Computational Techniques in

Neuroscience

The text discusses the techniques of deep learning and machine learning
in the field of neuroscience, engineering approaches to study the brain
structure and dynamics, convolutional networks for fast, energy-efficient
neuromorphic computing, and reinforcement learning in feedback control.
It showcases case studies in neural data analysis.

Features:

• Focuses on neuron modeling, development, and direction of neural

circuits to explain perception, behavior, and biologically inspired
intelligent agents for decision making
• Showcases important aspects such as human behavior prediction
using smart technologies and understanding the modeling of nervous
systems
• Discusses nature-inspired algorithms such as swarm intelligence, ant
colony optimization, and multi-agent systems
• Presents information-theoretic, control-theoretic, and decision-theoretic
approaches in neuroscience
• Includes case studies in functional magnetic resonance imaging (fMRI)
and neural data analysis

This reference text addresses different applications of computational neuro-

sciences using artificial intelligence, deep learning, and other machine learning
techniques to fine-tune the models, thereby solving the real-life problems
prominently. It will further discuss important topics such as neural rehabili-
tation, brain-computer interfacing, neural control, neural system analysis, and
neurobiologically inspired self-monitoring systems. It will serve as an ideal
reference text for graduate students and academic researchers in the fields of
electrical engineering, electronics and communication engineering, computer
engineering, information technology, and biomedical engineering.
Computational Methods for Industrial Applications
Series Editor: Bharat Bhushan

In today’s world Internet of Things (IoT) platforms and processes in con­

junction with the disruptive blockchain technology and path-breaking
artificial intelligence (AI) algorithms lay out a sparking and stimulating
foundation for sustaining smarter systems. Further computational intelli­
gence (CI) has gained enormous interests from various quarters in order to
solve numerous real-world problems and enable intelligent behavior in
changing and complex environment. This book series focuses on varied
computational methods incorporated within the system with the help of AI,
learning methods, analytical reasoning and sense making in big data. Aimed
at graduate students, academic researchers and professionals, the proposed
series will cover the most efficient and innovative technological solutions
for industrial applications and sustainable smart societies in order to alter
green power management, effect of carbon emissions, air quality metrics,
industrial pollution levels, biodiversity and ecology.

Blockchain for Industry 4.0: Emergence, Challenges, and Opportunities

Anoop V.S, Asharaf S, Justin Goldston and Samson Williams

Intelligent Systems and Machine Learning for Industry: Advancements,

Challenges and Practices
P.R Anisha, C. Kishor Kumar Reddy, Nguyen Gia Nhu, Megha Bhushan,
Ashok Kumar, Marlia Mohd Hanafiah
Computational
Techniques in
Neuroscience

Edited by
Kamal Malik
Harsh Sadawarti
Moolchand Sharma
Umesh Gupta
Prayag Tiwari
First edition published 2024
by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-2742

and by CRC Press

4 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN

CRC Press is an imprint of Taylor & Francis Group, LLC

© 2024 selection and editorial matter, Kamal Malik, Harsh Sadawarti,

Moolchand Sharma, Umesh Gupta and Prayag Tiwari; individual chapters,
the contributors

Reasonable efforts have been made to publish reliable data and information,
but the author and publisher cannot assume responsibility for the validity of
all materials or the consequences of their use. The authors and publishers have
attempted to trace the copyright holders of all material reproduced in this
publication and apologize to copyright holders if permission to publish in this
form has not been obtained. If any copyright material has not been acknowledged
please write and let us know so we may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be
reprinted, reproduced, transmitted, or utilized in any form by any electronic,
mechanical, or other means, now known or hereafter invented, including
photocopying, microfilming, and recording, or in any information storage or
retrieval system, without written permission from the publishers.

For permission to photocopy or use material electronically from this work, access
www.copyright.com or contact the Copyright Clearance Center, Inc. (CCC),
222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. For works that
are not available on CCC please contact [email protected]

Trademark notice: Product or corporate names may be trademarks or registered

trademarks and are used only for identification and explanation without intent
to infringe.

ISBN: 978-1-032-46128-1 (hbk)

ISBN: 978-1-032-50343-1 (pbk)
ISBN: 978-1-003-39806-6 (ebk)
DOI: 10.1201/9781003398066

Typeset in Sabon
by MPS Limited, Dehradun
Dr. Kamal Malik would like to dedicate this book to her father,
Sh. Ashwani Malik, her mother, Smt. Shakuntla Malik, and her brother,
Dr. Shiv Malik, for their constant support and motivation; I would also
like to give my special thanks to the publisher and my other co-editors
for believing in my abilities. Above all, a humble thanks to the Almighty
for this accomplishment.
Dr. Harsh Sadawarti would like to dedicate this book to his father,
Sh. Jagan Nath Sadawarti, his mother, Smt. Krishna, and his wife,
Ritcha, for their constant support and motivation; I would also like to
thank the publisher and my other co-editors for having faith in my
abilities. Above all, a humble thanks to the Almighty for this
accomplishment.
Mr. Moolchand Sharma would like to dedicate this book to his father,
Sh. Naresh Kumar Sharma, and his mother, Smt. Rambati Sharma, for
their constant support and motivation, and his family members, including
his wife, Ms. Pratibha Sharma, and son, Dhairya Sharma. I also thank
the publisher and my other co-editors for believing in my abilities.
Dr. Umesh Gupta would like to dedicate this book to his mother,
Smt. Prabha Gupta, and his father, Sh. Mahesh Chandra Gupta, for their
constant support and motivation, and his family members, including
his wife, Ms. Umang Agarwal, and son, Avaya Gupta. I also thank the
publisher and my other co-editors for believing in my abilities. Before
beginning and after finishing my endeavor, I must appreciate the
Almighty God, who provides me with the means to succeed.
Dr. Prayag Tiwari would like to dedicate this book to his father &
his mother for their constant support and motivation and his family
members. I also thank the publisher and my other co-editors for
believing in my abilities.
Contents

About the Book ix

Preface xi
Editors xiii
Contributors xvii

1 Dynamic intuitionistic fuzzy weighting averaging

operator: A multi-criteria decision-making
technique for the diagnosis of brain tumor 1
VIJAY KUMAR, H D ARORA, AND KIRAN PAL

2 Neural modeling and neural computation in a

medical approach 19
SIMRAN SINGH, ANSHIKA GUPTA, AND KALPANA KATIYAR

3 Neural networks and neurodiversity: The key

foundation for neuroscience 42
HERA FATMA, HARSHIT MISHRA, AND KALPANA KATIYAR

4 Brain waves, neuroimaging (fMRI, EEG,

MEG, PET, NIR) 67
SURBHI KUMARI AND AMIT KUMAR DUTTA

5 EEG: Concepts, research-based analytics,

and applications 83
RASHMI GUPTA, SONU PUROHIT, AND JEETENDRA KUMAR

vii
viii Contents

6 Classification of gait signals for detection of

neurodegenerative diseases using log energy
entropy and ANN classifier 101
PRASANNA J, S. THOMAS GEORGE, AND M.S.P SUBATHRA

7 An optimized text summarization for healthcare

analytics using swarm intelligence 113
REKHA JAIN, PRATISTHA MATHUR, AND MANISHA

8 Computer aided diagnosis of neurodegenerative

diseases using discrete wavelet transform and
neural network for classification 127
PRASANNA J, S. THOMAS GEORGE, AND M.S.P SUBATHRA

9 EEG artifact detection and removal techniques:

A brief review 143
SANDHYALATI BEHERA AND MIHIR NARAYAN MOHANTY

10 Analysis of neural network and neuromorphic

computing with hardware: A survey 188
MANISH BHARDWAJ, KAILASH NATH TRIPATHI,
YOGENDRA NARAYAN PRAJAPATI, AND ANALP PATHAK

11 Analysis of technology research and ADHD

with the neurodivergent reader: A survey 207
MANISH BHARDWAJ, JYOTI SHARMA, ANALP PATHAK,
VINAY KUMAR SHARMA, AND MAYANK TYAGI

Index 221
About the Book

Computational neuroscience is an interdisciplinary field that studies the

evolution of the nervous system and its working principle, and how it
thinks! In computational neuroscience, multi-scale models look at how the
brain works, from molecules, cells, and networks to cognition and
behavior. It is a way to learn how the nervous system grows and works
at different structural levels. Several practitioners and researchers must
provide an edge toward artificial intelligence and machine learning concepts
to deal with complex real-life problems. Biological neuroscience concepts
can be included in different computational paradigms to provide a unique
and exciting feature.
Research in computational neuroscience and psychology helps the research
community to understand how people act with each other. A psychologist
generally perceives a human’s feelings or attitude, which decides the person’s
behaviors. According to computational neuroscientists, the behavior of
humans is caused by how well a group of neurons spreads information in a
certain way to the brain. As per their understanding, the brain is considered a
black box where all connections and processes are abstract. Many inputs and
environmental variables are to be fed inside and then processed. As a result,
the behaviors of humans will be in action mode. Computational neuroscience
tries to discover how the brain works, so identifying people’s behaviors
before acting will be easy.
Machine learning studies how statistical models and algorithms can make
computer systems do things without being directed on how to act. Some
computational neuroscientists have explored machine-learning approaches
for their research work. However, the research’s complexity and cost still need
to be improved for machine learning. For example, many computational
neuroscience researchers have devised dynamic neural network models over
human behavior datasets. The use of machine learning is very insightful in
studying how the brain processes and stores information, but soft computing
approaches must be noticed for a computational neuroscience study. So, this
book will give the ensemble learning of different soft computing approaches
and machines and a deep understanding of complex human behavior datasets.

