find that in case of H/O micro-
patterns the cells colonize preferentially hydrophilic
(O-terminated) stripes forming confluent arrays with
sharp edges separating O- and H- terminated
regions. The cells generally did not show any
decreased viability, however some of them
(preferentially on hydrophobic region) remain
rounded for an extended period of time exhibiting
poor cell-substratum-compatibility (Liu, 2007;
Michalikova, 2008). Evolution of cell morphology
on hydrophobic surfaces is slower, but otherwise not
remarkably different than that observed for human
osteoblasts (hFOB) (Liu, 2007) or SAOS-2 on more
hydrophilic surfaces it is an example of the time-
cell-substratum-compatibility-superposition
principle.
Also noteworthy is bridging of unfavorable H-
terminated regions as illustrated in Figure 4(c). This
is obviously enabled by connection to the cells on
the O-terminated regions because solitaire cells on
the H-terminated regions exhibit bad adhesion and
reduced metabolic activity (Kalbacova, 2007a;
Kalbacova, 2008). To reach the optimal status on
unfitting surface, cells will communicate with each
other, exchanging growth factors and various stimuli
as well as produce extracellular matrix (ECM) and
thus modify the surface with proteins and
proteoglycans underneath to overcome the
inhospitable environment. It is known that proteins
adsorbed onto the substrate surface do not become
permanently immobilized. They will be
enzymatically degraded, denatured, they undergo
conformational and configuration changes and will
even be replaced by other proteins (Zeng, 1999).
However, when more cells are able to gently attach
to hydrophobic surface in a specific pattern (forming
a bridge between two hydrophilic stripes) then these
cells may form ECM faster due to support from their
proliferating neighbors, thus masking unsuitable
properties of the surface. This may be very useful
mechanism for bio-electronic applications as it
enables to overgrow electrically conductive H-
terminated surface when it is surrounded by O-
terminated regions at small enough dimensions.
As the cell adsorption is protein mediated, a
question arises whether the specific cell adsorption
is due to direct effect of diamond surface dipoles on
the cells or due to differences in protein adsorptions
on the micro-patterns. Figure 5 clearly demonstrates
that cells plated without protein (FBS-free medium)
do not sense any chemical micro-patterning, whereas
cells plated in FBS-supplemented medium clearly
follow the hydrophilic patterns. It proves that the
cell selectivity is driven by the FBS protein
adsorption. Since protein adsorption is much more
rapid than the transport of cells to the surface, it is
expected that the interaction of host cells with the
material is determined by the nature of this adsorbed
protein layer.
AFM study of the protein layers revealed that
FBS adsorbs on both types of diamond surfaces.
This is in agreement with previous reports that
albumin adsorbs on both hydrophilic and
hydrophobic surfaces (Browne, 2004). Here, the
adsorbed thickness differs by few nm. It should be
noted that FBS layer is a soft matter so there is some
uncertainty in determining its thickness by AFM
because even in OM-AFM the height may be
underestimated (Rezek, 2007a; Rezek, 2007b).
Another influence on the observed step in the height
across the nanoshaved region may be wear of the
substrate material. As the flat bulk diamond is very
hard compared to proteins and its wear is extremely
low, only the FBS layer was penetrated and removed
by the nanoshaving forces applied here.
The cell selectivity is thus not determined merely
by FBS layer presence. More subtle differences must
be considered for explaining the selective
adsorption, such as protein denaturation on
hydrophobic surfaces (Ukraintsev, 2007; Zeng,
1999). Detailed studies of surface morphology
revealed clear differences in surface roughness,
morphological features and phase images between
the protein layers on H- and O-terminated diamond.
By comparison with the literature (Browne, 2004) ,
where similar difference in topography on
polystyrene substrates were shown, the most
important factor for the cell growth on diamond
seems to be the wetting property of the surface
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
352

rather than any other specific property of the
diamond films.
One has to critically consider that the actual
composition of the adsorbed layers may be different
on H- and O-terminated diamond because various
proteins (albumin, fibronectin, vitronectin, etc.) from
FBS may influence the cell adhesion in different
ways. Further experiments are needed to elucidate
these details.
5 CONCLUSIONS
Chemical patterning of diamond films by hydrogen
and oxygen surface atoms enables self-assembly of
human osteoblastic cell micro-arrays. The cell
adhesion and assembly on diamond can be further
controlled and optimized by biochemical factors.
The cells strongly prefer O-terminated patterns. The
best selectivity is achieved for lower initial cell
concentrations (2,500 cells/cm
2
), regardless of
surface geometry and commonly used protein (FBS)
concentrations (5 to 15%). Widths of the patterns
affect the shape of adhered cells in the following
way: i) good cell spreading with a sharp boundary
was observed on broader stripes and ii) elongated
cell chains were observed on stripes which were
narrower than the cell size. Higher initial
concentration of cells enables colonization of less
favorable H-terminated surface regions, which are
electrically conductive and can be employed in
electronic devices. A non-preferential cell adhesion
is found when the initial cell adhesion occurs
without the serum presence. Hence the cell
selectivity is driven by the FBS properties on H- and
O-terminated surfaces. AFM detected presence of
the FBS layer on both types of surfaces. However,
the layer thickness and microscopic morphology are
rather different. This may be the reason for the cell
selectivity. Further experiments are needed to
elucidate details of the selectivity, such as particular
composition of the adsorbed layers and so on.
Nevertheless, the presented data may already
provide valuable information for application of
diamond films in tissue engineering, implants, bio-
electronics, and biotechnology in general. We
speculate that similar cell behavior will occur also
when using other cell lines.



ACKNOWLEDGEMENTS
This work was supported by the Academy of
Sciences of the Czech Republic contracts
KAN400100701, AV0Z10100521, and Czech
Ministry of Education, Youth and Sport projects LC-
510 and MSM0021620806. The authors would like
to express their thanks to Ing. Vlastimil J urka, Zdena
Polkov, Dr. Zdenk Potmil (all Inst. Phys.
ASCR) for a kind assistance with photolithography,
and surface treatments, J ana Sovov (Inst. Inh. Met.
Disorders, 1st Fac. Med., Charles Uni.) for technical
assistance and Mgr. Veronika Bareov (Inst. Inh.
Met. Disorders, 1st Fac. Med., Charles Uni.) for a
kind assistance with fluorescent microscopy.
REFERENCES
Bacakova, L., Grausova, L., Vacik, J., Franczek A.,
Blazewicz, S., Kromka, A., Vanecek, M. & Svorcik,
V. (2007). Improved adhesion and growth of human
osteoblast-like MG 63 cells on biomaterials modified
with carbon nanoparticles. Diamond and Related
Materials, 16, 21332140.
Bajaj, P., Akin, D., Gupta, A., Sherman, D., Shi, B.,
Auciello, O. & Bashir, R. (2007). Ultrananocrystalline
diamond film as an optimal cell interface for
biomedical applications. Biomedical Microdevices, 9,
787-794.
Browne, M. M., Lubarsky, G. V., Davidson, M. R. &
Bradley, R. H. (2004). Protein adsorption onto
polystyrene surfaces studied by XPS and AFM.
Surface Science, 553, 155167.
Garrido, A.J ., Nebel, C. E., Todt, R., Roesel, G., Amann,
M. C. & Stutzmann, M. (2003). Fabrication of in-
plane gate transistors on hydrogenated diamond
surfaces. Applied Physics Letters, 82, 988-990.
Kalbacova, M., Kalbac, M., Dunsch, L., Kromka, A.,
Vanecek, M., Rezek, B., Hempel, U. & Kmoch, S.
(2007a). The effect of SWCNT and nano-diamond
films on human osteoblast cells. Physica status solidi
(b), 244, 43564359.
Kalbacova, M., Roessler, S., Hempel, U., Tsaryk, R.,
Peters, K., Scharnweber, D., Kirkpatrick, C. J ., &
Dieter, P. (2007b). The effect of electrochemically
simulated titanium cathodic corrosion products on
ROS production and metabolic activity of osteoblasts
and monocytes/macrophages. Biomaterials, 28, 3263
3272.
Kalbacova, M., Michalikova, L., Baresova, V., Kromka,
A., Rezek, B. & Kmoch, S. (2008). Adhesion of
osteoblasts on chemically patterned nanocrystalline
diamonds. In press in Physica status solidi (b).
Kozak, H., Kromka, A., & Rezek, B. (2008). Enhancing
nanocrystalline diamond surface conductivity by
deposition temperature and chemical post-processing.
Submitted in Physica status solidi (a).
SELECTIVE OSTEOBLASTIC CELL MICRO-ARRAYS ON DIAMOND FILMS
353

Kromka, A., Rezek, B., Remes, Z., Michalka, M.,
Ledinsky, M., Zemek, J ., Potmesil, J . & Vanecek, M.
(2008). Formation of continuous nanocrystalline
diamond layer on glass and silicon at low
temperatures. In press in Chemical Vapor Deposition.
Liu, X., Lim, J . Y. F., Donahue, H. J ., Dhurjati, R.,
Mastro, A. M. & Vogler, E. A. (2007). Influence of
substratum surface chemistry/energy and topography
on the human fetal osteoblastic cell line hFOB 1.19:
Phenotypic and genotypic responses observed in vitro.
Biomaterials, 28, 45354550.
Maier, F., Ristein, J . & Ley, L. (2001). Electron affinity of
plasma-hydrogenated and chemically oxidized
diamond (100) surfaces. Physical Review B, 64,
165411.
Michalikova, L., Rezek, B., Kromka, A., Kozak, H.,
Grausova, L., Bacakova, L., Vanecek, M., Kocka, J . &
Kalbacova, M. (2008). Selective Adhesion and
Arrangement of osteoblast-like cells on hydrophilic
and hydrophobic nanocrystalline diamond micro-
patterns. Submitted to Thin Solid Films.
Nebel, C. E. (2003). From gemstone to semiconductor.
Nature Materials, 2, 431-432.
Popov, C., Kulisch, W., Reithmaier, J ., Dostalova, T.,
J elinek, M., Anspach, N. & Hammann, C. (2007).
Bioproperties of nanocrystalline diamond/amorphous
carbon composite films. Diamond and Related
Materials, 16, 735-739.
Potocky, S, Kromka, A., Potmesil, J., Remes, Z., Vorlicek,
V., Vanecek, M., & Michalka, M. (2007).
Investigation of nanocrystalline diamond films grown
on silicon and glass at substrate temperature below
400C. Diamond and Related. Materials, 16, 744-747.
Rezek, B., & Nebel, C. E. (2006a). Electronic properties
of plasma hydrogenated diamond surfaces: A
microscopic study. Diamond and Related Materials,
15, 13741377.
Rezek, B., Shin, D., Nakamura, T. & Nebel, C. E. (2006b).
Geometric properties of covalently bonded DNA on
single-crystalline diamond. Journal of American
Chemical Society, 128, 3884-3885.
Rezek, B., Shin, D., Uetsuka, H. & Nebel, C. E. (2007a)
Microscopic diagnostics of DNA molecules on mono-
crystalline diamond. Physica status solidi (a), 204,
2888-2897.
Rezek, B., Shin, D.& Nebel, C. E. (2007b). Properties of
hybridized DNA arrays on single-crystalline undoped
and boron-doped (100) diamonds studied by atomic
force microscopy in electrolytes. Langmuir, 23, 7626-
7633.
Rezek, B., Shin, D., Watanabe, H. & Nebel, C. E. (2007c).
Intrinsic hydrogen-terminated diamond as ion-
sensitive field effect transistor. Sensors and Actuators
B, 122, 596-599.
Shakenraad, J . M., & Busscher, H. J . (1989). Cell-polymer
interactions:the influence of protein adsorption.
Colloids and Surfaces, 42, 331343.
Song, K. S., Hiraki, T., Umezawa, H. & Kawarada, H.
(2007). Miniaturized diamond field-effect transistors
for application in biosensors in electrolyte solution.
Applied Physics Letters, 90, 063901.
Tachiki, M., Kaibara, Y., Sumikawa, Y., Shigeno, M.,
Banno, T., Song K. S., Umezava, H. & Kawarada, H.
(2003). Diamond nanofabrication and characterization
for biosensing application. Physica status solidi (a),
199, 39-43.
Tang, L., Tsai, C., Gerberich, W. W., Kruckeberg, L. &
Kania, D. R. (1995). Biocompatibility of chemical-
vapour-deposited diamond. Biomaterials, 16, 483-488.
Ukraintsev, E. V., Kiselev, G. A., Kudrinskii, A. A.,
Lisichkin, G. V. & Yaminskii, I. V. (2007). Formation
of lysoyzme fibrils on a solid support. Polymer
Science, 49, 6-9.
Yang, W., Auciello, O., Butler, J . E., Cai, W., Carlisle, J .
A., Gerbi, J . E:, Gruen. D. M., Knickerbocker, T.,
Lasseter, T. L., Russell, J . N. J r., Smith, L. M. &
Hamers, R. J . (2002). DNA-modified nanocrystalline
diamond thin-films as stable, biologically active
substrates. Nature Materials, 1, 253-257.
Yang, W., Butler, J . E., Russell, J . N. & Hamers, R. J .
(2004). Interfacial electrical properties of DNA-
modified diamond thin films: Intrinsic response and
hybridization-induced field effects. Langmuir, 20,
6778-6787.
Zemek, J ., Houdkova, J . Lesiak, B., J ablonski, A.,
Potmesil, J ., & Vanecek, M. (2006). Electron
spectroscopy of nanocrystalline diamond surfaces.
Journal of Optoelectronics and Advanced Materials,
8, 21332138.
Zeng, H., Chittur, K. K. & Lacefield, W. R. (1999).
Analysis of bovine serum albumin adsorption on
calcium phosphate and titanium surfaces.
Biomaterials, 20, 377384.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
354
ON-DETECTOR ELECTRONICS
OF THE CLEAR PEM SCANNER
E. Albuquerque
1
, V. Bexiga
1
, R. Bugalho
2
, B. Carrio
2
, C. S. Ferreira
2
, M. Ferreira
2
, J . Godinho
3

F. Gonalves
1
, C. Leong
1
, P. Lous
3
, P. Machado
1
, R. Moura
2
, P. Neves
3
, C. Ortigo
2
, F. Piedade
3

J . F. Pinheiro
2
, P. Relvas
3
, A. Rivetti
4
, P. Rodrigues
2
, J . C. Silva
2,6
, M. M. Silva
1
, I. C. Teixeira
1,5

J . P. Teixeira
1,5
, A. Trindade
2
and J . Varela
2,5,6

1
INESC-ID - Inst. de Eng de Sistemas e Computao, Investigao e Desenvolvimento, Lisboa, Portugal
{edgar.albuquerque,fernando.goncalves,manuel.silva}@inesc-id.pt, [email protected]
[email protected], {jct,isabel.teixeira}@inesc.pt, [email protected]
2
LIP - Lab. de Instrumentao e Fsica Experimental de Partculas, Lisboa, Portugal
{joao.varela,jc.silva}@cern.ch, {psilva,miguel,moura,bruno,jfilipe,andreia,ortigao,claudia}@lip.pt, [email protected]
3
INOV Inst. de Novas Tecnologias, Lisboa, Portugal
{paulo.relvas,pedro.lousa,joaquim.godinho,pedro.neves,fernando.piedade}@inov.pt
4
INFN - Istituto Nazionale di Fisica Nucleare, Torino, Italia
[email protected]
5
also with Inst. Superior Tcnico, Univ. Tcnica de Lisboa, Portugal
6
also with CERN, Geneva, Switzerland
Keywords: PET, Scanner, Breast, Cancer, SPECT.
Abstract: A Portuguese consortium has developed a PET scanner dedicated to breast cancer detection (Clear-PEM
scanner) within the framework of the international Crystal Clear Collaboration at CERN. In the construction
of this scanner several challenges have been addressed, from the design of the photons detector, front-end
electronics and data acquisition systems up to the image reconstruction algorithms. In this paper we describe
the development of the electronics in the detector heads needed to read-out and filter the data from 12288
detector channels, as well as to provide regulated high-voltages, low voltage power and control signals, and
also to monitor the environment in the detector heads. The scanner is currently in its final phase of
integration and will soon be installed in the department of Nuclear Medicine of Hospital Garcia de Orta and
Instituto Portugus de Oncologia (Porto) were clinical trials will be conducted.
1 INTRODUCTION
The Clear-PEM detector is a Positron Emission
Mammography scanner that was developed by
several Portuguese institutions within the framework
of the international Crystal Clear Collaboration at
CERN (Abreu, 2006).
The detector assembly is based on two detecting
planar heads. The detection heads are mounted on a
robotized mechanical system, enabling the exam of
both breasts, one at a time, as well as the axillary
lymph nodes.
The basic element of the scanner is the detector
module (Fig. 1). Each module has 32 LYSO:Ce
crystals with 2x2x20 mm
3
assembled in a matrix
coupled on both ends to avalanche photo-diodes
(APD) (Amaral, 2007). Twelve of these detector
modules are assembled in a mechanical structure
placed between two Frontend boards forming a
Supermodule (Fig. 2).

Figure 1: Detector module assembly scheme.
355
The traditional readout based on photomultipliers
is replaced by multi-pixel APDs. Due to its
compactness, it is possible to read each single crystal
with one APD pixel on each end, and to use the
relative amplitude of the two signals to estimate the
longitudinal coordinate of the photon interaction.
The individual 1:1 crystal-APD pixel coupling
leads to 12 288 detector channels, with a density at
about 13 channels per centimeter square. The limited
available space in the detector heads demand that all
the processing electronics must have a strict limited
power consumption budget. This and the low gain
(100) of the APDs, has lead to the development, for
the Clear-PEM scanner, of specifically tailored low-
noise Application-Specific Integrated Circuit (ASIC).
Frontend electronics boards based on the Clear-
PEM ASICs provide the first level of signal
processing, including readout, amplification,
sampling and storage in analogue memories of the
APD array signals.
The ASICs output pulses are digitized by free-
sampling ADCs in the Frontend boards. LVDS data
links are used to transmit the detector data to the off-
detector electronics system, which implements the
first-level trigger, and data concentration and
transmission to the data acquisition server
(Albuquerque, 2008).
Auxiliary service boards located in the detector
heads are needed to provide regulated high-voltages
for each of the APD arrays and to distribute low
voltage power as well as control and clock signals.
The electronics is also responsible to monitor the
detector heads environment (temperature and
pressure).
Each Clear-PEM detector head has a total 3072
crystals grouped in 96 detector modules and 8
Supermodules. The detector head includes also one
Service Board, one high-voltage connection Matrix
Board and one clock fan-out unit. These electronics
boards are described in the following sections.
2 THE FRONT END
ELECTRONICS SYSTEM
The Frontend electronics system is one of the most
challenging and innovative sub-systems of the
Clear-PEM detector. It is composed by the Frontend
boards which interface directly with the APD arrays
assembled in the detector modules and are connected
to the Auxiliary Boards in the detector head.
The system, physically located on the detector
heads, performs signal amplification, channel
selection and analog multiplexing, analog to digital
conversion and parallel-to-serial translation.
A frontend ASIC has been developed for readout
of the multi-pixel S8550 Hamamatsu APDs.
Themixed-signal ASIC incorporates 192 low-noise
charge pre-amplifiers, shapers, analogue memory
cells and digital control blocks. Pulses are
continuously stored in memory cells at clock
frequency. Channels above a common threshold
voltage are readout for digitization by off-chip free
sampling ADCs. The number of output channels of
the frontend ASIC is two, still allowing for the
readout of two-hit Compton interactions in the
detector. The ASIC has a size of 7.3 mm x 9.8 mm
and was designed in 0.35 m CMOS technology.
The Frontend Board (FEB) integrates two 192
channels ASICs and two dual free-sampling 10-bit
ADC chips working at frequencies up to 100 MHz.
The digitized data is transmitted to the off-detector
data acquisition system by LVDS serial links at 600
MHz.


Figure 2: Supermodule structure assembling 12 modules,
each with 32 LYSO:Ce crystals and two 32-pixel APD
arrays in double readout. Each Frontend board has two
ASICs with 192 input channels.
Two FEBs are used to mount one supermodule
structure with a total of 768 electronic channels and
dimensions of 12x4.5 cm
2
as illustrated in Fig. 2.
The frontend electronics must have low-noise
due to the initial reduced charge at the amplifier
input, which for a 511 keV photon energy deposit is
around 30fC (maximum value). The frontend ASIC
amplifies this charge by about three orders of
magnitude, while complying with the low-power
dissipation requirements (5 mW/channel),
compatible with a compact water based cooling
system that allows to operate the detector at 18
o
C. A
temperature stability of the order of 0.1
o
C

is
required since the LYSO:Ce light yield and APD
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
356
gain are inversely dependent on the temperature (2-
3%/
o
C).
The Frontend Boards have a mixed analog-digital
environment and therefore special care is needed
regarding the correct conditioning of the digital,
large amplitude, high-frequency clock and other
periodic signals. Noise pickup in the PCB traces that
connect the APD outputs to the frontend chips must
also be minimized.
The performance of the Supermodules was
evaluated in a setup that integrates all the electronics
and data acquisition sub-systems as it will operate in
the final detector. A cooling system based on
controlled flux of cold water was build to cope with
the power dissipation from the Frontend Boards
during the qualification tests.
Acquisition runs with
22
Na and
137
Cs radioactive
sources, as well as background acquisitions from the
176
Lu natural radioactivity of the LYSO:Ce crystals
have been performed (R. Bugalho, 2008).
The energy and time resolutions, energy linearity
of detector readout chain and output channels
occupancy are the parameters under evaluation for
each supermodule.
The noise of the detector channels, defined as the
equivalent noise charge (ENC) at the amplifier
inputs, is around 1300 e
-
RMS. This noise
contributes less than 2% to the energy resolution,
which at 511 keV is dominated by the fluctuations of
the scintillating light signal in the crystals. This
noise level implies a RMS time resolution of
individual 511 keV photons of the order of 1 ns.
3 THE AUXILIARY BOARDS
Several auxiliary boards have been designed to
provide regulated high-voltages, low voltage power,
as well as clock and control signals to the
Supermodules, and also to monitor the temperature
and pressure inside the detector heads.
These boards are placed in the back side of the
detector head, leaving the front side free of obstacles
for the detection of the PET photons.
The main auxiliary board is the Service Board
(SB) shown in figure 3.
Dedicated circuits in the SB including remotely
controlled Digital to Analog Converters (DAC) are
used to regulate the high-voltage (HV) lines (350-
500V) needed for the polarization of the APDs.
Plugging directly on the SB, the HV Matrix is
another auxiliary board that provides the
connections of the HV lines to the APD arrays
(figure 4).

Figure 3: View of the Service Board.
The connection matrix distributes 32 different
high-voltage references to the 384 APDs sub-arrays
(16 pixels) in the detector head. The APD gain
variation is of the order of 6%/V which requires a
stability of the biasing voltages better that 0.1 V.
The ripple of the high-voltage lines is less than 0.02
V. Before assembly each HV channel is calibrated in
order to guarantee that the gain of the APDs is
100.External power supplies provide independent
low-voltages for the analog and digital sections of
the detector electronics. The SB organizes the
distribution of these voltages to the Frontend Boards.
The control of the ASICs reset sequence uses an
Altera FPGA in the Service Board, and the setting of
the voltage thresholds of signal detection in the
ASICs is done by DACs controlled remotely via I2C.
The measurement of the temperature in the
detector modules uses PT100 sensors placed in
contact with the APD arrays, coupled to dedicated
signal conditioning circuits followed by ADCs
accessed remotely by the I2C control lines. The
pressure inside the detector head is also measured to
assure that slightly over-pressured nitrogen fills the
detector heads avoiding water condensation.
The system clock and a synchronization signal
are distributed to the frontend electronics. To avoid
interferences of the clock with the high-voltage lines,
a dedicated board is used to fan-out the clock and
sync signals.
Finally the detector head are closed using a
specifically designed patch panel board equipped
with vacuum tight connectors to insure the detector
box hermeticity (fig 4).

ON-DETECTOR ELECTRONICS OF THE CLEAR PEM SCANNER
357

Figure 4: View of a detector head, with the hermetic patch
panel on bottom, during the cabling phase.
4 CONCLUSIONS
In this paper, the detector electronics of the Clear-
PEM scanner, composed of the Frontend Boards
connecting directly with the detector modules and of
the Auxiliary Boards in the detector heads, was
presented. The detector electronics system is one of
the most challenging and innovative sub-systems of
the Clear-PEM scanner.
This electronics was validated in an experimental
setup that includes detector supermodules and all the
external sub-systems developed for the Clear-PEM
scanner, including the data acquisition electronics
and computing systems. The measured electronic
noise levels in the detector are below the
requirements set by the Clear-PEM specifications.
REFERENCES
Abreu et al. Design and Evaluation of the ClearPEM
Scanner for Positron Emission Mammography,
IEEE Trans. Nucl. Sci. (2006), Vol.53(1), pp 71-7.
Amaral et al. Performance and quality control of Clear
PEM detector modules, Nucl. Instrum. And Meth. In
Phys. Res. (2007) A580, 1123-1126.
R. Bugalho et al. Validation of Clear-PEM Data
Acquisition Electronics and Operation Software sub.
IEEE NSS-MIC 2008.
Albuquerque et al. Performance Evaluation of a Highly
Integrated APD/ASIC Double-Readout Supermodule
with 768 Channels for Clear-PEM sub. IEEE NSS-
MIC 2008.


BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
358
HARDWARE IMPLEMENTATION FOR EDGE
DETECTION IN CDNA MICROARRAY IMAGES
Bogdan Belean, Monica Borda
Technical University of Cluj-Napoca, Faculty of Electronics Teleomunications and Information Technology
Comunications Department, 26 28 George Baritiu, Cluj-Napoca, Romania
[email protected], [email protected]
Albert Fazakas
Technical University of Cluj-Napoca, Faculty of Electronics Teleomunications and Information Technology
Electronics Basics Department, 26 28 George Baritiu, Cluj-Napoca, Romania
[email protected]
Keywords: cDNA Microarray, Image Processing, Parallel Processing, FPGA Technology.
Abstract: The present paper proposes hardware implementation strategies for cDNA microarray image processing in
order to overcome the main disadvantage introduced by the existing computational tools, the increased
processing time. A hardware implementation of the Canny edge detection algorithm for microarray spots is
described. The implementation takes advantage of spatial and temporal parallelism offered by FPGA
technology. Results of the hardware implementation which prove time and cost efficiency are presented.
1 INTRODUCTION
Microarray experiments are providing genome wide-
data on gene expression patterns. Different
techniques including SAGE, differential display,
oligonucleotide array and cDNA microarrays have
been developed. The techniques offer the possibility
of mRNA expression to be assessed on a global
scale, allowing the parallel assessment of gene
expression for thousands of genes in a single
experiment. Gene expression represents the
transformation of genes information into proteins.
The most common use of these techniques is for the
determination of patterns of differential gene
expression, comparing differences in mRNA
expression levels between identical cells subjected
to different stimuli or between different cellular
phenotypes or developement stages (Bajcsy, 2004).
The hardware processing techniques described in
this paper will focus on processing images obtained
as a result of cDNA microarray experiment. cDNA
microarrays represent gene specific probes arrayed
on a matrix such as a glass slide or microchip.
Usually, samples from two sources are labelled with
two different fluorescent markers and hybridized on
the same array (glass slide). After the hybridization,
the array is scanned using two light sources with
different lengths (red and green) to determine the
amount of labelled sample bound to each spot
through hybridization process. The light sources
induce fluorescence in the spots, which is captured
by a scanner and a composite image is produced.

Figure 1: Microarray image with underlined independent
sub-image. Stanford University Medical Centre,
Department of Genetics.
359

Further on, image processing techniques are used
to quantify gene expression levels present in the
captured microarray image, in order to identify a
gene in a biological sequence and to predict the
function of the identified gene. The flow of
processing a microarray image (Yang, 2001) is
generally separated in the following tasks:
addressing, segmentation, intensity extraction and
pre-processing to improve image quality and
enhance weakly expressed spots. The first step
associates an address to each spot of the image. In
the second one, pixels are classified either as
foreground, representing the cDNA spots, or as
background. The last step calculates spot intensities
and estimates background intensity values.
The major tasks of microarray image processing,
which contributes in fulfilling the last mentioned
steps, are to identify the array format including the
array layout, spot size and shape, spot intensities and
distances between spots. The main parameters taken
into consideration in microarray image processing
are accuracy and time. The accuracy is given by the
quality of image processing techniques. The second
parameter, time, is critical due to the large amount of
data contained in a microarray image. A regular
microarray image has up to hundreds of MB, and it
can be divided in independent sub-images, which
consists in a compact group of spots as it can be seen
in figure 1. Sophisticated computational tools are
available for microarray image processing but, their
main disadvantages are the increased computational
time and the user intervention needed in processing.
To overcome the previous disadvantages,
microarray images are analyzed and processed using
FPGA technology. The hardware implementations
of microarray image processing techniques make use
of the features of FPGA, which allow accessing at
the same time hundreds of memory addresses. In this
way, calculations specific to microarray image
processing are made in parallel, increasing the
processing speed. For this kind of processing, image
acquisition is mandatory and its description is
presented in the following papers (Belean, 2008).
This paper proposes hardware strategies for
microarray image processing (paragraph 2), and also
an implementation of spot border detection
algorithm using the Canny filter.
2 FPGA & MICROARRAYS
FPGA technology uses pre-built logic blocks and
programmable routing resources to configure these
chips and to implement custom hardware
functionality. Their main benefits are the low cost,
the short time to market and also the increased
performance due to their structure which is able to
exploit spatial and temporal parallelism. These
advantages will be used to develop a system on a
chip in order to process the microarray images in a
manner that does not need user intervention. Also,
time being critical in microarray image processing,
using FPGA technology will decrease the
computational time due to its parallel computation
capabilities. The main goal of this approach for
microarray image processing is to obtain a device
which will be able to extract and quantify gene
expression a lot more faster than existing
computational tools, so that microarray analyses to
be easily performed on a large number of subjects
(thousands of patients and different diseases). This
task can hardly be achieved with the existing tools
which are time consuming and which also need user
intervention.
2.1 Hardware Implementation for
Microarray Image Processing
The hardware implementation strategies for
microarray image processing techniques presented in
this paper takes advantage of the structure size and
shape of a microarray image. As in figure 1, a piece
of the microarray image is delimited, and represents
an independent microarray sub-image. Taking into
account that the microarray image composed by
independent sub-images is written in a RAM
memory after the acquisition, a first step in
processing is to crop the image by determining the
address and dimension of each sub-image. Once
these parameters are known, the independent sub-
images are copied one by one in the Block RAM
memory inside the FPGA for further processing
which consists in addressing each spot of the image.
This type of memory offers multi-port and high
speed access needed for the next step of processing
which aims to determine the spot contour and extract
the spot intensity for quantifying gene expression.
The image being cropped and copied into the
Block RAM memory, hardware design techniques
such as parallelism and pipelining can be developed
using FPGA technology. These techniques will be
presented in the next subchapters.
2.1.1 Hardware Algorithms
Image processing operations like median filters and
gradient calculation are based on the convolution,
which is included in a class of algorithms called
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
360

spatial filters. Convolution is used for implementing
image operators which have as output pixel value a
linear combination between pixels of the original
image. Conceptually, each pixel in the output image
is produced by sliding an NxM window over the
input image and computing an operation according
to the input pixels under the window and the chosen
window operator. The hardware approach for
convolution is presented as follows: the entire input
image is stored into a frame buffer; each time the
window is moved, MxN pixels values are required to
calculate the resulted pixel value. Memory bandwith
constraints make obtaining all these pixels each
clock cycle impossible, so local caching is
performed. In this way, N-1 rows are cached using a
shift register which leads to the next block diagram:

Figure 2: Block diagram for hardware implementation of
convolution operator (J ohnston, 2004).
In this case, instead of sliding a window across
the image, the implementation feeds the image
through the window.
2.1.2 Temporal Parallelism
Based on the possibility of cropping a microarray
image into independent sub-images, a strategy for
microarray image processing emerges. The idea is to
design an architecture which processes in parallel
two or more independent sub-images. Once the
address and dimension of each independent sub-
image is determined, two ore more microarray sub-
images can be copied into the FPGA block RAM
depending on the capacity of block RAM. In figure
3 a cropped microarray image which consists in a
matrix of independent sub-images called A(i,j) is
presented. The numbers from 1 to 3 and from 1 to
3 represent the operations applied on two
independent microarray sub-images. This way,
operation 1 stands for copying the sub-image from
RAM to the FPGA block RAM. Operation 2
represents the processing of the sub-image using
FPGA technology and spatial parallelism. Operation
3 copies the microarray processed sub-image back
into the RAM, on the exact place from which it was
transferred into FPGA. The operations 1, 2 and 3 are
done for each microarray sub-image. In the end we
will have in the RAM memory a processed
microarray image.



Figure 3: Block diagram for pipeline architecture of
microarray image processing.
Time beeing critical in microarray image
processing, the previous approach reduces the
processing time to half of its regular value. As an
exemple we can mention that, while a sub-image is
copied into the block RAM, image processing
techniques are applied on another independent sub-
image, in case that there is place for more than one
microarray independent sub-image in RAM. A view
in time of the current architecture can be seen in the
next figure:

Figure 4: Time representation of the operation 1,2 and 3 in
pipelining.
2.1.3 Spatial Parallelism
To start with, it is to be mentioned that processors
divide computation across time, while dedicated
logic divides it across space, which highly decreases
computational time. Taking into consideration the
large amount of data contained in a microarray
image and also the repetitive nature of this type of
data, FPGA dedicated logic was chosen for the
implementation of microarray image processing
techniques in order to reduce computational time.
The use of hardware description language (HDL)
allows a description of a design with parallel data
paths and simultaneous computation.
Some of the major tasks in microarray image
processing are to extract spot intensities and to
determine spot contour and dimension. The
dimension of a regular spot is around 20-40 pixels
height and length. The reduced dimension of a
A(0,0) A(0,1)
A(1,0) A(1,1)
FPGA
Block RAM
A(i,j) A(i,j)+1
Cropped microarray image
representation
in RAM memory
1
2
3
1'
2'
3'
HARDWARE IMPLEMENTATION FOR EDGE DETECTION IN CDNA MICROARRAY IMAGES
361

microarray spot offers the possibility to exploit
spatial parallelism capabilities of FPGAs. Each spot
is copied into the distributed RAM of the FPGA, so
spatial computation can be applied, thanks to the
simultaneous access to each pixel.
3 HARDWARE
IMPLEMENTATION
This paper presents a hardware implementation of an
adaptive edge detection filter using FPGA, which
provides the necessary performance for fast
microarray image processing. In microarray image
processing, edge detection is a fundamental tool
used as a precursor step to intensity extraction and
spot segmentation. Edges occur at images location
with strong intensity contrast. For edge detection a
high-pass filter in Fourier domain can be applied, or
convolution with an appropriate kernel (Sobel,
Prewitt etc.) in the spatial domain is useful.
Convolution in the spatial domain has been chosen
for implementation because it is computationally
less expansive and offers better results.
The algorithm used for the hardware
implementation is Canny algorithm, which is
considered to be optimal, based on the following: it
finds the most edges, marks the edge as close as
possible to the actual edges, provides sharp and thin
edges. The filter that meets all the criteria mentioned
above can be efficiently approximated using the first
derivative of a Gaussian function. So the first two
steps in applying the Canny algorithm would be
smoothing the image and differentiating the image in
two orthogonal directions. The next step, non-
maximum suppression, computes the gradient
direction and magnitude in order to eliminate the
pixels that represent false edges.
I
n
p
u
t
O
u
t
p
u
t

Figure 5: Block diagram for Canny algorithm.
The previously described algorithm is applied on
a microarray spot. Due to its small dimension
(approximately 40x40 pixels), the whole spot is
copied into the FPGA, so hardware processing
strategies (spatial parallelism) described in the
previous paragraph can be applied. Further on, the
hardware implementations for each of the previous
steps are presented.
Smoothing operation is done using the following
convolution mask, because it contains a division by
2
8
that is easily done with an 8 bits shift operation:
21 31 21
31 48 31
21 31 21
256
1

The effect of the previous Gaussian convolution is to
blur the image, eliminating noise. As it can be seen
in the next diagram, for processing 1 pixel only one
clock cycle is needed, because all the neighborhood
pixels used in computation are accessed at the same
time.

Figure 6: Gaussian filter implementation.
After smoothing the image, the next step is
gradient calculation in order to find the edge
strength of the spot. To determine the orthogonal
gradient at each pixel location, the following
convolution masks are used:
0 0 0
1 0 1
0 0 0
and
0 1 0
0 0 0
0 1 0

In this case, only arithmetic operations of addition
are used, so for the whole microarray spot written in
FPGA distributed RAM, a single clock cycle and
2x40x40 adders are enough to determine the resulted
image after applying the convolution masks.
The sign and value of the orthogonal components
of the gradient determined before are used in
estimating the magnitude and the direction of the
gradient. Once the direction of the gradient is
known, pixels values around the pixel being
analyzed are interpolated. The pixel that does not
represent a local maximum is eliminated, by
comparing it with its neighbors along the direction
of the gradient (non maximum suppression).
Combinational logic and basic operations of addition
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
362

and shifting are used in implementation, so each
clock cycle one pixel its processed.
3.1 Experimental Results
The hardware platform used for implementation
includes the XC3S5000 and a quartz oscillator with
a frequency of 50 MHz used to generate the clock
signal. Summing up the time needed for each step of
the edge detection algorithm applied on a 20x20
pixels spot, a processing time of 3.2 s is obtained.
This results leads to a processing rate of
approximately 250 Mbytes/second.
The next table presents the FPGA resource usage
for such an implementation.
Table 1: Statistics of FPGA (XC3S5000) resource usage
for edge detection of microarray spot.


(a) (b) (c)
Figure 7: Preliminary results obtained on a 20x20
microarray spot; a) original image, b) smoothed image,
c) edge detection.
The results of the implementation applied on a
microarray spot are presented in the previous figure.
4 CONCLUSIONS
The hardware implementation strategies for image
processing presented in this paper represent an
essential step for microarray image processing on a
hardware platform. The results of the hardware
implementation for edge detection in cDNA
microarray image emphasize the importance of
using hardware methods in cDNA microarray image
processing. The technology chosen to implement a
digital system for microarray image processing was
FPGA, due to its parallel computation capabilities
and to the possibility of reconfiguration.
To sum up, the experimental results made this
hardware technology a solution for realizing a fast,
cost efficient and accurate automated system for
cDNA microarray image processing.
ACKNOWLEDGEMENTS
This work was supported by PNII IDEI Nr.
332/2007 grant, code 909 and also by CNCSIS BD
scholarship.
REFERENCES
Bajcsy, P., 2004. An Overview of DNA Microarray Image
Requirements for Automated Processing, IEEE
Transactions on Image Processing, Vol. 13, No. 1, pp.
15-25
Yang, Y., Buckley, M., Dudoit, S., and Speed, T., 2001.
Comparison of methods for image analysis on cdna
microarray data, Department of statistics - University
of California Berkeley, Technical Report 584
Li, Q., Fraley, C., Baumgarner,R., Yeung, K., and Raftery,
E., 2005. Donuts, scratches and blanks: Robust model-
based segmentation of microarray images,
Department of statistics - University of Washington.
Technical Report 473.
Belean, B., Borda, M., Fazakas, A., 2008, Adaptive
Microarray Image Acquisition System and Microarray
Image Processing Using FPGA Technolog. KES2008,
Springer (in print).
J ohnston, C.T., 2004. Implementing image processing
algorithms on FPGA, Institute of Information Sciences
and Technology, Palmerston, New Zeeland
Blekas, K., Likas, A., Legaris, I., 2005. Mixture model
analysis of dna microarray images, IEEE Transactions
on medical imaging, pp. 901-909.
HARDWARE IMPLEMENTATION FOR EDGE DETECTION IN CDNA MICROARRAY IMAGES
363
NEW FAST TRAINING ALGORITHM SUITABLE FOR
HARDWARE KOHONEN NEURAL NETWORKS DESIGNED
FOR ANALYSIS OF BIOMEDICAL SIGNALS
Rafa Dugosz
Swiss Federal Institute of Technology in Lausanne, Institute of Microtechnology
Rue A.-L. Breguet 2, CH-2000, Neuchtel, Switzerland
[email protected]
Marta Kolasa
University of Technology and Life Sciences, Institute of Electrical Engineering, ul. Kaliskiego 7, 85-791, Bydgoszcz, Poland
[email protected]
Keywords: Kohonen Neural Network, Optimized learning process, Linear and nonlinear filtering, WBSN applications.
Abstract: A new optimized algorithm for the learning process suitable for hardware implemented Winner Takes Most
Kohonen Neural Network (KNN) has been proposed in the paper. In networks of this type a neighborhood
mechanism is used to improve the convergence properties of the network by decreasing the quantization
error. The proposed technique bases on the observation that the quantization error does not decrease
monotonically during the learning process but there are some activity phases, in which this error decreases
very fast and then the stagnation phases, in which the error does not decrease. The stagnation phases
usually are much longer than the activity phases, which in practice means that the network makes a progress
in training only in short periods of the learning process. The proposed technique using a set of linear and
nonlinear filters detects the activity phases and controls the neighborhood R in such a way to shorten the
stagnation phases. As a result, the learning process may be 16 times faster than in the classic approach, in
which the radius R decreases linearly. The intended application of the proposed solution will be in Wireless
Body Sensor Networks (WBSN) in classification and analysis of the EMG and the ECG biomedical signals.
1 INTRODUCTION
Application of Artificial Neural Networks (ANNs)
in medical diagnostic tools may be observed for
many years, e.g. in classification of biomedical
ECG, EMG signals (Osowski, 2001), (Ghongade,
Ghatolfor, 2007), segmentation and analysis of brain
or mammography images (Wismller et al., 2002)
and many others.
In most cases ANNs are realized using software
platforms that is very convenient, but such networks
cannot be used in low power diagnostic devices.
Authors recently designed an experimental
Kohonen network in CMOS technology that enables
parallel data processing (Dugosz et al. 2008),
(Dugosz and Kolasa 2008). Networks of this type
may be hundreds times faster than networks realized
in software, consuming much less energy.
Various optimization techniques of the learning
algorithm for KNN have been proposed (Zeb Shah,
Salim, 2006), (McInerney, Dhawan 1994) but these
techniques are not suitable for hardware networks
due to large computational complexity. In this paper
a new optimized learning algorithm is proposed that
bases on simple filters, which makes this technique
much easier to implement in hardware. This
technique will be used in a next prototype of the
KNN in analysis of biomedical ECG and EMG
signals in Wireless Body Sensor Networks (WBSN).
Extracting the useful features of the ECG and the
EMG signals for use with ANNs is the problem
itself, which has been addressed by many papers e.g.
(Ghongade, Ghatolfor, 2007). In this paper example
simulation results are presented for selected training
data that are representative for different applications,
but can easily be adopted to biomedical data.
364

2 KOHONEN NEURAL
NETWORK
Kohonen neural networks typically consist of a
single layer of neurons arranged as a map, with the
number of outputs equal to the number of neurons,
and the number of inputs equal to the number of
weights in neurons. In practical applications 2-D
maps are the most commonly used, since they allow
for good data visualization (Kohonen 2001).
Training data files in such networks consist of m
n-elements patterns X, where n is the number of the
network inputs. The competitive learning in KNN is
an iterative process. During each iteration, called an
epoch, all m vectors are presented to the network in
a random order. The full learning process requires
even hundreds epochs, which means even several
hundreds thousands presentations of a single pattern.
Once a single pattern is presented to the network,
several calculation steps may be performed by the
network. In the first step a distance between a given
pattern X and the weights vector W in every neuron
in the map is calculated using, for example, the
Euclidean or the Manhattan metric. In the next step
the winning neuron is identified, and this neuron in
the following step is allowed to adapt its weights.
In the Winner Takes All (WTA) learning method
only the winning neuron, whose vector W is the
closest to the pattern X, is allowed to adapt the
weights, while in the Winner Takes Most (WTM)
approach also neurons that belong to the winners
neighborhood change the weights.
The WTA algorithm offers poor convergence
properties, especially in case of large number of
neurons. In this approach some neurons remain dead
i.e. they absorb the computational resources, but
never win and never become representatives of any
data class. The WTM algorithm, on the other hand,
is more complex, since it additionally involves the
neighborhood mechanism, which increases the
computational complexity, but this mechanism
usually activates all neurons in the network (Mokri
2004), thus minimizing the quantization error. This
error is defined as follows (w
w
are weights of the
winning neuron):
m
w x
Qerr
m
j
n
l
wl
l
j
= =

=
1 1
2
) (

(1)
The main problem in the WTM algorithm is very
large number of operations, especially in case of
large number of patterns and epochs. In hardware
implementations effective methods to minimize the
number of operations are therefore required.
3 THE PROPOSED TECHNIQUE
In a typical WTM learning algorithm the neighbor-
hood radius R at the beginning of the training
process is set up to the maximum possible value so it
covers an entire map. After each epoch the radius R
decreases linearly by a small value to zero. In
practice, as number of epochs usually is much larger
than the maximum value (R
max
) of the neighborhood
radius R, therefore the radius decreases always by
1 after the number of epochs equals to:
( )
MAX MAX
round R l l =

(2)
In equation (2) l
max
is the total number of epochs in
the learning phase. Value of the l parameter usually
is in the range between 20 and 200, depending on the
networks dimensions. In case of an example map
with 10x10 neurons and the rectangular neighbor-
hood R
max
equals to 19 (Dugosz and Kolasa, 2008).
To verify this linear approach authors designed
a software model of the WTM KNN. Simulations
have been performed for different network dimen-
sions and different training data files. Observation of
the quantization error in the time domain shows that
the linear approach is not optimal. The example
illustrative waveforms of the Q_error are in this case
shown in Fig. 1 for an example training data file
with 1000 patterns X, for selected network dimen-
sions 20x20, 10x10 and 4x4 neurons. The quantiza-
tion error is calculated after each epoch i.e. always
after presentation of 1000 training patterns X. This
does not increase significantly the computational
complexity of an entire learning process.
The first important observation is that when the
neighborhood radius R is larger than some critical
value, the quantization error does not decrease,
which means that in this period the network does not
make the progress in training. This critical value is
usually small, between 4 and 7 for different network
dimensions as illustrated in Fig. 1. The important
conclusion at this point is that the learning process
may start with the value of the radius R, which is
much smaller than the maximal value R
max
. This
significantly shorts the entire training process.
The second important observation is that the
error does not decrease monotonously with time, but
there are some distinct activity phases, just after the
radius R is switched to the smaller value, in which it
decreases abruptly and then some stagnation phases,
in which it does not decrease. The length of a single
activity phase usually is between 2-4 epochs inde-
pendently on the network dimensions.
The optimization technique proposed by authors
eliminates these stagnation phases by incorporation
of the multistage filtering of the quantization error in
NEW FAST TRAINING ALGORITHM SUITABLE FOR HARDWARE KOHONEN NEURAL NETWORKS
DESIGNED FOR ANALYSIS OF BIOMEDICAL SIGNALS
365

time domain and a special decision mechanism that
automatically switches over the radius R just after a
given activity phase is finished. This starts a new
activity phase, but for smaller value of the radius R.

Quantization error [10E-3] (20x20 neurons)
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0 200 400 600 800 1000 1200 Epoch No.
No progress in training
R =8
R =0

Quantization error [10E-3] (10x10 neurons)
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0 200 400 600 800 1000 1200 Epoch No.
No progress
R =6
R =0

Quantization error [10E-3](4x4 neurons)
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0 200 400 600 800 1000 1200 Epoch No.
R =3
R =0
Activity phase
Stagnation phase

Figure 1: Typical linear training process: (top) an e.g.
training data file with 1000 vectors X and final placement
of neurons in the map (below) quantization error as a
function of the Epoch No., for selected maps dimensions.
In our proposed method three filters have been
applied. The entire cycle starts with a lowpass finite
impulse response (FIR) filtering that smoothes out
the initial error waveform. This process is illustrated
in Fig. 2 (a). In this case a simple Butterworth flat
filter has been used with the following coefficients:
h
LPi
={0.125, 0.375, 0.375, 0.125}.
Original Q_ERROR [10E-3] and after LOWPASS FILTERING
0.01
0.03
0.05
0.07
0.09
0.11
0.13
0.15
0 200 400 600 800 1000 1200 Epoch No.
error 'noise'
(a)

HIGHPASS FILTERING
-0.01
0
0.01
0.02
0.03
0 200 400 600 800 1000 1200 Epoch No.
Activity phase
Stagnation phase
"noise" spike
(b)

MEDIAN NONLINEAR FILTERING
-0.006
-0.004
-0.002
0
0.002
0.004
0.006
0.008
0.01
0.012
0 200 400 600 800 1000 1200 Epoch No.
threshold level
Activity phase
"noise" spike (c)

Figure 2: Proposed 3-stage error filtering: (a) the original
waveform and the lowpass, (b) the highpass, (c) the
nonlinear median filtering.
The next step is the highpass filtering operation
that detects edges in the smoothed error waveform.
This filter may be very simple, with the length not
exceeding 4. In presented example a filter with the
coefficients h
HPi
={1, 1, -1, -1} has been employed.
The resultant waveform is illustrated in Fig. 2 (b).
The spikes in this waveform indicate the activity
phases. The problem here is that the noise present
in the initial error waveform is a source of additional
undesired spikes, which often are as high as the
activity spikes, although usually are narrower than
the activity spikes. To overcome this problem a
nonlinear median filter has been additionally
applied. The length of this filter has been selected in
such a way to even the height of the activity spikes
and to eliminate the noise spikes. For example, the
length of 5 allows to eliminate the noise spikes
with the width equal to 2, as shown in Fig. 2 (c).
Both the highpass and the median waveforms are
then used by a decision mechanism that automati-
cally switches over the radius R to the smaller value.
The decision procedure starts when the value of the
median waveform becomes larger than some
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
366

threshold value, which is high enough to exclude the
noise spikes. The decision about switching over is
made when the signal in the highpass waveform
becomes falling, which means that the training
process is just entering the stagnation phase.
It is worth noticing that the proposed algorithm
must cooperate with the classic linear method. This
is necessary in a situation, in which an activity spike
in the median waveform would be to small to
activate the decision procedure. In this case the
linear method will switch over the radius R after l
iterations that will stop a given stagnation phase.
The illustrative simulation results in case of the
optimized training process are shown in Fig. 3 for an
example network with 10x10 neurons. In this case
the entire training process has been shorten 16 times
from initial 1000 epochs to 60 epochs.
Original Q_ERROR [10E-3] and after LOWPASS FILTERING
0.01
0.03
0.05
0.07
0.09
0.11
0.13
0.15
0 20 40 60 80 Epoch No.
(a)

HIGHPASS FILTERING
-0.02
-0.01
0
0.01
0.02
0.03
0 20 40 60 80 Epoch No.
Activity phase
(b)

MEDIAN NONLINEAR FILTERING
-0.004
0
0.004
0.008
0.012
0.016
0 20 40 60 80 Epoch No.
threshold level
(c)

Figure 3: Training process after optimization, for an e.g.
map with 10x10 neurons: (a) original signal and after the
lowpass, (b) the highpass and (c) the median filtering.
4 CONCLUSIONS
A new simple learning algorithm for WTM KNNs
designed for low-power devices has been described
in the paper. The proposed technique bases on the
multistage filtering of the quantization error, which
is calculated after each epoch of the training process.
The proposed algorithm detects the periods in the
training process, in which the error decreases i.e. in
which the network makes a progress in training and
then automatically switches over the neighborhood
radius R just after the training process enters the
stagnation phase, thus shortening this phase.
The simulations show that this technique is able
to shorten the training process by more than 90%.
The proposed algorithm will be used in hardware
KNNs, designed for analysis of biomedical the ECG
and the EMG signals in Wireless Body Sensor
Network (WBSN) applications.
ACKNOWLEDGEMENTS
The work is supported by EU Marie Curie Outgoing
International Fellowship No. 021926
REFERENCES
Osowski, S., Tran Hoai Linh, ECG beat recognition using
fuzzy hybrid neural network, IEEE Transactions on
Biomedical Engineering, Vol.48, Issue 11, Nov. 2001
Wismller A., Lange O., Dersh R. D., Leinsinger
G. L., Hahn K., Ptz B., and Auer D. Cluster Analy-
sis of Biomedical Image Time-Series, International
Journal of Computer Vision, Vol. 46, No. 2, Feb. 2002
J ehan Zeb Shah, Naomie bt Salim, A Fuzzy Kohonen
SOM Implementation and Clustering of Bio-active
Compound Structures for Drug Discovery, IEEE
Symposium on Computational Intelligence and
Bioinformatics and Computational Biology (CIBCB),
28-29 September 2006
M. McInerney, A. Dhawan, Training the self-organizing
feature map using hybrids of geneticand Kohonen
methods Neural Networks, IEEE World Congress on
Computational Intelligence, Vol.2, 27 J un-2 J ul 1994
Rajesh Ghongade, A. A. Ghatol, Performance Analysis
of Feature Extraction Schemes for Artificial Neural
Network Based ECG Classification, International
Conference on Computational Intelligence and
Multimedia Applications (ICCIMA), 2007, Vol. 2
T. Kohonen, Self-Organizing Maps, third edition, Springer
Berlin, 2001
Mokri, R. Forg, Decreasing the Feature Space
Dimension by Kohonen Self-Orgaizing Maps, 2
nd

Slovakian Hungarian Joint Symposium on Applied
Machine Intelligence, Herany, Slovakia 2004
Dugosz R., Talaka T., Dalecki J ., Wojtyna R.,
Experimental Kohonen Neural Network Implemented
in CMOS 0.18m Technology, 15th International
Conference Mixed Design of Integrated Circuits and
Systems (MIXDES), Pozna, Poland, J une 2008
Dugosz R., Kolasa M., CMOS, Programmable,
Asynchronous Neighborhood Mechanism For WTM
Kohonen Neural Network, 15th International
Conference Mixed Design of Integrated Circuits and
Systems (MIXDES), Pozna, Poland, J une 2008
NEW FAST TRAINING ALGORITHM SUITABLE FOR HARDWARE KOHONEN NEURAL NETWORKS
DESIGNED FOR ANALYSIS OF BIOMEDICAL SIGNALS
367
WRIST-WORN FALL DETECTION DEVICE
Development and Preliminary Evaluation
Mattia Bertschi and Leopoldo Rossini
Swiss Center for Electronics and Microtechnology, Jaquet-Droz 1, Neuchtel, Switzerland
[email protected], [email protected]
Keywords: Elderly, Monitoring, Portable device, Wearable device, Automatic fall detection, Accelerometers.
Abstract: Falls are the most important cause of accidents for elderly people and often result in serious physical and
psychological consequences. The rapid growth of the elderly population increases the magnitude of the
problem as well as the generated costs. In order to take care of old people living by themselves or in care
centres and to reduce the consequences of a fall, various technological solutions have been studied, however
none led to a commercial product fulfilling user requirements. In this work we present an automatic fall
detector in the form of a wrist watch which could lead to better life conditions for the elderly. Our device
implements functionalities such as wireless communication, automatic fall detection, manual alarm
triggering, data storage, and simple user interface. Even though the wrist is probably the most difficult
measurement location on the body to discern a fall event, the proposed detection algorithm shows
encouraging results (90% sensitivity, 97% specificity) with the signals of our database.
1 INTRODUCTION
Falls are the most widespread domestic accidents
among the elderly. Their consequences often give
rise to impairments to the health and lifestyle of the
victims (Prolle et al., 2004). In many cases,
physical after-effects and other injuries are direct
consequences of these accidents and result in
significant medical costs.
Furthermore, it frequently happens that elderly
people who have previously experienced a fall fear a
new fall and sink gradually into inactivity and social
isolation. The reduction in their mobility leads
progressively to an increase in the risk of a fall
(Doughty et al., 2000). Hence, given the growing
part of the elderly people in our modern societies,
the socio-economic impact that this self-imposed
isolation may have should not be neglected.
The most widespread solution for limiting the
apprehension of a fall is provided by social alarms
consisting of portable devices. These are generally
equipped with an alarm triggering button and
endowed with telecommunication means suitable for
alerting the care centre. Nevertheless, because of the
fall, the person may not be able to actuate the button
and to trigger the alarm (unconsciousness, state of
shock, broken arm, etc.).
To alleviate this drawback, autonomous fall
detectors have been developed and are capable of
triggering an alarm automatically without any
intervention of the victim and transferring this
information to a remote site (Doughty et al., 2000).
These autonomous detectors operate essentially on
two principles. The detector is either sensitive to the
person's appearance or impact on its environment
and is based on video (CCD or IR camera) or
vibratory type sensors (acoustic or piezoelectric
layers on the ground) placed in the usual
surroundings of the subject. The benefit of these
devices is that they do not have to be worn. Instead,
they are fixed and integrated in a given spot and
cannot be moved easily when the person changes
location. Moreover, in the case of a video sensor the
person will have the impression of being supervised
and feel inconvenienced. The major drawback of
acoustic based sensors is that they are surface
dependent, while those based on vibration are fragile
and expensive. The detector can also be worn by the
person and thus detect a fall directly as soon as it
occurs, triggering an immediate alarm. In this case,
the information provided by inclinometers,
gyroscopes or accelerometers is exploited. These
devices are generally compact, inexpensive, fairly
non-obtrusive, easy to use, and can be worn at
various body locations.
368

Devices worn close to the centre of gravity
(Depeursinge et al., 2001) are the most reliable ones,
but also the least convenient to wear on a daily basis,
in particular while performing common daily
activities. A device having the form of a wristwatch
would be well tolerated in all situations, despite the
challenge to detect a fall due to changes of position
and accelerations that the wrist experiences during
everyday actions (Degen et al., 2003). Furthermore,
the inclination measurement of the forearm cannot
give reliable information about the person's position.
Generally, fall detectors placed on the wrist give rise
to a large number of false alerts, and this would
generate significant and unnecessary costs.
Our goal is to develop a small, comfortable, and
user friendly device, as well as an automatic fall
detection algorithm that will help elderly people to
handle this problem.
2 METHOD AND MATERIALS
The fall detector that we have developed is a device
capable of automatically detecting various body falls
and sending an alarm to a remote terminal. The user
can manually generate the alarm signal in case of
necessity or, inversely, he can cancel an
automatically generated alarm in case of a false fall
detection.
2.1 Fall Detection System
The detection system is integrated in the case of a
wrist watch. A picture of our prototype is depicted in
Figure 1.

Figure 1: Wrist fall detection system.
The core of the wrist-located device consists of a
microprocessor (MSP430) and two MEMS sensors
arranged perpendicularly to allow the measurement
of acceleration along three axes with a range equal
to 18g (ADXL321). The user interface is simple
and comprises a small LCD screen (Nokia 3310), a
vibrator which advises the user that a fall has been
detected and an alarm will soon be sent, and a push
button on the front panel to manually trigger the
alarm signal. Data can be transmitted from fixed
and/or mobile devices over short distances utilizing
a short-range communication technology (Bluetooth
protocol).
For experimental purposes, the acceleration signals
can also be stored in a flash memory card. The serial
RS232 port as well as three additional buttons are
also available for debugging and test.
The device is powered by a 3.7 volts rechargeable
Lithium-Cobalt-Polymer battery. The battery life of
the device varies from about 15 days to one month,
depending on the sampling frequency and the details
of the implemented data handling and storage
functionalities.
2.2 Fall Detection Algorithm
The three-axes acceleration signals are recorded and
stored in the flash memory, with a sampling
frequency of 910 Hz and 12-bit resolution. Although
the algorithm has not been implemented yet in real-
time, it has been tested offline using Matlab.
Initially, the device is slowly rotated a few times
in order to project the gravity vector on the three
axes in various configurations. The resulting
acceleration signals roughly define the surface of a
sphere in a three dimensional space and are used to
calibrate the sensors. We define the time-dependent
acceleration vector ( )
T
z y x
a a a
~
,
~
,
~
= A whose entries
are defined as
i i i
c a a =
~
, for z y x i , , = , where
i
a
represents the acceleration in the i-direction while
i
c
is the i-coordinate of the centre of the sphere
corresponding to a zero acceleration. Therefore, the
equation of the sphere is defined using Cartesian
coordinates as:
0
2
2
2
= r A (1)
where the radius r corresponds to the gravity
acceleration and defines the sensor gain while the
centre (c
x
,c
y
,c
z
) defines the accelerometer offset.
Notice that one can use the Gauss-Newton method
to estimate (c
x
,c
y
,c
z
,r) by minimizing the left-hand
part of equation (1) and use those parameters to
calibrate the acceleration signals (see Figure 2).
The three-axes acceleration experienced at the
wrist of the person wearing the device can not be
directly linked to the specific posture of the body.
However, it has been observed that the distribution
WRIST-WORN FALL DETECTION DEVICE - Development and Preliminary Evaluation
369

of the acceleration norm
2
A experienced at the
wrist over a time window T provides a particularly
reliable signature, allowing the identification of a
given event happening to the person wearing the
device.
-6
-3
0
3
6

x

(
g
)
-6
-3
0
3
6

y

(
g
)
0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
-6
-3
0
3
6

z

(
g
)
Time (s)

Figure 2: Typical recording of a fall event. Each signal
corresponds to the acceleration in the x, y, or z direction.
We define S
t
as the signature at instant t that
represents the distribution of the acceleration norm
2
A over a time window T. The signature S
t
can
be extracted via a simple amplitude threshold and
compared to a reference signature S
ref
. To obtain S
ref
,
a subset of all the fall signatures recorded in the
database is used to obtain statistics in the form of
means and variances. Specifically, a particular value
of each signature in the subset is identified so as to
carry out an alignment, for example on the
maximum. This particular value is used to align the
signatures, so as to obtain a group of aligned fall
signatures. On the basis of the various aligned
signatures, the reference signature S
ref
is obtained by
averaging the values of the aligned signatures.
In order to construct a similarity measure d
between S
t
and S
ref
, the squared value of the
acceleration due to gravity is subtracted to remove
its influence. For this application, d is calculated
according to the Mahalanobis distance definition.
The variances of the aligned signatures are placed on
the diagonal of a matrix that is used together with
S
ref
to estimate if the instantaneous signature S
t

arises from a fall:
) (
1
) (
) , (
ref
S
t
S
T
ref
S
t
S
e
ref
S
t
S d

=

(2)
The influence of the least reliable samples is reduced
due to the use of the variances. One can see that the
value of d is in the interval [0,1] and therefore can
be considered to be a fall probability. A threshold
was used to discriminate between the two classes:
fall and no fall.
2.3 Experimental Setup
A preliminary validation has been performed with
three adult healthy adult volunteers to assess the
reliability of our system for fall detection tasks. The
study consists in recording the three-axial
accelerometer signals (wrist-located) during
controlled fall events and daily life activities. Each
case was repeated three times and stored in a
separate file. In our database we have documented
180 situations by recording acceleration signals and
video sequences. The latter are particularly useful to
understand the dynamics of the wrist movement.
To evaluate the sensitivity of the fall detection a
first subset consisting of six kinds of falls is created.
The kind of falls selected for this subset are: front,
back, left, right, falling backwards while sitting
down, and falling forward while standing up. Each
fall was intentional and a mattress was used to
protect volunteers from injuries.
A second subset was used to estimate the
specificity of the fall detector. It includes 14
common situations representing challenges for the
detection algorithm (slow/fast walking, walking
upstairs/downstairs, hit on a table with the fist,
sitting down, lying down, applauding).
3 RESULTS
To generate the reference signature of a fall event
we have taken 8 recordings representing fall events
randomly selected from the 54 falls of the database
and we have computed the mean and the standard
deviation for each event. Figure 3 shows a typical
reference signature that we obtain with the signals of
our database. The mean and standard deviation
values of the reference signature are then used to
compute a normalized similarity measurement for
each recording of the database and a binary classifier
is used to separate falls from other events.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
370

0 0.5 1 1.5
0
2
4
6
8
10
12
14
Time (s)
A
c
c
e
l
e
r
a
t
i
o
n

n
o
r
m

(
g
)
0 0.5 1 1.5
0
2
4
6
8
10
12
14
Time (s)
A
c
c
e
l
e
r
a
t
i
o
n

n
o
r
m

(
g
)

Figure 3: Reference signature in terms of mean and
standard deviation.
A binary classifier system is generally assessed
in terms of sensitivity and specificity. A useful
graphical tool to represent the sensitivity versus (1-
specificity) for a binary classifier system as its
discrimination threshold is varied is the receiving
operating characteristic (ROC). Figure 4 shows the
ROC curve obtained with the signals of the database.
0 2 4 6 8 10
60
65
70
75
80
85
90
95
100
1 - Specificity (%)
S
e
n
s
i
t
i
v
i
t
y

(
%
)
0 2 4 6 8 10
60
65
70
75
80
85
90
95
100
1 - Specificity (%)
S
e
n
s
i
t
i
v
i
t
y

(
%
)

Figure 4: The receiver operating characteristic shows the
relationship between the sensitivity and (1-specificity).
In order to become independent from the
recordings used to generate the reference signal, we
repeated the same procedure ten times: random
selection of 8 fall events, estimation of the mean and
the standard deviation, classification of the 172
recordings, and estimation of the ROC curve. The
circles in the plot of Figure 4 are thus a mean value
of the ten simulated results.
One can see that there is a tradeoff between
sensitivity and specificity: the larger the sensitivity,
the smaller the specificity. We can decide to favour
one characteristic over the other, but a good balance
is normally to privilege the upper-left corner of the
ROC curve. In the present case, this particular point
allows us to detect about 90% of the falls while
keeping the false alarm rate below 3%.
4 CONCLUSIONS
We have developed a small, light, and comfortable
fall detector device which is worn at the wrist like an
ordinary watch, removing the social stigma of
wearing a medical device. The real advantage of
fixing the fall detector at the wrist is the possibility
of wearing the device at night, when falls can also
occur. The device is thus easy to wear continuously
without any specific constraints. The major
drawback is the signal processing challenge of
estimating a fall from wrist acceleration data, due to
the strong accelerations experienced by the forearm
during daily-life activities.
The proposed algorithm to detect falls is based
on accelerometric signals and has the advantage to
be simple and robust showing promising results with
our present database. Results demonstrate high
sensitivity (90%) as well as high specificity (97%)
for the detection of intentional falls.
REFERENCES
Prolle, G., Snchez, D., Abarrategui, M. I., Eizmendi, G.,
Buiza, C., Etxeberria, I., Yanguas, J . J ., 2004. Fall
Detection: Project of an Improved Solution. In
TELECARE 2004, 1st International Workshop on
Tele-Care and Collaborative Virtual Communities in
Elderly Care.
Doughty, K., Lewis, R., McIntosh, A., 2000. The Design
of a Practical and Reliable Fall Detector for
Community and Institutional Telecare. In Journal of
Telemedicine and Telecare.
Doughty, K., 2000. Fall Prevention and Management
Strategies Based on Intelligent Detection, Monitoring
and Assessment. In New Technologies in Medicine for
the Elderly.
Depeursinge, Y., Krauss, J ., El-Khoury, M., 2001. Device
for monitoring the activity of a person and/or detecting
a fall, inparticulaer with view to providing help in the
event on an incident hazardous to life or limb. In
Patent US 6,201,476 B1.
Degen, T., J aeckel, H., Rufer, M., Wyss, S., 2003.
SPEEDY: A Fall Detector in a Wrist Watch. In
IEEE2003, 7
th
International Symposium on Wearable
Computers.
WRIST-WORN FALL DETECTION DEVICE - Development and Preliminary Evaluation
371
A NEW LINEAR ARRAY IMAGING SYSTEM OF ELECTRICAL
AND ULTRASONIC PROPERTIES IN A LIVING BODY
Akira Kimoto, Yuuta Taninaka and Katsunori Shida
Faculty of Science and Engineering, Saga University, Honjyo 1,Saga, Japan
[email protected], [email protected]
Keywords: Piezoelectric ceramic transducer, Ultrasonic property, Electrical property.
Abstract: In this paper, a new linear array imaging system of ultrasonic and electrical properties in the living body is
proposed. The proposed imaging system measures not only the ultrasonic property of the living body using
the linear arrayed piezoelectric ceramic transducers, but also the electrical property using the surface
electrodes of each piezoelectric ceramic transducer. From these data, ultrasonic and electrical properties in
the same object space are simultaneously reconstructed. In the experiment, propagation time and electrical
voltage of the living body model are measured by the proposed imaging system based on linear arrayed
eight piezoelectric ceramic transducers. Ultrasonic and electrical properties are reconstructed from the
measurement values. It was found that the ultrasonic and electrical properties in the same space could be
reconstructed by the proposed imaging system. Therefore, it is suggested that the proposed imaging system
has potential for application although there are some problems that must be solved.
1 INTRODUCTION
Imaging techniques based on the ultrasonic property
(Opielinski and Gudra, 2000, Simaeys et al., 2000)
or electrical property (Holder et al., 1993, Barber
and Brown, 1984) of a living body are especially
important in medical field, and have been actively
researched. A non-invasive ultrasonic imaging
system using ultrasonic properties of the living body
has been studied for determining the blood flow
velocity distribution and internal organ imaging
(Nitta et al., 1996, Lopez et al., 1992). The electrical
impedance computed tomography using electrical
properties of the living body has also been
developed for imaging of the heart and lungs (Fuks
et al., 1991), temperature distribution measurements
(Conway et al., 1992) and so on.
The aim of our research is to establish a non-
invasive simultaneous imaging system of two
parameters in the living body such as temperature
and a body composition. To achieve it, we propose a
new linear array imaging system of ultrasonic and
electrical properties in the living body. In the
proposed system, the ultrasonic propagation time is
measured by the linear arrayed piezoelectric ceramic
transducers. In addition, the electrical potential is
measured by the surface electrodes of poizoelectric
ceramic transducers (Kimoto and Shida, 2001, 2002).
Therefore, it is possible to measure the ultrasonic
and electrical properties in the same object space
using the proposed imaging system. From these data,
ultrasonic and electrical properties in the living body
are reconstructed. Moreover, two parameters such as
temperature and composition are estimated from
their reconstructed distributions.
In this paper, the imaging system with the linear
arrayed eight piezoelectric ceramic transducers is
established. In the experiment, ultrasonic
propagation time and electrical voltage in 0.1 %
saline solution with acrylic as the living body model
are measured by the proposed imaging system, and
then, the reconstructions of ultrasonic and electrical
properties are demonstrated from their measuremet
values.
2 PRINCIPLE
The ultrasonic and electrical properties in the living
body are generally measured by different sensors. In
the proposed method, they are measured by the same
sensor. Figure 1 shows the measurement method of
the ultrasonic and electrical properties in the living
body. In this method, a linear arrayed piezoelectric
ceramics are used. In figure 1(a), an electrical signal
with a resonance frequency of the piezoelectric
372

ceramic transducer, S
t
, is applied to a piezoelectric
transducer, and the reflected wave, S
r
is measured by
the other piezoelectric ceramic transducer. From
these signals, the ultrasonic property of the object is
obtained. In figure 1(b), the constant current is
injected between the surface electrodes of a pair of
piezoelectric ceramic transducers and the induced
voltage between each surface electrodes are
measured. The electrical property is obtained from
the induced voltage and injected current. Therefore,
the ultrasonic and electrical properties in the same
object space are obtained using a pair of
piezoelectric ceramic transducers. Moreover, the
reconstructed distributions of the ultrasonic and
electrical properties are respectively obtained from
the measured values by using linear arrayed many
piezoelectric ceramic transducers.
Ultrasonic property
Time of flight
St Sr

(a)
Electrical property
V
I

(b)
Figure 1: Measurement method. (a) Measurement of
ultrasonic property. (b)Measurement of electrical property.
3 MEASUREMENT SYSTEM
Figure. 2 shows the schematic diagram of the
measurement equipment. The rectangular equipment
(4040100 mm
3
) was constructed by acrylic plate
and the eight piezoelectric ceramic transducers
(1051 mm
3
) with 2 MHz resonance frequency are
linearly arrayed with the gap of 1 mm at an acrylic
plate inside the equipment. It is filled with the 0.1%
saline solution as the living body model.
Figure 3 shows the outline of the measurement
system. In the ultrasonic measurement, the burst
wave of the five sinusoidal waves with the
amplitude of 10 V and the 2 MHz frequency as the
transmitted wave is given to one of the piezoelectric
ceramic transducers by the function generator. The
reflected wave is measured by each piezoelectric
ceramic transducer. Transmitted and reflected waves
are passed through the AD converter and stored at
the PC. From their waves, propagation time is
obtained.
100
120
40 60
10 10
10
10
25
5
10
Acrylic
Side view
Top view Side view
[mm]
Piezoelectric ceramic
transducer
1 8 2 3 4 5 6 7
0.1 % saline solution

Figure 2: Schematic diagram of experimental equipment
with linear arrayed eight piezoelectric ceramic
transducers.
1 8
Imaging area
I/O board
AD converter
GPIB board
VI C VI C VI C VI C VI C VI C VI C VI C
Switch1 Switch2 Switch3 Switch4 Switch5 Switch6 Switch7 Switch8
PC
Function
generator
Switch box
VI C:V-I converter

Figure 3: Schematic diagram of measurement system.
In the electrical measurement, the sinusoidal
current of 1mA, which sinusoidal voltage with the
amplitude of 1 V and 10 kHz frequency is converted
by V-I converter, is injected between the surface
electrodes of a pair of piezoelectric ceramic
transducers. The voltages induced on rest of surface
A NEW LINEAR ARRAY IMAGING SYSTEM OF ELECTRICAL AND ULTRASONIC PROPERTIES IN A LIVING
BODY
373

electrodes of piezoelectric ceramic transducers are
measured. Each voltage is also digitalized by the AD
converter and stored at the PC.
The ultrasonic and electrical measurements and
the selection of piezoelectric ceramic transducers are
changed using the switching system controlled by
the I/O signal.
4 EXPERIMENT
Figure 4 shows the experimental model. 0.1 % saline
solution model, which acrylic (1010 mm
2
) is
inserted at the position of 15 mm apart from the
piezoelectric ceramic transducers, was prepared. The
ultrasonic and electrical distributions are
respectively reconstructed from the measurement
values of ultrasonic propagation time and electrical
voltage. In this time, propagation time and voltage
for reconstructing the ultrasonic and electrical
distributions were measured as follows.
Propagation time as ultrasonic property was
obtained from transmitted and reflected waves
measured at each piezoelectric ceramic transducer
from No.1 to No.8. Therefore, ultrasonic distribution
was obtained from eight data. In electrical property,
the voltage induced at surface electrode of
piezoelectric ceramic transducer between a pair of
surface electrodes used as current electrodes was
measured. Table 1 shows the combinations of
current and voltage electrodes. Therefore, 56 voltage
values were used for the imaging of electrical
property. In this time, impedance distribution as
electrical property was reconstructed by the
measured data and the numerical calculation using
the finite element method (FEM).
Table 1: Combination of electrical measurement.
Current electrode
numbers
Voltage electrode
number
(1,3), (2,4), (3,5),
(4,6),(5,7),(6,8)
(2), (3), (4),
(5), (6), (7)
(1,4), (2,5), (3,6),
(4,7), (5,8)
(2,3), (3,4), (4,5),
(5,6), (6,7)
(1,5), (2,6),
(3,7), (4,8)
(2,3,4), (3,4,5),
(4,5,6), (5,6,7)
(1,6), (2,7), (3,8) (2,3,4,5), (3,4,5,6),
(4,5,6,7)
(1,7), (2,8) (2,3,4,5,6), (3,4,5,6,7)
(1, 8) (2,3,4,5,6,7)
Figure 5 shows the 26 unknown elements of
impedance estimated by FEM. Impedance values in
other regions are those of 0.1 % saline solution.
Impedance of each element was estimated by the
modified Newton-raphson method as the iterative
method (Kimoto and Shida, 2000).
Acrylic
0.1 % saline solution
Piezoelectric ceramic
transducer
1 8 2 3 4 5 6 7
15 mm

Figure 4: Reconstructed model.

Figure 5: Estimated elements of electrical property.

(a)
1.02
0.80
[%]

(b)
Figure 6: Reconstructed results. (a) The ultrasonic
propagation time. (b)The change ratio of electrical
impedance.
5 RESULTS
Figure 6 shows the reconstructed results of
ultrasonic and electric properties. Figure 6(a) shows
the positon of reflected wave obtained by the
propagation time and the sound speed. In this time,
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
374

sound speed was calculated by propagation time
measured in 0.1 % saline solution and 40 mm
distance of equipment. From figure 6(a), it is found
that the boundary of the target in 0.1 % saline
solution is obtained although it was difficult to
detect the construction of the target. Figure 6(b)
shows the impeadace change ratio between 0.1 %
saline solution with and without acrylic at three
iteration. From figure 6(b), it is found that impedace
in the part of acrylic decreased although the
estimated resolution is insufficient. In this system, it
is possible to reconstruct the ultrasonic and electrical
properties by measuremets of propagation time and
voltage. In addition, it is suggested that the
resolution of the reconstructed image would be
improved by combining the ultrasonic and electrical
reconstructed images because their reconstructed
distributions are different.
6 DISCUSSION
The accuracies of reconstructed distributions of
ultrasonic and electrical properties were insufficient.
They are mainly caused by measurement error of the
system and insufficient measured data number. The
measurement system, especially, the switching
system must be improved. Data number will be also
increased by using several measurement
combinations.
In this time, although ultrasonic and electrical
measurements are changed by the switching system,
simultaneous measurements of the ultrasonic and
electrical properties are possible in which one
electrical signal is created from the electrical signal
with the resonance frequency for ultrasonic
measurement and that of the electrical impedance
measurement and is applied to the electrode as an
alternating current.
7 CONCLUSIONS
A new linear array imaging system of ultrasonic and
electrical properties in the living body was proposed.
In the proposed imaging system, the ultrasonic
propagation time is measured by the linear arrayed
piezoelectric ceramic transducers and the electrical
potential is also measured by the surface electrodes
of poizoelectric ceramic transfuces. Therefore, it is
possible to measure the ultrasonic and electrical
properties in the same object space using the
proposed system. From these data, ultrasonic and
electrical properties in the living body are
reconstructed. In the experiment, the ultrasonic and
electrical properties in 0.1 % saline solution with
acrylic as the living body model were reconstructed
from propagation time and voltage measured by the
proposed imaging system. As a result, it was
suggested that the proposed imaging system has
potential for application although there are some
problems that must be solved.
REFERENCES
Opielinski K. J . and Gudra T., 2000, Ultrasound
transmission tomography image distortions caused by
the refraction effect, Ultrasonics, vol.38, pp.424-429
Simaeys B., Philips W., Lemahieu I. and Govaert P., 2000,
Quantitative analysis of the neonatal brain by
ultrasound, Computerized Medical Imaging and
Graphics, vol.24, pp.11-18
Holder D. S., Ed., 1993, Clinical and physiological
applications of electrical impedance tomography, UCL
Press London
Barber D.C. and Brown B.H., 1984, Applied potential
tomography, J. Phys. E: Sci. Instrum., vol.17, No.9,
pp.723-733
Nitta N., Hagihara K. and Shiina T., 1996, Experimental
Investigation of 3-D Blood Flow Velocity
Measurement, Jpn.J.Appl.Phys., vol.35, Pt.1, 5B,
pp.3126-3130
Lopez H., Loew M.H. and Goodenough D.J ., 1992,
Objective Analysis of Ultrasound Images by Use of a
Computational Observer, IEEE Trans. on Med. Img.,
vol.11, No.4, pp.496-506
Fuks L. F., Cheney M., Isaacson D., Gisser D. G. and
Newell J . C., 1991, Detection and Imaging of Electric
Conductivity and Permittivity at Low Frequency,
IEEE Trans. Biomed Eng., vol.38, No.11, pp.1106-
1110
Conway J ., Hawley M., Hangnall Y., Amasha H. and
VanRhoon G. C., 1992, Experimental assessment of
electrical impedance imaging for hyperthermia
monitoring, Clin. Phys. Physiol. Meas., vol.13,
Suppl.A, A185-A189
Kimoto A. and Shida K., 2001, Proposal of a New
Multifunctional Measurement Method Using
Piezoelectric Vibrator, Jpn. J. Appl. Phys. Vol.40,
Part1, 6A, pp.4258-4259
Kimoto A. and Shida K., 2002, A proposal of
measurements of conductance and propagation time
for determination of temperature and ingredient in the
living body model, Trans. IEE of Japan, vol.122-E,
No.6, pp.332-337 (in J apanise)
Kimoto A. and Shida K., 2000, Imaging of temperature-
change distribution in the brain phantom by means of
capacitance measurement IEEE Transactions on
Instrumentation and Measurement, Vol.49, Issue 3,
pp.591-595
A NEW LINEAR ARRAY IMAGING SYSTEM OF ELECTRICAL AND ULTRASONIC PROPERTIES IN A LIVING
BODY
375
MODELLING OF SAW BIOSENSORS
Marija Hribek, Slavica Risti, Zdravko ivkovi

Goa Institute, Milana Rakia 35, Belgrade, Serbia
[email protected]
Dejan Toi
Faculty of Electrical Engineering, Bulevar kralja Aleksandra 73, Belgrade, Serbia
[email protected]
Keywords: SAW devices, SAW filters, biosensors.
Abstract: New approach in surface acoustic wave (SAW) biosensors modelling is presented. Biosensor is modelled
as a three port network. The model is general and can be used also in the case of transponder type of sensor.
The closed form solutions for transfer function and input admittance at the electrical port of SAW devices
with uniform transducers based on complex equivalent circuit are presented. Transfer function and input
admittance in two different cases are calculated and compared with the experimental results showing very
close agreement.
1 INTRODUCTION
Surface acoustic waves (SAW) were discovered in
1885 by Lord Rayleigh, and are often named after
him: Rayleigh waves. A surface acoustic wave is a
type of mechanical wave motion which travels along
the surface of a solid material. Rayleigh showed that
SAWs could explain one component of the seismic
signal due to an earthquake, a phenomenon not
previously understood. The velocity of acoustic
waves is typically 3000 m/s, whish is much lower
than the velocity of the electromagnetic waves. A
basic SAW device consists of two interdigital
transducers (IDTs) on a piezoelectric substrate such
as quartz. The IDTs consist of interleaved metal
electrodes which are used to launch and receive the
waves, so that an electrical signal is converted to an
acoustic wave and then back to an electrical signal
(Morgan,1985)
The basic application of the SAW device is as
delay line. Central frequency and the bandwidth are
determined by the IDT`s geometry and the substrate
type. The IDT geometry is capable of almost endless
variation, leading to a wide variety of devices.
Starting around 1970, SAW devices were developed
for pulse compression radar, oscillators, and
bandpass filters for domestic TV and professional
radio. In the 1980s the rise of mobile radio,
particularly for cellular telephones, caused a
dramatic increase in demand for filters. New high-
performance SAW filters emerged and vast numbers
are now produced, around 3 billion annually. In the
last two decades SAW devices have found numerous
different applications outside their conventional
fields of application: communications and signal
processing. In the last decade considerable work has
been done in the development of SAW sensors of
different types of high quality. SAW filters are used
as temperature, pressure and stress sensors as well as
chemical and biosensors (Seifert, 1994, Pohl, 2000).
At Imperial College in London are working on
implantable and wearable SAW devices for long
term clinical monitoring. Saw sensors are also used
for wireless monitoring in harsh environment. There
are two different types of SAW sensors: transversal
and resonant. In liquids usually SH SAW type of
sensors are used. In the references only analyses in
time domain of the sensors exist. In the frequency
domain only resonant type of SAW sensors are
modelled (Campbell, 1989).
In this paper modelling of transversal SAW
sensors is presented. It is well known that the exact
analysis of SAW devices using surface wave theory
is very complex (Matthews, 1977). Because of that
approximate methods of analysis are developed. The
simplest method of analysis is using delta function
model. It gives the approximate results relatively
376

fast, but its use is limited to small loads and
substrates with lower coupling constants.
The better approximate methods use equivalent
models for IDTs, where the analysis tools known in
electrical engineering can be applied. In these
methods the accuracy depends on the complexity of
the model. The closed form solutions for transfer
functions and input admittances mostly only for
simple models are given (Matthews, 1977, Morgan,
1985, Smith 1, 1969, Smith 2, 1969, Debnath, 1983,
Hribek, 1983). In this new algorithm SAW
biosensor is modelled as three port network. The
IDT`s equivalent circuit based on Milsom`s and
Redwood`s equivalent model (Milsom and
Redwood, 1971) is most complex. Using the
symbolic analysis method (Hribek, 2007) the closed
form solutions for transfer function and input
admittance of the SAW biosensor are derived. The
algorithm is valid for both types of sensors: direct
and transponder types (Pohl, 2000). The algorithm is
verified by two examples: insertion loss and input
admittance are computed and compared with
experimental results. It is also shown that in the case
of the transponder straightforward dependence of the
loss and measured value is obtained.
2 THE MODEL
The transversal SAW biosensor can be
schematically presented as in Figure 1: between the
interdigital transducers (IDT) on the top of the
piezoelectric substrate the chemical or bio sensitive
layer is placed.

Figure1: The basic configuration of SAW biosensor.
The surface wave is induced by electrical signal
applied to the input IDT. The output signal (voltage)
is taken from the second IDT. The interdigital
transducers are wideband with uniformly spaced
electrodes. The configuration presented in Figure 1
can be modelled by equivalent electrical scheme
given in Figure 2 where IDT`s are three port
networks and the middle sensing part is a two port
network.

Figure 2: Equivalent electrical model of a saw sensor.
The characteristic acoustic impedance of the
unloaded substrate is Z
o
, and the electrical
impedances of the generator and load, are Z
g
and Z
p,

respectively. Each part is defined by its
corresponding admittance matrix Y
(1)
, Y
(2)
and Y
(3)
.
These matrices are calculated following the
procedure presented in (Hribek, Toi). In general
case matricies Y
(1)
and Y
(3)
are different, but usually
in sensors they are equal with small number of
electrodes, thus simplifying the calculations. Since
the circuit is passive for all matrices the passivity
condition: Y
ij
,=Y
ji
, for i j, is valid. F
i
`s and v
i
`s
denote the electrical equivalents of mechanical
forces and velocities. The elements of Y
(2)
of the
sensing part is given by:
s
jZ
g
y y
cot
) 2 (
22
) 2 (
11
= =
(1)
s
Z
ec j
y
cos
) 2 (
12
=
(2)
where
o
f f / = ,
o
f is the central frequency,
s
Z is the acoustic impedance of the sensing part,
and f is the frequency of the input signal.
Now the whole sensor can be represented as an
equivalent two port where one port is the electrical
input of the input IDT and the second port is the
electrical port of the output transducer. The transfer
function of the two port defined as:
g
o
V
V
T = (3)
can be expressed in the following form:
MODELLING OF SAW BIOSENSORS
377

g g
P
Z Y Y Z Y
Z
Y
Y
T
21 12 11 22
21
) 1 )(
1
( + +

=
(4)
where Y
ij
are the admittance parameters of the
equivalent two port. Therefore, for the transfer
function determination y parameters should be
found. They are found in several steps. In each step
one partial Y matrix is derived. In the first step
matrix Y`, which connects input voltage and current
with force F
2
and velocity v
2
, is found. Than the
matrix Y`` which gives the relationship between the
input signals and the F
3
and v
3
is determined.
Finally, the Y matrix is derived in terms of
parameters of the matrices Y
(1)
, Y
(2)
and Y
(3)
. The
expressions are in closed form, but very bulky and
that is the reason why they are omitted in this text.
The symbolic circuit analysis method is used. The
obtained relations are general. In the case of SAW
sensors they can be less complex if input and output
transducers are equal. Also if the simpler models of
IDTs are used, the calculation will be easier, but in
any case computer must be used.
For matching purposes, the input admittance
must be determined. It can be expressed in terms of
the y parameters of the input transducer as follows:
2
12 11 11
11 12 2
13 33
) )( (
2 2
y Y y Y y
Y Y y y
y y Y
s o
s o
i
+ +
+ + +
= (5)
Using the algorithm presented, the computer
program which calculates frequency dependence of
input conductance and susceptance and insertion
loss of SAW device with two uniform transducers
was made. The program was verified by two
examples taken from (Smith et al.1., Smith et al.2).
In these cases the substrate between transducers was
unloaded, e.g.
o s
Z Z = . The computed results are
presented in Figures 3. and 4., solid lines. To allow
comparison between the computed and measured
data, the experimental results are also presented in
the same figures, dotted lines.
From Figures 3 and 4 is obvious that
experimental and calculated data are in excellent
agreement, even better than in (Smith et al.1., Smith
et al.2).


Figure 3: Real (G
a
) and imaginary (B
a
) part of the input
admittance.