ix
x About the Book

This book explains the effective utilization of different basic computer-

devised methods, which will be considered to determine the efficient nervous
system. We’ll look at how computer science can help to understand visionary
control, sensory-motor control, learning, and memory-based tasks. In this
book, various specific emerging topics will cover neuron-related information
and its network process cycle with adaptability and learning capability.
We will also show how to implement using MATLAB, Octave, and Python
to make practical neuroscience applications. This will help the reader
understand neuroscience-related computing techniques’ basic concepts
and methods. The book is aimed mainly at third- or fourth-year students
in college and beginners-level researchers who want to learn more about how
the brain processes information. The edited chapters address action
and motor control, neuromodulation, reinforcement learning, vision, and
language—the core of human cognition. It provides all necessary neuro-
science foundations beyond neurons, synapses, and brain structure and
function. The book focuses on neuron modeling, development, and direction
of neural circuits to explain perception, behavior, and biologically inspired
intelligent agents for decision making.
Preface

We are delighted to launch our book entitled “Computational

Techniques in Neuroscience” under the book series Computational
Methods for Industrial Applications Series, CRC Press, Taylor &
Francis Group. Computational neuroscience is an interdisciplinary field
of neural computing, engineering, and artificial intelligence. Various
neuroscientists, cognitive researchers, psychologists, and data scientists
can publish their work that bridges the gap between neuroscientists
and data scientists. Innovative technologies such as artificial intelligence,
deep learning, machine learning, and Internet of Things influence
today’s modern world. This book presents the various approaches,
techniques, and applications in computational intelligence and neural
engineering. It is a valuable source of knowledge for researchers,
engineers, practitioners, and graduate and doctoral students working
in the same field. It will also be helpful for faculty members of graduate
schools and universities. Around 30 full-length chapters have been
received. Among these manuscripts, 11 chapters have been included in
this volume. All the chapters submitted were peer-reviewed by at
least two independent reviewers and provided with a detailed review
proforma. The comments from the reviewers were communicated to the
authors, who incorporated the suggestions in their revised manuscripts.
The recommendations from two reviewers were considered while
selecting chapters for inclusion in the volume. The exhaustiveness of
the review process is evident, given the large number of articles received
addressing a wide range of research areas. The stringent review process
ensured that each published chapter met the rigorous academic and
scientific standards.
We would also like to thank the authors of the published chapters
for adhering to the schedule and incorporating the review comments.
We extend my heartfelt acknowledgment to the authors, peer reviewers,

xi
xii Preface

committee members, and production staff, whose diligent work shaped

this volume. We especially want to thank our dedicated peer reviewers
who volunteered for the arduous and tedious step of quality checking and
critiquing the submitted chapters.
Editor(s)
Kamal Malik
Hash Sadawarti
Moolchand Sharma
Umesh Gupta
Prayag Tiwari
Editor(s)

Dr. Kamal Malik is currently working as a professor in CSE in the School of

Engineering and Technology at CTU Ludhiana, Punjab, India. She has
published scientific research publications in reputed international journals,
including SCI and Scopus-indexed journals such as Adhoc and Senior
Wireless Networks, 50, Engineering, Technology and Applied Sciences
Research, Journals of Advanced Research in Engineering, Research Journal
of Applied Sciences of Engineering & Technology (RJASET – Maxwell
Sciences), SSRN-Electronic Journal, Design Engineering, Indian Journal of
Science and Technology (IJST), International Journal of Computer
Applications (IJCA), and many more. She has also attended many
national and international conferences of repute like Springer, and
Elsevier in India. Her major research areas are artificial intelligence,
machine learning and deep learning, data analytics, computational
neurosciences, and bio-inspired computing. She has more than 13 years
of rich academic and research experience. She has guided 3 research
scholars and is currently guiding 8 research scholars at CT University,
Ludhiana. She has worked in renowned institutes and universities like
RIMT Mandigobindgarh, Maharishi Markandeshwar University, Mullana,
GNA University, Phagwara. She has also chaired various sessions in
Springer and Elsevier. She has also been awarded the preeminent
researcher award from Green Thinkerz Society at CII, Chandigarh. She
has completed her doctorate of philosophy in computer science from
IKGPTU Kapurthala in 2017, her masters and graduation from
Kurukshetra University, Kurukshetra, in 2009 and 2006, respectively.
Dr. Harsh Sadawarti is currently working as vice chancellor of CT
University and Professor of Computer Science and Engineering in the
School of Engineering and Technology at CTU Ludhiana, Punjab India. He
has published scientific research publications in reputed international
journals including SCI-indexed and Scopus-indexed journals such as
Journal of Intelligent and Fuzzy Systems, Journal of Applied Sciences,
International Journal of Scientific and Technology Research (IJSTR),
International Journal of Advanced Research in Engineering and

xiii
xiv Editor(s)

Technology (IJARET), International Journal of Light and Electron Optics

(Optik Elesevier Journal), International Journal of Computer Science and
Communication Engineering (IJCSCE), International Journal of Advanced
Research in Computer Science and Software Engineering, Maxwell’s
Sciences, and many more. Apart from it, he has attended many national,
international, and IEEE conferences of repute like Springer, Elsevier in India
as well as abroad. He has visited more than 20 countries in his academic
career for presenting his scientific research. His major areas of research are
machine learning, artificial intelligence, deep learning, parallel processing,
computational neurosciences, bio-inspired computing, and security in cloud
computing. He has more than 28 years of teaching, academic and research
experience in various reputed engineering institutions named as RIMT
Mandigobindgarh, Baba Banda Singh Bahadur Institute of Engg and
Technology. He has also chaired various international conferences of
Springer, Elsevier. He has guided 12 Ph.D. scholars and currently guiding
8 research scholars. He has also been awarded as a Punjab Ratan (Punjab
State Intellectuals honor) by All India Conference of Intellectuals at India
International Centre New Delhi on 26th of December, 2010. He is also an
eminent reviewer of many reputed journals like Elsevier, Springer, etc. He
has also been awarded as the best Young Vice Chancellor award from IARE
i.e., International Academic and Research Excellence Awards. He has done
his doctor of philosophy (Ph.D.) in computer science and engineering and
M.Tech (CSE) from Thapar University and his B.Tech(CSE) from Nagpur
University in 1999.
Mr. Moolchand Sharma is currently an assistant professor in the Department
of Computer Science and Engineering at the Maharaja Agrasen Institute of
Technology, GGSIPU Delhi. He has published scientific research publications
in reputed international journals and conferences, including SCI-indexed and
Scopus-indexed journals such as Expert Systems (Wiley), Cognitive Systems
Research (Elsevier),Physical Communication (Elsevier), Journal of Electronic
Imaging (SPIE), Intelligent Decision Technologies: An International Journal,
Cyber-Physical Systems (Taylor & Francis Group), International Journal of
Image & Graphics (World Scientific), International Journal of Innovative
Computing and Applications (Inderscience), and Innovative Computing and
Communication Journal (Scientific Peer-reviewed Journal). He has authored/
co-authored chapters with international publishers like Elsevier, Wiley, and
De Gruyter. He has authored/edited four books with a national/international
level publisher (CRC Press, Bhavya publications). His research areas include
artificial intelligence, nature-Inspired computing, security in cloud computing,
machine learning, and search engine optimization. He is associated with
various professional bodies like IEEE, ISTE, IAENG, ICSES, UACEE, Internet
Society, and life membership of the Universal Inovators research lab, etc. He
possesses teaching experience more than nine years. He is the co-convener of
the ICICC, DOSCI, ICDAM & ICCCN Springer Scopus-Indexed conference
 Editor(s) xv