Figure 4: Calculated and measured insertion loss.
Since the algorithm and computer programs are
verified than can be successfully used in the analysis
or prediction in any particular case in SAW
biosensors. In that case due to the loading of the
sensitive film the acoustic impedance will change
accordingly, and therefore Y
i
and T have to be
changed.
The algorithm is general. It can be also very
efficiently used in the frequency domain analysis of
SAW transponders (Pohl, 2000). SAW transponders
are SAW devices which do not have sensing film
between the transducers. They get the signal
obtained from the actual physical sensor on the
impedance Z
P
. In that case from relation (4) is
obvious that for one device all admittances are
constants and the only variable is Z
P
. Than, since
o g
Z Z = ,
o s
Z Z = , and
o o
R Z = , transfer
function can be expressed as:
P
o
P
Z
R
T Z T
+
=
1
1
) ( ) (
(6)
where ) ( T denotes the transfer function when Z
P

is infinite. Now two cases can be discussed: when Z
P

is real and when it is of capacitive or inductive type.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
378

If

Z
P
is real e.g.
P P P
G R Z / 1 = = , ratio of
) (
P
Z T and ) ( T can be represented as in Figure
5.
If the loading is purely inductive or capacitive ratio
of ) (
P
Z T and ) ( T can be expressed as follows:
o P
P
R jB
T B T
+
=
1
1
) ( ) (
(7)
1 2 3 4 5 6
G
p
G
0
0.2
0.4
0.6
0.8
1.0
T
T
inf

Figure 5: Relative amplitude versus resistive load.
In that case relative insertion loss can be represented
as in Figure 6.
1 2 3 4 5 6
Y
p
G
0
- 15
- 10
- 5
T
T
inf dB

Figure 6: Relative insertion loss versus load susceptance.
3 CONCLUSIONS
The new developed algorithm is general. It can be
used in frequency analysis of SAW based
biosensors, as well as of SAW transponders. The
efficiency of the presented algorithm is
demonstrated with calculations of frequency
dependence of input conductance and susceptance
and insertion loss of SAW devices with uniform
transducers. The results are compared with
corresponding experimental data showing very close
agreement.
ACKNOWLEDGEMENTS
The authors thank the Ministry of Science and
Technological Development of Serbia for financial
support under the project number 11026.
REFERENCES
Morgan D. P., 1985, Surface Wave Devices for Signal
Processing, Elsevier, London.
Seifert F., Bulst W.E., Ruppel C., 1994, Mechanical
sensors based on surface acoustic waves, in Sensors
and Actuators, A44, pp. 231-239.
Pohl A., 2000, A Review of Wireless SAW Sensors,in
IEEE Transactions on ultrasonics, ferroelectrics, and
frequency control, vol. 47, no. 2, pp. 317-332, (Invited
Paper).
Campbell C., 1989, Surface Acoustic Wave Devices and
Their Signal Processing Applications, Academic
Press, Boston.
Matthews, ed. 1977, Surface Wave Filters, J ohn Wiley,
New York.
Smith W.R., et al, 1, 1969., Analysis of interdigital surface
wave transducer by use of equivalent circuit model, in
IEEE Trans. on MTT, vol.17, No 11, pp.856-864.
Smith W.R., et al, 2, 1969., Design of surface acoustic
delay lines with interdigital transducers, in IEEE
Trans. on MTT, vol.17, No 11, pp. 865-873.
Debnath N., Ajmera J .C., Hribek M. F., Newcomb R.W.,
1983. Scattering and Admittance Matrices of SAW
Transducers, in Circuits, Systems and Signal
Processing, vol.2., No.2, pp. 161-178.
Hribek M., Toi D., 1983, An Improved Algorithm for
Analysis of Uniform SAW Transducers, in Proc. 26
th

Midwest Symposium On Circuits and Systems,
INAOE, Mexico.
Milsom R. F., Reddwood M., 1971, Interdigital
Piezoelectric Rayleigh Wave Transducer: an improved
Equivalent Circuit, in El. Letters, vol.7, p. 217.
Hribek M., Toi D., 2007, Symbolic analysis and design
of current-differencing-amplifier filters, in Scientific
Technical Review, vol. LVII, No. 2.
MODELLING OF SAW BIOSENSORS
379
A WIRELESS EEG ACQUISITION SYSTEM WITH
THERMOELECTRIC SCAVENGING MICRODEVICE
J . P. Carmo, L. M. Goncalves, R. P. Rocha and J . H. Correia
University of Minho, Dept. Industrial Electronics, Campus Azurem, 4800-058 Guimaraes, Portugal
[email protected]
Keywords: Wireless EEG, IrO
2
electrodes, Thermoelectric microsystems, Energy scavenging.
Abstract: This paper presents a wireless EEG acquisition system powered by a thermoelectric energy scavenger,
which was optimised to convert the small thermal power available in the human-body. The wireless EEG
system is composed with up to four EEG electrodes. A radio-frequency (RF) transceiver for operation in the
2.4 GHz ISM band, was optimised and fabricated in the UMC RF 0.18 m CMOS process. The receiver has
a sensitivity of -60 dBm and consumes 6.3 mW from a 1.8 V supply. The transmitter delivers an output
power of 0 dBm with a power consumption of 11.2 mW. Because listening and emitting are power-intensive
activities, innovative topics concerning efficient power management was taken into account during the
design of the RF CMOS transceiver.
1 INTRODUCTION
Energy scavengers are currently emerging for a
number of applications from automotive to
medicine. Micro energy scavengers are small
electromechanical devices which harvest ambient
energy and convert it into electricity. Energy
scavengers could harvest different types of energies.
Solar energy can be harvested with photovoltaic
solar cells, thermal energy can be harvested with
thermoelectric generators, mechanical energy can be
harvested with piezoelectric, electromagnetic or
electrostatic converters, and finally electromagnetic
energy can be harvested through RF resonators. It
exists two types of energy scavenging systems:
macro energy scavengers, typically in the cm
3
range,
and micro energy scavengers, typically in the mm
3

range and manufactured using micromachining
techniques. Micro energy scavengers are still in the
R&D phase. Direct thermal-to-electric energy
conversion without moving mechanical parts is
attractive for a wide range of applications because it
provides compact and distributed power, quiet
operation, and is usually environmentally friendly.
Thus, worldwide efforts are undertaken to expand
the technology of thermoelectric devices into the
field of micro-systems technologies (MEMS).
Previsions made by the specialists of the
microsystems area, shows that the most expected
growth of these devices, will be with medical
applications.
An emerging technology for ultra-low power
communication platforms triggered renewed interest
in power sources for wireless-sensor, in special
wireless-wearable-sensors, with power consumption
nodes of few mW. Today, almost all of these
platforms are designed to run on batteries which not
only have a very limited lifetime, but are also in
many areas a cost-prohibitive solution. An attractive
alternative is powering the sensors with energy
harvested from the environment. The developed
project aims at a solution for energy microgeneration
through energy harvesting by taking advantage of
temperature differences. A viable energy source for
low-powered devices such as micro sensor systems,
ZigBee chipsets, wearable electronics, implantable
medical devices, active RFID tags and many other
applications is proposed, provided a temperature
difference exists, between the two surfaces of a
thermoelectric microgenerator (in a wearable device,
the difference between the body and environment
can be tens of degree, depending on the environment
temperature). This temperature difference can be
converted into electrical energy using the Seebeck
principle. Since many of wireless sensors are
powered in a peak basis (e.g., the transmission of
data needs much more current than standby or
receiving mode) and the temperature gradient could
not always be present, the energy is stored in a
380

rechargeable thin-film battery of the Li-ion type
(integrated in the system). Ultra-low power
electronics performs DC-DC rectification with a
variable conversion factor and recharge the battery
on optimal conditions. Since a small volume is
required, integration into an IC is desirable. A
single-chip regulated thermoelectric power source is
the final goal to be achieved.
2 THERMOELECTRIC
GENERATOR
Future investigations, must prove if the operation
from low temperature gradients (a minimum
temperature difference of 3 C between ambient and
target thermo-source must provide an IC-compatible
voltage) and the power of 2 mW/cm
2
are possible to
be obtained. The fabrication of macro-scale
thermoelectric devices is based on standard
technologies for decades. Bismuth and antimony
tellurides was be used as thermoelectric materials
since these materials have the highest performance
figure-of-merit (ZT) at room temperature
(Gonalves et al, 2006). The co-deposition method
was used to fabricate these thermoelectric thin films.
A very stable evaporation rate of each element
(Bi/Te and Sb/Te for the bismuth telluride and
antimony telluride, respectively) allows the
deposition of polycrystalline n-type and p-type
materials, when the substrate is heated in the range
200-300 C. The design of a thermoelectric
microdevice, with vertical microcolumns, connected
in series by metal contact areas, requires the
application of microsystem technologies
(Gonalves et al, 2007). Reactive ion etching, lift-of
and wet-etching (hydrochloridric acid / nitric acid
solution) techniques were tested to create the vertical
columns. A thin (50 nm) layer of nickel (deposited
by e-beam) interfaces between the thermoelectric
material and the metal contact areas (1 m of
aluminum), preventing the diffusion of the metal
from the contacts into the thermoelectric film. The
contact resistance plays a major role in the
performance of the device, and a value smaller then
110
-6
ohm.cm
2
was achieved. A silicon substrate
was used for integration with microelectronics,
while at same time providing good thermal contact
with heat source and sink. The fabricated battery, as
well as, all the electronic circuitry to receive the
energy and to recharge the thin-film integrated
Li-ion battery (open circuit voltages between 1.5 V
and 4.2V, maximum current of few mA/cm
2
with a
charge-storage capacity around 100 Ah/cm
2
), was
placed on the bottom side of the generator. Thin-film
solid-state batteries show a very high life cycle and
are intrinsically safe (Bates at al, 2000).
The Figure 1 shows an artist impression of a
thermoelectric microdevice, with vertical
microcolumns, connected in series by metal contact
areas. On this thermoelectric generator will be
placed the thin-film integrated battery and all the
electronic circuitry to receive the energy and to
recharge the battery. The application of this
thermoelectric scavenging energy systems, is in the
powering of a wireless EEG acquisition system.
Hot side junctions
Microbattery
Cold side junctions

Figure 1: An artist impression of a thermoelectric
microdevice.
3 EEG ACQUISITION SYSTEM
The standard wireless EEG solutions use a braincap
with wires running from the electrodes position to a
bulky central unity (amplification, signal filtering
and analog-to-digital conversion) (IMEC 2003). A
more interesting solution is to use compact wireless
EEG modules, where the electronics, the antenna
and each electrode are mounted together. The power
supply for these modules is obtained locally from
the thermoelectric generator. Also, it is possible to
integrate additional electronics (amplification,
filtering and high-resolution digital conversion), for
local signal processing in these small-size individual
wireless EEG modules.
Bipolar or unipolar electrodes can be used in the
EEG measurement. In the first method the potential
difference between a pair of electrodes is measured.
In the second method, the potential of each electrode
is compared, either to a neutral electrode or to the
average of all electrodes. Figure 2 shows the full
block diagram of the wireless EEG module, where it
A WIRELESS EEG ACQUISITION SYSTEM WITH THERMOELECTRIC SCAVENGING MICRODEVICE
381

can be seen the electrode connected to an amplifier,
followed by an analog-to-digital converter (ADC).
In order to meet the EEG specifications, the
amplifier was designed to have enough gain, to
amplify signals with amplitudes of only 70 V. The
ADC must have at least a resolution of 22 bits and a
minimum sampling frequency of 2000 Hz.
S/H ADC
control
+
memory
transceiver
IrO
2
electrode
Analog-to-digital converter
Microchip 1
I
m
p
e
x
a

g
i
g
a
A
n
t
a
n
t
e
n
n
a
Microchip 2

Figure 2: Wireless EEG module. Note that the neutral
electrode, which is connected to the grounds of the module
it is not shown.
This wireless acquisition system uses a RF link to
communicate with an external base-station. It was
used the UMC RF 0.18 m CMOS process in the
design and fabrication of the RF transceiver.
Figure 3 shows the block schematic of this RF
transceiver, which consists of a receiver, a
transmitter, an antenna-switch and a
Phase-locked Loop (PLL) as a frequency
syntheziser. The RF transceiver was built to operate
at the 2.4 GHz ISM band. The receiver adopts a
direct demodulation, by means of envelope detection
and has a sensibility of -60 dBm for power
consumption of 6.3 mW from a 1.8 V supply. The
transmitter delivers an output power of 0 dBm with a
power consumption of 11.2 mW. These
characteristics fullfill the requirements for
communications up to ten meters, with a bit error
probability less than 10
-6
.
Moreover, without proper design, communication
tasks may increase network power consumption
significantly because listening and emitting are
power-intensive activities (Enz et al, 2004). Thus, in
order to optimise the power consumption, the RF
transceiver design predicted the use of control
signals. With these control signals it is possible to
enable and disable all the transceiver subsystems.
These signals allows, e.g., to switch off the receiver
when a RF signal is being transmitted, to switch off
the transmitter when a RF signal is being received,
and allows the transceiver to enter to sleep when RF
signals are neither being transmitted, nor being
received. The Figure 4 shows a die photograph of
the RF transceiver.
A
n
t
e
n
n
a
Filter
LNA
Post-amplifier
Envelope
detector
PA
On-chip
RF
switch
RXD
TXD
Local oscilator
2.4GHz
Enablereceiver
Enabletransmitter

Figure 3: The block schematic of the transceiver.

Figure 4: A die photograph of the RF transceiver.
The electrodes to acquire the EEG signals, with the
proposed wireless acquisition system, are of
sputtered iridium oxide (IrO
2
) type. The
experimental results shown a better performance of
the sputtered IrO
2
electrodes compared with the
standard sintered Ag/AgCl ring electrodes
(Slavcheva et al., 2004). These results promise a
new opportunity for the application of a dry IrO
2

electrodes in our wireless EEG modules, without the
need to use conductive gel in the interface between
the electrode and the skin. This will allow patients to
wear a brain cap with the electrodes and maintain
their mobility, while simultaneously having their
electrical brain activity monitored. The Figure 5
shows an artist impression of the thermoelectric
scavenging system and an wireless EEG module,
both attached to a cap (the zoomed part in that
Figure). The temperature gradient between forehead
and the environment will generate energy in the
thermoelectric microdevice and charge the solid-
state thin-film battery.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
382

The modules must offer the plug-and-play feature, in
order to mount distributed networks in the patients
head. Moreover, as the EEG data is periodically
acquired in all the modules, thus the latencies of data
transmissions are not allowed. The proposed EEG
modules uses a communication protocol that
overcomes these problems (Afonso et al, 2006). This
protocol combines the distributed and coordination
modes, e.g., when a new module is putted in the
head of patient, a contention based time interval is
used to make the registration request in the network.
A contention-less time interval, constituted by
time-slots, is granted to the new EEG module if the
registration is successful completed on the network.
The maximum number of simultaneous modules is
limited to the number of time-slots in the contention-
free interval.

Figure 5: An artist impression of the thermoelectric
scavenging system and an wireless EEG module.
4 CONCLUSIONS
This paper presented a thermoelectric scavenging
energy system, whose final goal is to supply a
wireless EEG system. The wireless EEG system is
composed by plug-and-play modules, one for each
electrode. Each wireless module is composed by an
electrode, processing electronics, a radio-frequency
transceiver and an associated antenna.
The RF transceiver was fabricated in the UMC RF
0.18 m CMOS process, for the operation in the
2.4 GHz ISM band, in order to optimise the
consumed power.
The use of microsystems techniques, makes possible
the integration of the whole thermoelectric
scavenging system, in an wireless EEG system.
ACKNOWLEDGEMENTS
The authors would like to thanks the Portuguese
Foundation for Science and Technology
(FCT/PTDC/EEA-ENE/66855/2006 project).
REFERENCES
Afonso, J ., et al, 2006, MAC protocol for low-power real-
time wireless sensing and actuation, in Proc. of the
13th IEEE International Conference on Electronics,
Circuits and Systems, Nice, France, pp. 1248-1251.
Bates, J ., et al, 2000, Thin-film lithium and lithium-ion
batteries, Solid State Ionics, Vol. 135, pp. 33-45.
Enz, C., et al, 2004, WiseNET: An ultralow-power
wireless sensor network solution, IEEE Computer,
Vol. 37, pp. 62-70.
Gonalves, L. M., et al, 2006, Thermoelectric properties
of Bi2Te3/Sb2Te3 thin films, Materials Science
Forum, Vols. 514-516, pp. 156-160.
Gonalves, L. M., et al, 2007, Fabrication of flexible
thermoelectric microcoolers using planar thin-film
thecnologies, J ournal of Micromechanics and
Microengineering, Vol. 17, pp. 168-173.
IMEC, 2003, IMEC press releases, Ambulatory EEG,
Human ++EU project, pp. 1-2.
Slavcheva, E., et al, 2004, Sputtered iridium oxide films as
charge injection material for functional
electrostimulation, J ournal of the Electrochemical
Society, Vol. 151, N7, pp. 226-237.

A WIRELESS EEG ACQUISITION SYSTEM WITH THERMOELECTRIC SCAVENGING MICRODEVICE
383
A NOVEL MOBILE MONITORING SYSTEM FOR FAST AND
AUTOMATED BACTERIA DETECTION IN WATER
Christoph Heller, Ulrich Reidt, Andreas Helwig, Florian Klettner, Gerhard Mller, Alois Friedberger
EADS Innovation Works, IW-SI Sensors, Electronics & Systems Integration, 81663 Munich, Germany
[email protected]
Leonhard Meixner, Karl Neumeier
FhG-IZM Fraunhofer-Institut fr Zuverlssigkeit und Mikrointegration, Hansastr. 27d, 80686 Munich, Germany
Petra Lindner, Ramona Molz and Hans Wolf
University of Regensburg, 93053 Regensburg, Germany
Keywords: Micromechanical filters, Micro fluidic system, Fluorescence detection, Bacteria sampling, Potable water.
Abstract: Standard detection methods for viable bacteria in potable water are time consuming due to a required
cultivation step. Fast and automated detection of water borne microorganisms with high sensitivity and
selectivity is still a challenge. We report on a novel biosensor using micromechanical filters with nano sized
pores to capture and enrich bacteria on the filter surface and subsequent detection using fluorescent probes.
The whole process is fully automated by integrating the sieves into a fluidic system together with a high
performance fluorescence detector. The results show the effective retention of bacteria on the filter surface,
which are then accessible for different staining procedures. As an example, we use special fluorescent dyes
that bind to or intercalate in the DNA molecules of the bacteria. After detection, the microfilters undergo
cleaning and conditioning steps to be ready for the next measurement.
1 INTRODUCTION
Under normal circumstances, our drinking water is
of very high quality. According to the current
directive of the European Community, no viable E.
coli may be present in a 250 mL sample of potable
water. To ensure this high quality, water works and
water suppliers are obliged to test the water quality
at regular intervals.
For these tests, classical methods with cultivation
on special growth media are still being used for
routine applications. The water samples are plated
on culturing plates and incubated at various
temperatures. Depending on the type of bacteria, this
cultivation step can take up to several days. After
incubation, the number of viable bacteria is
determined by visually counting the bacterial
colonies (Brenner, 1996; Gracias, 2004). Obviously,
this procedure is very tedious and time consuming
and requires skilled and experienced laboratory
personnel. The fact that a contamination with some
bacterial species (such as Legionella) will only be
detected after several days may lead to the outbreak
of epidemics. This can be a potential danger to
people with a weak immune system.
In the rare event of bacterial growth in parts of a
water supply network, all pipes, valves, pumps, etc.,
have to be thoroughly cleaned and re-examined
before they can be approved for further use. The
time consuming procedure of classical bacterial
testing will therefore lead to rather long down times
of a water supply system. The same is true for the
supply of clean production water in pharmaceutical
industry and for mobile water supply systems (e.g.
in ships).
To avoid such problems, a system is needed that
can continuously and automatically sample water
and test it quickly for the presence or absence of
bacteria or other pathogens.
Here, we describe such a system that has been
developed in our group. At the core of this device,
there is a micromechanical filter (sieve) having a
large number of pores (about 10
7
) with a diameter of
0,45 m (Figure 1). When passing water through
such filters, any bacteria present in the sample will
384

be effectively captured and retained on the filter
surface. Once on the filter, they are easily accessible
to any kind of specific or non-specific staining
method (e.g. with DNA intercalating dyes, enzyme
substrates, membrane stains, labelled antibodies or
hybridization probes). Fluorescent labels are used in
our monitoring equipment.
Micromechanical filters are highly discriminative
and show a much better filtration performance
(higher filtration rate) than conventional filters of the
same size. Their absolutely flat surface does not
allow any particle to penetrate deeper into the filters
structure. Any bacteria retained by the filter will be
present at the surface, allowing easy detection and
washing off afterwards. Some work has been
reported on using microfilters which successfully
separate particles by means of filtration (Hsiai,
2002; Xing, 1999). These filters have pore diameters
of 6-12 m. Ogura et al. reported the separation of
deformed red blood cells using microfilters with
larger pore size (Ogura, 1991a; b). Since the average
size of bacteria, for example E. coli, is
approximately 1m, theses filters are not suitable for
the discriminative enrichment of bacteria.
In this paper we describe the successful capture,
labelling and detection of E. coli on the surface of
micromechanical sieves. We have integrated the
MEMS filters into a microfluidic system, allowing
fully automated water analysis with this biosensor
system.
2 MATERIALS AND METHODS
2.1 Capture and Enrichment of
Bacteria
The micromechanical filters are composed of a
Si
3
N
4
membrane (5 x 5 mm, approx. 1 m thick)
supported by a silicon frame. The membrane is
perforated, each pore with a diameter of 450 nm
(Figure 1). In contrast to tissue or membrane based
filters, particles are not trapped inside a three
dimensional filter structure but on the surface.
Therefore, the particles are directly accessible.
Previous experiments show that bacteria trapped on
the surface can be easily removed after filtration
(Reidt, in press). A further advantage of the
micromechanical filters is based on the fact that all
pores are simultaneously etched after a
photolithographic patterning step. This ensures that
all pores have exactly the same diameter and their
spacing is completely homogenous (Figure 1).


Figure 1: REM picture of a micromechanical filter surface
with 0,45 m diameter pore size (top view) after filtration
of drinking water.
2.2 A Fully Automated System for
Water Analysis
Using the micromechanical filters (sieves), we have
developed a fully automated system for capturing,
labelling and detecting bacteria in water. The sieves
are housed in a microfluidic chamber with a glass
window in very close distance above the filter
surface. There is a water inlet close to the filter and
one outlet in opposite position. Filter and glass
window form a microfluidic channel, where liquid
can be pumped from the inlet to the outlet passing in
parallel to the filter surface (cross flow). A second
outlet is positioned on the back side of the filter,
allowing for flow through (dead end) filtration.
Using a double selector valve, inlet and outlets can
be activated to perform both types of flow. There is
also the possibility to swap inlet and outlet for back
flushing the filter (Figure 2). The maximum
differential pressure over the microsieve is limited to
2 bar to avoid breaking the filter.
For detection, we use a standard setup with
coaxial excitation and emission light paths. Light
from a high power LED is collected and passed
through an emission filter (485 +/-12 nm band pass).
It is then reflected at 45 angle by a dichroic mirror
(505 nm) and focussed onto the filter surface. The
fluorescent light passes the mirror, is filtered
through a 520 +/-15 nm band pass filter and
focussed onto a photomultiplier tube (Figure 3).
Device control and data acquisition is performed by
LabView software on a laptop computer.
The precision pump (0 - 10 bar, 0.01 - 70
mL/min) was purchased from HNP (Parchim,
Germany), whereas the selector valves were
obtained from Valco-VICI (Schenkon, Switzerland).
As detector we use a Perkin-Elmer MP962 photo
A NOVEL MOBILE MONITORING SYSTEM FOR FAST AND AUTOMATED BACTERIA DETECTION IN WATER
385

Figure 2: Scheme of the fluidic path in the automated detection system. L1 L8: different liquids; V1: selector valve, V2:
double selector valve, P: pump, W: waste, G: glass window, Ch: fluidic chamber, i: inlet, o1, o2: outlets, Det: detection
system, S: micromechanical sieve.

Figure 3: Scheme of the optical path in the automated
detection system. LED: high power LED with lens, L1
L3: lenses, A1, A2: apertures, Ex: excitation filter, Di:
dichroic mirror, Em: emission filter, PM: photo multiplier,
Ch: fluidic chamber.
multiplier; excitation is performed by a blue high
power LED (Luxeon). Optical filters were from
AHF Analysentechnik (Tbingen, Germany) and
LabView software as well as a DAQ board were
from National Instruments. Figure 4 shows a picture
of the complete biosensor system which is small
enough to be used as a mobile system.

Figure 4: Picture of the automated monitoring system for
bacteria in potable water.
2.3 Bacteria Labelling
For non specific labelling, 30 L of a 100 fold
dilution of SYBRGreen I (Invitrogen, Karlsruhe,
Germany) was added to 1 mL bacterial suspension
(about 10
8
cells/mL of heat inactivated E. coli
O157:H7 from KPL, Gaithersburg, USA) and
incubated for 15 min in the dark. From this stock
solution serial dilutions were made in water and the
exact numbers of bacteria were determined by




P
L1
V1
V2
Ch
G
Det
o2
o1
i
S
W
L2 L3 L4 L5 L6 L7 L8
L2
E
Di E L3 PM
L1
A1
A2
LED
Ch
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
386

counting in an improved Neubauer counting
chamber.
For immunodetection (specific detection) on the
filter surface captured bacteria were overlaid with
antibody solution specific for E. coli O157:H7
(polyclonal anti-E. coli O157:H7-Alexa 488 (ABL
Advanced Biomedical, UK) 2.4 g/ml) and
incubated for 15 min. Non-bound antibodies were
removed by an extensive washing step (100 ml,
PBS) and bacteria recognized by the antibodies were
identified by fluorescence microscopy (Reidt, in
press).
3 RESULTS
3.1 Suitability of Micromechanical
Filters for Capturing and
Enrichment of Bacteria
In a first step, the suitability of micromechanical
sieves for capturing bacteria was tested. 20 mL of
various dilutions (10
2
, 10
3
and 10
5
cells/mL) of a
freshly prepared overnight culture of E. coli XL1-
Blue were passed through the filter. After filtration,
bacteria were removed from the filter surface by
rinsing with 5 mL PBS (0.14 M NaCl, 0.01 M KCl,
0.008 M Na
2
HPO
4
, 0.0015 M KH
2
PO
4
, pH 7.4). In
each case 100 L of rinsing solution and filtrate
(flow through) were plated on LB agar (10 g
tryptone, 5 g yeast extract, 5 g NaCl to 1 liter H
2
O)
and incubated overnight at 37C. Cell counts
obtained by plating initially applied bacterial
solutions (original solution) and from rinsing
solutions correlate very well, indicating that bacteria
can easily and efficiently be removed from the filter.
A major finding of all filter experiments was that
in no case any bacteria could be detected in the 100
L samples from the flow-through. The
impermeability of the filters for bacteria was further
confirmed by examination of the complete filtrate
solution for the presence of bacteria. This
demonstrates a highly efficient and reliable filtering
process using the inorganic microsieves (Reidt, in
press).
3.2 Bacteria Detection
For unspecific detection we use fluorescent dyes that
bind to or intercalate in the DNA of the bacteria.
Upon binding, the fluorescent signal of the dyes
increases by a factor of 1.000-10.000.



Figure 5: Upper: Microscopic image of part of the
microsieve after filtering a water sample containing
SYBRGreen I stained bacteria. The width of an active
filter stripe is 180 m. Lower: Same area, with
fluorescence microscopy. The bright spots represent
bacteria.
In our case inactivated E. coli O157:H7 are
stained with SYBRGreen I. Water samples are
spiked with different amounts of fluorescently
labelled bacteria and analyzed in the monitoring
system. Figure 5 shows a microscopic image of the
filter after analysis and the corresponding
fluorescent image visualizing the bacteria. So far, a
limit of detection of 40.000 cells on a single
microfilter could be achieved (Figure 6). The
detection limit can be significantly improved as it is
currently limited by the optical detection system not
yet being optimized.
With the system, it is also possible to first pass a
water sample (with non labelled bacteria) through
the microsieve and then stain the cells on the filter
surface.
Obviously, it should be possible to perform other
specific or non specific labelling procedures. For
example, we have also tested the immunodetection
180 m
A NOVEL MOBILE MONITORING SYSTEM FOR FAST AND AUTOMATED BACTERIA DETECTION IN WATER
387

of non-pathogenic E. coli O157:H7 on
micromechanical filters. It could be demonstrated
that E. coli serotype O157:H7 could be detected
specifically by the antibodies while neither the
micromechanical filter surface nor the negative
control (E. coli DH5) interacted with the antibodies
(Reidt, in press).
0
1.000
2.000
3.000
4.000
5.000
6.000
7.000
0 200 400 600 800 1.000 1.200
Number of bacteria (x 1000)
S
i
g
n
a
l

(
k
H
z
)

Figure 6: Detection of fluorescently labelled bacteria with
a fully automated detection system.
Depending on the source, there may be potable
water available in some regions with a quite high
density of particles. This could cause problems in an
automated water analysis system. One concern could
be a background fluorescence signal originating
from particles. Although the optical system is not
optimized yet, we could not observe a significant
signal from particles.
Another issue might be the blocking of the filters
by the dirt. In fact, water flow rate of tap water
through the sieves is reduced in comparison to pure
water. However, a reasonable flow rate is still
sustained (Figure 7). If the system should be used
for water with a high density of particles, the
microfilter area can simply be increased (currently it
is 5 x 5 mm).

0
5
10
15
0 5 10 15 20
Volume (mL)
F
l
o
w
r
a
t
e

(
m
L
/
m
i
n
)

Figure 7: Flow rate in mL/min of tap water through a
microsieve (microfilter) with size 5 x 5 mm.
4 DISCUSSION
We have shown that micromechanical filters can
retain bacteria from water with high efficiency and
reliability. Also, bacteria can directly be detected on
the surface of the micromechanical filters using
fluorescently labelled antibodies or DNA binding
dyes. The microfilter platform is very flexible, i.e.
other specific or non specific labelling procedures
can be applied such as membrane probes, enzyme
substrates, nucleic acid hybridization probes or PCR.
Furthermore, the micromechanical filters have
been integrated into a fully automated fluid delivery
and detection system. Micromechanical filters with
appropriate pore sizes could also be an effective tool
for recovering other pathogens such as viruses and
protozoa.
During the last years, a few bacteria detection
methods have been developed as alternatives to
cultivation techniques, mainly based on flow
cytometry (e.g., Sakamato, 2005) or solid phase
cytometry (e.g., Aurell, 2004; Lisle, 2004). Flow
cytometry is fast, but is only capable of analyzing a
volume of a few microlitres per minute and thus
requires preconcentration of the drinking water
sample. In solid phase cytometry, the water is
filtered through a porous membrane which has then
to be transferred to a laser scanning device. Total
analysis time is about 4 hours.
In both cases, a number of manual steps is
required, making automation difficult, whereas our
method offers the possibility of fully automated and
rapid analysis of potable water samples.
ACKNOWLEDGEMENTS
We thank Erwin Yacoub-George and Waltraud Hell
(FhG-IZM) as well as Eberhard Rose and Thomas
Ziemann (EADS) for their valuable assistance. Part
of this work has been supported through the project
OptoZell funded by the German ministery for
education and research, BMBF (funding agency:
VDI).
REFERENCES
Aurell, H., Catala, P., Farge, P., Wallet, F., Le Brun, M.,
Helbig, J .H., J arraud, S., Lebaron, P., 2004. Rapid
Detection and Enumeration of Legionella
pneumophila in Hot Water Systems by Solid-Phase
Cytometry. Applied and Environmental Microbiology
70, 16511657.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
388

Brenner, K.P., Rankin, C.C., Sivaganesan, M., Scarpino,
P.V., 1996. Comparison of the recoveries of
Escherichia coli and total coliforms from drinking
water by the MI agar method and the U.S.
Environmental Protection Agency-approved
membrane filter method. Appl Environ Microbiol 62,
203-208.
Gracias, K.S., McKillip, J .L., 2004. A review of
conventional detection and enumeration methods for
pathogenic bacteria in food. Can J Microbiol 50, 883-
890.
Hsiai, T.K., Cho, S.K., Yang, J .M., Yang, X., Tai, Y.C.,
Ho, C.M., 2002. Pressure drops of water flow through
micromachined particle filters. Journal of Fluids
Engineering-Transactions of the Asme 124, 1053-
1056.
Lisle, J .T., Hamilton, M.A., Willse, A.R., McFeters, G.A.,
2004. Comparison of Fluorescence Microscopy and
Solid-Phase Cytometry Methods for Counting Bacteria
in Water. Applied and Environmental Microbiology 7,.
53435348.
Ogura, E., Abatti, P.J ., Moriizumi, T., 1991a.
Measurement of human red blood cell deformability
using a single micropore on a thin Si3N4 film. IEEE
Trans Biomed Eng 38, 721-726.
Ogura, E., Kusumoputro, B., Moriizumi, T., 1991b.
Passage time measurement of individual red blood
cells through arrayed micropores on Si3N4 membrane.
J Biomed Eng 13, 503-506.
Reidt, U., Chauhan, L., Mller, G., Molz, R., Lindner, P.,
Wolf, H., Friedberger, A., Reproducible Filtration of
Bacteria with Micromechanical Filters. Journal of
Rapid Methods and Automation in Microbiology, in
press.
Sakamoto, C., Yamaguchi, N. Nasu, M., 2005. Rapid and
Simple Quantification of Bacterial Cells by Using a
Microfluidic Device. Applied and Environmental
Microbiology 71, 11171121.
Xing, X., Yang, J . M., Tai, Y. C., Ho, C. M., 1999.
Micromachined membrane particle filters. Sensor.
Actuat. A-Phys 73, 184-191.
A NOVEL MOBILE MONITORING SYSTEM FOR FAST AND AUTOMATED BACTERIA DETECTION IN WATER
389
CONTROL OF CELL ADHESION AND FUNCTIONS
USING SELF-ORGANIZED HONEY COMB-PATTERNED
POLYMER FILMS
Masaru Tanaka
Institute of Multidisciplinary Research for Advanced Materials, Tohoku University, 2-1-1Katahira, Aoba-ku, Sendai, Japan
[email protected]
Akinori Tsuruma, Sadaaki Yamamoto
Sadaaki Yamamoto Creative Research Initiative Sousei (CRIS), Hokkaido University, N20W10 Kita-ku, Sapporo, Japan
[email protected]
Masatsugu Shimomura
Institute of Multidisciplinary Research for Advanced Materials, and World Premier International Research Center
Tohoku University, 2-1-1Katahira, Aoba-ku, Sendai, Japan
[email protected]
Keywords: Self-organization, Honeycomb, Scaffold, Tissue engineering, Cell adhesion, Stem cell.
Abstract: The design of nano- and microstructures based on self-organization is a key area of research in the search
for new biomaterials and biodevices, and such structures have a variety of potential applications in tissue
engineering scaffolds and medical implants. 3D scaffolds of appropriate pore size and porosities and with
interconnected pores are required to facilitate cell adhesion, proliferation, differentiation, and eventual tissue
regeneration in a natural manner. We have reported the honeycomb-patterned polymer film with highly
regular pores that is formed by self-organization. The honeycomb films exerted a strong influence on cell
morphology, proliferation, cytoskeleton, focal adhesion, and ECM production profiles. Our studies
demonstrated that the neural stem / progenitor cells morphology, proliferation, and differentiation are
controlled by the pore size of the honeycomb film. It is expected that the honeycomb films will have great
potentials as biomaterials for tissue regeneration in a growth factor free proliferation process of stem cells.
1 INTRODUCTION
It is well established that surface topography
influences implant integration. Many in vitro studies
have extended these observations to cells in culture,
demonstrating that scaffold architecture and surface
chemistry considerably influence cell behavior.
Therefore, cell adhesion, proliferation, and
differentiation can be regulated by controlling
surface topography.
We have reported a honeycomb-patterned
polymer film (honeycomb film) with regular pores,
which is formed by self-organization. The
honeycomb films strongly affected cell morphology,
proliferation, cytoskeleton, focal adhesion, and
extracellular matrix (ECM) production profiles
(Tanaka, 2006., Tanaka, 2007., Yamamoto, 2007.,
Mcmillan, 2007., Tanaka, 2008., Arai, 2008.,
Tsuruma, 2008.). These studies were performed on
cells cultured in the absence of growth factors.
In neural tissue engineering, the preparation of
neural stem/progenitor cells (NSCs) is required for
the treatment of diseases of the nervous system
(Steinman, 2003). NSCs are self-renewing,
immature, undifferentiated, and multipotent cells.
They can differentiate into cells constituting the
central neural system, such as neurons, astrocytes,
and oligodendrocytes (Roberti, 2000., Wurmser,
2004., Wang, 2006.). The use of NSCs is a potential
therapy for diseases of the nervous system. In order
to increase the feasibility of this technique, viable
method, the preparation, culture, and seeding of
NSCs are steps that have to be carefully controlled.
The main problem associated with the use of stem
cells is that when these cells are extracted from an
390

individual, they start differentiating (specializing
into a specific cell type); thus, they lose their stem
cell characteristics. Thus, the reintroduction of these
cells into patients is problematic. In order to
overcome this problem, a culture environment
wherein the stem cells can be maintained in the
undifferentiated state is required. The proliferation
of self-renewing NSCs is required for cell therapy.
We studied the effects of the pore size of the
honeycomb films on the proliferation and
differentiation of NSCs.
2 MATERIALS AND METHODS
Honeycomb films were fabricated using
biodegradable polymers poly(-caprolactone) (PCL)
and a copolymer of dodecylacrylamide and -
carboxyhexylacrylamide. The honeycomb film was
prepared on a glass substrate by employing a
previously described method (Sato, 2002., Tanaka,
2004., Tanaka, 2007). The flat film was prepared by
a spin coater in dry condition. NSCs were derived
from the cerebral cortex of embryonic 14 day mice.
The NSCs were seeded on the films at a density of 2
10
4
cells/cm
2
. NSCs were cultured in serum
medium (Opti-MEM, 10 % Fetal Bovine Serum, 55
M 2-mercaptoethanol) for 24 hr. After that, NSCs
were cultured in serum-free medium. The
morphologies of neurons were examined by a
scanning electron microscope (SEM) and a confocal
laser scanning microscope. NSCs were
immunostained for Nestin and BrdU. Cell number
was estimated by measuring of DNA concentration
from the extracted samples.
3 RESULTS AND DISCUSSION
Scanning electron microscopy (SEM) revealed a
highly regular hexagonal arrangement of pores
(honeycomb pattern), and a well-interconnected,
uniform pore structure (Fig. 1ad). NSCs cultured
on flat films differentiated into round neurons with
neurites that extended randomly on the film. The
morphology of cells on the honeycomb films
depended on the pore size of the films. On films
with a pore size of 1.5 m, the cell bodies were flat.
Their neurites extended randomly and jumped over
the pore of the film. On films with a pore size of 10
m, the cell bodies were round and the cells adhered
to the rims of the films. Their neurites extended
along the rims forming a simple network. The
honeycomb film provided positive cues to support
neurite extension. The cells on films with pore sizes
of 5 and 8 m exhibited a morphology similar to that
of cells on films with a pore size of 10 m.
Interestingly, on honeycomb films with a pore size
of 3 m, several large spheroids were observed. The
neurites gathered to form large bundles, which
radiated out from the periphery of the spheroids.
NSCs on honeycomb films with a pore size of 3 m
formed spheroids of diameter 30~90 m (Fig. 2);
such spheroid formation was not observed for NSCs
either on other honeycomb films (with pore sizes of
1.5, 5, 8, and 10 m) or on flat films.
Top
layer
Bottom
layer
Pillar
d
0 m 10
m
Tilted
5 m m 5 m
Closs
Top
Pillar
Bottom
10 m 10 m
Top
(a)
(b)
(c)
(d) Top
layer
Bottom
layer
Pillar
d
0 m 10
m 10
m m
Tilted
5 m m 5 m 5 m
Closs
Top
Pillar
Bottom
10 m 10 m
Top
(a)
(b)
(c)
(d) Top
layer
Bottom
layer
Pillar
d
0 m 10
m
Tilted
5 m m 5 m
Closs
Top
Pillar
Bottom
10 m 10 m
Top
(a)
(b)
(c)
(d) Top
layer
Bottom
layer
Pillar
d
0 m 10
m 10
m m
Tilted
5 m m 5 m 5 m
Closs
Top
Pillar
Bottom
10 m 10 m
Top
(a)
(b)
(c)
(d)

Figure 1: SEM images of the honeycomb film in different
views. (a) Surface, (b) tilted, and (c) cross-section. (d)
Schematic representation of the 3D structure of the
honeycomb film.
Cul ture ti me (days)
D
i
a
m
e
t
e
r

(

m
)
0
20
40
60
80
100
120
3 5 7 10
3 days
7 days
10 days
Cul ture ti me (days)
D
i
a
m
e
t
e
r

(

m
)
0
20
40
60
80
100
120
3 5 7 10
3 days
7 days
10 days

Figure 2: Spheroids diameter on the honeycomb film (pore
size:3 m). Fluorescent images indicate spheroids stained
for Nestin. Bar: 50 m.
In order to characterize the cells on the flat and
honeycomb films, the cells were immunostained for
nestin and microtubule-associated protein 2 (MAP2)
and were labeled with bromodeoxyuridine (BrdU).
Nestin and MAP2 are selective markers for NSCs
and neurons, respectively. Nestin expression
decreases when NSCs differentiate and mature into
neurons. BrdU is selectively incorporated into the
nuclei of proliferating cells, and thus indicates cell
CONTROL OF CELL ADHESION AND FUNCTIONS USING SELF-ORGANIZED HONEY COMB-PATTERNED
POLYMER FILMS
391

growth. Immunostaining for nestin and labeling with
BrdU revealed that the spheroids were aggregates of
self-renewed NSCs. The diameter of the spheroids
increased with the culture time (Fig. 2).

Figure 3: The removability and phenotype of the spheroids
on the honeycomb film (pore size: 3 m).

Figure 4: Comparison of our method for proliferation of
NSCs using honeycomb films with the conventional
neurosphere method.
These results implied that honeycomb films with
pore size less than the cell size promoted the
proliferation of NSCs, but prevented their
differentiation. We found that the number of total
neural cells increased after 3 days owing to the
maintenance of the undifferentiated state and to the
proliferation of NSCs.
In order to determine the removability of the
spheroids from honeycomb films and to ascertain the
phenotypes of the cells, cells obtained from the
spheroids were cultured in a standard culture dish
(Fig. 3). All cells of the spheroids adhered to the
culture dish. The cells extended neurites after 2 d
and were positive for MAP 2, suggesting their
differentiation into neurons. This result implied that
the cells in the spheroids could differentiate into
neurons and confirmed the finding of the
immunostaining experiment that the spheroids were
aggregates of self-renewed NSCs.
The conventional neurosphere culture method is
widely used for the proliferation of NSCs (Cattaneo,
1990., Louis, 2004). This technique involves the use
of serum-free culture medium supplemented with
growth factors (fibroblast growth factor-2: FGF-2
and/or epidermal growth factor: EGF) (Fig. 4). The
NSCs obtained by this method are expected to be
supplied to lost and dysfunctional nervous systems
in order to regenerate neural tissue. In this
technique, NSCs are cultured without the attachment
to a surface (floating culture) because the NSCs
immediately differentiate into neurons when they are
attached to the substrate surface. Improvements in
the neurosphere culture technique, which include the
use of U-bottomed wells, have recently been
reported (Mori, 2006). However, our technique
required neither growth factors nor the floating
culture system wherein the cells are not attached to a
surface.
In our technique, some NSCs were observed to
be encapsulated by the 3 m pores immediately after
cell seeding. At this adhesion arraignment, the NSC
contacts to the pore at around the cell body. Such
circular adhesion may result in a small adhesion
area; thus, the NSCs are in an environment similar to
that in the neurosphere method, that is, they are
suspended to prevent contact with surfaces. Thus,
cell encapsulation, in contrast to cell adhesion that is
observed on flat surface, is probably the reason for
the control of NSC proliferation by surface
topography; such encapsulation is characteristic to
the honeycomb film with a pore size of 3 m.
4 CONCLUSIONS
Our study revealed that the morphology,
proliferation, and differentiation of NSCs are
controlled by the pore size of the honeycomb film.
This is a novel approach to NSC culture in
regenerative medicine, wherein the proliferation and
differentiation of the NSCs are controlled by the
surface topography of scaffolds. Honeycomb films
are potentially useful biomaterials for neural tissue
regeneration, which can help in the proliferation of
NSCs in the absence of growth factors.



BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
392

ACKNOWLEDGEMENTS
This work is supported by Grants-in-Aid and
CREST from J apan Science and Technology
Corporation (J ST), and Special Coordination Funds
for Promoting Science and Technology of Ministry
of Education, Culture, Sports, Science and
Technology, J apan.
REFERENCES
Arai, K., Tanaka, M., Yamamoto, S., Shimomura, M.,
2008. Colloids Surf A, 313-314, 530.
Cattaneo, E., Mckay, R., 1990. Nature 347, 762.
Louis, S. A., Reynolds, B. A., 2004. Meth. Mole. Biol.
290, 265.
Mcmillan, J . R., Akiyama, M., Tanaka, M., Yamamoto, S.,
Goto, M., Abe, R., Sawamura, D., Shimomura, M,
Shimizu, H., 2007. Tissue Engineering, 13, 789.
Mori, H., Ninomiya, K., Kino-oka, M., Shofuda, T. 2006.
J. Neurosci. Res,. 84, 1682.
Roberti, Y.L., Ronald, D.G., 2000. J. Neurosci. 20, 3725.
Sato, K., Hasebe, K., Tanaka, M., Takebayashi, Nishikawa,
K., Kawai.T., Matsushita, M., Shimomura, M., and S,
Todo., 2002. Inter. J. Nanosci. 1, 689.
Steinman, L., 2003. Nature, 422, 671.
Tanaka, M. Takebayashi, M., Miyama, M., Nishida, J .,
Shimomura, M., 2004. Biomed. Mater. Eng., 14, 439.
Tanaka, M., Nishikawa, K., Okubo, H., Kamachi, H.,
Kawai, T., Matsushita, M., Todo, S., and Shimomura,
M., 2006. Colloids Surf A, 284285, 464.
Tanaka, M., Takayama, A., Ito, E., Sunami, H.,
Yamamoto, S., Shimomura, M., 2007. J. Nanosci.
Nanotech, 7, 763.
Tanaka, M., Yoshizawa, K., Tsuruma, A., Sunami, H.,
Yamamoto, S., Shimomura, M., 2008. Colloids Surf A,
313-314, 515.
Tsuruma, A., Tanaka, M., Yamamoto, S., Shimomura, M.,
2008. Colloids Surf A, 313-314, 536.
Wang, J . H., Hung, C. H., Young, T. H., 2006.
Biomaterials, 27, 3441.
Wurmser, A. E., Palmer, T. D., Gage, F. H., 2004.
Science, 304, 1253.
Yamamoto, S., Tanaka, M., Sunami, H., Ito, E., Yamashita,
S., Morita, Y., Shimomura, M., 2007. Langmuir, 23,
8114.
CONTROL OF CELL ADHESION AND FUNCTIONS USING SELF-ORGANIZED HONEY COMB-PATTERNED
POLYMER FILMS
393
ACOUSTIC THERMOAGITATION BASED ON PIEZOELECTRIC
-PVDF POLYMER FILMS
Potential Evaluation in Lab-on-a-Chip Applications
V. F. Cardoso,

G. Minas
University of Minho, Department of Industrial Electronics, Campus de Azurm, 4800-058, Guimares, Portugal
[email protected], [email protected]
P. Martins, J . Serrado Nunes,

L. Rebouta, S. Lanceros-Mndez
University of Minho, Department of Physics, Campus de Gualtar, 4710-057, Braga, Portugal
[email protected], [email protected], [email protected], [email protected]
G. Botelho
University of Minho, Department of Chemistry, Campus de Gualtar, 4710-057, Braga, Portugal
[email protected]
Keywords: Lab-on-a-chip, PVDF, Acoustic thermoagitation.
Abstract: This paper describes a lab-on-a-chip device with acoustic thermoagitation based on a piezoelectric -PVDF
polymer. The device is used for testing and monitoring biochemical parameters in biological fluids using
optical absorption spectrophotometry. Experimental results regarding the influence of the electrical signal
amplitude and frequency applied for the generation of acoustic thermoagitation is presented. The individual
contribution of the heating and the microagitation provided by the actuation of the piezoelectric film for the
fluids mixture was determined. The paper is completed with a study of the -PVDF degradation with
transparent conductive electrodes, ITO and AZO, when placed in contact with uric acid fluids. The final
goal of using this technique is the improvement of mixing and reaction time without interfering with
biochemical reactions and analytical measurements.
1 INTRODUCTION
For the development of a lab-on-a-chip device for
fluidic analysis, the microfluidics technologies have
been a vital tool allowing the fabrication of precise
and small structures. The huge interest in these
technologies is inherent to the performance
achieved: reduction on the sample quantity, high
integration and consequently high potential for
fluids automation and control in sub-microliter
volumes, decrease of response time, reduction of
chemical quantities stored and a large reduction in
total costs (Auroux et. al, 2002).
Lab-on-a-chip devices can perform several
analyses simultaneously. In order to get in-loco,
quick and reliable results, they need an automatic
system for microfluids control, which can cover all
the steps of a chemical or biological process.
Usually, a bio(chemical) process needs the
mixture of fluids. For that, devices based in MEMS
(Micro Electro Mechanical Systems) such as
micropumps (Reyes et. al, 2002) and microvalves
(Rife et. al, 2000) are used, but they increase the
device cost, need complex control systems and their
integration is complex. The mixture of the fluids
using just diffusion avoids these disadvantages.
However, when large molecules with small
diffusivities take part in the reaction, it is needed
high transit times of the molecules on the channels
and consequently long channels (Ottino et. al, 2004).
To overcome this high transit times, it is necessary
to develop alternative methods for improving
mixture.
The use of acoustic waves is an interesting
solution and it is one of the main issues of this
present work. The acoustics waves traveling inside
the fluid create differential pressures and induce the
so called acoustic propagation. For occurring the
394

acoustic microagitation of the fluids, the reaction
chamber must be coated with an electroactive
polymer. One example is the -PVDF (Lanceros-
Mendez et. al, 2006), which the use for fluid
microagitation purposes is the main innovation of
this work. Applying an a.c. voltage to the contacts of
the piezoelectric -PVDF film, it is produce
mechanical oscillations promoting the movement,
mixture and reaction of the fluids, as well the heat
generated by this technique, the so called acoustic
thermoagitation.
2 DEVICE DESCRIPTION
This paper describes the incorporation and
validation of an acoustic thermoagitator based on a
piezoelectric -PVDF polymer in a fully-integrated
disposable lab-on-a-chip for point of care testing and
monitoring of biochemical parameters in biological
fluids. This lab-on-a-chip has interesting
characteristics such as portability, low-cost and
disposability. Furthermore, it has a completely
automatic operation and uses optical absorption
spectrophotometry as analytical measurement
technique.
2.1 Biosystem Operation
The lab-on-a-chip is composed by two dies: the
fluidic die and the detection die (Figure 1).

Figure 1: Schematic representation of the lab-on-a-chip
structure with the -PVDF deposited underneath the
microfluidic structure.
The microfluidic die, fabricated in SU-8,
includes the microchannels and the reaction
chambers. Three reaction chambers are needed for
each analysis. One with a well-known concentration
sample of the biochemical parameter that is being
analysed, is used for the calibration of the device.
Other, for the mixture of the sample and the reagent
is needed, to perform the analysis of the coloured
mixed solution. Finally, a third one is used for the
chemical reagent in order to obtain the baseline
reference. Underneath the reaction chambers was
deposited the piezoelectric -PVDF polymer with its
corresponding electrodes.
The detection die includes the photodetectors and
the electronic components for signal actuation and
detection, all fabricated in CMOS technology.
Above the photodetectors, there are several high-
selective band-pass optical filters, deposited by Ion
Beam Deposition, that select the wavelength
according to the several biomolecules into analysis.
This optical filtering system allows the use of non-
calibrated external polychromatic light source.
2.2 Analytical Measurement Technique
Among the several analytical techniques available in
laboratories of clinical analysis, the
spectrophotometry is the most used. However, this
technique cannot be used directly since a high
number of biomolecules for clinical analysis do not
have chromophores which absorb the light in the
visible spectra. In order to overcome this limitation,
several specific chemical reactions allow to
transform these biomolecules in colored products
which absorbency is within the visible light spectra -
colorimetric reactions.
In the ideal colorimetric analysis, the mixture
coloration intensity is proportional to the
concentration of the biochemical parameter and can
be quantified measuring the optical absorption of the
mixture (reagent + biomolecule) at a specific
wavelength (Thomas, 1999).
2.3 Acoustic Streaming
The acoustic streaming is a steady flow generated by
the propagation of acoustic waves in a viscous fluid.
It arises from the transfer of momentum and energy
of the acoustic field to the medium, through its
acoustic attenuation.
Acoustic streaming offers several distinct
advantages for application in microfluidic devices
(Frampton et. al, 2004). In this work, the acoustic
streaming is due to the absorption of the acoustic
energy in the fluid itself. This absorption results in a
radiation pressure in the direction of the acoustic
propagation and is termed by quartz wind.
Quartz wind velocity and effective pressure are
limited by the heating and cavitation tolerance. A
small fraction of the incident acoustic energy goes
into kinetic energy of the fluid and the rest is
transformed into heat. In this way, the acoustic
thermoagitation generated by this effect becomes
Chip
Optical fi lters and
Photodetectors
Electroni cs
PVDF
Reagent
Mixture
Calibrator
Reagent
Sampl e
Gl ass
Gl ass
SU-8
ACOUSTIC THERMOAGITATION BASED ON PIEZOELECTRIC -PVDF POLYMER FILMS - Potential Evaluation
in Lab-on-a-Chip Applications
395

advantageous, once in some applications or
reactions, the increase of temperature is beneficial in
order to reduce the mixture time (Rife et. al, 2000).
2.4 Poly(vinylidene fluoride)
Nowadays, with the rapid development of polymer
processing methods and synthetic technology, the
industrial materials used in various manufacturing
fields are being substituted by polymer substances,
which have better properties than those of materials
previously used in many industrial areas. One of the
polymer that has received increased attention is the
poly(vinylidene fluoride), also known as PVDF. It
shows an unusual polymorphism in this class of
materials, showing four different crystalline phases.
From the technology point of view, the -phase is
the one which shows better properties to be applied
in sensors, actuators and transducers, due to its
higher piezo- piro- and ferroelectric properties.
Moreover, it shows an excellent combination of
processability, chemical agent resistance, lightness,
moldability and low-cost production. While ceramic
materials break easily and have hard and dense
structures, PVDF is flexible, has a low density and
can be easily produced into thin-films (Brown,
1992). This polymer also shows low acoustic and
mechanical impedance (Foster, 2000) crucial for
generating the acoustic waves that produce the
thermoagitation of the fluids, as is the purpose of
this study. Another interesting feature of the PVDF
is its transparency. Indeed, in the case of this
application, the analytical measurement by
spectrophotometry requires that the PVDF and the
corresponding conductive electrodes, deposited in
the reaction chamber, are transparent to visible light.
3 EXPERIMENTAL RESULTS
3.1 Acoustic Thermoagitation
The quantitative evaluation of the mixing process
was carried out using the Far Diagnostic kit and
standards of urine with 5 mg/dl of uric acid
concentration. The reagent reacts with the sample of
urine containing uric acid in a 40:1 ratio with a
maximum absorption at 550 nm.
The acoustic thermoagitation was studied for
various frequencies and amplitudes of the electrical
signal applied to the electrical contacts of a 110 m
thick -PVDF film. The system was calibrated and
the reaction was studied up to a maximum time of
20 minutes, which is the time of the complete
mixture without thermoagitation.
The thermoagitation was performed powering
the -PVDF, with a sinusoidal signal of 10 V
amplitude at different frequencies. After, a 10 MHz
frequency sinusoidal signal with various amplitudes
was applied. They results were processed to obtain
the results shown in Figure 2.
-2000 0 2000 4000 6000 8000 10000 12000 14000 16000
200
400
600
800
1000
1200
1400
Frequency signal
M
i
x
t
u
r
e

T
i
m
e

(
s
)
Frequency signal (kHz)
0 2 4 6 8 10
Amplitude signal
Amplitude signal (V)

Figure 2: Mixing time as a function of the frequency and
amplitude of the signal applied to the electrical contacts of
a 110 m thick -PVDF film.
It can be observed that with the application of
thermoagitation, the mixing occur in a faster way,
being the time necessary to obtain the complete
mixing, at 10 V amplitude and 10 MHz, only two
sevenths (351 s) of the complete mixing time
without thermoagitation (1200 s). In this way, the
resonance frequency of the used -PVDF film
(10 MHz) is the most efficient for the generation of
thermoagitation. This result is consistent with the
theory (Eiras, 2007). On the other hand, the mixing
time decreases as the signal amplitude increases.
According to previous results, it was determined
the individual contribution of the agitation and the
heating generated by the use of this acoustic
thermoagitation technique. For that, the temperature
profile of the sample was measured applying an
electrical signal of 10 V amplitude and 10 MHz to
the -PVDF film. Then, the sample was heated with
the same temperature profile obtained before with
the thermoagitation, using a temperature controller
(SHIMADZU TCC-260). It was observed that, for
the two sevenths in terms of gains achieved by the
application of the acoustic thermoagitation
technique, three-fifths are due to heating and two
fifths to agitation.
3.2 Transducer Degradation
As the first application of the present lab-on-a- chip
will be for measurement of uric acid concentrations,
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
396

a degradation study (using a high-performance
liquid chromatograph model 875 from Jasco) of the
110 m -PVDF film with ITO (Indium Tin Oxide)
and AZO (Aluminium doped Zinc Oxide)
electrodes, when placed in contact with the
biological fluids, was performed. The study was
done with the Far Diagnostic kit of uric acid,
described previously. The chemical reaction was
performed with and without acoustic
thermoagitation during 20 minutes. The
thermoagitation was set using an electrical signal at
10 V amplitude and 10 MHz. The results shown in
Figures 3 and 4 are representative for not degraded
and degraded electrodes, respectively.
2.00 2.05 2.10 2.15 2.20 2.25 2.30 2.35 2.40
0
2
4
6
8
10
12
14
16


I
n
t
e
n
s
i
t
y

(
a
.
u
.
)
Time (s)
Allantoin
AZO without thermoagitation
ITO without thermoagitation
ITO with thermoagitation

Figure 3: Chromatogram of the samples that did not
suffered degradation with a reaction time of 20 min.
2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4
-2
-1
0
1
2
3
4
5
6
7
8
9
10
11
12
Allantoin
AZO with thermoagitation

I
n
t
e
n
s
i
t
y

(
a
.
u
.
)
Time (s)
Figure 4: Chromatogram of the samples that suffered
degradation with the reaction time of 20 min.
It can be observed that the -PVDF films with
ITO electrodes did not influence the analyses results.
The opposite happens when the electrodes of the
-PVDF film include aluminium, even in small
concentrations, like the transparent conductive AZO.
4 CONCLUSIONS
The application of acoustic thermoagitation through
the -PVDF piezoelectric polymer is gainful when
fluids need to be mixed in a microfluidic device. It
reduces the mixing time resulting in a fast, complete
and homogeneous reaction of the reactants,
improving the global performance of the analysis
that is being performed.
With the optimization tests, it was proved that
better mixing results were obtained with the
electrical signal for actuation at the resonance
frequency of the film and with an amplitude of 10V.
Moreover, it was demonstrated that the improvement
of the mixture is mainly due to the heating of the
solution.
In the degradation study, it was verified that, for
all electrodes, the ITO is the one which does not
degrade in contact with the uric acid kit.
ACKNOWLEDGEMENTS
Work supported by the Portuguese Science
Foundation (grants PTDC/BIO/70017/2006,
PTDC/CTM/69362/2006).
REFERENCES
Auroux, P., Iossifidis, D., Reyes, D. R., Manz, A., 2002,
Micro Total Analysis Systems: Analytical Standard
Operations and Applications. Anal. Chem., 74, p.
2637-2652.
Reyes, D. R., Iossifidis, D., Auroux, P., Manz, A., 2002,
Micro Total Analysis Systems. 1. Introduction,
Theory, and Technology. Anal. Chem., 74, p. 2623-
2636.
Rife, J. C., et. al., 2000, Miniature Valveless Ultrasonic
Pumps and Mixers. Sensors and Actuators B., 86, p.
135-140.
Ottino, J . M., Wiggins, S., 2004, Introduction: Mixing in
Microfluidics. Phil. Trans. R. Soc. Lond. A. 362, p.
923-935.
Lanceros-Mendez, S., Sencadas, V., Gregorio Filho, R.,
Portuguese patent n. 103318.
Thomas, M., 1999, Ultraviolet and visible spectroscopy.
Analytical Chemistry by Open Learning, p. 2-47.
Frampton, K. D., Minor, K., Martin, S., 2004, Acoustic
streaming in micro-scale cylindrical channels. Applied
Acoustic, Vol. 65, p. 1121-1129.
Brown, L. F., 1992, Ferroelectric Polymers: current and
future ultrasound applications. In IEEE Ultrason.
Symp. Proc. p. 539-550.
Foster, F. S., 2000, A history of medical and biological
imaging with polyvinylidene fluoride (PVDF
transducers. In IEEE Transactions on Ultrasonics,
Ferroelectrics and Frequency Control. Vol. 47 N 6.
Eiras, J . A., 2007, Piezoelectric Materials. Departamento
de Fsica, Universidade Federal de So Carlos, Brasil.
ACOUSTIC THERMOAGITATION BASED ON PIEZOELECTRIC -PVDF POLYMER FILMS - Potential Evaluation
in Lab-on-a-Chip Applications
397
CONTACT LESS RADIO-FREQUENCIES BIOSENSOR FOR
BIOLOGICAL PARAMETERS ANALYSIS
K. Grenier
1
, D. Dubuc
1
, M. Kumemura
2
, H. Toshiyoshi
2
and H. Fujita
2

1
LIMMS/CNRS-IIS,
2
Institue of Industrial Science, The Univ. of Tokyo, 4-4-1 Komaba, Meguro-ku, Tokyo, Japan
[email protected], [email protected]
Keywords: Coplanar waveguides, Permittivity, Biomedical transducers, DNA, Sensor, Microwave, Biology.
Abstract: This paper presents a biosensor, which both takes benefits from RF/microwave detection scheme for
contactless ability and from microtechnologies fabrication potentialities for massive integration of sensors
into lab on chip. The proposed sensor is based on transmission line (which is the basis element of numerous
microwave circuits) associated with a biological micro volume in interaction with electrical microwave
fields. Thanks to the transmission characteristic of the line, we then detect the presence of DNA inside its
nominal liquid environment with a measured sensitivity compatible with RF/microwave measurements
capabilities. The results then demonstrated the potentialities of this approach for the analysis of biological
parameters of in vitro biological substance using microsystem integration facilities.
1 INTRODUCTION
In the past decades, the biological and medical fields
have been strongly modified with the emergence of
various bio and micro-sensors for the
characterization and quantification of biomolecules.
Such sensors permit indeed to reach important
parameters for biologists. Classical detection sensors
(Bashir, 2004) are very effective, but may
sometimes suffer from limitations such as the use of
markers, that may modify the studied biological
substance. They may also imply high cost and
volumic equipments especially with fluorescence, as
well as low sensitivity.

To overcome these drawbacks, the use of
electromagnetic field as transduction in the bio-
sensor may present great advantages. It could
provide a non-invasive detection as it is contact less
in a biological environment, as well as label free.
Some studies have already been performed, notably
for toxicology purpose, with the work of Sebastian,
2004 in order to be able to detect heavy metal
pollutants in tissues. Some others have focused on
the RF field effects on human or livings cells for
dosimetry aims (Gabriel, 1996 for example).

In RF measurements, large volume of liquids are
usually involved during the characterization to
achieve a high sensitivity (Liu, 2008). Thanks to the
integration capabilities of microtechnologies, we
demonstrate simultaneously high sensitivity (for
DNA detection notably) and high compactness,
which is compatible with future massive parallel
analysis.

Consequently, this paper deals with a very compact
RF based bio-sensor, which can serve to detect,
identify and quantify very low volumes (few l) of
biological substances. As an example, DNA solution
has been chosen as detected solution in order to
prove our RF biosensor concept.


Figure 1: Schematic view of the proposed RF biosensor.
The second part of this publication presents the
architecture of our proof of concept demonstrator
and describes the validation of the measurement and
modeling procedure specifically developed for liquid
parameters analysis. The third paragraph deals with
the experimental protocol used and measurements
obtained for DNA detection inside buffer liquid
which highlight the potentialities of our RF/
398

microwave based sensor for biological parameters
analysis as summarized in the conclusion.
2 RF BIOSENSOR DESCRIPTION
2.1 Device Topology
The investigated RF biosensor is presented in Figure
1. It is composed of a RF transmission line with
polymer walls realized on top, in order to delimitate
the liquid biological substance area.

A coplanar type of line (a signal path localized
between two ground planes) has been chosen, as it
permits to elaborate all signal and ground planes on
the same surface in one technological step.

Thanks to the polymer walls realized on top of the
line, it is possible to characterize various biological
substances in an appropriate environment and with
very small volume (few l range). The liquid is
inserted on top of the signal thanks to a micropipette
and well delimited with the polymer walls.

Figure 2: Electrostatic field distribution.
The principle of this biosensor is as follows: the
electromagnetic field propagates along the coplanar
line. The operating microwave frequency range
permits to assure a complete penetration of the field
into the biological substance, then enhancing the
sensitivity. The fluid presence translates into a
modification of the magnitude and phase of the
signal as the fluid presents different dielectric
characteristics than air. This electrical signature is
related to the biological substance properties, and
then permits its characterization/
identification/quantification.
The RF biosensor is consequently totally contact
less, which avoids any contamination of the
chemicals. Labels on the biomolecules are also
useless in this case.
Because of the coplanar topology of the line, the
electrical field is in consequence almost equally
distributed on both side of the line: in the substrate
and in the fluid. This phenomenon can easily be
noticed in Figure 2.

This permits to get a volumic interaction with the
biological substance and furthermore to enhance the
sensitivity of detection.
2.2 Calibration Protocol and
Validation
Test structures have thus been realized. Gold lines
have been deposited on top of a quartz substrate.
Polymer walls have then been performed on top to
elaborate the fluidic pool. The picture of the
obtained demonstrator is indicated in Figure 3.

Figure 3: Picture of the RF biosensor with liquid between
the polymer walls.
In order to calibrate the extraction procedure of the
liquid parameters, we have measured the
performance of the proposed test structure with DI
water. The Figure 4 presents the extracted
permittivity of the DI water (continuous line) and
the predicted one (dots) based on the cole-cole
theory (Buchner, 1999) (J iang, 2004). The predicted
effective permittivity is computed thanks :
to the equation 1, which considers that the
dielectric field is equally distributed in the
water (upper part of the coplanar line) and
the quartz material (lower part of the
coplanar line).

r,eff
=

r,quartz
+
r,Water
2
(1)
the classical cole-cole equation, described
by equation 2, where
0
,

and are
respectively the permittivity at low
frequency, the permittivity at high
frequency and the relaxation time.
CONTACT LESS RADIO-FREQUENCIES BIOSENSOR FOR BIOLOGICAL PARAMETERS ANALYSIS
399

r,Water
=
0
+

0
1+ ( )
2
(2)
The measured relative permittivity is extracted
from the phase of the transmission parameter
measured from 100MHz up to 40GHz.


Figure 4: Measured and modeled relative effective
permittivity of the test structure.
The good agreement (error lower than 15%)
observed between our proposed test structure and the
associated extracted procedure with theoretical data
taken in the literature then validates our approach
over a very broad bandwidth from 100 MHz up to
40GHz.

This validation step then opens the door to the
RF and microwave spectroscopy of biological
substance given electrical parameters as presented in
the next paragraph.
3 EXPERIMENTS
AND DISCUSSIONS
Both biological experimental protocol and results
are presented in this paragraph.
3.1 Biological Experimental Protocol
In order to perform the RF measurements, two
solutions have been prepared and characterized. The
first one is directly ready and commercialized for
characterization: a -DNA solution, which DNA
concentration corresponds to 0,37g/l. As
reference, the buffer solution in which the -DNA is
localized was also fabricated.
Both liquids have then been tested on our RF
biosensor. A micropipette was used to well localize
the liquid in the dedicated area, as shown in Figure
5.

Figure 5: Picture of the RF biosensor under test during
DNA solution incorporation with a micropipette.
This figure also presents the RF test setup: two
microwave coplanar probes connected to an network
analyzer (which operates from 100MHz up to
40GHz) by cables.
We would like to emphasize the fact that our
concept is fully compatible with the use of micro-
channel.
3.2 Discussions
The electrical parameter, which is monitored,
corresponds to the phase of the transmission
parameter of the coplanar line. As already reported
in this publication, this parameter is related to the
permittivity of the biological substance and then to
its intrinsic constitution.

Figure 6: Phase of the transmission parameter vs
frequency for the two biological substances: buffer
solution w/ (line with circle marks) and w/o (line with
square marks) -DNA.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
400

The Figure 6 presents the measured phase of the
transmission parameters of the tested structure for
two cases: (1) when the line is covered with the
buffer only (line with square marks) and (2) when
the line is covered with DNAs inside the buffer (line
with circle marks). As the electrical properties of the
DNA molecule differ from the buffer one, we can
observe on the RF/microwave measurement the
signature of the DNAs loading. We indeed obtain a
phase difference of 6 and 10 at 20GHz and 40GHz
respectively between the two cases, with and
without DNA, as summarized in the table I.
Table 1: Phase measurements summary.
Phase of the transmission
parameter in
20 GHz 40 GHz
Without DNA -97 -153
With DNA -103 -163
Difference 6 10

The observed phase difference, around 7% in our
case, is larger than the minimum detectable phase
shift that we have estimated around 1, but also
sufficiently high to envision microwave circuits with
enhanced sensitivity (7% multiply by the order of
the electrical function) like filter with operation
frequency sensitive to DNA concentration.

We also would like to outline that the use of
higher frequency increases the electrical signature
(the phase shift in our study) of DNA concentration,
and then to highlight one key point of our work,
which targets the integration of microwave (1-
30GHz) and even millimeterwave (30GHz to
110GHz or even higher) circuit for biological
analysis.
4 CONCLUSIONS
This publication presents a proof of concept
demonstrator that emphasizes the potentialities of
RF/microwave detection and quantification of DNA
in its nominal biological environment. We take
benefit from the RF/microwave fields to achieve a
contact less detection scheme opening the door to in
vitro analysis. We have first calibrated/validated
our RF sensor, and associated parameters extraction
procedure, with known liquid (D.I. water). We have
then detected the presence of DNA inside on host
liquid with a sensitivity, which is compatible with
measurement capabilities and future circuit design of
dedicated function.
Besides this electrical characterization, we
would like to outline that the fabrication of our
sensor is fully compatible with microtechnologies
and then inherit of their great interest, notably the
possible integration into lab-on-chip.
Furthermore, we think that it will be possible to have
access to the intrinsic DNA electrical parameters and
density, opening the door to the
quantification/discrimination of DNA inside its
biological environment.
REFERENCES
Bashir, R., 2004. Advanced Drug Delivery Reviews, 56
Buchner, R., Barthel, J ., Stauber, J., 1999. Chemical
Physics Letters 306, 57-63
Gabriel, S., Lau, R. W., Gabriel, C., 1996. Phys. Med.
Biol. 41, 2251-2269
J iang, J. H., Wu, D. L., 2004. Atmos. Sci. Let. 5: 146-151
Khan, U., & al., 2007. IEEE Int. Microwave Symp., June
Liu, C., Pu, Y., 2008. IEEE Microwave and Wireless
Components Letters, Vol. 18, n4, April
Sebastian, 2004. IEEE European Microwave Conference


CONTACT LESS RADIO-FREQUENCIES BIOSENSOR FOR BIOLOGICAL PARAMETERS ANALYSIS
401
WEARABLE TECHNOLOGY
Development of Polypyrrole Textile Electrodes for Electromyography
S. Rodrigues, R. Miguel, J . Lucas, C. Gaiolas
Textile Science and Technology Department, Textile and Paper Materials R&D Unit, University of Beira Interior
Rua Marqus dvila e Bolama, Covilh, Portugal
[email protected], [email protected], [email protected], [email protected]
P. Arajo
Computer Science Department, University of Beira Interior, Covilh, Portugal
[email protected]
N. Reis
Biosurfit S.A., Lisbon, Portugal
[email protected]
Keywords: Wearable Technologies, Electromyography, Smart Structures Monitoring, Textile Electrodes, Polypyrrole
Conducting Polymer.
Abstract: Following the work already done, in particular the textile arm" by the authors, incorporating metallic
filaments in the fabric during the weaving process with the aim of capturing surface electromyography
signals, we concluded that there was a need to maximise the area of contact with skin to improve the
obtained electrodes signals. The main objective of this work is the development of electrodes embedded into
textile materials able to capture electromyography signals, using the polypyrrole conducting polymer, both
in coating of electrodes directly into the textile fabric and preparing a conductor yarn with this polymer to
obtain a knit structure. This work allowed developing wearable technology textile structures for muscle
activity monitoring, through the measurement of electromyography signals, keeping simultaneously comfort
and easy-care textile characteristics. The signals obtained with this approach had a good definition
compared with commercial electrodes and with the great advantage of this procedure can be applied in
industrial production.
1 INTRODUCTION
Through times, many have been the fields of science
and technology that progressed in separate ways.
During the last years, a considerable convergence
appeared among some of these fields, with
surprising results. The smart technology for
materials and structures is one of these results
(Lucas, 2007).
1.1 Wearable Technology
The smart textiles concept was firstly defined in
J apan in 1989 (Langenhove, 2007). With the
discovery of shape memory materials in 19600 and
intelligent polymeric gels in the 1970 decade, it was
accepted the birth of really smart materials. These
have been introduced in textiles only in the 90s
(Langenhove, et al., 2006). Smart textiles can be
described as textiles that can feel stimuli from
environment, react and adapt themselves to them by
integrating functionalities in the textile structure
(Langenhove, et al., 2006).
The stimulus, as well as the response may be
electrical, thermal, chemical, magnetic or of other
kind. The degree of intelligence can be divided into
three subgroups:
Passive smart textiles that only feel, being
sensors;
Active smart textiles that perceive external
stimuli and react to them, besides being
sensors, they act also as actuators;
402

Truly smart textiles, being one step ahead
and able to adapt its behavior to the
circumstances (Zhang, 2001).
It would be wonderful to have clothes like our
skin, which is a smart material layer. Our skin has
sensors to detect pressure, pain, heat, etc. and,
together with the brain, can work in a smart manner
with environmental stimuli, generating large
amounts of perspiration to cool our body when in a
hot surrounding and stimulating blood circulation
when it is cool. It changes colour when exposed to a
high sun radiation level, in order to protect the body,
It is breathable and allows moisture penetration,
avoids the entry of undesired species, it can be
damaged, repaired and regenerate itself. Thus, to
study and develop a smart material like our skin is a
big challenge (Boczkowska, Leonowics, 2006).
To Bonato (2005), when looking for a better way
of life, more and more solutions have been
investigated to contribute to a comfort improvement
and well-being. The research field is going on and
several research groups have already demonstrated a
relevant application and the potential impact of this
technology is remarkable in clinic practices
concerning physical and rehabilitation treatments.
Presently, the main use of clinic techniques resides
on methods that embed devices into smart clothes on
in the patients body. The wearable technology has
a large potential to redirect these techniques into
functioning during the normal day life (Bonato,
2005).
The development of small size sensors that can
be connected to the body or can be a part of clothes,
embed elements into cloth fabrics, open a large
number of possibilities to monitor patients during
long periods of time. This is of particular relevance
to the practice of physical and rehabilitation
treatment. The wearable technology allows to
clinical staff the treatment of data, acquired at home
or by direct observations relating to the impact of
clinic interventions in mobility, thus reaching a
higher independence level and a better quality of
(Bonato, 2005).
1.2 Conductive Polymers
The firs conductive polymer was discovered
accidentally by Hideki Shirakawa in 1971
(Shirakawa, Ikeda, 1971) during a reaction with
acetylene. From this, it began research to understand
the charge transfer mechanisms in these (Neto,
2007).
The key to get a conductive polymer is the
presence of conjugate double bonds in the polymer
main chain. During conjugation, links between
carbon atoms are alternately single and double
bonds. Each double bond has a localized bond that
gives a strong attraction ( bond) and a weaker one
( bond). The electrical conductivity in polymers is
an effect due to electron displacement in alternate
sequence of single and double bonds (Lucas, 2007),
as seen in Figure 1.


Figure 1: Illustration of electrons displacement in
alternate sequence electrical conduction (Lucas, 2007).
Conjugation is not enough to assure conduction
in a polymer. The doping function is now of an
agent seen as polymer charge carrier, as extra or
missing electrons (hole). One of these holes can be
filled by an electron jumping from a neighbor
position, creating a new hole and, thus, successively,
allowing a moving charge through a long (Lucas,
2007).
Polypyrrole is one of the most studied
conductive polymers due to its high conductivity,
good environmental stability, harmless and easy to
synthesize. Consequently, polypyrrole has
advantages in real applications in microelectronics,
informatics industry and biomedical information.
However, polypyrrole is not soluble in water and in
organic solvents, due to the strong intermolecular
and intramolecular interaction and crossed polymer
chains. For this reason, to overcome the non
solubility issue of polypyrrole, many research works
have been done (Pires, 2007).
Recently, soluble polypyrrole was synthesized
with large dope agents, such as dodecyl-benzene-
sulphonate acid (DBSA) and naphtale-sulphonate
(Lim, et al., 2005).
According to DallAcqua, L. et al. (2006), the
affinity of different types of fibers, yarns and fabrics
with doped polymers allows production of textile
composites having improved electrical properties.
Following the study of Lim., H. K. et al. (2005),
a new soluble and conductive polypyrrole, doped
with DBSA and combined with a polymeric agent,
the poly (ethylene glycol) (PEG).
WEARABLE TECHNOLOGY - Development of Polypyrrole Textile Electrodes for Electromyography
403

1.3 Electromyography Signal
Acquisition
According to J . Basmajian e C. De Luca (1985)
electromyography measures the electric activity
resulting from skeletal muscle activation, these
being responsible by the support and movement of
skeleton.
According to the Hospital Sant Pere Claver
Electromyography Unit at Barcelona,
electromyography is an electrophysiological study of
neuromuscular system; it is not a complementary
proof but an extension of a neurological study. This
basis of all electrophysiological research is the
registration of excitable cells potentials.
Electromyography deals with registering of such
potentials, self acting in the muscle and
electroneurography of evoked potentials, both on the
muscle and on nerve branches, by generally
electrical stimulation of nerves having anatomic
connection with the registering area. The electrical
properties of excitable nervous and muscle fibers
derive from a semi permeable membrane that
separates intracellular and extracellular fluids having
different ionic concentration origination a
transmembrane potential.
When measuring electromyography signals
(EMGs) electrodes can be used placed inside the
muscle (deep electromyography) or electrodes
placed on the skin (surface electromyography)
(Sousa et al., 2007).
According to the type of acquisition, electrodes
may be invasive or not. When an electrode becomes
into contact with the body fluid is called invasive,
which can be of the yarn or needle type and presents
characteristics such as low electrical impedance, this
is, it does not need of conductive gel, it acquires
larger amplitudes and its power spectrum goes up to
10KHz (Ricciotti, 2006). They are most indicated
for clinical analysis, where it is possible to detect the
acting potential of a moving unit, as well as
exploring an isolated of a deep muscle. The invasive
electrodes have several drawbacks, being among
them excellent sterilization, the danger of breakage
of yarns inside the muscle and, after all, the patients
(Ricciotti, 2006).
According to J acquelin Perry (1998), from the
28 main muscles that control each lower member,
most of them are surface muscles, being possible to
monitor their activity using surface
electromyography. During the last years, surface
electromyography (sEMG) has become very
frequent when studying and treating several muscle
dysfunctions, due to the fact that it is a secure
technique, easy to use and non-invasive. Most of
knowledge fields use EMG signals as starting point
to several human muscle system function and
dysfunction analysis. Recent studies relate the
muscle fatigue process to a displacement of signal
spectral contents. Other investigations emphasized
the importance of these signals in neuromuscular
reeducation of patients having hyperactive muscles
by electromyography biofeedback (Sartori et al.,
2004).
There several ways to configure EMG signal
acquisition through electrodes, the most common
being those requiring one (monopolar) and two
(bipolar) capturing points. The monopolar
configuration deals with the potential difference
between two points (a capturing and a reference
point). The bipolar configuration is characterized by
the potential difference between to acquisition
points, measured respecting to a third point
(reference). Electrodes are normally made of silver
coated with silver chloride (Ag-AgCl), since it is a
non polarizable noble metal.
The surface electrodes allow capturing signals
representing the activity of a muscle as a whole or a
group of muscles in a global manner and can be
subdivided into passive and active electrodes
(Barros, 2005 in: Ricciotti, 2006). The passive
electrode is constituted by a metal disk (in general,
Ag-AgCl), that must be placed over the skin (Figure
2). To decrease the contact impedance between
electrode and skin, it may be required a tricotomy
(pile removal from the region where the electrode
will be placed), the use of conductive abrasives and
gel or paste (Ricciotti, 2006).


a) Adhesive for electrode fixation;
b) Reusable electrodes;
c) Reusable electrodes detail;
d) Reusable electrodes rear view;
e) Reusable electrodes front view.
Figure 2: Examples of commercial passive electrodes
(Ricciotti, 2006).
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
404

The active electrodes have a magnifying circuit
encapsulated near the electrode acquisition place.
Generally, these are bipolar electrodes, this meaning
that the amplifier used is of differential type. The
active electrodes, since they are made of a
differential amplifier, require a reference electrode
placed ion an inactive region to avoid measurement
interference (Basmajian, De Luca, 1985). Such
electrodes are also called dry electrodes, since
generally do not need to use conducting gel, paste,
abrasives or pile (Ricciotti, 2006).
The textile support represents the first
functionality level of a wearable technology
architectures system, since it comprises the infra-
structure to embed the basic functions as energy and
data transmission or the sensorial (Rodrigues et al.,
2008).
Following work already done, namely the
textile arm (Rodrigues et al., 2008), where
conductive metallic yarns are incorporated into the
fabric during weaving with the aim to acquire
surface electromyography signals, it was observed
the need to optimise the contact region with the skin
to improve electrode signal acquisition. Aiming to
contribute to problem solving, it is studied now the
applicability of polypyrrole as a conductive polymer.
2 MATERIALS AND METHODS
2.1 Printed Polypyrrole Electrodes
2.1.1 Textile Substrate
As textile substrate the dobby fabric developed in
the textile arm research work was used (Rodrigues
et al., 2008), having a compound weave based on
heavy twill 3 (Figure 3).
The main fabric characteristics are:
Cover Factor: 95.3%;
Weft density (DT): 19 yarns/cm;
Warp density (DB): 24 yarns/cm;
Warp yarn 2/36Nm (45WO/53PES/2EA)
Weft yarn 2/50Nm (45WO/55PES)
2.1.2 Reagent Preparation
The pyrrole solution was prepared with: 9% of
DBSA doping agent that reduces inter and intra
interactions in polypyrrole chains, improving pyrrole
solubility in organic solvents; 9% of PEG
polymeric additive and 82% distilled water.
These reagents are mixed at room temperature
with slow stirring (10-15minutes). After, pyrrole is
added on a 0.11ml/g ration of the prepared solution.
After dissolving completely, the resulting solution
was isolated from air. Due to the need of a viscous
solution, for coating, a printing paste with only a
thickener (6% Sodium Alginate) in water was
prepared.


Figure 3: Compound weave module based on 2/1 twill 3
with metallic yarn (Rodrigues et al., 2008).
Since the pyrrole solution is quite unstable, the
oxidizing solution must be prepared before, which is
more stable. The oxidizing solution was prepared
with 20% of APS oxidizing agent and 80%
distilled water.
2.1.3 Electrode Coating Methods
Following, the sodium alginate thickener was added
to the pyrrole solution on a 1/1 proportion. After a
complete mixing, the resulting paste was isolated
from air and kept at low temperature.
The fabric was pre-scoured in a bath containing
1g/l of Invadine LU (CIBA) wetting agent, at 40C
during 20 minutes.
For electrode coating the respective area was
delimited and, to achieve a good electrode area
definition, the liquid oxidising solution was applied
first in excess followed by the pyrrole solution
thickened with sodium alginate (Figure 4). The
coated fabric was kept at room temperature for 15 to
30 minutes and then placed in a woven during 2
WEARABLE TECHNOLOGY - Development of Polypyrrole Textile Electrodes for Electromyography
405


Figure 4: Electrode coating device.
minutes at 100C.
To take out polypyrrole in excess from fabric
surface and, thus, to improve washing fastness, a
1g/l Reoklen detergent scouring at 50C and during
20 minutes was carried out.
2.2 Embroidered Polypyrrole
Electrodes
2.2.1 Reagent Preparation
A pyrrole solution was prepared having a ratio of
9ml of pyrrole to 100ml of DBSA and PEG in
distilled water, in the same way as described before
in 2.1.2. The oxidising solution used was also the
same as that employed in electrode coating.
2.2.2 Conductive Yarn Preparation
The chemical polymerization was achieved by
making the yarn go through the pyrrole solution and
after, through the oxidising solution, leaving
exposed at room temperature during 15 minutes and
after in a woven at 100C during 4 hours. To
eliminate excess of polypyrrole at the knit surface
and improve washing fastness, a detergent scouring
was performed.
2.2.3 Textile Substrate
As textile substrate a jacquard knit was used, being
the polypyrrole conductive yarn inserted into the
knit structure as shown in Figure 5. For knitting, a 3
ply 100% cotton yarn of 2/32Nm count was used. At
the electrode region, a polypyrrole treated 3 ply
100% wool yarn of 2/36Nm count was employed.


Figure 5: J acquard knit structure showing electrode
positioning.

In Figures 6 and 7 knit images are shown, being
emphasized conductive yarn and electrodes.


Figure 6: Knit right face with electrodes.

Figure 7: Knit back face with conductive yarns.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
406

3 RESULTS AND DISCUSSION
3.1 Electromyography Signal
Acquisition
To acquire surface electromyography signals
(EMG), an experimental system was developed
made by a lap top computer, a differential amplifier
and software for measuring physiological signals
(Elektor Electronics, 2007), as shown in Figure 8.


Figure 8: Experimental system for sEMG signal
acquisition.
To acquire surface electromyography signals
(aEMG), electrodes were placed on the arm and,
when hand squeezing a ball, the stimulated muscle
contracted. At the beginning, commercial electrodes
were used, these being placed on the arm to test the
equipment (Figure 9).


Figure 9: sEMG captured signal using commercial
sensors.
Figure 10 shows the signal captured in the case
of textile arm produced with metallic filament and
embroidered electrodes using conductive yarn
(Rodrigues et al., 2008). In fact, the signal obtained
has a good definition and shows that this method is
adequate and can be used to measure the muscle

Figure 10: sEMG captured signal using hand embroidered
electrodes with metallic yarns (Rodrigues et al., 2008).
activity, for example, in rehabilitation.
However, this method doesnt show a good
industrial feasibility, mainly because electrodes have
to be hand embroidered. For this reason, it became
necessary to develop embedded electrodes into
textile materials able to capture electromyography
signals and, at the same time, maximise the area of
contact with skin to improve the acquired signals by
electrodes. These signals captured by the textile
arm made with printed polypyrrole electrodes are
shown in figure 11. Also, the obtained signals with
the electrodes of the knit structure made with
polypyrrole treated yarns are shown in figure 12.


Figure 11: sEMG captured signal using polypyrrole coated
electrodes.