series and ICCRDA-2020 Scopus-Indexed IOP Material Science &

Engineering conference series. He is also the organizer and co-convener of
the International Conference on Innovations and Ideas towards Patents
(ICIIP) series. He is also the advisory and TPC committee member of
the ICCIDS-2022 Elsevier SSRN Conference. He is also the reviewer
of many reputed journals like Springer, Elsevier, IEEE, Wiley, Taylor &
Francis Group, IJEECS, and World Scientific Journal, and many Springer
conferences. He is also served as a session chair in many international springer
conferences. He is a doctoral researcher at DCR University of Science &
Technology, Haryana. He completed his Post Graduation in 2012 from SRM
UNIVERSITY, NCR CAMPUS, GHAZIABAD, and graduated in 2010
from KNGD MODI ENGG. COLLEGE, GBTU.
Dr. Umesh Gupta is currently an assistant professor at the department of
Computer Science, SR University, Warangal, Telengana, India. He received a
doctor of philosophy (Ph.D.) (machine learning) from the National Institute
of Technology, Arunachal Pradesh, India. He has awarded a gold medal for
his master of engineering (M.E.) from the National Institute of Technical
Teachers Training and Research (NITTTR), Chandigarh, India, and bachelor
of technology (B.Tech.) from Dr. APJ, Abdul Kalam Technical University,
Lucknow, India. His research interests include SVM, ELM, RVFL, machine
learning, and deep learning approaches. He has published over 35 referred
journal and conference papers of international repute. His scientific research
has been published in reputable international journals and conferences,
including SCI-indexed and Scopus-indexed journals like Applied Soft
Computing (Elsevier) and Applied Intelligence (Springer), each of which is
a peer-reviewed journal. His publications have more than 158 citations with
an h-index of 8 and an i10-index of 8 on Google Scholar as of March 1, 2023.
He is a senior member of IEEE (SMIEEE) and an active member of ACM,
CSTA, and other scientific societies. He also reviewed papers for many
scientific journals and conferences in the United States and abroad. He led
sessions at the 6th International Conference (ICICC-2023), 3rd International
Conference on Data Analytics and Management (ICDAM 2023), the 3rd
International Conference on Computing and Communication Networks
(ICCCN 2022), and other international conferences like Springer ETTIS
2022 and 2023. He is currently supervising two Ph.D. students. He is the co-
principal investigator (co-PI) of two major research projects. He published
three patents in the years 2021–2023. He also published four book chapters
with Springer, CRC.
Dr. Prayag Tiwari received his Ph.D. degree from the University of Padova,
Italy. He is currently working as a postdoctoral researcher at Aalto
University. Previously, he worked as a Marie Curie Researcher at the
University of Padova, Italy. He also worked as a research assistant at
the NUST “MISiS”, Moscow, Russia. He has several publications in top
journals and conferences, including Neural Networks, Information Fusion,
xvi Editor(s)

IPM, IJCV, IEEE TNNLS, IEEE TFS, IEEE TII, IEEE JBHI, IEEE IOTJ,
IEEE BIBM, ACM TOIT, CIKM, SIGIR, AAAI, etc. His research interests
include machine learning, deep learning, quantum machine learning,
information retrieval, healthcare, and IoT. He is also associated with one
funded-based project named “Data Literacy for Responsible Decision-
Making,” short title (STN LITERACY/Marttinen). He is also the reviewer
of many reputed journals like Springer, Elsevier, IEEE, Wiley, Taylor &
Francis Group, IJEECS, and World Scientific Journal, and many Springer
conferences.
Contributors

Amit Kumar Dutta, Amity Institute of Biotechnology, Amity University

Jharkhand, Ranchi, India
Anshika Gupta, Department of Biotechnology, D. Ambedkar Institute of
Technology for Handicapped, Kanpur, Uttar Pradesh, India
Analp Pathak, Department of Information Technology, KIET Group of
Institutions, Delhi-NCR, Ghaziabad, India
H D Arora, Amity University, Noida, INDIA
Hera Fatma, Department of Biotechnology, Dr. Ambedkar Institute of
Technology for Handicapped, Kanpur, Uttar Pradesh, India
Harshit Mishra, Department of Biotechnology, Dr. Ambedkar Institute of
Technology for Handicapped, Kanpur, Uttar Pradesh, India
Jeetendra Kumar, Atal Bihari Vajpayee University, Bilaspur, Chhattisgarh,
India
Jyoti Sharma, Department of Information Technology, KIET Group of
Institutions, Delhi-NCR, Ghaziabad, India
Kalpana Katiyar, Department of Biotechnology, D. Ambedkar Institute of
Technology for Handicapped, Kanpur, Uttar Pradesh, India
Kailash Nath Tripathi, Department of AIML, ISBM College of Engineering,
Pune, India
Kiran Pal, Delhi Institute of Tool Engineering, Delhi, India
Manisha, Dept of Computer Science, Manipal University, Jaipur, India
Manish Bhardwaj, Department of Computer Science and Information
Technology, KIET Group of Institutions, Delhi-NCR, Ghaziabad, India
Mihir Narayan Mohanty, Department of Electronics and Communication
Engineering, ITER, Sikhsha ‘O’ Anusandhan (Deemed to be University),
Bhubaneswar, India

xvii
xviii Contributors

M.S.P Subathra, Dept. of Robotics Engineering, Karunya Institute of

technology and sciences, Coimbatore, Tamil Nadu, India
Mayank Tyagi, Department of Information Technology, KIET Group of
Institutions, Delhi-NCR, Ghaziabad, India
Prasanna J, Dept. of Biomedical Engineering, Karunya Institute of
Technology and Sciences, Coimbatore, India
Pratistha Mathur, Department of Information Technology, Manipal
University Jaipur, India
Rashmi Gupta, Atal Bihari Vajpayee University, Bilaspur, Chhattisgarh,
India
Rekha Jain, Department of Computer Applications, Manipal University
Jaipur, India
Sandhyalati Behera, Department of Electronics and Communication
Engineering, ITER, Sikhsha ‘O’ Anusandhan (Deemed to be University),
Bhubaneswar, India
S Dhawan, Department of Mathematics, University of Ladakh, Ladakh,
India
Surbhi Kumari, Amity Institute of Biotechnology, Amity University
Jharkhand, Ranchi, India
Sonu Purohit, Atal Bihari Vajpayee University, Bilaspur, Chhattisgarh, India
Simran Singh, Department of Biotechnology, D. Ambedkar Institute of
Technology for Handicapped, Kanpur, Uttar Pradesh, India
S. Thomas George, Dept. of Biomedical Engineering, Karunya Institute of
Technology and Sciences, Coimbatore, India
Vijay Kumar, Manav Rachna International Institute of Research Studies,
Faridabad, Haryana, India
Vinay Kumar Sharma, School of Computing Science and Engineering,
Galgotias University, Greater Noida, India
Yogendra Narayan Prajapati, Department of Computer Science and
Engineering, Ajay Kumar Engineering, Ghaziabad, India
Chapter 1

Dynamic intuitionistic fuzzy

weighting averaging operator
A multi-criteria decision-making technique
for the diagnosis of brain tumor
Vijay Kumar1, H D Arora2, and Kiran Pal3
1
Manav Rachna International Institute of Research Studies,
Faridabad Haryana, India
2
Amity University, Noida, India
3
Delhi Institute of Tool Engineering, Delhi, India

1.1 INTRODUCTION

Since the inception of computers, many scientific as well decision support

tools have been developed that make it easier for decision makers to make
decisions under unfavorable circumstances. Many soft computing tech-
niques such as fuzzy theory and its generalization provide hand held sup-
port to decision theory and allow the discipline to grow exponentially in
every sphere of the universe. Fuzzy set and its generalizations have appli-
cations in many domains, such as decision theory, information theory,
medical diagnosis, pattern recognition, etc. Moreover, it acts as an interface
and contributes to solving real-life problems with uncertain and imprecise
information. Real-life situations are always encountered with deterministic
and non-deterministic processes. Out of this, a non-deterministic process is
challenging and is taken up by statistics, if found stable. The bitter truth is
that many real-life problems are not well defined and are vague; therefore,
they are beyond the control of precise notations of mathematics. For such
vague situations, Zadeh (1965) introduced the concept of fuzzy set theory,
over the generalization of classical set theory, which has the inbuilt ability
to represent incomplete information.
From the past five decades, fuzzy theory has grown many folds. Immense
research growth has been noticed in the theory of fuzzy sets. Consequently,
fuzzy set theory and its generalized versions have emerged as a potential
area of interdisciplinary research. Among this, intuitionistic fuzzy set (IFS)
theory, introduced by Atanassov (1986), is one such generalization, which
is characterized by membership as well as non-membership grades respec-
tively. It is known that IFSs describe the fuzzy character more compre-
hensively in some special situations and have a variety of applications in
decision theory. The complexity of the human body system may not be
discussed through traditional approaches, and the behavior of such systems

DOI: 10.1201/9781003398066-1 1
2 Computational Techniques in Neuroscience

has a certain degree of fuzziness. Zadeh (1969) anticipated that fuzzy

theory handles the problems of medical diagnosis very well, as it is a
computational tool for dealing with imprecision and uncertainty of human
reasoning. The primary characteristic of fuzzy theory is its interpretability
that accepts the knowledge in verbal/linguistic ways and allows the system
to describe by simple human-friendly rules, which is the key factor in
medical discipline. In almost any diagnostic process, medical knowledge is
usually imprecise and uncertain and is a relationship between the symptom
and the disease. Doctors gather information about the patient from past
history, clinical examination, pathological test results, and other investi-
gative procedures such as radiological tests. The knowledge gathered from
these sources has a certain degree of incompleteness, and certain things
are overlooked or wrongly interpreted during investigation, resulting in
wrong treatment of the actual disease. Many fuzzy-based models intervene,
varying as per important symptoms, symptom patterns of different diseases,
the relationship between diseases and the hypothesis of disease stages,
preliminary diagnosis, and final diagnosis. These models form the initial
basis for the diseases, which might be iterated and configured as per the
requirement. Sanchez (1979) proposed the fuzzy-based model, which rep-
resents the knowledge base by establishing a fuzzy max-min relation
between the symptom and the disease. In this chapter, medical decision
making of various life-threatening diseases has been considered that sup-
port doctors and potential patients.