Figure 12: sEMG captured signal using polypyrrole
knitted electrodes.
WEARABLE TECHNOLOGY - Development of Polypyrrole Textile Electrodes for Electromyography
407

This data shows that we can obtain good definition
signals with the advantage that may have industrial
applicability.
4 CONCLUSIONS
In this research work electrodes for wearable
technology were developed, enabling simultaneously
a high level of integration, to keep textile properties,
the conductive polymer properties and the signal
capturing through substrate functionality. Analysing
the clarity of acquired signals it can be concluded
that they are similar, the solutions found presenting,
compared to commercial sensors, the same
advantages than the hand embroidered electrodes in
the textile arm (Rodrigues et al., 2008), this is,
textile comfort, mobility, cleanness and
maintenance, etc. Besides these, they also exhibit
advantages at industrial feasibility level.
The acquired results are satisfactory, considering
the possible advantages of these may present sport
and clinical rehabilitation level and even considering
research work on neuromuscular development. Thus,
this work presents the development of textile
solutions that, being improved, both from the textile
point of view, through incorporating electronic
devices, and counting on the informatics
contribution, gather potentialities to perform in
future an important role in peoples quality of life.
ACKNOWLEDGEMENTS
The Authors wish to tank to Eduardo J orge de J esus,
technician at the Knitting Laboratory of Textile
Science and Technology Department of University
of Beira Interior, for his collaboration concerning
weaving knit substrates.
REFERENCES
Basmajian, J ., De Luca,C., 1985, Muscles Alive: Their
functions revealed by electromyography, 5th ed.
Baltimore: Williams & Wilkins.
Boczkowska, A., Leonowicz, M., 2006, Intelligent
Materials for Intelligent Textiles, Fibres & Textiles in
Eastern Europe J anuary / December 2006, Vol. 14,
No. 5 (59).
Bonato, P., 2005, Advances in wearable tecnhnology and
applications in physical medicine and rehabilitation,
J ournal of NeuroEngineering and Rehabilitation, 25
Fevereiro.
DallAcqua, L., Tonin, C., Varesano, A. Cametti, M.,
Porzio,W., e Catellani, M., 2006, Vapour phase
polimerisation of pirrolrrole on cellulose-based textile
substrates, Syntthetic Metals.
Langenhove, L.Van, 2007, Smart textiles for medicine and
healthcare, Woodhead Publishing Limited, England.
Langenhove, L. Van, Puers, R., Matthys, D., 2006,
Intelligent Textiles for Medical Applications: An
Overview, in: Medical Textiles and Biomaterials for
Healthcare, edited by: Anand, S.C., Kennedy, J .F.,
Miraftab, M., Rajendran, S., Woodhead Publishing in
Textiles, 2006.
Lim, H.K., Lee.S.O.,Song K. J ., Kim, S.J ., Kim, K.H.,
2005, Synthesis and Properties of soluble
polypirrolrrole Doped whith Dodecylbenesulfonate
and combined with Polymeric Additive Poly(ethylene
glycol), J ournal of applied Polymer Science.
Lucas, J ., 2007, Txteis de Alta Tecnologia, Universidade
da Beira Interior, Departamento de Cincia e
Tecnologia Txteis, Covilh.
Neto, P., 2007, Foto-Gerao de Pares de Plarons em
Cadeias Acopladas de Poliacetileno, Dissertao de
Mestrado, Universidade de Braslia, Instituto de
Fsica, 26 J aneiro2007.
Perry, J ., 1998 The contribution of dynamic
electromyography to gait analysis, in Monograph 002:
Gait analysis in the science of rehabilitation, J .
DeLisa, Ed. Washington, DC: Department of Veterans
Affairs.
Pires, M.A., 2007, Obteno de propriedades antiestticas
numa malha de poliamida por polimerizao in-situ
de polipirrol, Dissertao apresentada para obteno
de grau de Mestre em Engenharia Txtil, Universidade
da Beira Interior, Departamento de Cincia e
Tecnologia Txteis, Covilh.
Revista Elektor Electronics, 2007, n265, J aneiro.
Ricciotti, A., 2006, Utilizao de Wavelets no
processamento de Sinais EMG, Dissertao para o
ttulo de Mestre em Cincias, Universidade de
Uberlndia, Faculdade de Engenharia Eltrica, Ps-
graduao em Engenharia Eltrica, Uberlndia.
Rodrigues, S., Miguel, R., Reis, N., Arajo, P., Lucas, J .,
2008, Wearable technology for muscle activity
monitoring, CONTROLO2008 - 8th Portuguese
Conference on Automatic Control, UTAD , Vila Real,
Portugal J uly 21-23, 2008.
Santori, G. F., Rocha, A. F., Gonalves, C,Veneziano, W.
H., 2004, Programa para Anlise de Sinais
Eletromiogrficos, IX Congresso Brasileiro de
Informtica em Sade Ribeiro Preto, Brasil, 07-10 de
Novembro de 2004.
Shirakawa, S., Ikeda,S., 1971, S. Polymer.
Sousa, D., Tavares, J ., Correia, M., Mendes, E., 2007,
Anlise Clnica da Marcha Exemplo de Aplicao em
Laboratrio de Movimento, 2 Encontro Nacional de
Biomecnica, vora, 8 e 9 Fevereiro 2007.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
408

Unidad de Electromiografia del Hospital Sant Pere Claver,
Eletromiografia EMG EMGEP Eletromigrafo,
Barcelona, in: http://www.biolinkbr.com/nleletro-
miografia/index.html
Zhang, X., To, X., Zhang,, 2001, Smart textiles:Passive
Smart, J unho 2001, in: Langenhove, L.Van(2007),
Smart textiles for medicine and healthcare, Woodhead
Publishing Limited, England.
WEARABLE TECHNOLOGY - Development of Polypyrrole Textile Electrodes for Electromyography
409
A SCALABLE AND OPEN SOURCE LINEAR POSITIONING
SYSTEMCONTROLLER
M. C. Medeiros, A. J. A. Fernandes, C. A. Teixeira and M. Graca Ruano
Centre for Intelligent Systems, University of Algarve, Gambelas, Faro, Portugal
[email protected], [email protected], [email protected], [email protected]
Keywords:
Automatic Positioning, Open Source Software, Scalable Systems, Ultrasound Measurements.
Abstract:
This paper is on the implementation of a dual axis positioning system controller. The system was designed
to be used for space-dependent ultrasound signal acquisition problems, such as pressure eld mapping. The
work developed can be grouped in two main subjects: hardware and software. Each axis includes one stepper
motor connected to a driver circuit, which is then connected to a processing unit. The graphical user interface
is simple and clear for the user. The system resolution was computed as 127m with an accuracy of 2.44m.
Although the target application is ultrasound signal acquisition, the controller can be applied to other devices
that has up to four stepper motors. The application was developed as an open source software, thus it can be
used or changed to t different purposes.
1 INTRODUCTION
In the scientic eld precision is everything. A
scientist cannot conduct a responsible scientic expe-
riment if the precision of the process is not controlled.
Laboratory experiments involving ultrasound deals
normally with space-dependent parameters, such as
time-of-ight or pressure. Manual adjustment can in-
duce human errors, that can be eliminated by using
a reliable and automatic positioning system. In the
scientic community, the open-source software con-
cept is often referred. It refers to computer appli-
cations whose source code is available and under an
open-source license to study, change or tune the soft-
ware, and to redistribute in the modied or unmodi-
ed form.
This paper presents the necessary hardware and
software to control a linear positioning system. The
main application of this system is the acquisition of
space-dependent ultrasound signals. The developed
controller is an evolution to the systemsupplied by the
manufacturer. The proposed controller is supported
in Unix/Linux operating systems (OSs) and the con-
nection with the personal computer is accomplished
through an USB interface, instead of the nowadays
rare parallel port (Arrick Robotics, nd).
Figure 1: Hardware interconnection.
2 HARDWARE
A schematic representation of the overall hardware
used is presented in Figure 1.
The main part of this project is the positioning
table (Arrick robotics, model XY-18). This table
has two belt-driven axis with a moving range of
410
464.8 mm. Each axis includes a stepper motor
(Telcomotion, 4H5618C2906), which is responsible
for the automatic motion of a top plate (mobile point).
Each stepper motor has a resolution of 0.9 degrees per
step with an accuracy +/- 5%. The top plate is where
the load (transducers, hydrophones, or a third axis)
is to be attached. Each axis has two limit-switches,
which prevents from equipment damage and are also
reference points (Arrick Robotics, nd).
2.1 Power Stage
Between the Processing Unit (PU) and each mo-
tor there is a driver circuit, responsible to adapt the
current sent by the PU (in the range of mA) to a
current necessary (in the range of A) to move the
motor with high torque. To mention that all stepper
motors work in half-step mode (Yeadon and Yeadon,
2001).
2.2 Processing Unit
For the Processing Unit (PU) we used three microcon-
trollers PIC 16F84A (Microchip Technology, 2001)
from Microchip, to control both stepper motors.
One of the microcontrollers is called the PIC
Master (PICM), which receives orders from the PC.
The information from the PC to the PU is sent in a
group of 10 Bytes. The information transmitted is:
target axis;
direction;
speed;
number of steps;
synchronization bytes.
After data processing the PICM starts the commu-
nication with the PIC Slaves (PICSs), that are
connected to the target stepper motors through the
driver circuit. Whenever a certain motor has to move
a step, PICM signals the corresponding PICS, which
only has to change the output pins state in order to
give a step.
If a switch (at the limits) is pressed, the PICS
that detected it, signals the PICM. When the move-
ment stops, the PICM sends a feedback byte to the
PC, on which it informs if the operation was or not
successful.
Whenever both stepper motors have to move the
same distance (same number of steps), they move at
the same time. Otherwise, only one motor moves at a
time.
The PU is designed to allow to control up to four
axis. The PICM has some pins that are not connected
to anything and can be used to add a new axis. If
the new axis dont have any special requeriment, the
rmware of the PICSs will all be the same. This char-
acteristics makes our system scalable.
2.3 PC/PU interface
The interface between the PC and the PU is done
through a USB to parallel FIFO module (FTDI
UM245R) (FTDI Ltd, 2005), integrated in the PU.
The UM245R is a module with a lot of poten-
tial, since that its not needed to develop USB-specic
rmware to handle the USB protocol. Besides this,
its possible to communicate through a Synchronous
Bit Bang Mode that allows to keep a synchronous
communication.
3 SOFTWARE
The software for this project was developed in an
open source environment, so it can be used in other
applications that involves stepper motors.
In Unix/Linux environments as in others OS, the
hardware is controlled by specic programs called de-
vice drivers. In particular case of Linux, this pro-
grams can be loaded to kernel at run time, and they are
called kernel modules. With this modules its possi-
ble in user mode, to communicate with the hardware
without direct access, hiding all the low level func-
tions.
One of the requirements was to develop the soft-
ware as a Python module, to be integrated in other
Python programs (Teixeira et al., 2006). Since a func-
tional userspace library in C language and at the mo-
ment there isnt any Python module for FTDI modu-
les for Linux, it was necessary to create a C Python
Extension (van Rossum, 2008). This enabled the in-
tegration of the necessary C code to control the linear
positioning table.
Figure 2 presents the software structure. The main
program is called table.py and implements a state ma-
chine that handles computer and user events. The
interfaceGUI.py le is responsible to create all the
graphical interface and the methods to access to all
the values that the user can change or congure. A
Python module called pylab included in matplotlib
library (Matplotlib Project, nd) was used to display
the motors movement. The options.py le includes
classes and methods for all the stepper motors op-
tions.
The session is saved in two data les, in order to
keep the data for next sessions. The options le has
the characteristics of each motor, and the settings le
A SCALABLE AND OPEN SOURCE LINEAR POSITIONING SYSTEM CONTROLLER
411
Figure 2: Software structure.
has the present positions of the motors after the move-
ment.
As mentioned before, in order to interface the C
libraries, a le named ftdipymodule.c was created as
an CPython Extension. This le is a wrapper between
C and Python languages.
4 GRAPHICAL USER
INTERFACE
The user interface was developed in GTK+, because
its portable, and his use with Python is easy with the
pygtk module (GNOME Project and PyGTK Team,
2006). The application front-end is presented in Fi-
gure 3.
We can dene the amount of movement for each
motor or simply scroll the bar to the position, we can
also control the movement with the start, stop and
home buttons. In the messages box, it can be seen
a small report of what is happening, and check the
device status (connected or disconnected).
From the Options Tab (see Figure 4), its possible
to choose the metric units, move type, the step mode,
number of motors, speed, acceleration and the axis
size.
The output can be seen in Output Tab (see Figure
5), where its represented the table (X and Y axis) on
the left side, and a vertical bar (Z axis) on the right
side. When the top plate is moving the output is ge-
nerated according to the amount of movement.
Figure 3: Control tab.
Figure 4: Options tab.
Figure 5: Preview tab.
5 RESULTS & DISCUSSION
In order to determine how precise the system is, travel
distances for different number of steps were assessed
in order to verify how the system behaves. Six diffe-
rent number of steps were tested: 100, 200, 400,
600, 800 and 1000. For each one, ten measurements
(N = 10) were taken using a caliper ruler (accuracy
of 20 m). The ratio between a measure (X) and the
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
412
number of steps (n), corresponds to the length of a
single step (X
N
). So, the mean (X
N
) of N values of
X
N
, is assumed to be a normalized mean and can be
computed by Equation 1.
X
N
=

N
k=1
X
N
k
N
(1)
After the X
N
values of each test have been deter-
mined, their Standard Deviation () can be computed
by Equations 2.
=

1
N1
N

k=1
(X
N
k
X
N
) (2)
The Variance (
2
) can easily be computed as the
positive square-root of .
Table 1: Results.
n (steps) X
N
(m) (m)
2
(m)
100 125.94 1.22 1.49
200 126.38 0.64 0.41
400 126.64 0.49 0.24
600 126.52 0.43 0.19
800 126.84 0.36 0.13
1000 126.76 0.23 0.05
From the values in Table 1, it can be seen that
presents a stable value on all the tests, with an accep-
table difference between them. Also, as the number
of steps increases, the values of and
2
decreases.
This happens becasuse of the initial inertia and im-
precisions of the system. The mechanical parts of the
table applies some forces to the system (belt has ten-
sion, top-plate has weight and the steel shafts have
friction) that will result in the resistance to a change
on its state of motion.
According to the data presented in Table 1, the re-
solution of each axis of the system can be dened has
127 m. Due to the real application of the system,
small travel distances (n < 100 steps) will be often
used and the worst case (n = 100 steps) is used to de-
termine the accuracy value. For a 95 % condence in-
terval (Normal distribuition) the accuracy is 2.44 m
(2 for n = 100 steps).
6 CONCLUSIONS & FUTURE
WORK
The system developed works as a linear positioning
system using step motors. It was created a hardware
driver in a modular way, which enables the addition
of a new axis easily. The software is open-source and
can be used and modied according to the user needs.
The motor movements have an acceptable resolution
and accuracy. The developed controller is an actu-
alized version of the system supplied with the linear
table. In the future this application can be used for ul-
trasound signal processing, and new improves can be
made, like operation in full-step mode and adding up
to four axis.
REFERENCES
Arrick Robotics (n.d.). X and XY Linear Positio-
ning Tables. Retrieved May 6, 2008, from
http://www.arrickrobotics.com.
FTDI Ltd (2005). UM245R USB-Parallel FIFO
Development Module. Retrieved from
http://www.ftdichip.com.
GNOME Project and PyGTK Team (2006). PyGTK:
GTK+ for Python. Retrieved May 12, 2008, from
http://www.pygtk.org/.
Matplotlib Project (n.d.). Matplotlib / pylab - matlab
style python plotting. Retrieved May 12, 2008, from
http://matplotlib.sourceforge.net.
Microchip Technology (2001). PIC16F84AData Sheet. Re-
trieved from http://www.microchip.com.
Teixeira, C. A., Ruano, M. G., Ruano, A. E., and Pereira,
W. C. A. (2006). Open source data sensing software
for ultrasonic non-invasive temperature estimation. In
5
o
Congresso Ibero-Americano de Sensores - Ibersen-
sor 2006. Montevideo, 27-29 September. Uruguay.
van Rossum, G. (2008). Extending and Embedding the
Python Interpreter. Retrieved May 12, 2008, from
http://www.python.org.
Yeadon, W. H. and Yeadon, A. W. (2001). Handbook of
small electric motors. McGraw-Hill Professional.
A SCALABLE AND OPEN SOURCE LINEAR POSITIONING SYSTEM CONTROLLER
413
DESIGN OF A BIO-INSPIRED WEARABLE EXOSKELETON
FOR APPLICATIONS IN ROBOTICS
Michele Folgheraiter, Bertold Bongardt, Jan Albiez and Frank Kirchner
German Research Center for Articial Intelligence DFKI Bremen
Robotics Lab Robert-Hooke-Strae 5D-28359 Bremen, Germany
{Michele.folgheraiter, bertold.bongardt, jan.albiez, frank.kirchner}@dfki.de
Keywords:
Haptic Interface, Bio-Inspired Device, Biorobotics, Exoskeleton.
Abstract:
In this paper we explain the methodology we adopted to design the kinematics structure of a multi-contact
points haptic interface. We based our concept on the analysis of the human arm anatomy and kinematics with
the intend to synthesize a system that will be able to interface with the human limb in a very natural way. We
proposed a simplied kinematic model of the human arm using a notation coming from the robotics eld. To
nd out the best kinematics architecture we employed real movement data, measured from a human subject,
and integrated them with the kinematic model of the exoskeleton , this allow us to test the system before its
construction and to formalize specic requirements. We also implemented and tested a rst passive version of
the shoulder joint.
1 INTRODUCTION
In this paper a bio-inspired design approach to syn-
thesize the kinematics structure of a haptic interface is
introduced. The system presented is mainly intended
for applications in the eld of the tele-robotics sys-
tems; nevertheless can also be effectively employed
as a sophisticated haptic interface during the interac-
tion with a virtual environment or during a training or
rehabilitation session.
In interaction with a human an exoskeleton can
provide two main functionalities: First, it can be used
as an assistive device, e.g in physiotherapy an ex-
oskeleton can be adopted for movement enhancement
(Gupta et al., 2008) (Carignan et al., 2005), in other
scenarios it can be used for performance augmenta-
tion (Dollar and Herr, 2008).
Second, an exoskeleton can be used as an input
device, enabling a human operator to manipulate ei-
ther a virtual or a real target system. The latter use-
case which is called teleoperation surrenders in any
situation where work has to be done in regions, in
which it is unfavorable or even impossible to work as
a human. Possible applications for teleoperation oc-
cur in telesurgery (Bar-Cohen et al., 2001), aerospace
(Schiele and Visentin, 2003) and underwater (Kwon
et al., 2000).
The primary aim of an intuitive teleoperation is to
allow a human operator to see and feel the remote en-
vironment, as a secondary aim he further should be
able to attribute himself with the target robot (IJssel-
steijn et al., 2006). Ageneral overviewabout the topic
of teleoperation is given in (Hokayem and Spong,
2006).
In case of teleoperation an exoskeleton acts as an
human-robot haptic interface which is nothing but
a bidirectional mechanical transducer(Hayward and
Astley, 1996). Ciriteria for the quality of haptic feed-
back are given in (Hayward and Astley, 1996), com-
parative studies are are presented in (Grifn et al.,
2005) or (Yu, 2003).
The hardware of existing constructions of ex-
oskeletons differ in their degree of activity: On the
one hand pure passive devices were developed by
(Song et al., 2005) and (Chen et al., 2007, ZJUESA).
On the other hand empowering exoskeletons were
built up, see (Dollar and Herr, 2008). Between
these extrema one nds exoskeletons acting as force-
reecting controlling devices. These can be further
categorized into solutions which are xed to an exter-
nal basis (Mistry et al., 2005) and (Perry et al., 2007)
and those remaining wearable. The latter is described
by the study in (Kim et al., 2001).
In the next following section we introduce the
model used during the design process, we also
presents some results from the study we conducted
on a real human subject in order to analyse the impor-
tance of the Clavicle-Scapula articulation during the
414
shoulder movement. Section 3 deals with the kine-
matic model of the exoskeleton, in particular here we
reported only the system that is supposed to be cou-
pled with the shoulder of the user. In section 4 we pro-
pose a possible design for the exoskeleton and present
some preliminary results. Finally section 5 draws out
the conclusions and the future developments.
2 HUMAN ARMSTUDY
AND KINEMATIC MODEL
The Human Arm represents one of the most advanced
manipulation system we can nd in nature. It is the
product of an evolutionary process lasted 3.7 billion
years (origin of life on the Earth). Its kinematics is
dened by the conguration of different bones and
articulations, these elements represent the structural
components of the limb. Grossly we can divide the
arm in two different parts: the upper arm and the fore-
arm. The upper armis represented by the segment that
goes from the shoulder to the elbow, the forearm the
segment that goes from the elbow to the hand.
Starting from the sternum (see picture 1), that for
us represent the reference base, and moving toward to
the distal part of the limb, we can encounter the fol-
lowing bones : Clavicle, Scapula, Humerus, Radius,
Ulna, Carpus Bones, Metacarpus Bones, Phalanxes.
Figure 1: Representation of the skeleton of the human
arm and its 10 DOF simplied kinematics model using a
robotics notation.
In literature we can nd different kinematic mod-
els for the human arm (Klopcar and Lenarcic, 2005),
(Schiele and van der Helm, 2006), each one oriented
to describe certain aspects rather than the others; As
in (Schiele and van der Helm, 2006), to represent the
kinematic model of the human arm, we used a no-
tation related with the robotics eld this in order to
couple it more easily with the kinematic model of the
exoskeleton. Of course we introduced numerous sim-
plications and assumed the articulations like joints
with a well dened geometry, nevertheless we think
that for our study this ts well.
The model formalized is represented in gure 1
(right side), again we can separate the kinematic ar-
chitecture in two different parts: Shoulder Kinemat-
ics, Arm Kinematics. The shoulder Kinematics is
composed by four joints, three spherical (3DOF) and
one planar. More in detail the planar joint can be de-
composed in two prismatic and one rotational joint,
however in this rst representation we preferred to use
a compact notation. It also should be noted that the
shoulder kinematics can be further separated in two
other parts: one that is a closed kinematic chain, and
the other that is an open kinematic chain represented
by a spherical joint located in the proximal part of the
upper arm link. The closed chain is formed by three
links and three joints, joint-1 (spherical) is located be-
tween link-0 (the link with the inertia reference sys-
tem) and link-1, joint-2 (spherical) is located between
link-1 and link-2, and nally joint-3 (planar) close the
kinematics chain connecting link-2 and link-0.
The rst consideration we can do on this kine-
matic chain is about the overall mobility. The three
joints have a total of 9 DOFs, however because of its
parallel nature there are some constrains that limit the
mobility. We can dene q as the conguration vari-
able, this is a vector with m components (q R
m
)
that dene unambiguously the position and the ori-
entation of the all rigid bodies that compose the kine-
matic chain. We consider in this case only minimal
congurations, this means that is not possible dene
unambiguously the system with less than m scalars.
Given the kinematic chain we can calculate the
dimension for q applying the Kutzbach-Gr ubler for-
mula (Zhao et al., 2004) :
m = 6(n g 1) +
g

i=1
f
i
(1)
where n is the number of links present in the kine-
matic chain, g is the number of joints, and f
i
is the
number of degrees of freedom for the i
th
joint. If we
apply this equation to our specic case we obtain:
m = 6(3 3 1) +
3

i=1
f
i
=6 +9 = 3 (2)
This mean that this chain has overall three de-
grees of freedom, therefore to dene unambiguously
DESIGN OF A BIO-INSPIRED WEARABLE EXOSKELETON FOR APPLICATIONS IN ROBOTICS
415
its kinematic conguration we can just dene only
three scalars. The model differs from (Schiele and
van der Helm, 2006) for the presence of an addi-
tional DOF that allows to represents better the human
anatomy. The question that arise now is: which joints
variables we have to chose to dene the congura-
tion of the shoulder, in theory it is possible to chose
just three variables from the nine we have. In practice
we will see that there are some choices that are bet-
ter than others, this especially if we need to measure
these quantities in a real system.
Starting from the joint-4 the human arm can be
represent as a open kinematic chain. As we can
see from picture 1 joint-4 (lower part of the shoul-
der) connects link-2 to link-3. This joint has a total
of three DOF and allows movements of extension-
exion, adduction-abduction and rotation around the
upper arm axis.
Moving toward the distal part of this model we
encounter joint-5 (the elbow) that connect link-3 with
link4, this is a one DOF rotational joint that allow
forearm exion and extension. Finally we have joint-
6 (rst degrees of freedom for the wrist) that connect
link-4 with link-5. In comparison with the human arm
anatomy this represent a simplication, indeed in hu-
man beings it is a complex movement of both radio
and ulna bones that allows the wrist rotation. Any-
how in a rst approximation this simplication is not
so critical for our purposes. A more accurate model
will be formalized in the case the results obtained will
be not satisfactory.
In picture 1 we represented also the other joint for
the wrist, joint-7, this has a total of two degrees of
freedomthat in the human armallowthe wrist exion-
extension and adduction-abduction. At the moment
the hand kinematics is not considered in our study.
2.1 Analysis of the Arm Movements
In order to better understand the kinematic of the hu-
man Arm, and to start the validation of the proposed
model, we conducted a rst experiment where we ac-
quired the trajectories of different points of interest
located on the surface of the Arm. We applied a total
of 19 markers (see picture 2) on a male subject (height
1.7m.) 3 along the spinal cords, 3 on the scapula, 1
on the top of the shoulder, 3 for the shoulder ring, 1
in the middle of the upper arm, 3 in the elbow ring , 1
in the middle of the forearm, and 4 in the wrist ring.
We wanted to acquire the trajectories of all the parts
of the Arm that are involved during the performance
of an arbitrary movement. In the rst experiment we
asked the subject to perform a movement of exion
and extension of the shoulder. The rotation was rela-
tive to an hypothetical axes orthogonal to the sagittal
plane, of course due to the complex kinematic of the
shoulder this axes is not xed, but changes according
with the shoulder movements.
Figure 2: 19 Markers were xed on the surface of the sub-
jects arm.
The Experiment where done using a commercial
motion tracking system by Qualysis

, we employed
the version with three cameras.
We chose an arrangement in order to avoid as
much as possible the landmarks occlusion during the
planned movement.
During the acquisition subject was located near
the reference system, and was asked to perform the
movement of the arm trying to keep xed as much as
possible the rest of the body.
After data acquisition and post-processing, it was
possible to analyze the results visually, and to ex-
port the data in text format for a further elabora-
tion by Matlab. The analysis of the movement by
the QualysisTrackManager

showed many interest-

ing features and behaviors. We noted that from the
rst phase of the extension movement all the bones
of the upper shoulder are involved. We could recog-
nize this by observing the trajectory of the markers
located on the top of the shoulder, and in proximity of
the scapula (gure 3a ).
This means that it is very difcult to separate the
movement of the lower shoulder from the movement
of the upper shoulder, it turns out that we need to con-
sider the entire shoulder kinematics from the begin-
ning of the exoskeleton design. In picture 3 (b) we
can see the trajectories followed by the markers for
quite the complete extension movement (90%), it ap-
pears that many of them have a circular pattern, this
is natural if we think at the kinematic structure of the
human shoulder.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
416
a. b.
Figure 3: (a)First part of the movement, it possible to note
that the markers on the top of the shoulder start to move (b)
Trajectories followed by the markers during the 90% of the
extension movement.
3 THE EXOSKELETON
KINEMATIC
The exoskeleton kinematics is strongly inuenced
both by the human arm anatomy and the perfor-
mances we want to reach. The central idea is to try
to restrict the mobility of the users arm as less as
possible when he is wearing the exoskeleton. Other
desiderable requirements for the overall system are:
Lightweight construction
A system easily wearable
Multi-contact points haptic feedback
Modular design
All these goals are important to synthesize the
kinematic structure for the exoskeleton, but for an ini-
tial analysis, the overall mobility constrains and the
necessity to have multi-contact point haptic interface,
represent our most relevant aims.
If we want to reduce the users mobility limita-
tions due to the exoskeleton we can x the following
kinematic requirements:
The upper arm coupled with the upper part of the
exoskeleton should have a total of 3 DOFs
The forearm coupled with the lower part of the
exoskeleton should have a total of 2 DOFs
In order to provide the user a broad haptic feed-
back, our exoskeleton will transmit forces and torques
via multiple contact points. One of these we have
already dened locating the exoskeleton-shoulder on
the user-shoulder, the other will be located in the mid-
dle of the user upper armand the last one in the middle
of the user forearm. This locations are optimal in the
sense that they reduce the interference with the hu-
man articulation during the user movements. One can
see these three different contact-points represented in
gure 4.
Cont act Poi nt 1
Cont act Poi nt 2
Cont act Poi nt 3
Figure 4: Contact points between the exoskeleton and the
human arm.
3.1 Coupling the Exoskeleton with the
Human Arm
It is important now to do some considerations about
the overall kinematics that we obtain combining the
arm with the exoskeleton. This will also help us to
setup how many degrees of freedoms are required for
the exoskeleton.
The next paragraph will concentrate only on the
exoskeletons kinematics that deals with the shoulder
and the upper arm of the user, even if we have al-
ready started to extend the analysis also for the fore-
arm. This omission is also justies by the fact that at
the moment we tested on a human subject only this
part of the exoskeleton.
3.1.1 The Upper Shoulder Joint
Here we the term joint we want to refer to the en-
tire kinematic structure for the mechanical systemthat
is charged to deal with the upper shoulder (clavicle-
scapula articulation). How many degrees of freedom
should have this joint? From the motion analysis on
the human arm it comes up that the upper shoulder
has a total of 3DOF, but again in order to design the
exoskeleton joint it is necessary to do some assump-
tions about its kinematics. A possible conguration
is presented in gure 5 (upper part) here we can see
that now we have a complex structure with different
closed paths.
In order to study the system we can do a rst sim-
plication substituting the upper shoulder kinematic
with a single joint with 3 DOF, in this case we have a
simplied structure showed in gure 6.
Now it is possible to apply the theory in order to
nd the overall mobility of this conguration, we can
calculate the number of DOFs with equation 3.
m = 6(ng1) +
g

i=1
f
i
= 6(551) +9 = 3 (3)
DESIGN OF A BIO-INSPIRED WEARABLE EXOSKELETON FOR APPLICATIONS IN ROBOTICS
417
3 D OF
3 D OF
J 1
e s
J 2
e s
J 3
e s
3 D OF
L 0
s
L 0
s
L 0
e s
L 2
e s
L 3
e s
Figure 5: Closed kinematic chain formed between the ex-
oskeleton shoulder joint and the upper shoulder kinematic
chain.
2 D OF
3 D OF
L 0
s
L 0
e s
J 1
e s
L 2
e s
J 3
e s
L 1
e s
J 2
e s
2 D OF
1 D OF
L 2
s
L 1
s
1 D OF
Figure 6: Equivalent model.
As is represented in gure 6 the exoskeleton has
a total of 6 DOF; because we need to actuate a 3
DOF kinematic chain, it means that only 3 of the 6
DOF must be actuated and sensed. From a mechan-
ical point of view it is more suitable to actuate the
joints (J
es
1 and J
es
2) that is near to the barycenter of
the body, in this way the actuation system is not re-
quired to move also the weigh of the actuators itself,
this come clear if you think to the torque that can ex-
ercise a weight localized near the joint J
es
3 to the joint
J
es
1.
3.1.2 Model Simulations
To x some specications for the actuation system of
the exoskeleton we formilized a kinematic-dynamic
model of the exoskeleton shoulder-joint using the
toolbox SimMechanics in Matlab-Simulink environ-
ment. The system is composed of a spherical joint
and a prismatic joint (see picture 7).
The initial point for the exoskeleton joint is coin-
cident with the middle landmark along the spinal cord
(see section 2.1).
In order to analyze the motion in a realistic way,
we constraint the point P (see picture 7 ) to lie on a
trajectory. We imposed as a trajectory the one we ob-
tained fromthe motion analysis of the human armper-
forming an extension-exion movement of the shoul-
der (underlined Marker in picture 2). To constrain
the point on the desired trajectory we apply a force
eld directly on the point that was generated using a
MIMO (Multi Input Multi output) PID controller, pa-
Figure 7: Simulation for the Shoulder Joint, measures in
meters.
rameters of this controller were not optimized because
we wanted only to perform a kinematical test of the
system. The PID in question can be represented using
a diagonal matrix:

P=0.4,I=0.1,D=0.3 0 0
0 P=0.4,I=0.1,D=0.3 0
0 0 P=0.4,I=0.1,D=0.3

This means that only the position error along the

X-axis will effect the X-component of the force, and
the same for error along Y- and Z-axis. The results of
this simulation is reported in picture 8, as it possible
to see the force eld generated is able to constrain
the point trajectory (thicker line) near the reference
trajectory (thin line).
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0.1
0.08
0.06
0.04
0.02
0.55
0.6
0.65
0.7
0.75
m
m
m
Figure 8: The trajectory followed (thicker line) by the point
located on the upper shoulder, the thin line is the reference
trajectory acquired from a human subject.
Once we are sure that point P is well constrained
we can monitoring the position of each joint of the ex-
oskeleton in order to evaluate the range of its move-
ment, this is very useful to obtain same specications
for the design of the real system. In the graph of pic-
ture 9 we can see the linear position of the prismatic
joint, how it is possible to note the range for this sub-
ject is about 0.08m(8cm). Of course it is necessary to
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
418
perform this analysis on different subjects if we want
to have a system that can adapt to different arm sizes,
this will be the subject of future work.
0 50 100 150 200 250 300
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
Sample
P
o
s
i
t
i
o
n

[
m
m
]
Figure 9: Range for the prismatic joint.
Figure 10 reports the angular positions of the
spherical joint in convention of the Euler angles (roll,
pitch and yaw). Therefore if we want to obtain the an-
gular position of the exoskeleton we should do in se-
quence three rotations along X,Y and Z respectively.
Again we can evaluate the excursion for each sin-
gle angle, from gure 10 we can see that roll range is
about 30
o
, the pitch 50
o
and nally the yaw 15
o
.
0 50 100 150 200 250 300
200
150
100
50
0
50
100
150
200
Sample
A
n
g
u
l
a
r

p
o
s
i
t
i
o
n

[
D
e
g
r
e
e
s
]
X
Y
Z
Figure 10: The three Euler Angles for the spherical joint.
3.1.3 Lower Shoulder Joint
Considering the lower shoulder (upper arm) com-
bined with the upper part of the exoskeleton (gure
11) one can identify a closed kinematic chain. It
starts at the link L
a
0, crosses join J
a
1 (the user shoul-
der), goes toward the contact point one (where the ex-
oskeleton is xed with the user upper arm) and then
encounters in the order J
ea
3, J
ea
2 and J
ea
1 that belong
to the exoskeleton. Finally the kinematic chain ends
with the link L
ea
0 that in this case coincides with L
a
0.
We have also to observe that in picture 11 L
a
1 and L
3
1
should be considered as a single link. Supposing now
that J
ea
1 is a 2DOF rotational joint, J
ea
2 is a 1DOF
prismatic joint and J
ea
3 a spherical joint we can cal-
culate the overall degrees of freedom of this closed
kinematics.
m = 6(ng1) +
g

i=1
f
i
= 6(441) +9 = 3 (4)
Equation 4 shows that combining the exoskeleton
with the arm brings to a system that has three degrees
of mobility (user upper arm), of course this do not
guarantee that the mobility we obtain is similar to the
mobility of the user upper arm. This is due to the
fact that Kutzbach-Gr ubler formula do not takes in ac-
count the conguration of the joint, but only the total
number of joints and degrees of freedom. For a more
precise analysis, a possible solution is to performa se-
ries of simulations, this will hallow us also to explore
different congurations with different parameters.
L 0
a
L 1
a
J 1
a
L 0
e a
L 1
e a
J 1
e a
J 2
e a
L 2
e a
J 3
e a
L 3
e a
Figure 11: Kinematic representation of the exoskeleton
coupled with the user upper arm.
4 EXOSKELETON DESIGN
From the data analysis of the extension/exion move-
ments of the shoulder it clearly appears that it is nec-
essary to consider the overall shoulder complexity in
order to dene the exoskeleton kinematic.
However, in order to simply the design process,
it is still possible to assume the point dened on the
upper-lateral part of the shoulder as a starting point
where to x the kinematic structure that will follow
the lower shoulder movements.
Therefore we separated the exoskeleton design in
two different parts: one that deals with the upper and
the other with the lower shoulder. In the following
we explain a possible solution for the upper shoulder
joint.
4.1 Upper-shoulder Joint
The mechanical structure is composed by four joints:
a sequence of two rotational, one prismatic and one
spherical joints. In gure 12 we can see a rst con-
cept for this structure, were we can note that there are
two connection structures: one that is intended to be
xed to the user pelvis (the belt), and the other that is
DESIGN OF A BIO-INSPIRED WEARABLE EXOSKELETON FOR APPLICATIONS IN ROBOTICS
419
intended to be connected with the top side of the user
shoulder. We want to employ rigid materials for these
two parts in order to have a stable connection with
the human body, but of course, we need also to shape
these parts in order to be comfortable for the user.
2 D OF
3 D OF
L 0
s
L 0
e s
J 1
e s L 2
e s
J 3
e s
L 1
e s
J 2
e s
2 D OF 1 D OF
L 2
s
L 1
s
1 D OF
Ac t i v e DOF s
Pa s s i v e DOF s
Figure 12: The Upper-Shoulder joint Exoskeleton concept
and its kinematic structure.
exoskeleton. It is quite easy to x the belt to the
pelvis.
To design the device we should also take into ac-
count some important parameters:
The exoskeleton length: this parameter should be
adjustable in order to t with different user sizes.
Fortunately the length depends on the linear posi-
tion of the prismatic joint, this means that the joint
movement can be used to control the position of
the user shoulder but also to adjust the device to
the user size.
The distance between the exoskeleton and the user
back: this is important because if the exoskeleton
is too near to the user back collisions will occur
during the shoulder movements.
The Shoulder-Connection dimension: this also
depend on the size of the user shoulder, in this
case it is necessary to build up a mechanism
adaptable. A possible solution is to use an inat-
able device, even if this will decrease the stability
of the contact point.
Furthermore in order to keep low the inertia and
the torque requirements of the actuation system, we
can think to actuate the rst three DOF and let pas-
sive the last three (Spherical Joint). This solution is
also optimal for the mass distribution, in this case the
barycenter is more near to the user spinal cords (the
endoskeleton that sustains all the upper body weight).
To evaluate if the kinematic structure we assumed
for the exoskeleton is suitable and efcacious, we de-
cided to build a passive version of the system. This,
depicted in gure 13, reproduces the same mechanical
functionality of the system until the lower-schoulder-
Joint, but has only sensory capabilities, indeed no ac-
tuators are mounted on the joints. We tested the de-
vices on different subjects and we get a rst impres-
sion on how the system works.
Figure 13: The passive version of Upper and Lower
Shoulder-joint.
From this rst qualitative analysis we obtain a
very usefull feedback in order to guide the next steps
for the design process. We noticed that the dimen-
sions of the different exoskeltons links are not only
important to t the size of different users, but are
also critical to keep the joints movements in a proper
range. For example, if the link L
s
2 (picture 12) is
too long the prismatic joint is always completely re-
tracted. This initial condition brings the system to
lose some degrees of mobility. A solution for this
problem is to dimension the exoskeleton in a manner
that each joint, in its initial state, assumes a position
in the middle of the possible range.
5 CONCLUSIONS AND FUTURE
WORK
In this paper we described the design methodologywe
adopted to develop a multi contact point haptic inter-
face. We introduce a kinematic model for the human
arm and combine it we real motion data in order to
synthesize the exoskeleton. We show by realistic sim-
ulations that the kinematic conguration we chose for
the shoulder-joint ts with the human arm anatomy
and do not restrict the shoulder movement. Future
work will be nalized to better study the kinematics
of the system that will deal with lower shoulder and
the forearm and to test a complete arm-prototype. We
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
420
need also to solve the problem of designing a stable
interface between the exoskeleton and the arm and to
develop a proper actuation system. At the movement
we are dealing with the experimentation of a light hy-
brid hydraulic-pneumatic actuator that will be able to
nally control the force feedback and to change the
impedance actively.
REFERENCES
Bar-Cohen, Y., Mavroidis, C., Bouzit, M., Dolgin, B.,
Harm, D. L., Kopchok, G., and White, R. (2001).
Virtual reality robotic telesurgery simulations using
memica haptic system. In Electroactive polymer ac-
tuators and devices, SPIE proceedings series, volume
4329, pages 357363.
Carignan, C., Liszka, M., and Roderick, S. (2005). Design
of an arm exoskeleton with scapula motion for shoul-
der rehabilitation. Advanced Robotics, 2005. ICAR
05. Proceedings., 12th International Conference on,
pages 524531.
Chen, Y., Zhang, J., Yang, C., and Niu, B. (2007). The
workspace mapping with decient-dof space for the
puma 560 robot and its exoskeleton arm by using or-
thogonal experiment design method. Robot. Comput.-
Integr. Manuf., 23(4):478487.
Dollar, A. and Herr, H. (2008). Lower extremity exoskele-
tons and active orthoses: Challenges and state-of-the-
art. Robotics, IEEE Transactions on, 24:144158.
Grifn, W. B., Provancher, W. R., and Cutkosky, M. R.
(2005). Feedback strategies for telemanipulation with
shared control of object handling forces. Presence:
Teleoper. Virtual Environ., 14(6):720731.
Gupta, A., OMalley, M. K., Patoglu, V., and Burgar, C.
(2008). Design, control and performance of ricewrist:
A force feedback wrist exoskeleton for rehabilitation
and training. Int. J. Rob. Res., 27(2):233251.
Hayward, V. and Astley, O. (1996). Performance mea-
sures for haptic interfaces. In Giralt, G. and Hirzinger,
G., editors, Robotics Research: The 7th International
Symposium, pages 195207. Springer Verlag.
Hokayem, P. and Spong, M. (2006). Bilateral teleoperation:
An historical survey. Automatica, 42:20352057.
IJsselsteijn, W. A., de Kort, Y. A. W., and Haans, A. (2006).
Is this my hand i see before me? the rubber hand illu-
sion in reality, virtual reality, and mixed reality. Pres-
ence: Teleoper. Virtual Environ., 15(4):455464.
Kim, Y. S., Lee, S., Cho, C., Kim, M., and Lee., C.-W.
(2001). A new exoskeleton-type masterarm with force
reection based on the torque sensor beam. Robotics
and Automation Proceedings 2001 ICRA. IEEE In-
ternational Conference on, 2623:26282633.
Klopcar, N. and Lenarcic, J. (2005). Kinematic model
for determination of human arm reachable workspace.
Meccanica, (40):203219.
Kwon, D.-S., Ryu, J.-H., Lee, P.-M., and Hong, S.-W.
(2000). Design of a teleoperation controller for an
underwater manipulator. In Robotics and Automa-
tion, 2000. Proceedings. ICRA apos;00. IEEE Inter-
national Conference on, pages 31143119.
Mistry, M., Mohajerian, P., and Schaal, S. (2005). An ex-
oskeleton robot for human arm movement study. In
Intelligent Robots and Systems, 2005. (IROS 2005).
2005 IEEE/RSJ International Conference on, pages
4071 4076.
Perry, J., Rosen, J., and Burns, S. (2007). Upper-limb pow-
ered exoskeleton design. Mechatronics, IEEE/ASME
Transactions on, 12:408 417.
Schiele, A. and van der Helm, F. (2006). Kinematic design
to improve ergonomics in human machine interaction.
Neural Systems and Rehabilitation Engineering, IEEE
Transactions on, 14(4):456469.
Schiele, A. and Visentin, G. (2003). The esa human arm
exoskeleton for space robotics telepresence. In 7th
International Symposium on Articial Intelligence,
Robotics and Automation in Space, iSAIRAS.
Song, D. H., Lee, W. K., and Jung, S. (2005). Control
and interface between an exoskeleton master robot
and a human like slave robot with two arm. In Ad-
vanced Intelligent Mechatronics. Proceedings, 2005
IEEE/ASME International Conference on, pages 319
324.
Yu, A. B. (2003). A taxonomy and comparison of haptic
actions for disassembly tasks.
Zhao, J.-S., Zhou, K., and Feng, Z.-J. (2004). A theory of
degrees of freedom for mechanisms. Mechanism and
Machine Theory, 39(6):621643.
DESIGN OF A BIO-INSPIRED WEARABLE EXOSKELETON FOR APPLICATIONS IN ROBOTICS
421
EXTENDED HEALTH VISIBILITY IN THE HOSPITAL
ENVIRONMENT
H. Fernndez Lpez, J . A. Afonso, J . H. Correia
1
Industrial Electronics Engineering Department, University of Minho, Campus de Azurm, 4800-058, Guimares, Portugal
{hlopez, jose.afonso, higino.correia}@dei.uminho.pt
Ricardo Simes
2
Technology School, Cvado and Ave Polytechnic Institute, Urbanizao das Calada
Edifcio Galo - SN R/C, 4750-117, Arcozelo, Barcelos, Portugal
3
Institute of Polymers and Composites, University of Minho, Campus de Azurm, 4800-058 Guimares, Portugal
[email protected]
Keywords: Vital signals monitoring, Health monitoring device, Wireless sensor networks, ZigBee.
Abstract: Wireless sensor networks can help healthcare providers enhance patient monitoring and communication
capabilities. This paper describes the present state of the development of a vital signal monitoring network
applied to the hospital environment. The proposed network is based on non-obstructive sensors able to
communicate through a low power wireless sensor network based on the ZigBee protocol. This network
enables continuous patient monitoring, creating entirely new mechanisms for providing healthcare under a
plethora of cases (e.g. post-op, continuous care, and chronic diseases). The main advantages of this system
include increased patient mobility, faster detection of potential problems, real-time feedback to caregivers
and patients, and faster response to emergency situations.
1 INTRODUCTION
We have specified a system that facilitates vital
signals gathering in a hospital environment that
resembles the basic configuration analyzed in
(Cypher, Chevrollier, Montavont, & Golmie, 2006)
with some improvements. We have added a
mechanism to send messages back from the
hospitals monitoring system to the patient. With
this system we want to introduce the concept of
extended health visibility, where non-critical
patients can be continuously and unobtrusively
monitored while simultaneously being informed
about their condition and receive messages from the
hospital caregivers (e.g., a doctor appointment, a
reminder to take their medicine, or a visitor arrival
notice). Through this system, instead of collecting
vital signals only a few times along the day,
hospitals will be able to maintain continuous patient
records which will provide better emergency
condition detection and response, enhanced
diagnosis capabilities, and promote, among the
patients, a better appreciation of the care being
provided.
Despite the evident benefits that can result from
the adoption of wireless technologies, there are still
many concerns that limit the widespread application
and challenge researchers to devise potential
solutions. It is essential to satisfy demanding
requirements in terms of quality of service such as
sustainable throughput, bounded delay and reliable
packet delivery. At the same time, it is necessary to
guarantee that the power consumption of the sensor
nodes is small, since they are powered by batteries,
in order to increase their autonomy. Another
difficulty arises from the fact that some sensors must
be sampled quite often, generating a large amount of
data and, consequently, requiring the network to
operate under high load, which is not common in
typical wireless sensor network scenarios. Several
strategies are been used to overcome those problems
and they include the use of techniques to compact
data and the efficient use of the communications
channel.
In order to adequately introduce and describe our
solution, we present a quick review of related work.
The scenario envisioned in (O'Donoughue, Kulkarni,
& Marzella, 2006) differs from ours because the
422

developed system aims at transmitting vital signals
from in-patients in a very restricted area, an
intensive care unit. In our case, patient mobility
poses another challenge. It will be necessary to
provide means of efficiently collecting the signals
throughout the hospitals targeted areas without
losing information.
Two different scenarios are also considered in
other related works. In case of mass casualty,
overwhelming quantities of patients need to be
monitored and triaged. AID-N is a real-time patient
monitoring system that integrates vital signs and
location sensors, ad-hoc networking, electronic
patient records, and Web portal technology to allow
remote monitoring of patient status (Tia, Greenspan,
Welsh, J uang, & Alm, 2005). CodeBlue also
considers critical care environments (Lorincz et al.,
2004). It is a common protocol and software
framework which allows wireless monitoring and
tracking of patients and healthcare professionals.
Our work shares the basic concerns of systems
planned to operate in those scenarios, namely device
discovery, naming, routing and security.
Nevertheless, we are not concerned with data
prioritization but in assuring the reliability of the
data transmission.
Telediagnosis and teleconsulting are considered
in the AMBULANCE project (Pavlopoulos,
Kyriacou, Berler, & Koutsouris, 1998). A portable
device, carried inside an ambulance, allows
telediagnosis, long distance support and
teleconsultation by specialized physicians. It allows
the transmission of vital signals and still images of
the patient from the incident place to the hospital.
While AMBULANCE uses GSM, a system
primarily designed to handle voice communications,
we have chosen to use ZigBee (ZigBee.Alliance,
2007), an emerging wireless sensor network (WSN)
protocol primarily conceived to be applied in low
traffic scenarios.
In the next section, we present some tests results
considering the electrocardiogram data acquisition
and its wireless transmission. The conclusions and
future work will be presented in the last section.
2 EXTENDED HEALTH
VISIBILITY
Gathering current patient medical data promptly and
accurately is vital to proper health care. Continuous
monitoring offers the capability of identifying and
responding to events as they occur, possibly
preventing a dangerous condition, instead of simply
allowing its diagnosis after the danger has taken
place.
2.1 The Envisioned Solution
The envisioned system is not intended to eliminate
current existing devices or routines but to improve
patients monitoring capabilities. Under this
assumption, it is not intended, for instance, to
replace wire-based continuous monitoring devices
from intensive care units. Urgency, observation,
ward and recovery from procedures that require
local anesthesia or sedation are possible scenarios.
The system in development is based on a ZigBee
ad-hoc wireless sensor network with routers and
gateways distributed all over the hospital spaces
intended for monitored patients. This network will
be responsible for receiving data from sensor units
and transmitting this information to the main
application running in the nurses station through a
ZigBee to Wi-Fi gateway. The data sent to the main
application will be processed and will become
available, through a web portal, to registered
healthcare providers carrying a portable device (e.g.,
a Personal Digital Assistant, PDA), as depicted in
Figure 1, where only one ZigBee network is shown.