1.2 MULTI-CRITERIA DECISION MAKING

Multi-criteria decision making (MCDM) is the branch of decision making

for designing computational tools to support the evaluation of perform-
ance criteria with an aim to solve real-world problems of varied disci-
plines. For the past several decades, MCDM techniques have grown
multifold and have provided hand held support to decision makers to
classify factors into manageable groups and rank them according to their
preferred order. Traditional decision-making problems have certain
courses of action with an aim to select the best course of action and are
structured into three phases, as given in Figure 1.1.
Atanassov (1986) proposed the concept of IFS, which is the generaliza-
tion of Zadeh’s fuzzy set (1965), characterized by the membership and
non-membership functions. Later, Atanassov (1999) and Bustince et al.
(2007) developed the theory to handle more complex decision-making
problems. De et al. (2000) developed new categories of IFS operators,
such as concentration, dilation, and normalization. Based on vague sets and
IFS, Chen and Tan (1994), Hong and Choi (2000), Szmidt and Kacprzyk
(2002) and Atanassov et al. (2005) developed some approximate techniques
for handling MCDM problems under the fuzzy environment. Xu and
 Dynamic intuitionistic fuzzy weighting averaging operator 3

Identification and Structuring of Problem

Development of Model and its Use

Development of Action Plans

Figure 1.1 Illustrates phases of MCDM problems.

Yager (2006) developed some aggregation operators, which include the

intuitionistic fuzzy weighted average (IFWA) operator, intuitionistic fuzzy
weighted geometric (IFWG) operator, intuitionistic fuzzy ordered weighted
geometric (IFOWG) operator, and intuitionistic fuzzy hybrid geometric
(IFHG) operator for handling MCDM problems under the IFS environ-
ment. Liu and Wang (2007) developed an evaluation function, commonly
known as score functions, to measure the degree to which alternatives
satisfy and do not satisfy the requirement. Xu (2007) develop a method
for MCDM problems by proposed new IFS preference relations along with
their properties. Researchers such as Delgado et al. (1998), Bordogna and
Pasi (1993), Fisher (2003), Herrera et al. (1997, 2000a, 2000b, 2000c,
2001), Herrera and Martinez (2000d), Karsak and Tolga (2001), Law
(1996), Lee (1996), Roubens (1997), Sanchez (1996), and Yager (1995,
2001) used the application of fuzzy sets in the MCDM problems related
to clinical diagnosis of diseases, risk assessment in the development of
software, the environment, and manufacturing for decision making. The
MCDM problem is a trade-off among the set of factors with respect to
the given evaluating conditions. Experts from the same domain involved
in the decision making find the best factor among the available ones based
on some decision tree under specified criteria with the aim to get the per-
formance rating of the factors obtained by each expert. MCDM models
with fuzzy techniques have been proposed by several researchers: Chen
and Hwang (1992), Kacprzyk et al. (1992), Fodor and Roubens (1994),
Herrera et al. (1996), Bordogna et al. (1997). Hwang and Yoon (1981)
pointed out that right assessment of criteria weights is very important in
MCDM problems, as the variation of weight values affects the ranking
of alternatives.
The information-based computational techniques optimize decisions in
each step to bridge the gaps that occur during treatment. These techniques
take care of all types of complexities and navigate for patients and doctors
by offering other alternatives. Any form of illness present in the human
body is deadly, and the trajectory of the disease from diagnosis to treatment
4 Computational Techniques in Neuroscience

passes through certain challenging decisions. Information-based manage-

ment systems could be the key factor in understanding the diagnosis of
diseases, making decisions, and if possible, reconfiguring treatment.
Many decision-making models focus on one criterion and have limited
applications; therefore, they fail in most real-life situations. Diaby and
Goeree (2014) proposed three stages of the MCDM model, similar to the
stages defined in Figure 1.1. Roy (1996) addressed the four types of MCDM
problems as: Choice, Sorting, Ranking, and Description. Technique-Ordered
Preference by Similarity to the Ideal Solution (TOPSIS) gives promising
results and has applications in manufacturing, health care systems, medical
decision making, decision making, etc. MCDM strategies tackle complex
situations and provide robust solutions to the problem at hand. The TOPSIS
method states that the optimal point should be closest to the positive ideal
solution and farthest from the negative ideal solution; it yields profit as well
as risk as much as possible. TOPSIS is a compensatory method, in which
a bad result in one criterion is compensated by a good result in another cri-
terion, with an assumption that each criterion has either a monotonically
increasing or decreasing preference function.

1.3 AGGREGATION

The process of representing a collection of data with respect to one repre-

sentative value is called aggregation. The representative value could be some
average maximum or minimum value. In many complex situations, different
types of fuzzy and generalized fuzzy-based aggregation operators, such as
ordered weighted averaging (OWA) operator, ordered weighted geometric
(OWG) operator, weighted harmonic mean (WHM) operator, ordered
weighted harmonic mean (OWHM) operator, hybrid harmonic mean
(HHM) operator, etc., have been used extensively in different domains of
research. For the last couple of years, many MCDM theories and methods
have been proposed by using various types of aggregation operators under
fuzzy environments with the hypothesis that both the attributes and the
decision makers are at the same level of priority. However, in reality, attri-
butes and decision makers have different priority levels. Yager (2008, 2009)
proposed a new type of aggregation operator known as prioritized weighted
average (PWA) operator. In this chapter, priority operators are used for the
diagnosis of the type of brain tumor present in patients, respectively.

1.4 DECISION MAKING

Decision making is a framework by which selection of the best alternative

from a finite number of available alternatives has taken place. Moreover, it is
an art and a common activity used in every sphere of human functionality and
 Dynamic intuitionistic fuzzy weighting averaging operator 5

has applications across domains. Decision making receives great interest from
researchers across various disciplines, but in real life, the information avail-
able is ambiguous or imprecise. To solve such problems of decision making
with vague or imprecise information, fuzzy set and IFS theory have emerged
as powerful tools. In order to define the entropy function, fuzzy set and IFS
theoretic approaches are useful in many real-life situations.

1.5 MEDICAL DIAGNOSIS

Globally, thousands of people die every year due to errors in diagnosis

of diseases. Like other domains, the medical domain is characterized by an
exponential evolution of knowledge. There are many computational tools
related to medical diagnosis, which try to reduce the risk of error and
have many advantages. Medical diagnosis begins when a patient consults
with a doctor. The doctor evaluates the whole situation of the patient
and prepares a knowledge base to prescribe a suitable treatment. The whole
process might be iterated and, in each iteration, the diagnosis might be re-
configured, refined, or even rejected. Medical diagnosis is only possible
through intensive collaboration between physicians and mathematicians.

1.6 FUZZY THEORY

Fuzzy sets were introduced by Zadeh (1965) as an extension of the classical

notion of a set and are defined as:
A fuzzy set A defined in a discrete universe of discourse X = {x1, x2, ….., xn}
is given as: A = {< x, A (x) > : x X}
Where, A : X [0, 1] is the membership function of set A and A (x) is
called the grade of membership of x X in A.
Fuzzy sets have many generalizations like fuzzy soft sets, IFSs, in-
tuitionistic fuzzy soft sets, etc. In this work, IFSs, have been used for the
purpose of decision making.

1.7 INTUITIONISTIC FUZZY SETS (IFS)

IFSs were introduced by Atanassov (1986). They are quite useful and
applicable, and are defined as:
An IFS A in X = {x1, x2, ….., xn} is given as: A = {< x, A (x), A (x) > | x X}
described by membership function A (x): X [0, 1] and non-membership
function A (x): X [0, 1] of the element x X where the function
A (x) = 1 A (x) A (x) is defined as an intuitionistic index or hesitation
index of x in A. In the limiting case, if A (x) = 0, IFS reduces automatically
to a fuzzy set.
6 Computational Techniques in Neuroscience

1.8 INTUITIONISTIC FUZZY VARIABLE

For a time variable t , (t) = ( (t) , (t) ) is called the intuitionistic fuzzy
variable, as proposed by Xu and Yager (2008).
If t = t1, t2, …, tp, then (t1), (t2), …, (tn) indicate q intuitionistic fuzzy
numbers (IFNs) collected at q different periods, where, (t) , (t) [0, 1]
Some operations of IFNs are as:
Let (y1) = ( 1 (y1), 1 (y1), 1 (y1) ) and (y2) = ( 2 (y2), 2 (y2), 2 (y2) ) be two
IFNs, then

• (y1) (y2) = ( (y1) + (y2) (y1) (y2),(1 (y1) )(1 (y2) )

(y1) (y2) )
• (y1) = (1 (1 (y1) ) , (y1) , (1 (y1) ) (y1) ), > 0.

1.9 INTUITIONISTIC FUZZY NUMBER (IFN)

AND ITS OPERATIONS

A fuzzy number is a fuzzy set A R satisfying the following conditions

i. A is a convex fuzzy set, i.e., x1 + (1 ) x1 A where x1, x2 and

[0, 1]
ii. A is a normalized fuzzy set, i.e., max A (x) = 1.
iii. Membership function A (x) is piecewise continuous over A R.
iv. It is defined in a real number.

For x X , let A = ( A , A) be the IFN with 0 A (x) + vA (x) < 1, where

A , vA [0, 1]
The operations of IFNs A1 = ( A1 , A1), A2 = ( A2 , A2) and A3 = ( A3 , A3)
are defined by Yazdani et al. (2016, 2017)

• A1 + A2 = ( A1 + A2 A1 A2 , A1 A2)
• A1 × A2 = ( A1 A2 , A1 + A2 A1 A2)
• A = (1 (1 A ) , A ); > 0
• A = ( A , 1 (1 A) ); >0

1.10 DYNAMIC INTUITIONISTIC FUZZY WEIGHTED

AVERAGING (DIFWA) OPERATOR

Let (y1), (y2), …., (yq) be a collection of IFNs collected at q different

periods and T be the weight vector of the periods yk (k = 1, 2, ….., q), then
 Dynamic intuitionistic fuzzy weighting averaging operator 7

DIFWA (y) ( (y1), (y2), ……, (yq)) = (y1) (y2) ……. (yq) (yq) is called
dynamic intuitionistic fuzzy weighted averaging (DIFWA) operator.

DI (y) ( (y1), (y2), …., (yq))

q q q q
= 1
i =1
(1 (y ) )
i
(y ) ,
i
i =1
(yi)
(yi) ,
i =1
(1 (y ) )
i
(y )
i
i =1
(yi)
(yi)

q
where, (yi) 0; i =1 (yi) = 1

1.11 MEDICAL DIAGNOSIS OF THE TYPE OF BRAIN TUMOR

As the mechanization of human life activities increases at exponential pace,

the risk for the rise of cancer increases. If the treatment of the disease is not
started in time, this may be lethal for patients, and the diagnosis of cancer is
really a tedious job for doctors. In the era of technology, many evolutionary
innovative technologies tackle the problem of diagnosis of diseases and
thus increase the efficiency of doctors. Medical diagnosis with the help of
computational techniques is a very useful tool for doctors for the purpose
of better diagnosis. The idea of MCDM has been extensively used in real-
life situations. When it is taken with IFSs, it gives more strength to the
concept. Xu and Yager (2006) established certain aggregation operators:
IFWG, IFOWG, and IFHG to tackle the problems of multi-attribute deci-
sion making. Yager (2004a, 2008, 2009) modeled the concept of prioritized
operators to streamline the concept of decision making. Wei (2012) pre-
sented the generalized version of prioritized aggregation operators as
explained by Yager (2004b, 2009) in a hesitant fuzzy environment and
developed some operators of hesitant fuzzy prioritized aggregation. Zhang
(2014) presented some prioritized operators under an intuitionistic fuzzy
environment, but these operators have certain drawbacks under certain
levels. To overcome these limitations, Xu and Yagar (2008) studied the
multi-attribute decision-making problems and expressed IFNs at different
periods. For the sake of simplicity, such kinds of difficulties as DIF-MADM
is governed by the DIFWA operator. In this chapter, an algorithmic
approach of the DIFWA operator as proposed by Xu and Yagar (2008) has
been discussed and used with ITrFN for the diagnosis of the type of brain
tumor over certain attributes. These attributes are given in the form of
IFNs. Once the disease has been diagnosed, it is difficult for a doctor to
diagnose the type of disease from the available set of diseases. The present
work helps the doctors to diagnose the disease and execute the right
treatment for the brain tumor and rank it on the basis of available infor-
mation. For this purpose, we have developed a hypothetical case study to
explain the algorithm.
8 Computational Techniques in Neuroscience

Figure 1.2 Illustrates different types of brain tumors present in human body.
https://www.brainhealthdoctor.com/wp-content/uploads/2018/07/shutterstock_
230256940-sm-768×x706.jpg accessed on 20-01-2023.

There are many types of primary brain tumors, named according to the
type of the cells or part of the brain in which they grow (Figure 1.2). The
most common types are: acoustic neuroma, astrocytoma, brain metastases,
choroid plexus carcinoma, craniopharyngioma, embryonal tumors, epen-
dymoma, glioblastoma, glioma, medulloblastoma, meningioma, oligoden-
droglioma, pediatric brain tumors, pineoblastoma, haemangioblastoma,
lymphoma, pineal region tumors, spinal cord tumors, pituitary tumors,
germ cell tumors, and many more. Out of these, 78% of malignant tumors
belong to gliomas, as they arise from supporting cells of the brain called the
glia. Glioma is comprised of glial cells, which protect and support neurons
and are of malignant nature, usually located in the brain and spinal cord.
Ependymoma is a subtype of glioma generally found in children, which
comprises ependymal cells that line the brain’s ventricles. According to
the National Cancer Institute, approximately 30% of this type of tumor is
found in children of age 0–14 years by National Cancer Institute (2018).
Based on factors like: most malignant, aggressive, widely infiltrative, rapid
recurrence, and necrosis prone, WHO developed a grading system to judge the
malignancy of the tumor as: low grade (Grade I and Grade II) and high grade
(Grade III and Grade IV) tumors. Grade I and Grade II tumors generally grow
 Dynamic intuitionistic fuzzy weighting averaging operator 9

Acoustic Neuroma

Craniopharyngioma

Haemangioblastoma

Pineal region tumors

Ependymoma

Spinal Cord Tumors

Types of Brain Tumor
Glioma

Medulloblastoma

Meningioma

Oligodendroglioma

Lymphoma

Pituitary Tumors

Figure 1.3 Illustrates types of brain tumors.

slower than Grades III and Grade IV tumors and are referred to as low-grade
and high-grade tumors. Over a time, a low-grade tumor becomes a high-
grade tumor. The grade of a tumor refers to the structure of the cells under a
microscope. Brain tumors of any type are treated with surgery, radiation, and/
or chemotherapy, either alone or in various combinations. In this chapter,
12 types of brain tumors have been considered and given in Figure 1.3.

1.12 PROPOSED MEDICAL MAKING ALGORITHM:

DYNAMIC INTUITIONISTIC FUZZY WEIGHTED
AVERAGING (DIFWA) OPERATOR

In this chapter, the DIFWA operator has been used to evaluate the type of
brain tumor.
Assumptions:

I. Let F = {f1, ……, fn} be the set of n alternatives.

II. Let S = {S1, …, Sm} be the finite set of attributes, which are collected from
different experts and are articulated in IFNs, with w = (w1, w2, …, wm)T as
weight vectors, where wj 0; m j =1 wj = 1
10 Computational Techniques in Neuroscience

III. Let yk (k = 1, 2, .…, q) be the q periods whose weight vector is

q
(y) = ( (y1), …, (yq))T as weight vectors; where (yk) 0; k =1
(yk) = 1
IV. Let R (yk) = (tij (yk))n × m be an IF decision matrix of the period yk,
where tij (yk) = ( rij (yk), rij (yk), rij (yk) ) is an attribute value defined
by IFN.

Xu and Yagar (2008) proposed the DIFWA operator for solving MCDM
problems and is defined as:

12.1 Using the operator(10)

The aggregation of all the IF decision matrix is:

T (yk) = (tij (yk))m × n

(yk)
where, tij = ( ij , ij , ij), ij =1 k= q
k =1 ( 1 tij (yk) ) (yk);
ij = k=q
k =1 tij (yk)
;
(yk)
ij =1 k=q
k =1 (1 tij (yk) ) (yk) k= q
k =1 tij (yk)

12.2. Let i+ = ( 1+, … m + T

) and T
i = ( 1 , … m) be the IF positive ideal
solution (IFPIS) and IF negative ideal solution (IFNIS), respectively, where
+
i = (1, 0, 0) and i = (0, 1, 0), (i = 1, 2 …. m) be the m largest and m
smallest IFNs, respectively. Furthermore, let the alternatives fi such
as fi = (ti1, …, tim)T , (i = 1, 2…. . n).