Figure 1: Envisioned system architecture.
Five vital signals will be monitored: cardiac rate,
electrocardiogram (ECG), arterial pressure, pulse
oximetry and temperature. Medical sensors will be
designed to be minimally obtrusive.
The main application will analyze patient data
and, based on thresholds defined by the physician,
generate alerts to healthcare providers if a critical
condition occurs. Furthermore, it will be possible to
alter threshold values individually. Additionally,
portable devices will allow healthcare providers to
remotely access patient information and change
monitoring configurations at any time. In the patient
perspective, the system will be able to provide a
mechanism for communication between patients and
EXTENDED HEALTH VISIBILITY IN THE HOSPITAL ENVIRONMENT
423

the nurses station. Nurses will be able to send
messages to patients and to receive assistance
requests.
2.2 Current Project Status
ZigBee is a wireless network standard which
resulted from the collaborative efforts of a
consortium of companies known as the ZigBee
Alliance. IEEE 802.15.4 (IEEE, 2003) is a standard
defined by the IEEE for low-rate, wireless personal
area networks which specifies the physical (PHY)
and the medium access control (MAC) layer used by
ZigBee. The specification for the physical layer
defines a low-power spread spectrum radio
operating at 2.4 GHz with a bit rate of 250 kbps.
There are also PHY specifications for 915 MHz and
868 MHz that operate at lower data rates.
Even though ECG data rates are relatively high
compared with data rates generated by other sensors,
they are well below the bit rate of a ZigBee network
operating at 2.4 GHz, so it is expected that ZigBee
will be able to carry the generated traffic using its
CSMA/CA protocol, provided that high traffic loads
are avoided, in order to limit the percentage of
packets lost due to collisions. ZigBee was chosen
because it allows the construction of networks
composed by a large number of nodes that can cover
large areas through the use of multihop
communication, in contrast, for example, with
Bluetooth, which supports only seven end nodes per
piconet in a short-range star topology. Additionally,
ZigBee is a standard based protocol that operates on
a license-free ISM band and provides low-cost, low-
power and small form factor modules.
The ZigBee network developed is based on
J N5139 modules manufactured by J ennic (J ennic,
2008). Basic network functionalities have already
been tested using several end devices that
continuously send simulated ECG data in a star
network topology.
2.2.1 The Wireless Network for ECG
Monitoring
The ECG application was chosen as the first one to
be assessed because it is the most demanding one.
The network operates in the 2.4 GHz band and the
access to the radio channel is gained using the
unslotted CSMA/CA. The chosen sample rate is the
same used by commercial monitoring ECG
equipments, 120 Hz (Fulford-Jones, Gu-Yeon, &
Welsh, 2004).
2.2.2 Power Consumption Estimation and
Measurement
Power consumption is an important issue. It is
desirable that end devices consume the least possible
power to allow for continuous operation for long
periods without batteries replacement. To estimate
the battery lifetime, we must consider that the CPU
is always active because frequent ADC
measurements are required. During these periods,
the module consumption is equal to 9.21 mA. In
reception and transmission modes the current
consumption reaches 37mA (J ennic, 2007).
The module reads the ECG sensor thirty times,
once every T
S
=8.33 ms (which is the inverse of the
sampling rate, 120 Hz) using an internal 12-bit
resolution ADC, and then, every 250 ms, sends a
data packet with the measurements. One module
cycle is depicted in Figure 2, where current
consumption values are shown for each activity:
ADC reading (ADC), backoff (BP), clear channel
access (CCA), data transmission (T
DF
),
acknowledgment waiting (t
ack
) and acknowledgment
reception (T
ACK
).


Figure 2: Module current estimation.
Taking those operation parameters into
consideration, it is possible to determine that the
average current over the 250 ms period is equal to
9.84 mA. The total operating time considering the
module is powered by two AAA 1200 mAh batteries
is approximately 122 hours or 5 days.
The current drained by the module as a function
of time was measured to confirm the power
consumption estimation using a data acquisition
board driven by a MATLAB program. In Figure 3 it
is possible to observe the current variation over a
period of approximately 300 ms where ADC sensor
readings and backoff, CCA, data frame transmission,
t
ACK
and acknowledge reception are shown.

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
424


Figure 3: Module current measurement.
2.2.3 Network Operation Tests
We have tested the packet error rate by considering
just one end device transmitting a 2-byte payload
data frame to the coordinator. Five runs were
executed involving the transmission of a total of
40,000 packets at different intervals. No packet was
lost in any run but we have observed packets
retransmissions during the longer tests probably due
to interference from Wi-Fi traffic. The number of
packets that required retransmission in each run is
shown on Table 1. The results confirm the
satisfactory coexistence between ZigBee and Wi-Fi
networks.
Table 1: Transmission test results.
Interval between
packets
Number of retransmitted packets in
each run
1 2 3 4 5
100 ms 6 5 6 0 0
20 ms 1 0 0 0 0
10 ms 0 0 0 0 0
5 ms 0 0 0 0 0

A second test was performed using five ZigBee
modules to simulate a small ZigBee star network.
Four modules were used as end devices to gather
and transmit ECG signals using the unslotted
CSMA/CA mode. Every 250 ms a 45-byte payload
data packet was sent to the fifth module, which was
programmed to perform as the network coordinator.
It also checks if any data packet from any end device
is lost and reports it. Ten runs involving the
transmission of at least 2400 packets by all four end
devices were executed. No packet was lost in any
run, which can be considered a promising result.
3 CONCLUSIONS
Wireless sensor networks promise ubiquitous vital
signal monitoring that might completely transform
the way health care is provided.
We have proposed a system that facilitates vital
signal gathering in a hospital environment for what
we call extended health visibility. Instead of
collecting vital signals only a few times along the
day, healthcare professionals will have access to
continuous patient records which will provide
several improvements compared to current
capabilities. Power consumption estimation and
measurements, as well as network performance test
results were discussed. These preliminary results
seem to validate the potential applicability of the
proposed system for the intended application.
ACKNOWLEDGEMENTS
This work has been supported by the Portuguese
Foundation for Science and Technology and the
POCTI and FEDER programs.
REFERENCES
Cypher, D., Chevrollier, N., Montavont, N., & Golmie, N.
(2006). Prevailing over wires in healthcare
environments: benefits and challenges. Comm.
Magazine, IEEE, 44(4), 56-63.
Fulford-J ones, T. R. F., Gu-Yeon, W., & Welsh, M.
(2004). A portable, low-power, wireless two-lead EKG
system. IEMBS '04. 26th Annual International
Conference of the IEEE.
IEEE Std 802.15.4-2003IEEE Standard for Information
Technology Part 15.4, (2003).
J ennic. (2007). Application Note JN-AN-1001-1v3
Calculating JN5121 Power Consumption.
J ennic. (2008). Product Brief J N5139-xxx-Myy
IEEE802.15.4/ZigBee Module Family.
Lorincz, K., Malan, D. J ., Fulford-J ones, T. R. F., Nawoj,
A., Clavel, A., Shnayder, V., et al. (2004). Sensor
networks for emergency response: challenges and
opportunities. Pervasive Computing, IEEE, 3(4), 16-23.
O'Donoughue, N., Kulkarni, S., & Marzella, D. (2006).
Design and Implementation of a Framework for
Monitoring Patients in Hospitals Using Wireless
Sensors in Ad Hoc Configuration. 28th IEEE EMBS
Annual International Conference
Pavlopoulos, S., Kyriacou, E., Berler, A., & Koutsouris,
D. (1998). A mobile system for emergency health care
provision via telematics support-AMBULANCE,
Information Technology Applications in Biomedicine,
1998. ITAB 98. Proceedings. 1998 IEEE International
Conference on (pp. 150-154).
Tia, G., Greenspan, D., Welsh, M., J uang, R. R., & Alm,
A. (2005). Vital Signs Monitoring and Patient
Tracking Over a Wireless Network, Engineering in
Medicine and Biology Society, 2005. IEEE-EMBS
2005. 27th Annual International Conference of the
(pp. 102-105).
ZigBee Alliance Document 053474r17, ZigBee
Specification, v. 1.0 r17, (2007).
EXTENDED HEALTH VISIBILITY IN THE HOSPITAL ENVIRONMENT
425
MICROFLUIDIC CELL STIMULATOR USING BEAD IMPACT
Young-Hun Kim, Tae-J in Kim, Hyung-J oon Kim, Min-Suk Park and Hyo-Il J ung
School of mechanical engineering, Yonsei University, Seoul, Republic of Korea
[email protected], [email protected], [email protected]
[email protected], [email protected]
Keywords: Mechanical stimulation, Micro device, Rolling bead, CPAE, PDMS.
Abstract: Recently many researchers are focused on cell stimulation regarding observation of cells to specific
stimulation factors. We introduce new mechanical stimulation method using micro beads without any
chemical reagents. CPAE (calf pulmonary artery endothelial cell) were cultured in PDMS
(polydimethylsiloxane) microfluidic device. After starvation process, sterilized 10m glass beads were
rolled by only gravitational force for 10 minute. To find optimal stimulation time, 16 devices were made by
PDMS and each device was slanted every hour. Results show that cell exposed under micro bead
stimulation perform at a higher growth rate than normal conditions and 1 hour stimulation time represents
more effective than other stimulation times. This new cell stimulation method will be able to help make
artificial organ such as blood vessels in the future.
1 INTRODUCTION
Studies about observation of cell behaviour are
performed to explore factors which are related with
inhibition and promotion of cell growth. It is true
that cells in human and animal are exposed under
various stimulation such as mechanical (Matteucci,
2007), chemical (Nakashima, 2005), and electrical
stimulation (Mattei, 2004). These studies were tried
to find alternative stimulation of cell growth, thus
could contribute tissue regeneration. However the
effect of mechanical stimulation is not to be
accomplished and possibility of stimulation is still
opened.
Miniaturized bioreactor for cell stimulation was
developed by soft lithography and PDMS
(polydimethylsiloxane) molding technique. It is
considered inexpensive and time saving method and
makes disposable device and requires small culture
media and other reagents. Moreover PDMS has bio-
compatible and gas permeability. By using PDMS
device, we could fabricate micro cell stimulation
system with a small amount of culture media and
other reagents (Kim et al., 2008). Various methods
for physical cell stimulation such as fluidic shear
stress (Brown., 2008), pneumatic pressure (Sim,
2006), microfluidic motion (Pioletti, 2003), and etc.
have developed. In light of this perspective; we
expect that micro-beads can be one of effective
sources for physical stimulation to improve cell
growth.
Now, we demonstrate a novle stimulation
method using micro-bead impact and microfluidic
device. The mechano-stimulation by bead impact
will be one of the efficient stimulation methods.
2 MATERIAL AND METHOD
2.1 Fabrication of Stimulation Device
Various experiments were conducted in order to find
an optimal condition for cell culture in microfludic
culture chambers, and the first step is to determine
the appropriate method to fabricate devices. Our
choice was to use soft lithography and PDMS to
easily acquire the required microfluidic devices.
A set of single straight channels were created on
a Silicon wafer using negative photoresist Su-8
mold. After treating a bare silicon wafer with
acetone, methanol and DI (deionized) water,
respectively, Su-8 2050 (MicroChem, MA, USA)
negative photoresist was spin-coated at 1,300 rpm
for 30 seconds to acquire a channel height of 100
m. After the soft baking process, the wafer was
exposed under UV light followed by additional heat
treatment for post exposure baking process. The
wafer was etched using Su-8 developer and was
426

rinsed using iso-propyl alcohol (Figure 1).

Figure 1: Soft lithography fabrication process using Su-8
photoresist.
2.2 Cell Culture
CPAE (calf pulmonary artery endothelial cell) was
selected to investigate the change in growth rate
when exposed to micro bead impact. These cells are
thought to be exposed to blood cells impact in a
blood vessel. It can be thus, mimicry of blood cells
impact that micro beads collide against the surface
of CPAE. Sterilization was performed under UV for
24 hour 70% ethanol. The cells were seeded into the
microfluidic chamber with a concentration of
1.65x10
6
/ml and were incubated overnight. Before
experiments, starvation process was conducted with
0.5% FBS RPMI media for 24 hours to fix cell phase
in G1.
2.3 Experimental Setup
In previous work, two types of cell lines, HeLa cell
and MC3T3 cell, were selected to investigate the
changes in growth rate when exposed to micro-bead
impact. The cells were seeded into the microfluidic
chamber with a concentration of 5 x 10
5
cells/ml and
were incubated overnight. Fresh DMEM was
supplied in HeLa cell cultured chambers and aMEM
in MC3T3 cell cultured chambers every 12 h. Since
the concentration of 11m micro-beads in culture
medium is 10
6
/ml, and the flow rate is 3l/min, the
number of 11m micro-beads passing through a
single straight cell culture chamber per minute is
calculated as 3 x 10
3
beads/min (Figure. 2)
However, this simulation system was designed
except important aspects. It had two mechanical
stimulus factors, flow and bead impact and it is not
considered about cell cycle which the series of
events that take place in a cell leading to its
replication. Therefore we developed idea to
minimize other stimulus sours and maximize bead
impact to cells. Also we performed experiment with
considering cell cycle to find optimal stimulation
time and to get effect of the number of micro-bead
rolling in cell growth.
Sixteen micro devices have been fabricated
because cell cycle is about 16 hour and each device
has 10 cell culture chamber which dimensions are
height = 100m, width = 40m, length = 80m
(Figure 3). There were two inlets, one for cell
seeding and media and other for beads. The
concentration of 10 m glass beads is 1.9x10
5
/ml.
Before tilting device, micro beads were gathered on
inlet for bead. Slating process was performed for 10
minutes in incubator.











Figure 2: Photograph and schematic diagram of previous
work about cell stimulation.

Figure 3: Schematic diagram and photograph of device.
MICROFLUIDIC CELL STIMULATOR USING BEAD IMPACT
427

3 RESULT
The micro-bead and cell culture medium mixture
was introduced into the microfluidic chamber using
syringe pump infusion mode, and deflections in the
micro-bead pathway were observed as they flowed
over or by the adhered cells. This motion indicates
that micro-beads flow along the adhered cells
boundary layer and the outer membrane of the cells
are stimulated through this manner. The cell
population of HeLa and MC3T3 cells under different
micro-bead stimulation was counted through
microscopic observation at a 12-h interval, and the
proliferation rate was then achieved by dividing the
cell number in each time interval by the initial
population. The cell increase rate of HeLa cells in
the experimental chamber with 11-lm micro-bead
stimulation performed the highest growth rate,
whereas 2m micro-bead stimulation performed the
lowest growth rate. For the case of MC3T3 cells, the
cell increase rate of the 2m micro-bead stimulation
device was significantly higher than that of the 11-
lm micro-bead stimulation device, while the control
group performed the lowest stimulation rate (Fig. 4).
Cell growth was measured in 16 type device:
control device and other related with each
stimulation time. The cell population of HeLa and
MC3T3 cells in the control and experimental
chambers was observed through an inverted optical
microscope (Olympus, J apan) and compared after 16
hours. All fluids were supplied by capillary force to
remove effect of flow. Figure 5 shows increase of
CPAE cell according to the stimulation time. The
proliferation rate (Figure 6) was then achieved by
dividing the cell number in each time interval by the
initial population.
As mentioned above, cells are divided by cell
cycle which consists of G1, S, G2, and M phase.
First, G1 phase is marked by synthesis of various
enzymes that are required in S phase, mainly those
needed for DNA replication. Next, S phase is related
with DNA synthesis. Third, G2 phase significant
protein synthesis occurs which are required during
the process of mitosis. Finally in M phase the cell is
divided by two cells.
According to figure 6, we could conclude that
micro bead stimulation was more effective at 1 hour
data than other stimulation time. We analyzed that 1
hour stimulation time was involved in G1 phase
because cells were fixed by starvation process. G1
phase takes part in synthesis of diverse proteins
about DNA synthesis. Therefore mechanical
stimulation by rolling micro bead plays an important
role to proliferation of cells. Following this data, we
performed other experiment to verify effects of the
number of rolling micro-beads. Figure 7 shows that
more frequent micro-beads stimulation could be
better effect in cell growth.

Figure 4: Cell number increase comparison of control
group. : (a) HeLa cells and (b) MC3T3 cells.

Figure 5: Photograph of CPAE cell in micro chamber : (a)
control device, (b) device at 1 hour stimulation time,(c)
device at 15 hour stimulation time.

Figure 6: Comparison of growth rate of CPAE cells under
different stimulation time: blue bar for data at 0 hour and
yellow bar for data at 16 hour.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
428


Figure 7: Comparison of growth rates of CPAE cells under
different the number of beads rolling.
4 CONCLUSIONS
A new method for stimulating cells has been
introduced. Previous our work was conducted with
two types of polystyrene micro-beads: 2 and 11m
micro-beads. The cell population increase rate was
the highest when stimulated by 11 m beads in the
case of HeLa cells and 2 m beads in the case of
MC3T3 cells. From the comparison between control
device and experimental device, it can be seen that
cells in the experimental group performs at a higher
growth rate. This suggests that the growth rate is
accelerated when stimulated by micro-beads. We
designed new device to maximize micro bead
stimulation and minimize other factors: beads were
rolled by only gravitational force. Thus, we get data
about more effective simulation time, 1 hour which
is involved in G1 phase and cells exposed under
micro bead impact represent better proliferation.
Moreover, the number of bead impact is one of the
important factors in cell stimulation and more
frequent impacts show better effects to the cells.
When our new devices which were designed to
maximize beads impact was compared with previous
our device, cell proliferation data by only beads
impact represent similar increase rate (about
10%~40%). Therefore we could conclude that
micro-beads stimulation can be one of the effective
physical stimulation to enhance cell proliferation.
Now, we are conducting additional experiments
to convincing our experiments including cell
viability test and we will perform other analyzing
test to compare with other research data such as
protein analysis.
ACKNOWLEDGEMENTS
This work was supported by National Core Research
Center (NCRC) for Nanomedical Technology of the
Korea Science & Engineering Foundation (Grant no.
R15-2004-024-01001-0), Seoul Research &
Business Development (R&BD Program, 11128)
and Korea Research Foundation Grant funded by the
Korean Government (MOEHRD) (KRF-2007-313-
D00073).
REFERENCES
Matteucci M, 2007, Preliminary Study of
Micromechanical Stress delivery for Cell Biology
studies. Microelectron, Eng 84:1729173
Nakashima Y, 2005, Microfluidic device for axonal
elongation control, Transducers, Seoul, Korea, 59
J une
Mattei MD, 2004, Effects of Physical Stimulation with
Electromagnetic Field and Insulin growth factor-I
treatment on proteoglycan synthesis of bovine
articular cartilage. Osteoarthritis Cartilage 12:793
800
Tae-J in Kim, Su-J in Kim, Hyo-Il J ung, 2008, Physical
stimulation of mammalian cells using micro-bead
impact within a microfluidic environment to enhance
growth rate, Microfluid Nanofluid on-line published
Brown TD, 2000, Techniques for mechanical stimulation
of cell in vitro: a review, J Biomech, 33: 3-14
Sim WY, 2006, A pneumatic micro cell simulator for the
differentiation of human mesenchymal stem cell
(hMSCs), MicroTAS, Tokyo, J apan,5-9
Pioletti DP, 2003, Effect of micromechanical stimulations
on osteoblast: development of a device stimulating the
mechanical situation at the bone-implant interface, J
Biomech 36: 131-135
MICROFLUIDIC CELL STIMULATOR USING BEAD IMPACT
429
A LOW-COST EEG STAND-ALONE DEVICE FOR BRAIN
COMPUTER INTERFACE
Alexandre Ribeiro, Ant onio Sirgado, Jo ao Aperta, Ana Lopes
Jorge Guilherme, Pedro Correia, Gabriel Pires and Urbano Nunes
Department of Electrical Engineering, Polytechnic Institute of Tomar, Tomar, Portugal
[email protected]
Institute for Systems and Robotics, University of Coimbra, Coimbra, Portugal
Keywords:
EEG, Stand-alone device, BCI, SSVEP.
Abstract:
This paper describes the ongoing development and design of a portable and stand-alone EEG device to be
used as a Brain Computer Interface (BCI) for wheelchair steering. The overall system comprises the EEG
(electroencephalogram) amplier, data acquisition, communications and stimuli generator. The main attrac-
tive feature of this system is its stand-alone operation which makes it independent of a computer for signal
processing and visual stimuli presentation. Preliminary results obtained from experiments exploring alpha
rhythms and steady state visual evoked potentials attest the overall functionality and robustness of the system.
1 INTRODUCTION
EEG portable devices are mainly used for long term
signal recording or for remote real-time monitoring
(Jiang and Wang, 2005). They are usually small sized
and low-cost and therefore suitable for a daily using.
EEG-based BCI systems are however almost always
dependent of a personal computer (PC) and its screen.
This is mostly because they rely on real-time heavy
processing algorithms and because a computer screen
is needed to provide visual feedback information or
visual stimuli presentation.
Current main non-invasive BCI systems based on
EEG data are divided in three main classes according
to the type of neuromechanisms: event related syn-
chronization and desynchronization (ERD/ERS) of
and rhythms usually associated with motor imagery
(Pfurtscheller et al., 1998), P300 evoked potentials
(Donchin et al., 2000), and nally, steady-state visual
evoked potentials (SSVEP) (Gao et al., 2003). The
rst approach is usually used for 1D or 2D cursor con-
trol on a computer screen to select targets or letters (as
a spelling device), or for playing simple games. The
large number of required EEG channels and the heavy
processing needed for EEG pre-processing, feature
extraction and classication make this BCI approach
highly dependent on a computer system. The visual
P300-based BCI approach depends on the user at-
tentional focus to an infrequent specic visual stim-
ulus that occurs among other random stimuli (Pires
et al., 2008). The stimuli can be generated with
LEDs or with a low cost display controlled by a mi-
crocontroller, without resorting to a computer and
screen. Assuming that many EEG epochs are aver-
aged, then low computational demanding feature ex-
traction techniques can be implemented on a micro-
controller or DSP. However, the use of a large number
of epochs slows down the BCI transfer rate and it re-
quires a buffer of several seconds. In the last BCI ap-
proach, the SSVEP results from a stimulus ickering
at a constant frequency. The icker stimuli are eas-
ily implemented with some LEDs, and signal feature
extraction and classication can rely essentially on a
FFT (fast fourier transform) that can be computed on
a simple DSP.
Therefore, the SSVEP BCI approach is the only
one that can be easily implemented on a stand-alone
device. Some works have already been developed fol-
lowing a similar approach. The work presented in
(Gao et al., 2003) describes a stand-alone device DSP-
based. A 48 LED stimuli generator is used to control
external devices (TV, videos, etc.) through the clas-
sication of SSVEP signals recorded at the occipital
brain region. The LEDs are ickering at frequencies
between 6 and 15 Hz with a 0.2 Hz resolution.
The use of a stand-alone device is of particular
importance for us since it is intended to be applied
as a BCI to steer a wheelchair (Pires and Nunes,
430
2002). Therefore, computer independence, portabil-
ity and low power consumption are required criteria.
BCIs based on SSVEP are not considered true BCIs
because the users have to gaze the specic stimulus
and therefore have to move the eyes. Notwithstanding
this, the interface can be suitable for people with se-
vere motor disabilities and unable to control standard
interfaces such as joysticks, head switches, voice or
eye trackers, but still able to perform small eye move-
ments.
The remaining sections of the paper describe the
design and implementation of the system still under
development, and also some preliminary results al-
ready reached.
2 MODULE ARCHITECTURE
The diagram on Figure 1 illustrates the proposed ar-
chitecture. The rst module is a 6 channel EEG
amplier that receives EEG signal picked from non-
invasive electrodes. The output signal is in the range
of 0 5V. The data acquisition module is a PIC
18F4550 (Microchip, 2008), which was congured to
sample the signal at a 512 Hz rate and was imple-
mented based on a real-time interrupt. The A/D con-
verter has a 10 bit resolution. The digital signal is
then sent via USB to a PC for data recording, mon-
itoring or real-time signal processing. If working on
stand-alone mode, digital data is sent via SPI to the
dsPIC 30F3013. The dsPIC is able to perform some
statistical computation, digital ltering and FFT com-
putation to extract signal features for classication.
The output from decoded patterns can be directly sent
to a control unit on the wheelchair. The last module is
a stimuli generator implemented on the PIC 18F458.
It controls a set of LEDs that can be activated sequen-
tially, randomly, or at a given frequency (Figure 5).
The module is therefore suitable to generate stimuli
for SSVEP and for P300. In the case of P300 opera-
tion, each stimulus code is sent to the dsPICvia SPI or
to the PC via USB or RS232. Table 1 presents some
of the main specications of the system.
On the present development stage, only one am-
plier channel was implemented. The signal process-
ing algorithms are not running at the dsPIC, but only
at the PC after signal recording.
2.1 Amplier
EEGsignals are characterized by having very lowam-
plitude values, typically in the range of 5-100 V, re-
quiring a very accurate conditioning in order to am-
plify input signals and reject the existing noise or in-
Figure 1: System architecture.
Table 1: Overall system specications.
Number of channels 6
Resolution 10 bit
Min input voltage step detect 195nV
Input voltage full scale 200 V
Input frequency range 0.25-45 Hz
CMRR 100-120 dB
Noise spectral density 339 nV/

(Hz)
Min and Max Gain 66-91 dB
Current per channel 6 mA
terferences. A high input impedance of the condition-
ing module avoid signal distortion due to loading ef-
fects.
An EEG amplier module was developed for
signal conditioning and amplication. The mod-
ule design follows the OpenEEG project approach
(OpenEEG-Project, 2008). The main difference was
the insertion of a notch lter to reject the 50 Hz from
power line interference. This improvement made pos-
sible the use of standard electrodes, whilst the Ope-
nEEGproject needs non-standard shielded electrodes.
The amplier module is shown in Figure 2.
The EEG signals acquired from the electrodes pass
through the protection circuit who acts as a voltage
and current limiter. The instrumentation amplier
with 28dB of gain is implemented with the INA128P
circuit that is characterized by having low offset volt-
age, typically 50 V, and high Common Mode Re-
jection Ratio (CMRR) (between 100-120 dB). One
second-order high-pass lter with cut-off frequency
at 0.16 Hz and gain 15 to 40dB follows the INA128P
in order to increase the input signal amplitude and
to reject the DC component imposed by the ampli-
er offset voltage. Then, the signal passes through
an anti-aliasing lter that is a third-order Bessel low-
pass lter with cut-off frequency of 45Hz and gain
22.4dB. In order to reject the 50Hz component, the
A LOW-COST EEG STAND-ALONE DEVICE FOR BRAIN COMPUTER INTERFACE
431
Figure 2: Block diagram of EEG Amplier.
Figure 3: Notch lter circuit.
EEG amplier stage includes one notch lter imple-
mented using the GIC (Generalized Immittance Con-
verter) topology with adjustable central frequency, f
0
,
varying between 44Hz and 58Hz and 50dB of atten-
uation at f
0
. The notch lter circuit implementation
and EEG amplier frequency response simulation are
shown in Figures 3 and 4. The signal conditioning
system has a noise spectral density of 339nV/

(Hz),
which achieves a signal-to-noise ratio higher than the
maximum required by the A/D converter of 61.96 dB
for a N = 10 bit converter (Eq. 1).
SNR = 6.02N+1.76dB (1)
The circuit dynamic range has to take into account the
signal dynamics of a crest factor of about 6dB plus
the required SNR of the ADC. Therefore the required
circuit minimum dynamic range is 67.96dB. The cir-
cuit SNR, considering a noise bandwidth of 256 Hz is
about 120dB which is more than enough for the appli-
cation. The sampling frequency of the ADC was set
to 512Hz that is a decade above the aliasing lter cut
off frequency. The lter attenuation at half the sam-
pling frequency is not critical since the type of signal
detected by the sensor do not fall in this frequency
zone.
Figure 4: Amplier gain obtained from PSPICE simulation.
3 EXPERIMENTS AND RESULTS
Two experiments were performed with two subjects
to evaluate the performance of the system. The EEG
activity was recorded from Ag/Cl electrodes at posi-
tion Oz according to the internacional 10-20 standard
system (Figure 5). The electrode was referenced to
the right mastoid and the ground was placed at AFz.
The cap and electrodes are g.tec products.
The rst experiment consisted on the analysis of
the alpha rhythms (typically in the range 8-12 Hz).
It is known that there is an increase of the amplitude
of alpha rhythms when the eyes are closed and a de-
crease when the eyes are open (Feige et al., 2005).
Figure 6 (top) presents the recorded signal, bandpass
ltered between 8 and 12 Hz, during the two situ-
ations. It was asked the user to be relaxed and to
close the eyes and open them (without any kind of vi-
sual stimuli in front) alternately during some seconds.
During the open-eyes period there is a clear decrease
of alpha amplitude, and an increase during the closed-
eyes period. In Figure 6 (bottom), the FFT of the non-
ltered recorded signal shows a peak at the 10 Hz
when the eyes are closed. To compare and evaluate
the results, similar experiments were performed with
a certied commercial amplier, reaching the same
results.
The second experiment was performed to validate
simultaneously the stimuli generator and the SSVEP
BCI approach. The user was asked to gaze alternately
two LEDs stimuli blinking at frequencies 15 Hz and
16 Hz. Figure 7 shows the SSVEP signals recorded
at Oz channel. The two frequencies were clearly dis-
criminated. The plot correponds to the application of
the FFT to a time window of 1 s (1 Hz resolution).
These results attest the validity of the overall system
and also of the SSVEP BCI approach.
4 CONCLUSIONS
This paper describes the implementation of a low cost
EEG system for BCI. The amplier, digital data ac-
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
432
Figure 5: Subject with electrodes cap and stimuli presenta-
tion.
Figure 6: Top) Alpha rhythms recorded with open eyes vs.
close eyes; Bottom) FFT of recorded signal with open eyes
vs. close eyes.
quisition and USB communications were already im-
plemented, tested, and validated. A full stand-alone
operation was already achieved, through the connec-
tion of the dsPIC via SPI, even though signal pro-
cessing algorithms are still in the early steps of de-
velopment. Also, the batteries supply system is cur-
rently under development. The proposed SSVEP BCI
was also tested reaching 15 and 16 Hz discrimination
applying a FFT. For comparing purposes the experi-
ments were also performed with a commercial EEG
system. The results showed a similar performance.
ACKNOWLEDGEMENTS
This work has been in part supported by Fundac ao
para a Ci encia e Tecnologia (FCT), under Grants
POSC/EEASRI/58279/2004 and PTDC/EEA-
ACR/72226/2006.
Figure 7: FFT of the SSVEP signal for a 15 Hz and 16 Hz
stimuli.
REFERENCES
Donchin, E., Spencer, K., and R., W. (2000). The mental
prosthesis: Asensing the speed of a p300-based brain-
computer interface. IEEE Trans. on Rehabilition Eng.,
8(2):174179.
Feige, B., Schefer, K., Esposito, F., Salle, F. D., Hennig, J.,
and Seifritz, E. (2005). Cortical and subcortical corre-
lates of electroencephalographic alpha rhythm modu-
lation. Journal of Neurophysiology, 93:28642872.
Gao, X., Xu, D., Cheng, M., and Gao, S. (2003). A bci-
based environmental controller for the motion dis-
abled. IEEE transactions on Neural Systems and Re-
habilitation Engineering, 11(2):137140.
Jiang, X. and Wang, X. (2005). Development of ultra small
two-channel system of eeg radio telemetry. First Int.
Conf. on Neural Interface and Control, pages 6063.
Microchip (2008). http://www.microchip.com/.
OpenEEG-Project (2008). http://openeeg.sourceforge.net/.
Pfurtscheller, G., Christa, N., Scholgl, A., and Lugger, K.
(1998). Separability of eeg signals recorded during
right and left motor imagery using adaptive autore-
gressive parameters. IEEE Trans. on Rehab. Engineer-
ing, 6(3):316324.
Pires, G. and Nunes, U. (2002). A wheelchair steered
through voice commands and assisted by a reactive
fuzzy logic controller. Journal of Intelligent and
Robotic Systems, 34(3):301314.
Pires, G., Nunes, U., and Castelo-Branco, M. (2008). Use of
a P300-based BCI to Steer a Wheelchair: a Bayesian
Approach. In 30th Annual Int. Conf. of the IEEE Engi-
neering in Medicine and Biology Society - EMBC08,
pages 658661.
A LOW-COST EEG STAND-ALONE DEVICE FOR BRAIN COMPUTER INTERFACE
433
SPECIAL SESSION ON
ACTIVE MATERIALS
FOR MEDICAL DEVICES

CHAIR:
ANDRES DIAZ LANTADA

CHARACTERISATION AND MEDICAL APPLICATIONS OF
MAGNETORHEOLOGICAL FLUIDS
J avier Echavarri Otero, Andrs Daz Lantada, Pilar Lafont Morgado, J uan Manuel Munoz-Guijosa
J os Luis Muoz Sanz, Hctor Lorenzo-Yustos and J ulio Muoz-Garca
Grupo de Investigacin en Ingeniera de Mquinas, E.T.S.I. Industriales, Universidad Politcnica de Madrid
c/ Jos Gutirrez Abascal, n 2. 28006, Madrid, Spain
[email protected]
Keywords: Magnetorheology, Damper, Characterisation, Application, Medical.
Abstract: Magnetorheological fluids are composed by magnetic particles suspended within a carrier fluid which allow
to change their rheology when subjected to a magnetic field. This property can be controlled very accurately
by varying the magnetic field intensity which provides a rapid and effective system to control the response
of very diverse fluid-mechanical devices.The characterisation of two new magnethoreological fluids is
presented. The force response when the magnetic field varies is studied in static and dynamic
conditions.This paper sets out some of the opportunities and advantages provided by this fluids both for
medical and non-medical applications.
1 INTRODUCTION TO THE
MAGNETORHEOLOGICAL
FLUIDS
Magnetorheological fluids (MR-fluids) are capable
to change their rheology when subjected to a
magnetic field. This property can be controlled very
accurately by varying the magnetic field intensity,
which provides a rapid and effective system to
control the response of very diverse fluid-
mechanical devices.
These kinds of fluids are composed by magnetic
particles suspended within a carrier fluid. Thus, they
are typically composed by micrometer or nanometer
scale spheres distributed in a mineral oil.
The response of the magnetorheological fluids is
the result of the induced distribution of the magnetic
suspension particles (usually in the 0.1-10 m range)
within the fluid when an external magnetic field is
applied (usually about 100 mT in air). The
microscopic particles align themselves along the
lines of magnetic flux. Thus, the resulting chains of
particles restrict the movement perpendicular to the
direction of the lines of flux, leading to an
anisotropic behaviour and to an increase in the
viscosity of the fluid (Torcuato, 2006).
This phenomenon is considered completely
reversible and almost instantaneous (response time
is about 10 milliseconds) for these fluids.
The magnetically controllable properties of
magnetorheological fluids depend on the carrier
fluid and its additives, the magnetic field applied,
the working condition and the size, shape,
concentration, density and distribution of the
particles.
The first patent related to MR-technology was
issued to inventor J acob Rabinow in the 1940s,
though this technology remained with little practical
use. However, in the last twenty years there is a
serious research in this field, especially when other
technologies began to converge that made their use
practical and a real possibility.
Numerous research efforts have been carried out
in order to improve the magnetorheological fluids, in
specialised companies like Lord, Repsol-YPF and in
different universities.
Although MR-fluids have many potential
applications, they are limited by aspects like their
high density and wear in materials due to presence
of iron, very high cost or precipitation of particles.
437

2 CHARACTERISATION OF NEW
MAGNETORHEOLOGICAL
FLUIDS
The Machines Engineering Division (DIM)
belonging to the Technical University of Madrid
(UPM) has collaborated with the company Repsol-
YPF and the Applied Physics Department of the
University of Granada in the development and
characterisation of new magnetorheological fluids
for shock-absorbers used for different applications.
Two oils have been developed: MR-Lub1 and
MR-Lub2. The carrier fluid is the same for both of
them but the second one includes a higher
percentage of magnetic particles. The cited
characterisation was carried out in a commercial
damper mounted in a MTS damper test system
shown in figure 1.

Figure 1: MTS Damper Test System.
The first characterisation includes an static test
(velocity of the damper below 1mm/s) to determine
the effect of the electric intensity (which produces a
magnetic field) in the increase of the damping force
in the damper.
The results for MR-Lub1 are depicted in figure
2, which shows a non-linear behaviour.

Figure 2: Damping force vs. Intensity for MR-Lub1.
The reversible magnetorheological behaviour is
shown in figure 3, obtained for a of the damper at a
velocity of 130mm/s and a stroke of 25mm. The
traction and compression cycles are depicted for the
MR-Lub1. The force obtained at each point when
the intensity is growing is very similar to the force
measured when intensity is decreasing.
Nevertheless, the results show a slight remaining
magnetisation when low intensities around 1A are
applied.

Figure 3: Static force vs. Intensity for MR-Lub1 in traction
and compression cycles.
Figure 4 shows a comparison between dynamic
tests in the shock absorber with MR-Lub1, MR-
Lub2 and two commercial fluids by Lord at a
velocity of 1m/s. The higher quantity of magnetic
particles in MR-Lub2 than in MR-Lub1 leads to an
increase in the sum of the traction (T) and
compression (C) forces. A similar Force-intensity
behaviour is observed in all the tests performed.

Figure 4: Dynamic force vs. Intensity.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
438

3 POSSIBLE NON-MEDICAL
APPLICATIONS FOR
MR-FLUIDS
Nowadays, the main manufacturers of
Magnetorheological fluids (Lord, 2008) offer
specific solutions for many applications mainly
within the following sectors:
-Suspension systems
-Fan Clutches
-Crash-protection systems
-Buildings and bridges
-Medical applications
In the near future many other applications are
expected to be available, because their possibilities
are almost endless.
The most extended application for the fluids
developed is the use in a magnetorheological
suspension, where a controllable fluid replaces
traditional hydraulic oil in each shock absorber. As
sensors monitor road and vehicle conditions, a
controller modifies the intensity applied and
therefore the damping characteristics are adjusted in
real-time. This enables remarkable improvements in
both ride comfort and handling.
Magnetorheology has had a large role in recent
advances in military and automotive industries. The
called MR-Technology has been integrated not only
in military tactical and combat vehicles but also in
primary suspension systems of high performance
vehicles designed by Cadillac, Audi, Ferrari, Honda
and others. In addition, these new technologies are
used in seat and cab suspension designs of
agricultural and other off-highway vehicles in order
to improve operation environment.
Apart from the use for suspension applications,
there are a lot of patents concerning MR-fluids for
fan clutches. The main aims of these inventions are
to provide smooth, efficient torque-transfer in clutch
devices. The designs include a large range of
applications in order to improve the controllability
and the high off-state drag.
Another present application for MR-fluids is
obtained by combining the variable control
magnetorheology with advanced sensors in
passenger protection systems. In these cases, the
protection systems can be adjusted to provide the
perfect resistance based on the impact severity and
the passenger size.
There is also a civil and structural application in
skyscrapers and long bridges protection, which are
susceptible to vibrations induced by high winds and
seismic activity. In order to mitigate their effect,
large dampers are built into their design, which
protect against shocks through a continuously
controllable and cost-effective solution. Figure 5
shows a Rheonetic Seismic Damper by Lord.