12.3. Determine the distance between the alternative fi and the IFPIS( i+)
and distance between the alternatives fi and the IFNIS ( i ), respectively

m m
+ +
(fi, i) = wj (tij , j) = wj (1 ij ) (1.1)
j =1 j =1

m m
(fi, i )= wj (tij , j )= wj (1 ij) (1.2)
j =1 j =1

12.4. The closeness coefficient of each alternative has been calculated as:

(fi , i )
C (fi) = +
, i = 1, ……, n (1.3)
(fi , i ) + (fi , i )

Since,

m
+
(fi, i) + (fi, i )= wj (1 + ij) (1.4)
j =1
 Dynamic intuitionistic fuzzy weighting averaging operator 11

then

m
j =1 wj (1 ij )
C (fi) = m
, i = 1,…., n (1.5)
j =1 wj (1 + ij)

12.5. To rank alternative fi , based on the closeness coefficients C (fi),

the greater the value C (fi) gives the better alternative.

12.6. End.

1.13 EVALUATION OF CASE STUDY

Let fi (i = 1, 2, …., 11) be the available set of various brain tumors

described in Figure 1.3. The diagnosis for different brain tumors has been
demonstrated on the basis of the prescribed symptoms C = (c1, …, c5)
such as different patterns of headaches, unexplained nausea, vision prob-
lems, speech difficulties, hearing problems, etc. Also, the decision for the
final diagnosis will be made from the panel of decision makers
as D = (d1, d2, d3). Let the weight vector for each decision maker be
1 2 3
(t) = ( 6 , 6 , 6 )T and the weight vector of the symptoms be
w = (0.1, 0.15, 0.2, 0.25, 0.3, 0.4)T . The evaluation of the diagnosis for
various brain tumor among fi (i = 1, 2, …, 11) is done on the basis of
proposed algorithm.
The opinions collected from various experts of the same domain are
articulated in IFNs and are listed in the following Tables 1.1–1.3.

Table 1.1 Results of IFN information provided by the expert d1

Treatment/ c1 c2 c3 c4 c5
Symptom
f1 (0,8,0.1,0.1) (0.9,0.1,0.0) (0.7,0.2,0.1) (0.7,0.2,0.1) (0.2,0.4,0.4)
f2 (0.7,0.3,0.0) (0.6,0.2,0.2) (0.6,0.3,0.1) (0.5,0.2,0.3) (0.2,0.7,0.1)
f3 (0.5,0.4,0.1) (0.7,0.3,0.0) (0.6,0.1,0.3) (0.4,0.6,0.2) (0.1,0.8,0.1)
f4 (0.9,0.1,0.0) (0.7,0.1,0.2) (0.8,0.2,0.0) (0.7,0.1,0.2) (0.5,0.1,0.4)
f5 (0.6,0.1,0.3) (0.8,0.2,0.0) (0.5,0.1,0.4) (0.2,0.4,0.4) (0.4,0.5,0.1)
f6 (0.3,0.6,0.1) (0.5,0.4,0.1) (0.4,0.5,0.1) (0.2,0.7,0.1) (0.5,0.5,0.0)
f7 (0.5,0.2,0.3) (0.4,0.6,0.0) (0.5,0.5,0.0) (0.1,0.8,0.1) (0.8,0.2,0.0)
f8 (0,8,0.1,0.1) (0.9,0.1,0.0) (0.7,0.2,0.1) (0.7,0.2,0.1) (0.5,0.4,0.1)
f9 (0.7,0.3,0.0) (0.6,0.2,0.2) (0.6,0.3,0.1) (0.5,0.2,0.3) (0.4,0.6,0.0)
f10 (0.5,0.4,0.1) (0.7,0.3,0.0) (0.6,0.1,0.3) (0.4,0.6,0.0) (0.6,0.1,0.3)
f11 (0.9,0.1,0.0) (0.7,0.1,0.2) (0.8,0.2,0.0) (0.7,0.1,0.2) (0.3,0.6,0.1)
12 Computational Techniques in Neuroscience

Table 1.2 Results of IFN information provided by the expert d2

Treatment/ c1 c2 c3 c4 c5
Symptom
f1 (0.9,0.1,0.0) (0.8,0.2,0.0) (0.8,0.1,0.1) (0.6,0.3,0.1) (0.4,0.3,0.3)
f2 (0.8,0.2,0.0) (0.5,0.1,0.4) (0.7,0.2,0.1) (0.4,0.3,0.3) (0.7,0.1,0.2)
f3 (0.5,0.5,0.0) (0.7,0.2,0.1) (0.8,0.2,0.0) (0.7,0.1,0.2) (0.3,0.5,0.2)
f4 (0.9,0.1,0.0) (0.9,0.1,0.0) (0.7,0.3,0.0) (0.3,0.5,0.2) (0.7,0.2,0.1)
f5 (0.5,0.2,0.3) (0.6,0.3,0.1) (0.6,0.2,0.2) (0.6,0.1,0.3) (0.8,0.2,0.0)
f6 (0.4,0.6,0.0) (0.3,0.4,0.3) (0.5,0.5,0.0) (0.2,0.3,0.5) (0.7,0.3,0.0)
f7 (0.3,0.5,0.2) (0.5,0.3,0.2) (0.6,0.4,0.0) (0.1,0.5,0.4) (0.5,0.1,0.4)
f8 (0.9,0.1,0.0) (0.8,0.2,0.0) (0.8,0.1,0.1) (0.6,0.3,0.1) (0.7,0.2,0.1)
f9 (0.8,0.2,0.0) (0.5,0.1,0.4) (0.7,0.2,0.1) (0.4,0.3,0.3) (0.9,0.1,0.0)
f10 (0.5,0.5,0.0) (0.7,0.2,0.1) (0.8,0.2,0.0) (0.7,0.1,0.2) (0.5,0.5,0.0)
f11 (0.9,0.1,0.0) (0.9,0.1,0.0) (0.7,0.3,0.0) (0.3,0.5,0.2) (0.9,0.1,0.0)

Table 1.3 Results of IFN information provided by the expert d3

Treatment/ c1 c2 c3 c4 c5
Symptom
f1 (0.7,0.1,0.2) (0.9,0.1,0.0) (0.9,0.1,0.0) (0.6,0.1,0.3) (0.4,0.5,0.1)
f2 (0.9,0.1,0.0) (0.6,0.2,0.2) (0.5,0.2,0.3) (0.5,0.2,0.3) (0.7,0.1,0.2)
f3 (0.4,0.5,0.1) (0.8,0.1,0.1) (0.7,0.1,0.2) (0.3,0.3,0.4) (0.8,0.2,0.0)
f4 (0.8,0.1,0.1) (0.7,0.2,0.1) (0.9,0.1,0.0) (0.4,0.4,0.2) (0.5,0.4,0.1)
f5 (0.6,0.3,0.1) (0.8,0.2,0.0) (0.7,0.2,0.1) (0.5,0.5,0.0) (0.9,0.1,0.0)
f6 (0.2,0.7,0.1) (0.5,0.1,0.4) (0.3,0.1,0.6) (0.1,0.4,0.5) (0.6,0.1,0.3)
f7 (0.4,0.6,0.0) (0.7,0.3,0.0) (0.5,0.5,0.0) (0.2,0.3,0.5) (0.3,0.6,0.1)
f8 (0.5,0.4,0.1) (0.7,0.3,0.0) (0.6,0.1,0.3) (0.4,0.6,0.2) (0.1,0.8,0.1)
f9 (0.9,0.1,0.0) (0.7,0.1,0.2) (0.8,0.2,0.0) (0.7,0.1,0.2) (0.5,0.1,0.4)
f10 (0.6,0.1,0.3) (0.8,0.2,0.0) (0.5,0.1,0.4) (0.2,0.4,0.4) (0.4,0.5,0.1)
f11 (0.3,0.6,0.1) (0.5,0.4,0.1) (0.4,0.5,0.1) (0.2,0.7,0.1) (0.5,0.5,0.0)

The collective results of the experts are listed in Table 1.4.

Using the proposed algorithm, the values of Closeness Coefficients C (fi)
for each type of tumor are given in Table 1.5.
The value of the Closeness Coefficient C (fi) represents the type of brain
tumor.
Table 1.4 Results of collective IFN information provided by the experts D = (d1, d2, d3)
Treatment/Symptom c1 c2 c3 c4 c5
f1 (0.80,0.1,0.094) (0.87,0.12,0.0) (0.84,0.11,0.039) (0.61,0.16,0.219) (0.37,0.40,0.223)
f2 (0.84,0.15,0.0) (0.56,0.15,0.272) (0.59,0.21,0.192) (0.46,0.22,0.30) (0.64,0.13,0.21)
f3 (0.45,0.48,0.06) (0.75,0.15,0.09) (0.72,0.12,0.14) (0.48,0.23,0.28) (0.61,0.34,0.04)
f4 (0.85,0.1,0.04) (0.79,0.14,0.06) (0.83,0.16,0.0) (0.43,0.34,0.22) (0.57,0.25,0.17)
f5 (0.56,0.21,0.21) (0.74,0.22,0.02) (0.64,0.17,0.18) (0.49,0.28,0.22) (0.83,0.16,0.0)
f6 (0.28,0.64,0.06) (0.44,0.2,0.35) (0.39,0.22,0.38) (0.15,0.39,0.45) (0.62,0.18,0.18)
f7 (0.38,0.47,0.14) (0.60,0.33,0.06) (0.53,0.46,0.0) (0.15,0.41,0.43) (0.49,0.27,0.23)
f8 (0.74,0.20,0.05) (0.78,0.21,0.0) (0.69,0.11,0.19) (0.53,0.39,0.07) (0.43,0.44,0.11)
f9 (0.84,0.15,0.0) (0.62,0.11,0.26) (0.74,0.21,0.04) (0.58,0.16,0.25) (0.69,0.13,0.16)
f10 (0.55,0.21,0.23) (0.75,0.21,0.03) (0.64,0.12,0.22) (0.45,0.27,0.28) (0.47,0.38,0.14)
f11 (0.73,0.24,0.02) (0.73,0.20,0.06) (0.60,0.36,0.03) (0.35,0.45,0.19) (0.69,0.30,0.0)
Dynamic intuitionistic fuzzy weighting averaging operator
13
14 Computational Techniques in Neuroscience

Table 1.5 Results of closeness coefficients of various types of brain tumors

Type of Tumor Closeness Coefficients C (f i)
Glioma (f1) 0.690
Ependymoma (f 2 ) 0.672
Oligodendrogliom (f 3) 0.654
Meningioma (f 4 ) 0.692
Pituitary Tumors (f 5) 0.705
Acoustic Neuroma (f 6 ) 0.539
Craniopharyngioma (f 7 ) 0.518
Haemangioblastoma (f 8 ) 0.630
Lymphoma (f 9 ) 0.727
Pineal Region Tumor (f10 ) 0.622
Spinal Cord Tumors (f11) 0.625

1.14 RESULT

This section provides the results with discussion of closeness coefficient for
various types of brain tumors, as shown in Figure 1.4.
Thus, the ranking of various brain tumors based on available information
is given below:

f9 f5 f4 f1 f2 f3 f8 f11 f10 f6 f7

Type of Brain Tumor

0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Closeness coefficient

Figure 1.4 Illustrates closeness coefficient for different brain tumors.

 Dynamic intuitionistic fuzzy weighting averaging operator 15

1.15 RESULT DISCUSSION

Thus, a better alternative for the above study is f9 i.e., lymphoma.

This study initially computes the overall collective information of deci-
sion makers D = (d1, d2, d3) as given in section 1.5. We find intuitionistic
fuzzy ideal solutions, positive as well as negative to calculate distance
between IFPIS and IFNIS. To get the optimal decision, closeness coefficient
is calculated as given in Table 1.5. The highest value of the closeness
coefficient represents the decision value. The result obtained by using
algorithm (1.4) is given in Figure 1.4, which gives a clue to the decision
makers to make the right decision.

1.16 CONCLUSION

The MCDM technique has been proposed in this chapter for the diagnosis
of a type of brain tumor in the context of IFNs. DIFWA operator has been
used to aggregate IFNs for each of the alternatives. The optimal decision is
evaluated by ranking the alternatives based on their closeness coefficient
value. With the help of the given operator, one can easily rank the type of
brain tumor present in the body of the patient as per their attributes under
the intuitionistic fuzzy environment. This model can be utilized for other
methods to get a better decision without doing any additional medical tests,
as the proposed algorithm is sufficient enough to perform an initial inves-
tigation of the disease. The proposed model may help doctors make better
decisions under uncertainty.

REFERENCES

Atanassov, K. “Intuitionistic fuzzy sets”. fuzzy sets and systems 20(1), 87–96.
DOI: 10.1016/S0165-0114 (86)(1986): 80034-3.
Atanassov, Krassimir T., and Krassimir T. Atanassov. Intuitionistic fuzzy sets.
Physica-Verlag HD, 1999.
Atanassov, Krassimir, Gabriella Pasi, and Ronald Yager. “Intuitionistic fuzzy
interpretations of multi-criteria multi-person and multi-measurement tool
decision making.” International Journal of Systems Science 36, no. 14 (2005):
859–868.
Bordogna, Gloria, and Gabriella Pasi. “A fuzzy linguistic approach generalizing
boolean information retrieval: A model and its evaluation.” Journal of the
American Society for Information Science 44, no. 2 (1993): 70–82.
Bordogna, Gloria, Mario Fedrizzi, and Gabriella Pasi. “A linguistic modeling of
consensus in group decision making based on OWA operators.” IEEE
Transactions on Systems, Man, and Cybernetics-Part A: Systems and Humans
27, no. 1 (1997): 126–133.
16 Computational Techniques in Neuroscience

Bustince, Humberto, Francisco Herrera, and Javier Montero, eds. Fuzzy sets and
their extensions: Representation, aggregation and models: Intelligent systems
from decision making to data mining, web intelligence and computer vision.
Vol. 220. Springer, 2007.
Chen, Shu-Jen, Ching-Lai Hwang, Shu-Jen Chen, and Ching-Lai Hwang. Fuzzy
multiple attribute decision making methods. Springer Berlin Heidelberg, 1992.
Chen, Shyi-Ming, and Jiann-Mean Tan. “Handling multicriteria fuzzy decision-
making problems based on vague set theory.” Fuzzy sets and systems 67, no. 2
(1994): 163–172.
De, Supriya Kumar, Ranjit Biswas, and Akhil Ranjan Roy. “Some operations on
intuitionistic fuzzy sets.” Fuzzy sets and Systems 114, no. 3 (2000): 477–484.
Delgado, Miguel, Francisco Herrera, Enrique Herrera-Viedma, and Luis Martinez.
“Combining numerical and linguistic information in group decision making.”
Information Sciences 107, no. 1-4 (1998): 177–194.
Diaby, Vakaramoko, and Ron Goeree. “How to use multi-criteria decision analysis
methods for reimbursement decision-making in healthcare: a step-by-step
guide.” Expert review of pharmacoeconomics & outcomes research 14, no. 1
(2014): 81–99.
Fisher, Bernard. “Fuzzy environmental decision-making: applications to air pollu-
tion.” Atmospheric Environment 37, no. 14 (2003): 1865–1877.
Fodor, Janos C., and M. R. Roubens. Fuzzy preference modelling and multicriteria
decision support. Vol. 14. Springer Science & Business Media, 1994.
Herrera, F., E. Herrera-Viedma, and J. L. Verdegay. “A linguistic decision process in
group decision making.” Group Decision and Negotiation 5 (1996): 165–176.
Herrera, F., and E. Herrera-Viedma. “On the linguistic OWA operator and exten-
sions.” The Ordered Weighted Averaging Operators: Theory and Applications
(1997): 60–72.
Herrera, Francisco, and Enrique Herrera-Viedma. “Linguistic decision analysis:
steps for solving decision problems under linguistic information.” Fuzzy Sets
and Systems 115, no. 1 (2000a): 67–82.
Herrera, Francisco, and Enrique Herrera-Viedma. “Choice functions and mecha-
nisms for linguistic preference relations.” European Journal of Operational
Research 120, no. 1 (2000b): 144–161.
Herrera, Francisco, Enrique Herrera-Viedma, and Luis Martı́nez. “A fusion
approach for managing multi-granularity linguistic term sets in decision
making.” Fuzzy Sets and Systems 114, no. 1 (2000c): 43–58.
Herrera, Francisco, and Luis Martínez. “A 2-tuple fuzzy linguistic representation
model for computing with words.” IEEE Transactions on Fuzzy Systems 8,
no. 6 (2000d): 746–752.
Herrera, Francisco, Enrique Herrera-Viedma, and Francisco Chiclana. “Multiperson
decision-making based on multiplicative preference relations.” European
Journal of Operational Research 129, no. 2 (2001): 372–385.
Hong, Dug Hun, and Chang-Hwan Choi. “Multicriteria fuzzy decision-making
problems based on vague set theory.” Fuzzy Sets and Systems 114, no. 1
(2000): 103–113.
Kacprzyk, Janusz, Mario Fedrizzi, and Hannu Nurmi. “Group decision making
and consensus under fuzzy preferences and fuzzy majority.” Fuzzy Sets and
Systems 49, no. 1 (1992): 21–31.
 Dynamic intuitionistic fuzzy weighting averaging operator 17