Figure 5: Example of a Rheonetic Seismic Damper,
obtained from Lords website.
4 SOME POSSIBLE MEDICAL
APPLICATIONS FOR
MR-FLUIDS
The main medical applications for MR-fluids are
considered haptic devices, new prothesis
development and innovative cancer therapies.
4.1 Haptic Devices
Tactile and force feedback systems are used in
surgery training assisted by computer. Active
dampers have been developed based on MR-fluids
and their utilization in haptic systems provides a
very precise control which enhances the skills of the
surgeons (Bar-Cohen, Mavroidis et al., 2001;
Neelakantan, 2002; Rizzo, 2007).
In particular, by using sensorized surgical
instruments, suitable signals could be acquired and
used for controlling the haptic device. The surgeons
could use the surgical instruments to interact with
biological tissues and organs during a simulated
operation and, probe their compliance by touching
the haptic device (Scilingo, Sgambelluri et al, 2003).
4.2 Prosthetic and Rehabilitation
The company Lord, in collaboration with
Biedermann Motech, manufacturer of prosthetic
components, has developed a device that improves
the mobility of leg amputees. The new design,
shown in figure 6, is based in MR-dampers and
produces an increase in gait balance, stability and
energy efficiency.
CHARACTERISATION AND MEDICAL APPLICATIONS OF MAGNETORHEOLOGICAL FLUIDS
439


Figure 6: New prosthesis for leg amputees, obtained from
Lords website.
Other medical systems based on similar active
control elements are used in training devices for
rehabilitation purposes which allow an adjustable
resistance to the movement according to the
evolution of the patient (Dong, 2005).
4.3 Cancer Therapy
The biocompatibility and tolerability by humans and
animals to magnetorheological materials have been
tested successfully (Sheng, Flores and Liu, 1999).
Some techniques developed for cancer therapies
target chemotherapeutic agents to the tumor sites
employing magnetic nanoparticles as carriers is a
promising cancer treatment reducing side effects of
conventional chemotherapy.
Electron microscope investigations show that the
ferrofluids can be enriched in tumour tissue and
tumour cells (Alexiou, 2006). Therefore, these
studies show a remarkable improvement to the
conventional cancer therapy but still causes toxicity
to the body.
Other interesting studies are based in MR-fluids
introduced into the blood vessels supplying the
tumor. When a magnetic field is applied a seal is
formed, which blocks the blood flow and cut the
alimentation in oxygen, leading to tumor necrosis.
There are interesting in-vitro investigations on these
therapies (Sheng, Flores and Liu, 1999) concerning
the study of the seal kinetics and pressure
resistibility of the formed seal.
5 CONCLUSIONS
In this paper the characterisation of two new
magnethoreological fluids in a damper is presented.
The force response when the magnetic field varies is
studied both in static and dynamic conditions. The
remaining magnetisation when the external field
stops is analysed.
Different possible applications for the two fluids
developed are indicated for medical or non medical
purposes. In the future, the new possibilities of the
magnetorheological fluids are incalculable.
REFERENCES
Alexiou C. et al.; Targeting cancer cells: magnetic
nanoparticles as drug carriers, Eur. Biophys J, 2006.
Bar-Cohen, J ., Mavroidis, C. et al..- Virtual reality
robotic telesurgery simulations using MEMICA haptic
system. Proceedings of SPIEs 8th Annual
International Symposium on Smart Structures and
Materials, 2001.
Dong, S. Rehabilitation device with variable resistance
and intelligent control. Medical Engineering and
Physics, 2005.
Flores, G., Sheng, R. et al..-Medical applications of
magnetorheological fluids: A possible new cancer
therapy. 7th International Conference on
Electrorheological (ER) Fluids and
Magnetorheological (MR) Suspensions, 1999.
Neelakantan, V. et al.., Force feedback system using
magnetorheological fluids for telerobotic surgery.
International Congress on Industrial and commercial
applications of smart structures technologies, 2002.
Rabinow J ., The magnetic fluid Clutch, AIEE Trans 67,
1948.
Rizzo, R., Electromagnetic Modeling and Design of
Haptic Interface Prototypes Based on
Magnetorheological Fluids. IEEE Transactions on
Magnetics, 2007.
Scilingo E.P., Sgambelluri N., De Rossi D., Bicchi A.,
Haptic displays based on MR- uids: design,
realization and psychophysical validation,
Proceedings of the11th Symposium on Haptic
Interfaces for Virtual Environment and Teleoperator
Systems, 2003
Sheng, Flores and Liu, In vitro investigation of a novel
cancer therapeutic method using embolizing properties
of magnetorheological fluids, Journal of Magnetism
and Magnetic Materials 194, 1999.
Torcuato M., Estudio de suspensiones
magnetorreolgicas en medios no acuosos. Master
Thesis, 2006.
Website: Lord, 2008. Web-site: www.lord.com.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
440
COLLIMATION OF X-RAY DIAGNOSTIC BUNDLE BY MEANS OF
STEERING FERROFLUID
Andrzej Dyszkiewicz
1,2,3
, Pawe Poe c
1,3
, Jakub Zajdel
1,3
, Bartomiej Pawlus
3
Damian Chachulski
1,3
and Pawe Kpi nski
1,3
1
Laboratory of Biotechnology Cieszyn ul.Go zdzik ow, 2, Poland
2
Technical University of Opole, Faculty of Physical Education and Physiotherapy, Poland
3
Specialist Rehabilitation Department VIS Cieszyn ul. Bielska 3A, Poland
Keywords:
Incoherence x-ray beam, Collimation of x-ray beam, Ferrouids, Computer steering.
Abstract:
The study addressed the problem of the incoherence of an X-ray tube generated bundle. In spite of the large
progress made in the area of the durability and efciency of these devices, the problem of the inherent diver-
gence of the bundle hasnt yet been resolved in a satisfying way. In the face of difculties with concentrating
X-rays, the only practical way of eliminating non-axial rays is by applying lead collimators of a suitable length
and diameter as well as mechanical systems of movable distracting grids. An entirely new approach has been
the use of electro-magnetorheological uids as lters, which are capable of modifying their physical and quan-
tum properties contingently based on the parameters of the applied external electric current, centrifugal force
and magnetic eld. The experiments carried out showed, that the described changes also concern the absorb-
tion, distraction or change in the direction of the X-ray beam. In the constructed prototype device, interesting
and recurrent results in the modication or eleimination of non-axial rays were obtained, which resulted in the
improvement of the quality of bone images. The following research questions were posed: (1) are changes
in the photodensitometric parameters of joint (PIP) X-ray images containing soft and osseous tissues proof
of alterations of the physical parameters of a diagnostic X-ray beam? (2) is there a relationship between the
volume of ferrouid in a lens and its capability to alter the parameters of a permeating X-ray beam? (3) is there
a relationship between the r.p.m. of the collimator and its capability to alter the parameters of a permeating
X-ray beam, and does it have a linear character? Based on the conducted research a non-linear dependancy
was observed between the volume of ferrouid in the lens as well as its rotational speed and the capability of
the system to alter the parameters of the permeating X-ray beam.
1 INTRODUCTION
WilliamRoentgens pioneer works began a completely
new chapter in the history of medical diagnostics, en-
abling small invasive investigations of internal organs,
especially the osseous tissue. Progress in the eld
of electronics resulted in X-ray tubes becoming in-
creasingly more efcient, reliable and the repeatabil-
ity of emission in relation to a given electric param-
eter was largely improved (Bankier et al., Carlson,
Dyszkiewics, 2001, Dyszkiewics, Folster, Grampp et
al., 1997b, awniczak and Milecki, 1999). From
the dawn of the development of radio-photographic
technology, beginning with tubes with a motionless,
and later rotating and multi-focal anode, constructors
struggled with the problem of the originally divergent
diagnostic bundle, the geometry of which results from
the way an X-ray beam is formed as a consequence
of breaking the stream of accelerated electrons on the
anode. A divergent bundle produces geometrical dis-
tortions and leads to a considerable worsening of the
acutance of the photo. A large quantity of non-axial
rays is the reason for this phenomenon. Constructors
of X-ray tubes have been trying to achieve a larger co-
herence of the bundle for many years. As a result, the
use of collimators began, that is, thick-walled tunnels,
made of materials of a large thickness, able to absorb
non-axial rays. Applying tubes with a multi-focal
anode was another solution, which enabled a better
adaptation of the bundle geometry to the distance of
a photographed object. Still another solution was the
use of stable lters or oscillating metal grates placed
above the lm, distracting non-axial rays (Czerny et
al., 1997, Grampp et al., 1997a, Kainberger, et al.,
1997, Kollmann, et al., 1997, Kramer et al., 1997). In
spite of the results achieved in improving the quality
of X-ray images, it can be clearly observed that these
activities lead exclusively to the removal of part of the
441
rays from the bundle by using materials of large den-
sity, which only increase the difference between the
real and effective power of the tube and have a mea-
surable inuence on the weight and durability of the
device (Dyszkiewicz, 1999, Dyszkiewicz, 2001, Ma-
jumdar et al., 1997, Phule, Rand et al., 1997, Sasaki).
One new innovative solution used by scientists in
Geneva has been the application of ferrouids with
variable physical and chemical parameters, which, af-
ter being introduced into capillary tubes placed in an
X-ray beam axis allow the parameters of the beam to
be changed contingently.
1.1 Ferrouids
Ferrouids display the characteristics of both uids
and magnetic substances at a very wide temperature
range. With the absence of an external magnetic
eld they behave like normal newtonian uids, where
the shear stress is in direct proportion to the shear
velocity without displaying any signs of magnetisa-
tion. The viscosity of controlled ferrouids can be
adapted within a range of 5 - 25 000 cP. Analogi-
cally to electrorheological uids they are built of two
basic components: carrier uid (<85% ) and ferro-
magnetic particles covered with a surface layer (from
2% to 15% , for ferrouids, up to 80% for MRF).
The carrier uid is a non-magnetic substance; how-
ever the components of the particles of the suspen-
sion are solid soft magnetic materials which form mi-
cromagnets [27]. Removing the action of Van der
Waals forces and magnetic attraction forces (group-
ing) makes it possible to cover the particles with a
surface active agent (e.g. oleic acid). Depending on
the size of the particles two types of ferrouids can
be distinguished: (1) nanomagnetorheological also
called ferromagnetic; (2) magnetorheological uids
MRF. In ferromagnetic uids the diameter of the par-
ticles (most commonly Fe
3
O
4
) ranges from 3 15nm.
At smaller sizes they do not display any magnetic
properties. Normally, synthetic oil is used as a car-
rier, or (not so often) light mineral oil, esters, glycer-
ine, poliphenyl or water. The maximum magnetisa-
tion of 0,6T domains limits the magnetic absorption
of the uids (0,005 0,13 T) and the maximum shear
stress is less than 5 kPa. They can operate at tem-
peratures from -65 to 200 C, as it is within this range
that magnetisation is independent from temperature.
Their durability depends on the evaporative power of
the carrier uid, which should be as low as possible
whilst the conductivity of the entire system should
be the highest. As the particles are of small dimen-
sions they do not accumulate at the bottom of the
container even if the uid remains still for long pe-
riods of time (they are raised as a result of thermic
motions). In magnetorheological uids the size of the
particles ranges from 0,5 to 8 mm. Mineral or sil-
icon oil with a low evaporative power is used most
often as the carrier uid. In industrial applications the
magnetic induction can total 1,2T (max.2,15T). These
properties do not change within the temperature range
of -50 150C. The maximum shear stress at a mag-
netic eld strength of 150 250 kA/m is 50 to 250
kPa. A characteristic feature of magnetorheological
uids is their capability of rapidly (<10ms) changing
their viscosity after a relatively weak magnetic force
is applied. Leaving the magnetorheological uid still
leads to the accumulation of particles at the bottom of
the container, which is their drawback. However, their
production is much easier, making them signicantly
cheaper compared to ferromagnetic liquids. In the sit-
uation when an external magnetic eld is not present,
the magnetic moments linked to the particles are dis-
tributed at random, which leads to a complete dissipa-
tion of their forces. After applying an external force,
the magnetic moments of the particular particles line
up along the ux created by the force eld and thus
are no longer subject to thermal currents. This mech-
anism is similar to the mechanism of chain creation
in electrorheological uids. Magnetorhelogical uids
display considerably stronger magnetisation than fer-
rorheological uids. By controlling the strength of
the magnetic eld we can inuence the viscosity of
the uid, e.g. with a eld of ca. 200 kA/m it can
reach a value of 700 P for magnetorheological uids
and 50 P for ferromagnetic uids, which allows for
the application of shear stress of 100 kPa and 5 kPa
respectively.
1.2 New Methodology
The prototype device has been shown in g. 1a. It
consists of a ring with grooves adapted to the dimen-
sions of the electromagnets or solid magnets active
elements, which are synchronised geometrically and
parametrically to work at a desired angle within the
ring, where a suitable container with ferrouid has
been placed.
1.3 Collimator Function
A collimator (g. 1a) is composed of a cylinder lled
with ferrouid which serves as a diffraction crystal.
The ferrouid, being situated inside a ring contain-
ing several or more powerful magnets with alternate
or compatible polarities becomes spatially organised,
forming the shape of a convex lens with a relatively
large curvature and containing additional smaller lo-
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
442
Figure 1: Collimator for modifying the geometry and en-
ergy of the diagnostic X-ray radiation: (a) example arrange-
ment of xed magnets or electromagnets in induction ring;
(b) concept showing the relation between the induction of
the magnetic eld (increasing the thickness of the lens) and
centrifugal force in a rotating motion, the action of which
attens the curvature of the ferrouid lense.
Figure 2: Collimation system: (a) view of ferrouid inside
the inducting ring of the collimator at one magnet cong-
uration; (b) laboratory unit on a adjustable stand equipped
with a dental X-ray tube.
cal convexes. After putting the entire arrangement
into motion by an electric motor, the action of cen-
trifugal forces which act in an opposite direction to
the magnetic inductance vector decrease the size of
the lens curvature proportionally to the speed with
which the magnetic ring turns and in inverted propor-
tion with the value of the magnetic inductance (g.
1b).
The X-rays emitted from the tube, when passing
through a collimator are curved to the axis, concen-
trated in the focal point or are diverged away from the
axis (dispersed) depending on the type of ferrouid
used, the shape of the eld (depending on the num-
ber and induction of the used magnets or additional
electrodes) as well as on the rotational speed of the
collimator.This allows the rays which are outside the
axis to be aligned towards it or focused. Using a col-
limator also enables energetic ltration to take place
which eliminates image distortions resulting from an
uneven distribution of radiation on the images plane
(initially divergent beam) (g. 2a).
The operation of the new generation collimator
Patent P366266 is based on Lawrence Braggs law,
which explains the mechanism of how the path of
hard radiation is distributed in a crystal. Ray 1 falls
Figure 3: The central element of the collimator is its rotat-
ing ferrouid lens, the curvature of which is determined by
the relation between the centrifugal force and the magnetic
induction of the rings magnets. (a) the initially divergent X-
ray bundle is modied especially within the range of non-
axial beams (b) Braggs law describes this process.
onto the surface of the crystal at angle V and affects
the electron layer of atom C, whilst the second beam
which is parallel to beam 1 affects the layer of atom
M. The electron layers of the atoms scatter the X rays.
The symmetry straight AC which is perpendicular to
the incident rays constitutes the face of the incident
wave. Such is also line BC a wave scattered at an-
gle V. As a result the difference between the paths of
rays 1 and 2 is AM + MB. Triangle AMC produces
the following formula:
AM = dsinV(1)
where d is the distance between neighbouring planes
in the crystal. AM = MB, thus the difference between
the paths of beams 1 and 2 is:
2dsinV(2)
Amplication takes place when the difference be-
tween the paths of two rays equals an integral mul-
tiple of the the length of wave 1. Thus the condition
for amplication can be expressed by:
2dsinV = nl(n = 1, 2, ...)(3)
If this condition known as Braggs formula - is ful-
lled, then the scattered beams 1 and 2 become am-
plied and a reection will occur. It can be noticed
that the reection is a result of scattering and interfer-
ence (Bankier, et al.). Experiments have shown that
the computer controlled prototype collimator allows
a sharp focus of the divergent beam to be obtained in
a precisely determined point of a structure (g. 4a, b),
and the focus to be precisely moved within the tested
structure (scanning). Such a system creates the foun-
dations for gradient tomography without a need to use
the expensive gantry system and multislice scanners.
2 AIMOF WORK
The subject of the clinical tests was the RTG UDR1
prototype collimator. The tests were conducted
COLLIMATION OF X-RAY DIAGNOSTIC BUNDLE BY MEANS OF STEERING FERROFLUID
443
Figure 4: One of the rst images taken- the authors hand
showing two extreme parametrical compositions: (a) image
taken by axial beams at 770 r.p.m., displaying a sharp bone
trabeculae structure with an almost complete elimination of
soft tissue; (b) image taken with a non-axial beam com-
ponent at 650 r.p.m. displaying soft tissue with an almost
completely faded bone trabeculae structure.
among a group of healthy volunteers employed as re-
searchers at the Laboratory of Biotechnology. The
aim of the research was to answer the following ques-
tions:
(1) Does a ferrouid which is spatially organised in
a magnetic eld and put into a spinning motion
change the properties of a permeating diagnostic
X-ray bundle produced by a dental X-ray unit?
(2) Are changes in the photodensitometric parame-
ters of joint (PIP) X-ray images containing soft
and osseous tissues proof of alterations of the
physical parameters of a diagnostic X-ray beam?
(3) Is there a relationship between the volume of fer-
rouid in a lens and its capability to alter the pa-
rameters of a permeating X-ray beam?
(4) Is there a relationship between the r.p.m. of the
collimator and its capability to alter the parame-
ters of a permeating X-ray beam, and does it have
a linear character?
2.1 Subjects Tested and Method
Male white-collar volunteers aged 356, 7 , in good
health were qualied for the research.
Excluding Criteria. History of hand injury (bruises,
dislocation, fracture) treated by immobilisation, in-
ammation of the joints, osteoporosis, diabetes,
heavy physical work or extreme sports, contact with
vibration, ionising radiation.
Testing Apparatus. The tests were carried out in a
standard room of the Laboratory of Biotechnology in
Cieszyn under the supervision of a Radiology Protec-
tion Inspector (of the Central Laboratory for Radio-
logical Protection), on a prototype UDR1 collimator
linked to a small-size Siemens X-ray unit for standard
dental diagnostics. All photos were made with a 0.5
second exposure time using 50kV to power the tube.
The registration of the bone structure radiograms and
their conversion to digital form was carried out on a
DIGORA dental 30 x 50mm matrix.
Test Method. The aim of the experiments was to test
the possibilities of controlling an X-ray beam by us-
ing different rotational speeds of a ferromagnetic lens
composed of different volumes of ferrouid and also
to determine the applicability of the obtained photos,
which were modied in terms of quality, in diagnostic
medicine. The experiment schedule included:
1. Performing a series of shots of the PIP 3 (R) joint
with a lens containing 0.5 ml of ferrouid at 11
ascending and repeatable rotational speeds of the
collimator.
2. Performing a series of shots of the PIP 3 (R) joint
with a lens containing 1.5 ml of ferrouid at 11
ascending and repeatable rotational speeds of the
collimator.
3. Performing a selective densitometric analysis of
bone trabeculae groups in standard measurement
area locations (g. 5) by means of the Structure
1.0 software.
4. Performing an analysis in measurement areas lo-
cated diagonally to the length axis of the long
bone with visible soft tissue on one side of the
bone P(1) and on the other P(2) by means of
the authors trademark Density 1.0 software (g.
6,7).
Following a 5 minute acclimatisation in the labora-
tory, the tested persons were seated in a comfortable
position at station UDR1, placing their hands loosely
on the positioning pad. The trunk and legs of the pa-
tient were screened by a standard protective lead rub-
ber apron. The geometric axis of the lamp was di-
rected at the center of the picture converting matrix
size 20 x 30 mm, and the hand position pad forced
the third nger of the right hand to be situated so that
the axis also passed through the center of the PIP III
(R) gap. The distance between the bottom opening of
the X-ray tube collimator and the surface of the ma-
trix was 100mm. In the rst series the shot was taken
through a still lens containing 0.5 ml of ferrouid.
Then the collimator was put into a spinning motion
for the 10 subsequent rotational speeds and shots were
taken at approx. 30-40 second intervals following the
stabilisation of the successively set speed. In the sec-
ond series the lens was replaced with one containing
1.5 ml of ferrouid and all of the above procedures
were repeated for the 10 rotational speeds. A standard
system of DIGORA converting matrices was used to
register the images and enabled the X-ray shadow to
be saved directly and repeatably in the computers op-
erational memory in BMP format which, as a result,
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
444
Figure 5: Structure 1.0 graphical interface allowing for
the determination of the surface area of pixels belonging to
the visible bone trabeculae groups in measurement elds P
and D.
allowed for the complete elimination of errors which
could occur if the images were processed photochem-
ically.
X-ray Image Analysis
(1) Analysis of the visible bone trabeculae groups
surface area compared to the surface of the en-
tire image performed via the authors trademark
Structure 1.0 software implemented in the C++
environment. The standard area around the proxi-
mal phalangeal II (P) epiphysis and the distal pha-
langeal I (D) epiphysis, proximal interphalangeal
joint of the right hands third digit.
(2) Analysis of the X-ray shadows photdensitomet-
ric cross-section - was performed by the authors
trademark ensity 1.0 software implemented in
the Delphi 7.0 Professional environment. The
program enables a photodensitometric evaluation
to be made of the image in the measurement eld
situated crosswise to the long bone axis with con-
sideration given to the soft tissue background on
one side of the bone (P1), the bone area (P3) and
the soft tissues (P2) on the other side of the bone
area (g. 5,6).
2.2 Results
Next pictures shows the results of photodensitometric
tests of bone trabeculae groups visible macroscopi-
cally in X-ray shadows of the PIP III joints, made in
a stand which positioned the axis of the bone against
the X-ray beam, and processed by the Structure 1.0
software. The averaging results for the relative den-
sity in measurement eld P (proximal epiphysis) and
D (distal epiphysis) were based on the 11 tested col-
limator rotational speeds. The picture g. 8 presents
Figure 6: Density 1.0 graphical interface enabling the
preparation of a histogram displaying the optical density in
the measurement eld situated typically around the distal
phalangeal epiphysis of the right hands third digit. The
program performs a measurement in the soft tissue areas
P(1), P(2) and in the bone area.
Figure 7: Example histogram displaying the optical density
in area P(1), P(2), B(bone area).
differences between the average values in measure-
ment areas P and D.
Measurement data displayed in diagrams 8-10
show the non-linear dependency between the degree
of the modication of the X-ray beam and the rota-
tional speed.
3 CONCLUSIONS
(1) The ferrouid which has been spatially arranged
in a magnetic eld and put into a rotational motion
changes the properties of the permeating diagnos-
tic X-ray (with constant and repeatable parame-
ters), inducing changes in the clarity and optical
density of the soft tissue and bones.
(2) Changes in the optical clarity and translucency of
the analysed fragments of X-ray images of soft
tissues and bones obtained froma repeatable lamp
exposition at different collimator parameter set-
tings prove that changes in the physical properties
COLLIMATION OF X-RAY DIAGNOSTIC BUNDLE BY MEANS OF STEERING FERROFLUID
445
Figure 8: Averaging photodensitometric values of bone tra-
beculae groups in measurement areas P and D in X-ray
shadows made by the collimator at 11 rotational speeds of
the lens with ferrouid (signicant changes occurred at 650,
770, 1000, 1350, 1450 r.p.m).
Figure 9: Structure 1.0 diagram presenting the averaging
optical density values of structures belonging to macroscop-
ically visible bone trabeculae in areas (P1) and (P2), (D1)
and (D2) for images made by lenses containing 0.5 and 1.5
ml of ferrouid respectively.
of X-ray beams have occured
(3) A dependency has been noted between the vol-
ume of ferrouid in the lens and its capability of
changing the permeating X-ray beam
(4) A non-linear dependency has been noted between
the rotational speed of the collimator and its capa-
bility of changing the permeating X-ray beam
4 DISCUSSION
The original divergence of a diagnostic X-ray bundle
resulting from the design of the X-ray tube is an is-
sue which has been accompanying radiological diag-
nostics ever since the emergence of the method. The
heterogeneity of the beamleads to signicant geomet-
rical changes of the image and causes an uneven dis-
tribution of radiation on the surface of the registering
matrix. Attempts at a compensating this issue focus
on two main areas:
(1) computer image processing
(2) technical elimination of non-axial beams
The collimation of non-axial diagnostic X-ray beams
has for long been performed by means of moving l-
ters (grids) or by lead collimators with thick walls
and a narrow pass which allowed only axial beams
to pass through. One drawback of using this type of
collimator is facing a signicant loss of the tubes ef-
fective power as the axial beams constitute a small
share of the entire emission. Correcting an X-ray im-
age by homogenising the optical density and restoring
the geometric relations to a 1:1 status entail a com-
plete remodelling of the image matrix which results in
breaching the principle of ling patients test results
in lossless formats (Dyszkiewicz, Wrbel; ICXOM
Vienna 2001). An interesting approach to solving
this issue were attempts at modifying the direction
of non-axial beams (Atkinson K, Folkard M et all;
ICXOM Chammonix 2003) based on magnetorheo-
logical uids which made it possible to make imme-
diate changes to the parameters of the beam by using
lters of varying densities (Remesh, N; Malagodi D at
all; ICXOM Chammonix 2003). A completely differ-
ent approach, on the other hand, was the idea of con-
structing a collimator which would be able to mod-
ify the propagation angle of non-axial beams towards
the axis and even to focus them (Dyszkiewicz, Wrbel
ICXOM Chammonix 2003), which would result in an
insignicant loss of energy from the initially diver-
gent beam while maintaining relatively good control
over the parameters of the produced image (e.g. expo-
sition of soft tissue (g. 4b) or hard tissue (g. 4a)).
The conducted laboratory experiments have proved
that the ferrouid which has been spatially arranged
and put into motion by a magnetic eld changes the
properties of the permeating diagnostic X-ray beam
generated in a parametrically repeatable manner by
the tube (permanently coupled with the collimator),
located at a xed distance away from the tested dig-
its on the hand (positioned against the tubes axis by
means of a special stand), inducing changes in the
optical clarity and density of the contours of soft tis-
sue and bones. Changes in the contrast and thick-
ness of the edge contour of the bone tissues struc-
ture are accompanied by a change in the proportions
between the axial and non-axial rays in the diagnos-
tic beam, enabling or disabling the development of a
sharp and narrow border line in the presence of an
excess numer of rays. Changes in the optical clarity
and translucency of the analysed fragments of X-ray
images of soft tissues and bones obtained from a re-
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
446
peatable exposition at a xed distance from the object
and different collimator parameter settings prove that
changes in the physical properties of X-ray beams oc-
cur. Proof of the existence of a dependency between
the volume of ferrouid in the lens and the lens capa-
bility for modifying the parameters of the permeating
X-ray beam was of signicant importance in conrm-
ing the inuence of ferrouid on the physical proper-
ties of an X-ray bundle. This may in the future lead
to the creation of lenses for various practical applica-
tions. A non-linear dependency has also been found
between the collimators rotational speed and its abil-
ity to modify the parameters of the permeating X-ray
beam. This feature seems to be best evidence that
interference occurs at closely quantized electron or-
bits of atoms belonging to the rotating quasi-crystal
of the ferrouid as it should be remembered that if
the crystal was understood to operate as a simple stop,
the characteristics would most probably have a linear
character. Therefore the clear dependency between
the modifying properties of the collimator and the ro-
tational speed, exluding changes in the volume of fer-
rouid, seems to be of signicant value for the fu-
ture perspective of creating easily-controllable units,
serially produced devices broadening the diagnostic
capabilities of standard X-ray tubes or enhancing the
further development of work on gradeint tomography.
Although the research results have been quite interest-
ing, it is important to bear in mind that the presented
scientic evidence concerning the modication of an
X-ray bundle has an intermediate character, this due
to the fact that it deals with a secondary evaluation
of effects induced in the tested object following a
conversion of the rays energy into the visible light
range. The main reason why such an approach for
conducting this research was chosen is that the med-
ical community has become used to utilising X-ray
diagnostics in such a manner, as well as for its con-
siderably lower costs. Currently work is underway in
the Laboratory of Biotechnology on the next stage of
research which involves making direct measurements
of difraction and interference in the output X-ray bun-
dle after being modied by the collimator.
REFERENCES
Bankier A, Fleichmann D, Aram L, Heimberger K,
Schindler E, Herold C. Bildgebung in der In-
tensivmedizin Techniken, Indikationen, diagnos-
tische Zeichen. In: Bardenheuer H, Hilker O,
Larsen R, Radke J. Weiterbildung fr Ansthesisten.
Springer,Berlin, 15-48
Carlson J. D., Electrorheological Fluids, US Patent 4,772,
407
Czerny C. Steiner E., Gstttner W., Baumgartner WD.,
Imhof H. Postoperative radiographic assessment
of the Combi 40 cochlear implant. Am. Journ.of
Roentgenology 1997:169(6):1689
Dyszkiewicz A, Sapota G, Wrbel Z. Standaryzowana
fotometria zdj radiologicznych ukadu kostnego w
tworzeniu komputerowych algorytmw densytome-
trycznych. Konf. TIM 99, Jaszowiec 1999
Dyszkiewicz A, Wrbel Z. The procedure of supervising
treatment of thyroid gland with isotope j 131 and using
2d and 3d analysis of scintigraphical image. ICXOM
Vienna 2001
Dyszkiewicz A. Procedure for monitoring the evolution of
inammatory and degeneration changes in sacroiliac-
lumbar joints and correcting the rtg-picture density di-
vergence. ICXOM Vienna 2001
Dyszkiewicz A., Kolumna chromatograczna do sczenia
lub ltracji, patent PL 175577
Folster R. T., Magnetoorheological Fluids, US Patent
5,667,715
Grampp S. Steiner E. Imhof H. Radiological diagnosis of
osteoporosis. Eur J Radiol 1997a:7:2:S11-9
Grampp H. Genant A. Mathur Ph. Lang M. Jergas M.
Takada C. Gler Y. Lu, M. Chavez Comparison of
noninvasive bone mineral measurements in assessing
age-related loss, fracture discrimination, and diagnos-
tic classication. J Bone Miner Res 12 (5) (1997b):
697
Kainberger F. Mittermaier F. Seidl G. Parth E. Weinstabl
R. Imaging of tendonsadaptation degeneration rup-
ture. Eur J Radiol 1997:25(3):209
Kollmann, K. Turetschek, W. Backfrieder, G. Most-
beck Quantitative analysis with an amplitude-coded
color Doppler imaging system (in-vitro study). World
Congress on Medical Physics and Biomedical Engi-
neering. Nice, France Sept 14-19, 1997
Kramer J. Hofmann S. Plenk H. Schneider W. Engel A.
Imaging of avascular necrosis of bone. Eur Radiol
1997: 7:2:180
awniczak A., Milecki A., Ciecze elektro- i mag-
netoreologiczne oraz ich zastosowania w technice,
Wydawnictwo Politechniki Poznaskiej 1999
Majumdar, H. Genant, S. Grampp, D.C. Newitt, V. Truong,
J.C. Lin, A. Mathur Correlation of trabecular bone
structure with age, bone mineral density, and osteo-
porotic status: in vivo studies in the distal radius using
high resolution magnetic resonance imaging. J Bone
Miner Res 12 (1) (1997): 111
Phule P. P. Magnetoorheological Fluid, US Patent
5,985,168
Rand T. Seidl G. Kainberger F. Resch A. Hittmair K.
Schneider B. Gler CC. Imhof H. Impact of spinal de-
generative changes on the evaluation of bone mineral
density with dual x-ray absorptiometry (DXA). Calc.
Tissue 1997: 60:430
Sasaki M., Ishii T., Haji K, Electrorheological Fluid com-
prising lyotropic liquid crystalline polymer, US Patent
5,746,934
COLLIMATION OF X-RAY DIAGNOSTIC BUNDLE BY MEANS OF STEERING FERROFLUID
447
NOVEL COMBINED TEMPLATE FOR AMPEROMETRIC
BIOSENSORS WITH CHANGEABLE SELECTIVITY
J ulija Razumiene
1
, Vidute Gureviciene
1

, J urgis Barkauskas
2

Virginijus Bukauskas
3
and Arunas Setkus
3

1
Institute of Biochemistry, Department of Bioanalysis, Mokslininku 12, 08662 Vilnius, Lithuania
[email protected]
http://www.bchi.lt
2
Vilnius University, Department of General and Inorganic Chemistry, Naugarduko 24, 03225 Vilnius, Lithuania
[email protected]
3
Semiconductor Physics Institute, Sensors Laboratory, A. Gostauto 11, LT01108 Vilnius, Lithuania
{virgis, setkus}@pfi.lt
Keywords: Biosensors, Enzymes, Carbon Nnotubes, Medical Application.
Abstract: Present paper describes innovative approach in design of amperometric biosensors useful in various appli-
cations. Original template of the electrodes has been prepared on a base of carbon nanotube support layer
deposited on the polycarbonate membrane. Novel template and changeable enzyme layer give rise to crea-
tion of new family of biosensors acceptable for detection of wide range of carbohydrates. The morphology
and electric properties of the constituent parts of the template electrode are characterized by scanning probe
microscopy. The sensitivity, selectivity and stability are described for typical types of the biosensors.
1 INTRODUCTION
Electrochemical biosensors are the most common
class of biosensors in various practical applications
(A. Chaubey and B. D. Malhotra, 2002). In these
sensors, bio-interaction at specific sites of enzymes
results in extra electric charge. The extra charge can
be transferred from enzyme to electrode and de-
tected by an external electric circuit. Technology of
the electrodes and enzyme immobilization is crucial
for conversion of biochemical interaction into the
response signal and, still, is under intensive studies.
This type of sensors has well known advantages
such as acceptability for functioning in turbid media,
comparable instrumental sensitivity and amenability
to miniaturization, and etc. In this report, we present
an original approach in biosensor technology based
on immobilization of enzymes within special matrix
attachable to carbon nanotube electrode.





2 METHODS
AND EXPERIMENTS
We developed and tested an original technology
acceptable for production of a family of biosensors.
Selectivity of two component biosensors can be
changed simply by replacing special matrix contain-
ing enzymes attached to the single wall carbon na-
notube (SWCNT) based electrode. Therefore the
biosensors can be adjusted for detection of monosa-
charydes and disacharydes such as glucose, lactose,
galactose, maltose and et cetera.
Nanotube support layers on the polycarbonate
membranes were prepared from the industrial
SWCNT (Cheap Tubes Inc., USA). The main para-
meters of these nanotubes ware as follows: diameter
is 2 nm, 5.0 30.0 m length, specific surface area
400 m2/g, electrical conductivity 10-2 S/cm. The
SWCNT were functionalized with carboxyl groups
(2.73 wt%).
In this study we introduced an original protocol
for coating of flexible support by SWCNT layer
acceptable for biosensor electrode. The protocol
includes special filtration of aqueous suspension
448
through the isopore polycarbonate membrane and is
crucial for electrochemical properties of biosensors.
The prototype biosensors were based on three
types of enzymes, namely glucose oxidase from
Aspergillus niger and pyrroloquinoline quinone
(PQQ) dependent glucose dehydrogenases and
aldose sugar dehydrogenase. The soluble glucose
dehydrogenase (s-PQQ-GDH) from Acinetobacter
calcoacetics L.M.D. 79.41 was purified by the me-
thod reported in (A. J . A. Olsthoorn and J . J . Duine,
1996). The membrane-bound enzyme (m-PQQ-
GDH) was purified from Erwinia sp. 34-1
(Marcinkeviien et al., 1999). The water-soluble
aldose sugar dehydrogenase (s-PQQ-ADH) was
purified from Escherichia coli (Southall et al.,
2006). Each of the enzyme types was immobilized
on individual flexible support of polivinylalcohol
coated terylene. Adsorption and cross linking to the
support were the methods for immobilisation of
enzymes.
Our prototype amperiometric biosensors con-
sisted of SWCNT based electric charge drain and
changeable biosensitive detector. The sensor con-
struction is illustrated by a sketch in Fig. 1.
Surfaces of the sensor components, namely, elec-
trode support, SWNT coatings and matrix without
and with immobilized enzymes, were analyzed by
scanning probe microscope (SPM) D3100 / Nanos-
cope IVa (Veeco Instruments Inc.). Standard AFM
methods such as contact and tapping mode surface
scanning were used for visualization of the surface
morphology. The surface electrical characteristics
were evaluated from measurements of tunneling
current obtained in contact mode. Conductive probe
of the SPM was firmly pressed to the surface so that
it was not damaged. Special module SPM D3100
TUNA (Veeco Instruments Inc.) was used for these
experiments. The maps of the current and local point
volt-amperic characteristics (VACh) were obtained
for the components of the biosensor electrodes in
these experiments. The data and SPM mages were
processed by the NanoScope Software 6.14 (Veeco
Instruments Inc.).
Electrochemical experiments were performed us-
ing a conventional three-electrode system containing
a screen-printed carbon electrode as a working elec-
trode, a platinum wire as a counter electrode and an
Ag/AgCl in saturated KCl as a reference electrode
(all potential values presented in the text are vs. this
reference electrode). 0.05 M acetate buffer (pH 6.0)
containing 1 mM of Ca
2+
and 0.2 mM N-
methylphenazonium methyl sulphate was used as a
default buffer. Steady state currents of the biosen-
sors were recorded at 0.4 V using a polarographic
analyzer PARSTAT 2273 (Princeton Applied
Research, USA).

Figure 1: General side and top views (A) and the compo-
nents (B) of the biosensor: 1 insulating film, 2 enzyme
immobilized on terylene film, 3 contact zone, 4
SWCNT-polycarbonate membrane, 5 insulating film.
3 RESULTS AND DISCUSSIONS
Characteristics of the biosensor family were ob-
tained only for four types of the biosensors based on
the original prototype structure in present study. The
SWCNT layer on polycarbonate membrane and
changeable enzyme based detector are the most
important results of the sensor technology in present
study. It was proved by experiments with the proto-
type biosensors that SWCNT based structure is
acceptable for the sensor electrode and immobiliza-
tion of enzymes. The attachable enzyme detectors
were reproducible and stable for comparatively long
time.
3.1 Surface Properties of the
Electrodes
The morphology and electric properties were de-
scribed for separate components of the template
electrodes by the SPM experiments. The results
were obtained for the components at intermediate
stages of the technology.
Typical structure of the SWCNT coating is illu-
strated by a SPM image in Fig. 2. It is seen in Fig. 2
the SWCNT were found in vertical and horizontal
positions on the membrane. Since the membrane
contained the pores deep valleys were found in the
nanotube layer. It was revealed by high aspect ratio
SPM tests that SWCNT are in vertical position in
the areas corresponding to the pores in the mem-
brane. On the flat surfaces of the membrane there
were no preferable orientations of the SWCNT with
respect to the membrane surface. The SWCNT layer
NOVEL COMBINED TEMPLATE FOR AMPEROMETRIC BIOSENSORS WITH CHANGEABLE SELECTIVITY
449
was comparatively thick and at least several layers
of horizontal nanotubes were detectable.
It was proved by measurements of tunneling cur-
rent that electric conductivity highly depends on the
structure of the SWNT layer. The areas with vertical
nanotubes were more conductive that that with hori-
zontal SWCNT. Electrical properties of individual
areas of the SWCNT layer are compared in Fig. 3 by
typical voltamperic characteristics (VACh) that were
measured by special SPM module TUNA in contact
mode. The tunneling current was measured by the
SPM conductive cantilever tip diameter of which is
about 20 nm.

Figure 2: The SPM image of the surface of the SWCNT
coating on polycarbonate membrane. The surface area of
the image is 5x5 m
2
and the maximum height is 400 nm.
In Fig. 3, the surface areas with the lowest and
the highest conductivity are represented by the
VACh measured at the tip-points of individual sur-
face area. The lowest conductivity was obtained for
the tip attached to the horizontal SWCNT (2 in
Fig. 3). The highest conductivity of the SWCNT
layer was found in the areas corresponding to the
pores in the membrane (1 in Fig. 3).
Detailed distribution of the electrical conductivi-
ty over the surface of the SWCNT layer was visua-
lized by scanning of the surfaces with the SPM
TUNA. It was found that the spots of high conduc-
tivity are measured over the flat surface of the mem-
brane if vertical SWCNT are detected in this area.
Comparatively large areas were characterized by
intermediate electric conductivity. It was supposed
that only part of the SWCNT are connected so that
produce conductive mesh of the electrode. The ma-
jor part of the vertical SWCNT is only partly con-
nected to this mesh and, therefore, limits electric
charge transfer from the enzymes to the measure-
ment circuit. This electric limitation reduces the
effectiveness of the biosensor electrode.

Figure 3: Tunneling current versus dc-potential between
the SPM cantilever and the sample.
3.2 Biosensor Characteristics
Since the main advantage of PQQ-dependent dehy-
drogenases is functional independence of oxygen
these enzymes are highly attractive for development
of biosensors (Razumiene et al., 2005; Razumiene et
al., 2006). All these enzymes were chosen also due
to different ability to oxidise a number of carbohy-
drates. Thus, integration of these biosensors in
whole sensing system allows detecting broad range
of sugars, encompassing clinically important such as
lactose, galactose, maltose and et cetera that is
usually not detectable in body fluids although are
associated with several diseases. In spite of numer-
ous modifications of these enzymes that can be
acceptable for detection of various important com-
pounds we probed only a few types in this study.
Typical calibration curve for glucose obtained
using s-PQQ-GDH based biosensor is shown in
Fig. 4.
In Fig. 4, the current generated at the electrode
during electrocatalytic oxidation of glucose by the
enzymes was measured as a function of glucose
concentration in the solution. Similar dependences
were measured for all types of biosensors manufac-
tures and probed in this work.
Kinetic characteristics, namely the apparent Mi-
chaelis constant (K
M
app
) and maximum current (I
m-
ax
app
), calculated for each type of the biosensors are
summarized in Table 1.



BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
450
Table 1: Kinetic characteristics of SWCNT-based biosen-
sors with different enzymes.
Biosen-
sor type
K
M
app
,
mM
I
max
app
,
A
n r
2

s-ADH 305.4 42 7 0.9913
s-GDH 5 27 12 0.9945
m-GDH 0.11 10 8 0.9986
GOx 5.8 260 7 0.9870

Results in Table 1 shows that enzymes possess
different kinetics of action. From the functional
standpoint, there are also different, i.e. (s-) soluble
types operate in cytoplasma and (m-) membrane-
bound is tightly bound to the outer surface of the
cytoplasmic membrane (Matsushita et al., 2003). It
has been shown that they are different enzymes with
different pH-optima, molecular weights and sub-
strate specificity.

Figure 4: The calibration curves of the biosensors based
on direct immobilization of s-PQQ-GDH on SWCNT (1)
and with changeable s-PQQ-GDH film attached to
SWCNT electrode (2). C
g
is glucose concentration in the
solution.
In our previous paper (Laurinavicius et al., 2004)
has been demonstrated that due to the immobiliza-
tion the active center of enzyme can be distorted that
leads to different affinity to substrates as in the case
of the native enzyme. Aiming to evaluate the affinity
of enzymes operating in heterogeneous biosensing
systems, the selectivity to clinically important meta-
bolites such as glucose, lactose, galactose and mal-
tose were investigated for all types of the proposed
biosensors. The responses to individual metabolites
were represented by the signal ratio with respect to
the detection of 100 % glucose. The results are
summarized in Table 2.

In order to understand an influence of the en-
zyme immobilization method of the of s-PQQ-GDH
on the the selectivity and main kinetic parameters of
the biosensor we investigated the electrodes based
on SWCNT and carbon paste electrodes (CE) and
manufactured by the method previously described in
(Razumiene et al., 2006). The K
M
app
and I
max
app
pa-
rameters for three types of the electrodes are sum-
marized in Table 3.
The I
max
app
for all substrates can be explained by
the s-PQQ-GDH catalyzed oxidation of main sub-
strates such as glucose, lactose, and galactose with
almost the same rate ratio for all probed types of
biosensors (Table 3). However, the kinetic parame-
ters are individual for the biosensors with differently
immobilized enzymes (Table 3). The increase in
K
M
app
results in extension of the interval of the linear
calibration curve. We associate it with diffusion
limited access of the substrate.

The stability of the s-PQQ-GDH and GOx based
biosensors was investigated during couple of weeks.
The responses to the standard glucose solution (5
mM) were periodically recorded at room tempera-
ture equal to about 25 C during these experiments.
The residual response of the probed biosensors was
not less than about 80 % of initial magnitude over
the period of the tests.
4 CONCLUSIONS
Original technology was developed and probed for
manufacturing of prototype biosensors with change-
able selectivity. The template of the electrodes has
been prepared on the basis of SWCNT conductive
layer deposited on the polycarbonate membrane.
The attachable flexible matrix with immobilized
enzymes was proved functionally acceptable for
catalysis of biochemical reactions and detection of
these reactions. Vertical arrangement of the SWCNT
in the electrodes was related to the areas of high
electric conductivity of the electrodes that was as-
sumed essential for functioning of the biosensors.
Using glucose oxidase, two types of pyrroloqui-
noline quinone dependent glucose dehydrogenases
(namely s-PQQ-GDH, m-PQQ-GDH) and water-
soluble aldose sugar dehydrogenase s-PQQ-ADH
four versions of the prototype biosensors were
manufactured and investigated in this work. In the
tests, the responses to clinically important metabo-
lites such as glucose, lactose, galactose, arabinose,
manose and glucose-6-phosphate were measured. It
was
NOVEL COMBINED TEMPLATE FOR AMPEROMETRIC BIOSENSORS WITH CHANGEABLE SELECTIVITY
451
Table 2: Responses to different substrates of SWCNT-based biosensors.
Biosensor type Glucose Lactose Galactose Manose Arabinose Glucose-6-
phosphate
s-ADH 100 61 120 7 102 60
s-GDH 100 95 99 87 68 15
m-GDH 100 0 76 57 72 91
GOx 100 0 3 1 0 0
Table 3: Kinetic parameters of SWCNT-based and CE-based biosensors with differently immobilised s-PQQ-GDH.
Substrate
CE, enzymes on tery-
lene
SWNT, enzymes on
terylene
enzymes adsorbed on
CE
I
max
app
,
A
K
M
app
,
mM
I
max
app
,
A
K
M
app
,
mM
I
max
app
,
A
K
M
app
,
mM
Glucose 5.2 4.8 23.4 9 0.37 5.7
Lactose 4.7 8.1 21.7 7.2 0.3 8
Galactose 3.6 10.1 10.4 4.2 0.24 4

proved that the prototype biosensors are sufficiently
stable so that can be acceptable for practical use.
ACKNOWLEDGEMENTS
The study was partly supported by the Lithuanian
State Science and Studies Foundation contracts no.
N-08007 and N-09/2008. It was also partly sup-
ported by COST programme contract no. 31V-119.
REFERENCES
Chaubey, A., Malhotra, B.D.: Mediated biosensors. Bio-
sensors & Bioelectronics Vol. 17 (2002) 441456
Olsthoorn, A. J . A., Duine, J . J .: Production, characteriza-
tion and reconstitution of recombinant quinoprotein
glucose dehydrogenase (soluble type; EC 1.1.99.17)
apoenzyme of Acinetobacter calcoaceticus. Archives
of biochemistry and biophysics Vol. 336 (1996) 4248
Marcinkeviien, L., Bachmatova, I., Semnait, R.,
Rudomanskis, R., Branas, G., Mekien, R.,
Mekys, R.: Purification and characterization of PQQ-
dependent glucose dehydrogenase from Erwinia sp.
34-1. Biotechnol. Lett. Vol. 21 (1999) 187192
Southall, S.M., Doel, J .J ., Richardson, D.J ., Oubrie, A.:
Soluble Aldose Sugar Dehydrogenase from Escheri-
chia coli. A high exposed active site conferring broad
substrate specificity. J ournal of Biological Chemistry.
Vol. 281, 41 (2006) 3065030659
Razumiene J ., Barkauskas J ., Kubilius V., Meskys R.,
Laurinavicius V.: Modified graphitized carbon black
as transducing material for reagentless H2O2 and en-
zyme sensors. Talanta Vol. 67(2005) 783790


Razumiene J ., Vilkanauskyte A., Gureviciene V., Bar-
kauskas J ., Meskys R., Laurinavicius V.: Direct elec-
tron transfer between PQQ dependent glucose dehy-
drogenases and carbon electrodes: An approach for
electrochemical biosensors. Electrochimica Acta Vol.
51 (2006) 51505156
Matsushita, K., Toyama, H., Yamada, M., Adachi, O.:
Quinoproteins: structure, function, and biotechnologi-
cal applications. Appl. Microbiol. Biotechnol. 58, 13
22 (2002) 8. (3) Davis, J .J ., Coleman, K.S., Azamian,
B.R., Bagshaw, C.B., Green, M.L.H. Chemical and
Biochemical Sensing with Modified Single Walled
Carbon Nanotubes. Chem. Eur. J . Vol. 9 (2003) 3732
3739
Laurinavicius, V., Razumiene, J ., Ramanavicius, A.,
Ryabov, A.D.: Wiring of PQQ-dehydrogenases. Bio-
sensors and Bioelectronics Vol. 20 (2004)12171222



















BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
452
MODELLING AND TRIALS OF PYROELECTRIC SENSORS FOR
IMPROVING ITS APPLICATION FOR BIODEVICES
Andrs Daz Lantada, Pilar Lafont Morgado, Hctor Hugo del Olmo, Hctor Lorenzo-Yustos
J avier Echavarri Otero, J uan Manuel Munoz-Guijosa, J ulio Muoz-Garca and J os Luis Muoz Sanz
Grupo de Investigacin en Ingeniera de Mquinas E.T.S.I. Industriales Universidad Politcnica de Madrid
C/ Jos Gutirrez Abascal, n 2. 28006 Madrid, Spain
[email protected]
Keywords: Pyroelectricity, Ferroelectric Polymers, Sensors Behaviour, Medical Devices.
Abstract: Active or Intelligent Materials are capable of responding in a controlled way to different external physical
or chemical stimuli by changing some of their properties. These materials can be used to design and develop
sensors, actuators and multifunctional systems with a large number of applications for developing medical
devices.
Pyroelectric materials, with thermoelectrical properties coupling, can be used as temperature sensors with
applications in the development of several biodevices, including the combination with other thermally
active materials, whose actuation can be improved by means of precise temperature registration.
This paper makes an introduction to pyroelectricity and its main applications in the development of
biodevices, focusing also in the pyroelectric properties of polyvinylidene fluoride or PVDF and presenting
some results related with sensors behaviour modelling and characterization.
1 INTRODUCTION TO
PYROELECTRICITY
Pyroelectricity is the ability of certain materials to
generate an electrical potential when they are heated
or cooled. As a result of this change in temperature,
positive and negative charges move to opposite ends
through migration (the material becomes polarized)
and, therefore, an electrical potential is established.
This kind of phenomenon appears in dielectric
materials with spontaneous polarizations due to
dipole orientation within their structure. These
effects have been known to mankind even since
Antiquity, especially regarding ceramic materials
and metallic oxides.
The name of pyroelectricity was given by
Brewster in 1824. But investigations on polymer
pyroelectricity are more recent, starting around 1955
with some initial results, which were not
commercially promising.
New attention was given to this property with
the discovery of pyroelectric effects in
polyvinylidene fluorides (PVDF and copolymers) by
Bergman in 1971, after the discovery of
piezoelectricity in these materials by Kawai in 1969.
During the last decades important progress has
been made in creating artificial pyroelectric
materials, usually in the form of a thin film, out of
gallium nitride (GaN), caesium nitrate (CsNO
3
),
polyvinylidene fluorides (PVDF and copolymers),
derivatives of phenylpyrazine cobalt phthalocyanine
and other materials.
The main applications developed so far of these
materials in biomedical devices are explained below,
before paying attention to pyroelectricity in
polymers, modelling, signal conditioning and trials.
2 PYROELECTRIC MATERIALS
AND POTENTIAL BIODEVICES
The main industrial applications are related with the
development of temperature sensors, presence
sensors, humidity and leakage sensors and for
measuring other processes which mean a
temperature change. It can also be applied in
biological or medical context as explained in the
following examples.
453

Infrared Thermography Cameras. Infrared
Thermography is a technique for carrying out
inspections and non-destructive tests which has
multiple applications in the development of
machines and products, equipment and facilities
maintenance, and troubleshooting.
Since all bodies emit (according to their
temperature) infrared radiation, which increases in
intensity as the temperature rises, variations in this
intensity can be detected by using infrared sensors.
Thermal cameras can detect radiation in the
infrared range of the electromagnetic spectrum
(usually between a 900 and 14000 nm wavelength,
instead of operating in the visible range of 450 to
750 nm) and can produce images of this radiation.
These cameras are fitted with a sensor matrix
(called microbolometer) that can be developed using
pyroelectric materials. Depending on the intensity of
the radiation more or less current is sent to the
cameras control electronics, which with the aid of
specific software enables temperature maps to be
obtained.
Some of the fundamental advantages of the
technique are its speed and ease of use, easy to
interpret temperature map-based results and the fact
that it is a non-destructive technique that does not
damage the systems under study (Schindel, 2007,
Maldague, 2001).
Apart from these applications, its use as a
support tool for developing medical devices,
especially those based on the use of thermal
materials has also been proposed (Paumier, 2007).
Biometric Systems. Pyroelectric materials can also
be used as part of complex biometric systems for
real-time recognition of people inside a building or
room with security purposes, or in the medical field
for evaluating the progress of injuries that limit the
mobility of patients (Fang, 2007).
Aided Surgery. These materials have also been
proposed and tested for measuring blood
temperature during surgeries, such as coronary stent
placements, with the purpose of relating temperature
profile with the blood velocity field and using this
comparison as a method of controlling the surgical
procedure (Mochi, 2004).
Flow Sensors. Dymedix Co. has developed nasal
flow sensors using PVDF, that besides being
piezoelectric has also pyroelectric properties and can
be used as temperature sensor. This products allow
an active management of pathologies such as sleep
apnea or sudden death in children. Such devices are
placed adjacent to the nostrils and patient breath
induces charges to the sensor, with a typical and
recognisable pattern. When breathing ceases, the
pattern changes and the microcontroller detects such
problem and activates an alarm to alert both the
patient and his or her relatives.
X-Ray Intensity Sensors. Based on corporal
heating due to absorption of X-Ray (during
radiological explorations) pyroelectric sensors can
be used, so as to make an estimation of the dosage
received and in order to avoid risk situations. The
phenomenon has been proved in vivo during
mammography scans with positive results according
to precision and sensitivity (De Paula, 2005).
3 PVDF PYROELECTRIC
POLYMER SENSORS
Polyvinylidene fluoride or PVDF -(CH
2
-CF
2
)-
n
and
its co-polymers such as poly(vinilydenefluoride-
trifluoroethylene) or P(VDF-TrFE), are the polymers
of this kind with the largest number of industrial
applications. They posses partial crystalinity with an
inactive amorphous phase and an elastic modulus
close to between 1 and 10 GPa.
The ferroelectric structure makes this polymer both
piezoelectric and pyroelectric, which increases its
applications, not only as temperature and pressure
sensor, but also as actuator. Its use as actuators is
limited by the need to apply high electric fields
(around 20 V/m for a 3% deformation), but their
use as pressure sensors is taking the place of
traditionally used piezoelectric ceramic materials.
Regarding pyroelectricity its important value of
pyroelectric coefficient, together with its greater
resistance and sensitivity is displacing the use of
pyroelectric ceramics.

Figure 1: Metallized PVDF sheets. Piezotech S.A..
To make the sensors, we took PVDF 40 m
thick sheets from Piezotech S.A. with Au-Pt coated
electrodes. These sheets were cut, joined to the
connecting wires and suitably encapsulated into
flexible polyurethane layers to protect them.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
454

The sensors obtained and the main properties of
the PVDF used, along with some alternatives of the
same materials family, are shown below.

Figure 2: PVDF pyroelectric polymer sensor for fluidic
applications with polyurethane protecting layer.
Table 1: Pyroelectric polymers main properties.

d
33
pC/N
d
31
pC/N
d
32
pC/N

F/m
p3
C/m
2

C
Uni-ax.
PVDF
-20 18 3 1,11
-10
25
Bi-ax.
PVDF
-24 7 7 1,11
-10
25
P(VDF
-TrFE)
-24 7 7 0,91
-10
25
The sensors behaviour due to combined
piezoelectric and pyroelectric properties can be
studied in first approximation using the following
equations.
Figure 3 a) shows the sensor layout. The charge
displacement (produced when a force of temperature
change is applied to the piezoelectric sensor) can be
represented using the equivalent electric circuit
depicted in Figure 3 b).

a)

b)
Figure 3: a) Piezoelectric Sensor. b) Electrical behaviour
circuit diagram of the piezoelectric sensor.
Force F on the sensor acts as a generator of
intensity powering a C capacity condenser. Ec (1).

C =C(F) = (L
1
L
2
) / e (1)
Where:
.- The dielectric constant of the sensor.
L
1
L
2
.- The effective area of the sensor.
e.- The thickness of the sensor.
The thickness of the sensor, e, depends on the
initial thickness, e
0
, on the pressure applied, =F /
(L
1
L
2
), and the Young modulus of the material, E,
using the following expression Eq. (2):
e =e
0
(1 / E) (2)
Current intensity, I, generated by applying force,
F, depends on the transversal piezoelectric
coefficient of sensor d33 according to Eq. (3).
Q =d33 F I =dQ / dt =d33 dF / dt (3)
Current intensity, I, generated by applying
temperature chages, T, depends on the pyroelectric
coefficient p
3
and can be expressed as:
Q =p
3
(L
1
L
2
) T
I =dQ / dt =p
3
(L
1
L
2
) dT / dt (4)
The total amount of current intensity can be
obtained as addition of Eq. (3) and (4).
When the sensor is connected to an external
circuit, as is shown in Figure 3 b), it discharges in
accordance with the equivalent R resistance of this
external circuit (i.e. oscilloscope, charge amplifier).
The intensity is given by Eq. (5).
I =d33 dF / dt =U / R +C dU / dt (5)
The effect of thermal expansion has to be taken
also into account when the deformations produced
due to important temperature changes become
relevant, as consequence of the thermo-electro-
mechanical coupling in these materials.
Similar equations have been used to model and
simulate the behaviour of this kind of materials as
part of more complex devices. They can be used not
only for design purposes, but also as a way of
estimating adverse effects due to piezo and pyro
effects coupling, in applications that are intended to
use these materials only as pressure sensors or only
as temperature sensors.
In case of willing to obtain positive voltages
when using this sensors for measuring compressive
stresses a polarity change in the connections is
enough, but then temperature increases lead to
decreasing voltages (as happens in our PVDF trials
explained below).
Following chapter shows some characterization
trials of the pyroelectric behaviour of PVDF sensors
and the influence of parameter changes for its
applicability.
5 mm
MODELLING AND TRIALS OF PYROELECTRIC SENSORS FOR IMPROVING ITS APPLICATION FOR
BIODEVICES
455

4 TRIALS WITH PVDF
PYROELECTRIC SENSORS
For proving the pyroelectric response of PVDF
sensors a charge amplifier for signal conditioning
was designed. It has typical impedance for such
conditioning circuits of 10 T and includes a
voltage supply of 3.7 V.
The trial bench includes an additional
Measurement Computing LS1208 data acquisition
card, connected via USB to a personal computer.
The trials were carried out by introducing the
sensor from room temperature (26 C) into a water
vase with temperature control.
Several trials were made, changing water
temperature, before introducing the sensor into it, so
as to study the influence of T on the speed
response of the sensor, which can be measured as a
function of the origin slope of the function V(t).
The main results are shown below. Figure 4
represents the trials made using a sensor with just
one polyurethane protecting layer. Figure 5 was
obtained by using a sensor with three protecting
layers and shows a slower response, due to the effect
of thermal isolation produced by the polyurethane.
0,0
0,5
1,0
1,5
2,0
0 5 10 15 20 25 30 35 40
Time (s)
V
o
l
t
a
g
e

(
V
) 25C
26C
26,5C
27,5C
29C
32,5C
34C
36C

Figure 4: Voltage changes due to introduction of the
PVDF sensors in water at different temperatures. (One
protecting layer).
0,500
1,000
1,500
2,000
2,500
0 5 10 15 20 25 30 35 40
Time (s)
V
o
l
t
a
g
e

(
V
)
23C
25C
26,3C
27,3C

Figure 5: Voltage changes due to introduction of the
PVDF sensors in water at different temperatures. (Three
protecting layers).
These trials help to show the sensitivity of PVDF
as temperature measuring material, due to its
pyroelectric properties. The fact that temperature
changes can be detected, even using a 4.5 mm thick
polyurethane protecting layer, is important for
increasing the number of applications of such
materials, which can be used not only as surface
sensors, but also for measuring surface temperature
changes from the inside of a biodevice.
5 CONCLUSIONS
The work presented shows an introduction to
pyroelectricity and its main applications in the
development of biodevices, focusing also in the
pyroelectric properties of polyvinylidene fluoride or
PVDF.
Additionally, some results related with sensors
behaviour modelling, signal conditioning and
characterization trials are presented.
This type of materials can also be applied to
medical devices in combination with other active
materials, especially those based on thermal
activation.
Thus pyroelectric sensors could be used as a way
of monitoring temperature and optimising activation
of the active part of the device.
REFERENCES
S. Ashley, Artificial Muscles. Scientific American 2003.
Y. Bar-Cohen, Electroactive Polymer (EAP) Actuators as
Artificial Muscles. SPIE Press, Second Edition.
Washington 2004.
Y. Bar-Cohen, et al. Characterization of the
Electromechanical Properties of EAP Materials. J et
Propulsion Laboratory (J PL)/Caltech. SPIE. Newport,
CA, 2001.
B. Schindel, Thermographie in der Theorie und Praxis.
www.irPOD.net. 2007.
Maldague, X., 2001. Nondestructive evaluation of
materials by infrared thermography. John Wiley &
Sons Publishers.
G. Paumier, et al., Thermoresponsive Polymer-Based
Microdevice for Nano-Liquid Chromatography.
Biodevices 2008 International Conference on
Biomedical Electronics and Devices. IEEE
Engineering in Medicine and Biology Society.
INSTICC Press.
J .S. Fang, A Pyroelectric Infrared Biometric System for
Real-time Walker Recognition by use of a Maximum
Likelihood Principal Components Estimation
(MLPCE) method. Optics Express. 2007.
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
456

M. Mochi, et al., A Study for a Portable IR Sensor to
detect the Blood Temperature during Coronary
Bypass Implantation. CNR University Pisa. 2004.
M. De Paula, et al., Microcontrolled pyro-electric
instrument for measuring X-ray intensity in
mammography. Medical and biological engineering
and computing. 2005.
M. Hafez, Course on Polymer Based Actuators as
Artificial Muscles. FSRM (Swiss Foundation for
Research in Microtechnology). Zurich 2006.
Measurement Specialties, Inc., Piezo Film Sensors
Technical Manual. Sensor Products Division 1999.


MODELLING AND TRIALS OF PYROELECTRIC SENSORS FOR IMPROVING ITS APPLICATION FOR
BIODEVICES
457
POLYMERIC FILM SENSORS BASED ON PAH-PAZO IONIC
SELF-ASSEMBLED MULTI-NANOLAYERS
Celso Ribeiro, Paulo J . Gomes, Paulo A. Ribeiro, Maria Raposo
Centro de Investigao em Fsica Tecnolgica (CEFITEC), Physics Department, Faculdade de Cincias e Tecnologia
(FCT), New University of Lisbon (UNL), 2829-516, Caparica, Portugal
[email protected]
Hugo guas
Department of Material Science and CINEMAT, Faculdade de Cincias e Tecnologia (FCT), New University of Lisbon
(UNL), 2829-516, Caparica, Portugal
Pedro Santos, Beatriz Borges
Department of Electrical and Computer Engineering, Instituto Superior Tcnico (IST), 1049-001, Lisbon, Portugal
Pedro Brogueira
Department of Physics and ICEMS, Instituto Superior Tcnico (IST), 1049-001, Lisbon, Portugal
Keywords: Sensors, Biosensors, Electronic Tongue, Lab-on-chip, Polymers, polyelectrolytes, PAH, PAZO, Layer-by-
Layer, Nanotechnology, Nanofilms, Thin films, Impedance Spectroscopy, AFM, Roughness, Ellipsometry.
Abstract: A sensor platform composed by interdigitated electrodes covered by a tailored nanofilm of 5 bilayers of
PAH and PAZO polyelectrolytes was successfully built and tested in NaCl solution at relative low
concentrations (10
-5
-10
-6
M). The effect of temperature on the electrical measurements was addressed and
indicated its importance for to sake accuracy reduction in about one order of magnitude. The film
morphology was studied by using ellipsometry and atomic force microscopy techniques. The first inferred
precise and consistent values for the film thickness about 50 nm. With respect to the second, relative low
values of the surface roughness of about few nanometers were measured and also micrometer diameter
agglomerates and channels with a bilayer size depth were identified.
1 INTRODUCTION
Sensing, monitoring, and control are natural tasks
performed by any living organism and also an
increasing part of the humankind activity in which
technology is directly or indirectly involved in day-
to-day life. Numerous sensors are constantly being
developed and the advent of the nanomaterials
increases even further their variety and applicability.
A new branch of potential platform for sensors is
being widely researched by using polyectrolytes
nano-thin films ionically self-assembled in bilayers
via what is so-called layer-by-layer (LbL) technique
(Decher and Hong, 1991, Oliveira J r., 2001). The
production of a bilayer involves sequential
adsorption of two opposite charged polyelectrolytes
from an aqueous solution, one by one, onto an inert
substrate such as glass or metal. The process is
repeated for attaining multi-layers. This simple and
inexpensive technique is also expected to deliver
films with high thermal stability and robust to
critical solvent and alkaline environments which are
the ideal platform for sensors.
Among several polyeletrolytes bilayer pairs for
building this platform, the ones made by the
polyanic poly[1-[4-(3-carboxy-4-hydroxyphenylazo)
benzene sulfonamide]-1,2-ethanediyl, sodium salt]
(PAZO) and the polycationic poly (allylamine
hydrochloride) (PAH) have a great potential and
results have already been reported (Mermut and
Barrett, 2001 and Ferreira, 2007).
In addition, selective molecules (organic,
inorganic, enzymes, DNA, etc) can be equally
mobilized onto the last layers in order to react
chemically to a particular analyte of interest.
458

The interplay of a broad set of parameters related
to the measurement such as the temperature, pH,
hysteresis, and their time-domain dynamic (drift)
should be carefully quantified aiming to control or
more likely to correct theirs effects via a posteriori
software analysis. This optimization towards
reliability can enhance the enormous commercial
potential of the LbL sensors mainly as a monitor in
an environment with no lab-control.
This work reports preliminary results on the
temperature effects on a LbL film sensor plataform
built with PAH/PAZO bilayers and tested in rather
simple aqueous solutions containing sodium choride
(NaCl) via impedance spectroscopy technique. The
polyelectrolytye PAZO which is at the last surface
of the multilayers should attract electrostatically the
cation Na
+
, so the global electrically properties of
the multilayer are expected to change according to
the concentration of this cation and therefore NaCl
concentration is sensed. In addition to this,
prelimilary analysis on the sensor film thickness and
morphology are presented by using ellipsometry and
atomic force microscopy (AFM) techniques.
2 EXPERIMENTAL
The sensor was produced from polyelectrolytes LbL
films deposited onto substrates of BK7 optical glass
where gold interdigitated electrodes were deposited
by vacuum evaporation. The sensor effective area
was about 2x5mm
2
, the interspace between the lines
was 20m and their width and thickness were 2m
and 0.2m, respectively. These dimensions were
measured by a Dektak perfilomoter and an optical
microscopy Olympicus SZ-PT.
The polyelectrolytes poly [ 1-[4-(3-carboxy-4-
hydroxyphenylazo) benzenesulfonamido]-1,2-
ethanediyl, sodium salt] (PAZO) and the
poly(allylamine hydrochloride)(PAH) (average M
w
=
50,000-60,000g/mol) were acquired from Sigma-
Aldrich. The PAH polyelectrolyte aqueous solution
with concentration of 10
-2
M was prepared by
dissolving this polyectrolyte in deionised water with
a resistivity of 18.2Mcm supplied by a Millipore
system (Milli-Q, Millipore GmbH). The PAZO
aqueous solution also with a concentration of 10
-2
M
was obtained by dissolving this polyelectrolyte on an
aqueous buffer solution of pH=10. The
polyelectrolyte concentrations were based on the
molecular weight repeat unit and the buffer solution
was prepared mixing a 0.05M sodium hydrogen
carbonate (NaHCO
3
) aqueous solution with a 0.1M
sodium hydroxide (NaOH) solution in a proportion
of 500:107(v/v) (Ferreira (a), 2007). The PAZO
solution was also filtered with a 5 mm thick and 50
m porous diameter ceramic filter.
The LbL films were prepared by immersing the
substrate with the interdigitated electrodes into the
PAH solution for 5 minutes, washed 3 times into
water for a total of 10s, and then immersed into the
PAZO solution for the same 5 minutes and equally
washed but into the buffer solution instead of water.
This procedure leads to a production of a bilayer and
repeated until the 5 bilayers were obtained. Finally,
the thin film was dried with a nitrogen flux.
The sensor impedance measurements were
carried out by a Precision Impedance Analyser
Agilent 4294A (40Hz-110MHz, 1mHz resolution,
GPIB connection). The root mean square oscillator
voltage signal level was 50 mV.
The film thickness was measured using a
spectroscopic ellipsometer model HORIBA J obin
Yvon UVISEL. A three layer model was used
assuming the sensor is composed by a film layer on
the top of a 1 mm thick BK7 glass substrate and
another on its back. The spectral range used was 1.5-
6.5 eV (531-2302 nm) with a 0.025 eV increment.
The AFM measurements were performed by a
Dimension 3100 SPM with a Nanoscope IIIa
controller from Digital Instruments (DI) under
ambient conditions in tapping mode
TM
. A
commercial tapping mode etched silicon cantilever
probe from DI (constant force of 42N/m, resonance
frequency of 320kHz) and a 90x90 m
2
scanner
were used. The scan rate was 1.51 Hz. The image
resolution was fixed to 256256 pixels.
3 RESULTS AND DISCUSSION
The influence of the NaCl concentration on the
sensor impedance (Z=Z+iZ) were analysed. Real
(Z) and imaginary (Z) parts, module and phase of
Z, and Nyquist representation were all studied. The
clearest way for visualise that influence was by
using the Z part against the impedance analyser
frequencies which is presented in the graph of fig.1.
The sensor was immersed into solutions of different
NaCl concentrations from the lowest concentration,
virtually null corresponding to Milli-Q pure water,
to the highest of 1M. Impedance spectra
measurements within several weeks show systematic
drifts for all concentrations but not to a point to
overlap any two consecutive values, excepted at low
frequencies below 5 kHz.
POLYMERIC FILM SENSORS BASED ON PAH-PAZO IONIC SELF-ASSEMBLED MULTI-NANOLAYERS
459

10
2
10
3
10
4
10
5
10
6
10
7
10
8
10
100
1k
10k
100k
possible values to
discriminate the
concentrations

R
e
a
l

c
o
m
p
o
n
e
n
t

o
f

t
h
e

I
m
p
e
d
a
n
c
e

Z
'

(

)
frequency (Hz)
NaCl in M
1
5x10
-1
2x10
-1
10
-1
10
-2
10
-3
10
-4
10
-5
10
-6
10
-7
milli-Q
water

Figure 1: Real part, Z, of the sensor impedance versus the
impedance analyser frequencies for the sensor immersed
in NaCl aqueous solutions with different concentrations.
The overall Z curves show a complex behaviour
with the frequency and concentration leading to a
short window for an effective analysis which is
empirically delimited (e.g. by the two lines drawn in
this figure). In the lower limit at 8kHz, a
concentration up to 10
-5
M can be easily measured
and possibly 10
-6
M too if sufficient statistics and at
least temperature measurements of the aqueous
medium are available as will be seen further on.
At the upper frequency of 40 kHz the sensor is
robust since the Z values are well distinguished for
each concentration, in addition to be less sensitivity
to the frequency. Values of 10
-5
M can not be really
exceeded. However, in this region the sensor thermal
stability is very high as it is shown in the plot of
fig.2, where the impedance spectra are shown for
NaCl solutions of 10
-4
M and 10
-3
M concentrations
over a temperature range of 9 to 62C and 9.0 to
26C, respectively. Around 40kHz, Z variation is
almost the same for temperatures close to 9C (
2
)
or to 26C (
1
) for these concentrations. However, it
is clear that the 10
-4
M concentration can be
mistaken for the 10
-3
M if the right temperature
curve is not used. An overlap of these curves occurs
below 20kHz, even for small temperature variation.
Therefore, the possibility to measure lower
concentrations (e.g. down to 10
-6
M) really demands
the precise knowledge of the temperature and the
statistical values of the concentration in a laboratory
environment. A more feasible alternative to this
might be to increase the sensor sensitivity by
increasing the number of bilayers, for example.
Ellipsometry measurements inferred a precise
and consistent sensor film average thickness of
494nm. The ellipsometric measurements were
20k 30k 40k 50k 60k 70k 80k 90k 100k
400.0
600.0
800.0
1.0k
2.1k
2.2k
2.3k
2.4k
2.5k

2
possible value
to discriminate
the concentrations
R
e
a
l

c
o
m
p
o
n
e
n
t

o
f

t
h
e

I
m
p
e
d
a
n
c
e

Z
'

(

)
frequency (Hz)
Temperature (
o
C)
at 10
- 4
M
62
44
33
26
15
11
8.8
------------------------
26 (10
- 5
M)
26 (10
- 3
M)
9.0 (ibidem)

1

Figure 2: The real part, Z, of the sensor impedance versus
the impedance analyser frequencies for the sensor
immersed in NaCl aqueous solutions with few different
concentrations and temperatures.
carried out about 8 mm away form the interdigitated
electrodes where only the PAH/PAZO film
deposited onto the glass substrate was presented.
Topography measurements of the PAH/PAZO
films were obtained by AFM scanning over the film
sensor at 2 mm and 8 mm away from the
interdigitated electrodes, again in regions where only
the PAH/PAZO film deposited onto the glass
substrate was presented. In the fig.3 two typical 3-D
reconstructions of the topography, both with a scan
window of 5x5m
2
, are presented in a sensor used in
a NaCl aqueous solution. In the fig.3a, a relative
smooth surface is observed except for the presence
of agglomerates with a micrometer size diameter and
at least 60 nm height. These structures are possibly
made or a result of NaCl attachment to the surface.
The root mean square roughness (Rrms) varied over
7-13nm for this type of scan window. These values
further reduce to 3-4nm for scan windows of
11m
2
. In the fig3b, clear channels were shown at
the surface possibly due to the effect of the electrical
currents, since the electrical field created was
relatively high, i.e. of the order of 2.5kV/m. The
creation of such structure should be avoided in order
to keep the multilayer integrity and thus the sensor
reliability. A worthwhile attempt to minimise this
might be the reduction of oscillator voltage applied
by the impedance analyser. In addition, the reduction
of the upper frequency could also help since no
useful information can be extracted for values much
beyond the optimum frequency (by 40kHz in this
case) where Z substantially reduces and the current
flow through the sensor increases accordingly.
The channels depth can be roughly estimated as
10nm from the topographic AFM measurements
BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
460

shown in fig.3b. This corresponds to the size of one
bilayer assuming a linear scaling between the
number of bilayers with the film thickness.

Figure 3: AFM 3-D topography of the PAH/PAZO film:
(a) in a typical region; (b) where channels were formed.
As a remark, the AFM morphology of samples
that have not been immersed into a NaCl aqueous
solution showed no agglomerates such as the ones
discussed earlier and their Rrms values are lower,
regardless the scan window values used, i.e. they are
3-5nm, 1.5nm, and 1.6nm for 55m
2
, 22m
2
, and
11m
2
, respectively. The particular value of 1.6nm
here is about one magnitude smaller from that of
~13nm, obtained from previous studies (Ferreira,
2008). This indicates that filtering the polyetrolytes
solution such as for the PAZO here may play an
important role for reducing the roughness.
4 CONCLUSIONS
A sensor platform composed by an interdigitated
electrode covered by a tailored film of 5 bilayers of
polyelectrolytes PAH-PAZO was successful built
and tested in NaCl aqueous solutions at relative low
concentrations around few micro moles per litre.
The effect of temperature on the impedance
measurements was addressed in a limited range of
this variable. Nevertheless, the preliminary results
indicate that the knowledge of this parameter is vital
for attaining accuracy close to one order of
magnitude in useful range of concentrations.
The film thickness and morphology were
characterized by using ellipsometry and atomic force
microscopy techniques. The first allowed to obtain
precise and consistent values for the multilayer film
thickness of about few tens of nanometers. From the
second, relative low values of surface root mean
square roughness, about few nanometers, were
measured and channels with a bilayer size depth and
micrometer diameter agglomerates were also
identified.
ACKNOWLEDGEMENTS
The authors thanks the plurianual financial support
from "Fundao para a Cincia e Tecnologia-FCT",
Portugal. CR is supported by the programme
Compromisso com a Cincia also from FCT.
REFERENCES
Decher, G. and Hong, J .D., 1991. Ber. Bunsen-Ges. Phys.
Chem., 95, 1430.
Ferreira, Q., A., 2008, Estudo da Formao de Filmes
Nanoestruturados para Aplicao em Fotnica, PhD
thesis, FCT, New University of Lisbon, 97p.
Ferreira, Q., Gomes, P.J ., Raposo, M., Giacometti, J . A.,
Oliveira J r., O. N. and Ribeiro, P.A., 2007. J . Nanosci.
Nanotechnol., 7, 2659.
Ferreira, Q., Gomes, P.J ., Maneira, M.J .P., Ribeiro, P.A.,
Raposo, M., 2007 Sensors and Actuators B: Chemical,
126, 311.
Ferreira, Q., Gomes, P.J ., Nunes, Y., Maneira, M.J.P.,
Ribeiro, P.A., Raposo, M., 2007, Microelectronic
Engineering Journal, 84(3), 506.
Mermut, O. and Barrett C. J ., 2001, Analyst, 126, 1861.
Oliveira J r., O. N., Raposo, M. and Dhanabalan, A., in
Handbook of Surfaces and Interfaces of Materials, H.
S. Nalwa, Ed. Academic Press, New York (2001),
Vol. 4, Chapter 1. p.1-63.
(a)
(b)
POLYMERIC FILM SENSORS BASED ON PAH-PAZO IONIC SELF-ASSEMBLED MULTI-NANOLAYERS
461
AUTHOR INDEX

Afonso, J . ................................................. 422
guas, H. ................................................. 458
Albiez, J . .................................................. 414
Albuquerque, E. ....................................... 355
Amado, A. ............................................... 131
Ambrosini, E. .......................................... 245
Amini, N. ................................................. 220
Antunes, L. .............................................. 268
Aperta, J . ................................................. 430
Arajo, P. ................................................. 402
Arita, K. ................................................... 157
Arney, D. ................................................... 52
Arrigan, D. ................................................. 83
Artacho, I. .................................................. 17
Atassi, I. ..................................................... 75
Atherton, S. ............................................... 13
Atienza, D. ................................................. 88
Azeka, L. ................................................. 231
Bakounine, N. ............................................ 83
Ballesta-Claver, J . ................................... 343
Balluch, B. ................................................. 75
Baranauskas, G. ......................................... 67
Barkauskas, J . .......................................... 448
Baron, N. ................................................. 290
Bastos-Filho, T. ........................................... 7
Belean, B. ................................................ 359
Beni, V. ..................................................... 83
Benmakrouha, F. ..................................... 335
Bentley, W. .............................................. 109
Berhane, R. .............................................. 339
Bertschi, M. ..................................... 214, 368
Bexiga, V. ........................................ 192, 355
Biagetti, G. ................................................ 39
Bian, T. .................................................... 296
Blackman, M. .......................................... 173
Blanco, R. .................................................. 17
Bloebaum, R. ........................................... 178
Blondy, P. .................................................... 3
Bonfanti, A. ............................................... 67
Bongardt, B. ............................................ 414
Boquete, L. ................................................ 17
Borda, M. ................................................. 359
Borges, B. ................................................ 458
Borges, G. .................................................. 97
Borghi, T. .................................................. 67
Botelho, G. ............................................... 394
Branco, J . ................................................. 310
Brogueira, P. ............................................ 458
Bugalho, R. ................................ 31, 192, 355
Bukauskas, V. .......................................... 448
Cabodevila, G. ......................................... 290
Cabral, J . .................................................. 256
Capitan-Vallvey, L. ................................. 343
Cardoso, V. .............................................. 394
Carmo, J . .......................................... 281, 380
Carrio, B. .......................................... 31, 355
Carvalho, J . ................................................ 97
Chachulski, D. ......................................... 441
Chen, J . .................................................... 209
Chester, G. ............................................... 251
Choi, M. ................................................... 235
Correia, J . ........................... 61, 281, 380, 422
Correia, P. ................................................ 430
Correia, V. ............................................... 256
Cruz, A. .................................................... 186
Dabiri, F. .................................................. 220
Dalmay, C. ................................................... 3
Delis, A. ..................................................... 97
Daz, V. ...................................................... 88
Dib, L. ...................................................... 276
Dugosz, R. ........................................ 46, 364
Doyle, J . ..................................................... 83
Dubey, R. ................................................. 339
Dubuc, D. ................................................. 398
Dykstra, P. ....................................... 104, 109
Dyszkiewicz, A. ....................................... 441
Edetsberger, M. .......................................... 75
Elamary, G. .............................................. 251
Escalona, R. ............................................. 262
Esteban, V. ............................................... 137
Evans, C. .................................................... 13
Fadiga, L. ................................................... 67
Fazakas, A. .............................................. 359
Fernandes, A. ........................................... 410
Ferrante, S. ............................................... 245
Ferreira, A. ................................................... 7
Ferreira, C. ......................................... 31, 355
Ferreira, D. ................................................. 61
Ferreira, M. ................................ 31, 192, 355
Ferrigno, G. .............................................. 245
463

AUTHOR INDEX (CONT.)

Figueroa, M. ............................................ 262
Folgheraiter, M. ....................................... 414
Foursov, M. ............................................. 335
Friedberger, A. ........................................ 384
Fujita, H. .................................................. 398
Gaiolas, C. ............................................... 402
Garca, J . ...................................................... 7
Gaubitzer, E. .............................................. 75
Gaudet, V. ................................................. 46
Ghodssi, R. ...................................... 104, 109
Godinho, J . .............................................. 355
Goldman, J . ............................................... 52
Gomes, P. ................................................ 458
Gonalves, F. ........................................... 355
Goncalves, L. ........................................... 380
Gouveia, I. ............................................... 268
Grenier, K. ............................................... 398
Guilherme, J . ........................................... 430
Gureviciene, V. ....................................... 448
Gusmeroli, R. ............................................ 67
Gutirrez, L. .............................................. 88
Gvichiya, K. .............................................. 75
Heitzinger, C. ............................................ 24
Heller, C. ................................................. 384
Helwig, A. ............................................... 384
Henriques, M. ............................................ 61
Herrera, D. ............................................... 262
Hespel, C. ................................................ 335
Hespel, J . ................................................. 335
Hribek, M. .............................................. 376
Hughes, D. ................................................. 13
Isaacson, B. ............................................. 178
Ishii, D. .................................................... 113
J auberteau, M. ............................................. 3
J imnez, M. ............................................... 88
J onsson, F. ............................................... 209
J ung, H. ........................................... 286, 426
Kalbacova, M. ......................................... 347
Kameyama, K. ......................................... 326
Kaniusas, E. ............................................. 304
Khaled, N. ................................................. 88
Kim, H. ............................................ 286, 426
Kim, T. .................................................... 426
Kim, Y. .................................................... 426
Kimoto, A. ............................................... 372
Kirchner, F. .............................................. 414
Klettner, F. ............................................... 384
Knite, M. .................................................. 117
Koev, S. ........................................... 104, 109
Khler, G. .................................................. 75
Kolasa, M. ................................................ 364
Kpiski, P. ............................................... 441
Krauss, J . .................................................. 214
Kromka, A. .............................................. 347
Kumemura, M. ......................................... 398
Kurdthongmee, P. .................................... 151
Kurdthongmee, W. .................................. 151
Lacey, G. .................................................. 320
Lalloue, F. .................................................... 3
Lanceros-Mndez, S. ............................... 394
Lantada, A. ...................... 137, 145, 437, 453
Leal, G. .................................................... 314
Lee, H. ..................................................... 286
Lee, I. ......................................................... 52
Leong, C. ......................................... 192, 355
Lim, H. ..................................................... 286
Lindner, P. ............................................... 384
Lopes, A. .................................................. 430
Lpez, H. ................................................. 422
Lorenzo-Yustos, H. .......... 137, 145, 437, 453
Lous, P. .......................................... 192, 355
Lucas, J . ................................................... 402
Lucas, M. ................................................. 157
Machado, P. ..................................... 192, 355
MacLeod, R. ............................................ 178
Marasovic, T. ........................................... 300
Marco, M. ................................................ 226
Marioli, S. ................................................ 290
Martnez-Olmos, A. ................................. 343
Martins, G. ............................................... 239
Martins, M. .............................................. 256
Martins, P. ................................................ 394
Materna, T. .............................................. 304
Matthews, J . ............................................. 220
Mauser, N. ................................................. 24
McCaffrey, C. ............................................ 83
McHale, G. ................................................ 13
Medeiros, M. ............................................ 410
Meixner, L. .............................................. 384
Michalkov, L. ........................................ 347

BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
464

AUTHOR INDEX (CONT.)

Micheli, G. ................................................. 88
Miguel, J . ................................................... 17
Miguel, R. ................................................ 402
Mil-Homens, P. ....................................... 239
Minas, G. ........................................... 61, 394
Mohan, V. ................................................ 198
Molz, R. ................................................... 384
Moraes, R. ............................................... 161
Morgado, P. ..................... 137, 145, 437, 453
Morici, A. .................................................. 39
Moura, R. ........................................... 31, 355
Mller, G. ................................................ 384
Muoz Sanz, J . ........................................ 145
Muoz-Garca, J . ............. 137, 145, 437, 453
Munoz-Guijosa, J . ........... 137, 145, 437, 453
Music, J . .................................................. 300
Nascimento, F. ........................................... 97
Neasham, J . ............................................. 251
Neumeier, K. ........................................... 384
Neves, C. ................................................. 314
Neves, P. .......................................... 192, 355
Newton, M. ................................................ 13
Noshadi, H. .............................................. 220
Nunes, U. ................................................. 430
ONeill, J . ................................................ 310
Ogurtsov, V. .............................................. 83
Oliveira, R. ................................................ 61
Oliynyk, A. ................................................ 67
Olmo, H. .................................................. 453
Ortega, S. ................................................... 17
Ortigo, C. ......................................... 31, 355
Otero, J . ........................... 137, 145, 437, 453
Ouchi, K. ................................................. 326
Paiva, M. ................................................. 231
Palma, A. ................................................. 343
Palma, S. .................................................. 239
Park, M. ........................................... 286, 426
Pawlus, B. ................................................ 441
Payne, G. ......................................... 104, 109
Pazart, L. ................................................. 290
Pedrocchi, A. ........................................... 245
Pereira, L. ................................................ 131
Prez, J . ................................................... 186
Piedade, F. ....................................... 192, 355
Pieralli, C. ................................................ 290
Pinheiro, J . ................................. 31, 192, 355
Pires, G. ................................................... 430
Poe, P. ................................................... 441
Pothier, A. .................................................... 3
Quaresma, C. ........................................... 310
Queiroz, J . ................................................ 268
Quintas, M. .................................................. 7
Ramn-Azcn, J . ..................................... 226
Raposo, M. ............................................... 458
Razumiene, J . ........................................... 448
Rebouta, L. .............................................. 394
Recas, J . ..................................................... 88
Rego, J . .................................................... 192
Reidt, U. ................................................... 384
Reis, N. .................................................... 402
Relvas, P. ................................................. 355
Renevey, P. .............................................. 214
Rezek, B. .................................................. 347
Ribeiro, A. ............................................... 430
Ribeiro, C. ................................................ 458
Ribeiro, P. ................................................ 458
Ribeiro-Claro, P. ...................................... 131
Rimminen, H. .......................................... 125
Rincn, F. .................................................. 88
Ringhofer, C. ............................................. 24
Risti, S. ................................................... 376
Rivetti, A. ................................................ 355
Roach, P. .................................................... 13
Rocha, A. ................................................... 97
Rocha, J . .................................................. 256
Rocha, R. ................................................. 380
Rodrigues, P. .............................. 31, 192, 355
Rodrigues, S. ............................................ 402
Rodrguez, . ........................................... 226
Rodrguez, J . .............................................. 17
Rossini, L. ........................................ 214, 368
Rowe, P. ................................................... 198
Ruano, M. ................................................ 410
Rubloff, G. ............................................... 109
Sakale, G. ................................................. 117
Salgado, N. .............................................. 300
Sanchez, F. ............................................... 226
Snchez, J . ................................................... 7
Snchez-lez, M. ....................................... 88
Santos, P. ................................................. 458

xxxxx
465

AUTHOR INDEX (CONT.)

Sanz, J . ..................................... 137, 437, 453
Sarcinelli-Filho, M. ..................................... 7
Sarrafzadeh, M. ....................................... 220
Schauer, T. ............................................... 245
Secca, M. ................................................. 310
Selberherr, S. ............................................. 24
Sepponen, R. ........................................... 125
Serrado Nunes, J . ..................................... 394
Setkus, A. ................................................ 448
Shida, K. .................................................. 372
Shimomura, M. ................................ 113, 390
Shinonaga, Y. .......................................... 157
Silva, D. ................................................... 161
Silva, H. ................................................... 239
Silva, J . .............................................. 31, 355
Silva, M. .................................................. 355
Simes, R. ............................................... 422
Sirgado, A. ............................................... 430
Skelin, A. ................................................. 300
Smetana, W. .............................................. 75
Song, S. ................................................... 286
Spinelli, A. ................................................. 67
Stancic, I. ................................................. 300
Stinstra, J . ................................................ 178
Stork, M. .................................................. 204
Su, M. ...................................................... 167
Supuk, T. ................................................. 300
Suzuki, T. ................................................ 326
Sverdlov, V. ............................................... 24
Szeles, J . .................................................. 304
Takahashi, M. .......................................... 326
Tan, Y. ..................................................... 167
Tanaka, M. ............................................... 390
Taninaka, Y. ............................................ 372
Tavares, M. .............................................. 161
Teixeira, C. .............................................. 410
Teixeira, I. ....................................... 192, 355
Teixeira, J . ....................................... 192, 355
Tenhunen, H. ........................................... 209
Teteris, V. ................................................ 117
Theurillat, P. ............................................ 214
Toshiyoshi, H. ......................................... 398
Toi, D. .................................................. 376
Trefny, M. ............................................... 204
Trefny, Z. ................................................. 204
Trindade, A. ............................... 31, 192, 355
Tsuruma, A. ............................................. 390
Tupureina, V. ........................................... 117
Turchetti, C. ............................................... 39
Twomey, K. ............................................... 83
Ukraintsev, E. .......................................... 347
Vahdatpour, A. ........................................ 220
Valamatos, M. .......................................... 239
Valencia-Miron, M. ................................. 343
Valera, E. ................................................. 226
Valsan, G. ................................................ 198
Varcoe, B. ................................................ 173
Varela, J . .................................... 31, 192, 355
Varoneckas, G. ........................................ 304
Velde, F. .................................................. 131
Verjus, C. ................................................. 214
Vetter, R. .................................................. 214
Vieira, P. .................................................. 314
Vilario, F. ............................................... 320
Villa, P. ...................................................... 17
Wacogne, B. ............................................ 290
Wang, Z. .................................................. 167
Whitehead, S. ............................................. 52
Windbacher, T. .......................................... 24
Wolf, H. ................................................... 384
Yabu, H. ................................................... 113
Yamamoto, S. .......................................... 390
Yang, G. ................................................... 209
Ying, S. .................................................... 339
Yu, L. ....................................................... 104
Yulong, Z. ................................................ 296
Zajdel, J . .................................................. 441
Zambra, G. ................................................. 67
Zanchi, V. ................................................ 300
Zheng, L. .................................................. 209
Zhuangde, J . ............................................. 296
ivkovi, Z. ............................................. 376








BIODEVICES 2009 - International Conference on Biomedical Electronics and Devices
466