Karsak, E. Ertugrul, and Ethem Tolga. “Fuzzy multi-criteria decision-making

procedure for evaluating advanced manufacturing system investments.”
International Journal of Production Economics 69, no. 1 (2001): 49–64.
Law, Chiu-Keung. “Using fuzzy numbers in educational grading system.” Fuzzy
Sets and Systems 83, no. 3 (1996): 311–323.
Lee, Huey-Ming. “Applying fuzzy set theory to evaluate the rate of aggregative
risk in software development.” Fuzzy sets and Systems 79, no. 3 (1996):
323–336.
Liu, Hua-Wen, and Guo-Jun Wang. “Multi-criteria decision-making methods based
on intuitionistic fuzzy sets.” European Journal of Operational Research 179,
no. 1 (2007): 220–233.
National Cancer Institute. (2018) available at https://www.cancer.gov/types/
childhood-cancers/child-adolescant-cancers-fact-sheet
Roubens, Marc. “Fuzzy sets and decision analysis.” Fuzzy Sets and Systems 90,
no. 2 (1997): 199–206.
Roy, Bernard. Multicriteria methodology for decision aiding. Vol. 12. Springer
Science & Business Media, 1996.
Sanchez, Elie. “Truth-qualification and fuzzy relations in natural languages, appli-
cation to medical diagnosis.” Fuzzy Sets and Systems 84, no. 2 (1996):
155–167.
Sanchez, Elie. “Medical Diagnosis and Composite Fuzzy Relations. Gupta, MM;
Ragade, RK; Yager RR (Eds.): Advances in Fuzzy Set Theory and Applications.”
(1979): 437–444.
Wei, Guiwu. “Hesitant fuzzy prioritized operators and their application to
multiple attribute decision making.” Knowledge-Based Systems 31 (2012):
176–182.
Xu, Zeshui, and Ronald R. Yager. “Some geometric aggregation operators based
on intuitionistic fuzzy sets.” International Journal of General Systems 35,
no. 4 (2006): 417–433.
Xu, Zeshui. “Intuitionistic preference relations and their application in group
decision making.” Information Sciences 177, no. 11 (2007): 2363–2379.
Xu, Zeshui, and Ronald R. Yager. “Dynamic intuitionistic fuzzy multi-attribute
decision making.” International Journal of Approximate Reasoning 48, no. 1
(2008): 246–262.
Yager, Ronald R. “An approach to ordinal decision making.” International Journal
of Approximate Reasoning 12, no. 3-4 (1995): 237–261.
Yager, Ronald R. “Fusion of multi-agent preference orderings.” Fuzzy Sets and
Systems 117, no. 1 (2001): 1–12.
Yager, Ronald R. “Modeling prioritized multicriteria decision making.” IEEE
Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics) 34,
no. 6 (2004a): 2396–2404.
Yager, Ronald R. “OWA aggregation over a continuous interval argument
with applications to decision making.” IEEE Transactions on Systems,
Man, and Cybernetics, Part B (Cybernetics) 34, no. 5 (2004b): 1952–1963.
Yager, Ronald. “Prioritized aggregation operators.” International Journal of
Behavioral Science 16, no. 6 (2008): 538–544.
Yager, Ronald. “Prioritized OWA aggregation.” Fuzzy Optimization & Decision
Making 8, no. 3 (2009): 245–262.
18 Computational Techniques in Neuroscience

Yazdani, Morteza, Sarfaraz Hashemkhani Zolfani, and Edmundas Kazimieras

Zavadskas. “New integration of MCDM methods and QFD in the selection
of green suppliers.” Journal of Business Economics and Management 17,
no. 6 (2016): 1097–1113.
Yazdani, Morteza, Prasenjit Chatterjee, Edmundas Kazimieras Zavadskas,
and Sarfaraz Hashemkhani Zolfani. “Integrated QFD-MCDM framework
for green supplier selection.” Journal of Cleaner Production 142 (2017):
3728–3740.
Zadeh, Lotfi A. “Fuzzy sets.” Information and Control 8, no. 3 (1965): 338–353.
Zadeh, Lotfi A. “Biological application of the theory of fuzzy sets and systems.” In
The Proceedings of an International Symposium on Biocybernetics of the
Central Nervous System, pp. 199–206. London: Little, Brown and Comp.,
1969.
Zhang, Guangquan, and Jie Lu. “Using general fuzzy number to handle uncertainty
and imprecision in group decision-making.” Intelligent Sensory Evaluation:
Methodologies and Applications (2004): 51–70.
Chapter 2

Neural modeling and neural

computation in a medical approach
Simran Singh, Anshika Gupta, and Kalpana Katiyar
Department of Biotechnology, D. Ambedkar Institute of
Technology for Handicapped, Kanpur, Uttar Pradesh, India

2.1 INTRODUCTION

2.1.1 Introduction
As we all know, biological/spiking neuron models show certain cells in
the nervous system that generate sharp electrical potentials across cell
membranes. But in today’s epoch, neuroscientists have developed various
algorithmic and computational approaches to analyze their findings and
make forecasts and hypotheses based on the study’s results. However, it is
critical to differentiate between generic modeling and neuronal modeling
related to computational neuroscience. Neural modeling is a mathemat-
ical or computer methodology that utilizes a neural network, an artificial
intelligence (AI) technology that trains computers to interpret data in a
manner similar to that of the human brain. Deep learning is a machine
learning approach that engages linked nodes or neurons in a hetero-
structure similar to the human brain. Precise neural models make certain
assumptions according to the available explicit data, and the conse-
quences of these suppositions are quantified. Recent advances in neural
computation reflect multidisciplinary research in theory, statistics in
neuroscience, modeling computation, design, and construction of neurally
inspired information processing systems. Hence, this sector attracts psy-
chologists, neuroscientists, physicists, computer scientists, and AI inves-
tigators functioning on neural systems underlying perception, cognition,
emotion, and behavior and artificial neural systems that have similar
capabilities. Thus, advanced experimental technologies being developed
by brain initiatives will fabricate large, complex data sets and meticulous
statistical analysis and theoretical insight for a better understanding of
these data mean sets.

2.1.2 Why are neuron models better?

A neural or connectionist model is a network of functional processing units
whose local interactions contribute to global model behavior over time.

DOI: 10.1201/9781003398066-2 19
20 Computational Techniques in Neuroscience

It is easy to conceive a neural model as having three components: a net-

work, an activation rule, and a learning rule.
Opinions on what defines a decent neuron model differ greatly (Herz
et al., 2006). For a long time, two distinct perspectives on this have coex-
isted: comprehensive biophysical models (proposed in 1952 by neurobiol-
ogists Alan Hodgkin and Andrew Huxley) that classify ion channels on the
neuronal cell’s tree-like spatial configuration, and the “integrate-and-fire”
model is a simple mathematical framework used to describe the behavior of
neurons in the brain. This model is based on the observation that neurons
generate pulsatile electrical activity, also known as action potentials or
spikes, through a threshold process. Biophysical models are used by elec-
trophysiologists since they are acquainted with the idea of ion channels that
open and close (and hence affect cerebral activity) in response to environ-
mental influences.
On the other hand, theorists often favor uncomplicated neuron models
with very few criteria that can be mathematically analyzed. Following
prior attempts at the smaller-scale model comparison, the International
Neuroinformatics Coordinating Facility (INCF) sponsored an interna-
tional competition, which enabled a fair measure of neuron models,
earlier this year.
The INCF competition is based on the premise that a successful model
can predict neuronal activity using data that was not utilized for parameter
tweaking. Three in vitro and one in vivo data set were included in the
competition. The in vitro data sets were compiled using standard electro-
physiological tests in which a random electrical current was delivered into a
pyramidal cell and an interneuron through an electrode. Based on their
pursuit recorded during the first 38 seconds of data collection, the chal-
lenge was to predict the precise moment of a spike in neuronal electrical
activity induced throughout a 22-second time frame for 13 (or 9, respec-
tively) repeats of the identical current waveform injected. A basic integrate-
and-fire model using a sliding threshold, the winning entry confidently
predicted 59.6% (or 81.6%, correspondingly) spike timing of the two
neurons.
The majority of threshold models used to study neuronal activity are
point neuron models, which assume a simplified neuron construction
without any dendritic architecture. However, the INCF competition
included a third challenge that involved double-electrode testing to inves-
tigate the interaction between somatic and dendritic spike activity. This
challenge required the injection of current into both the nerve cell body
(soma) and the apical nerve fiber located 600 to 700 micrometers away
from the soma. Remarkably, the highest performing model in this challenge
was a threshold model that had been augmented with two dendritic
equations. The model was able to predict the timing of 90.5% of the spike
activity of a neuron in the lateral geniculate area of the brain, given its input
triggered by visual stimulation of the retina. This performance was 11%
 Neural modeling and neural computation in a medical approach 21

better than the previous data analysis of the same neuron, highlighting the
potential relevance of the competition’s results in re-evaluating previous
findings. Overall, this challenge demonstrated the importance of consid-
ering dendritic architecture in threshold models and the potential for im-
proving our understanding of neuronal activity by incorporating more
complex models (Sincich et al., 2009).
Threshold models describe brain function phenomenologically, but they
only provide a tenuous relationship to the fundamental biophysical reasons
for electrical activity. Threshold models have restricted ability that cannot
forecast the specific time course of the voltages before and after a spike, nor
can they predict the effect of temperature-dependent, chemical environment
changes, or pharmacological modifications of ion channels, whereas
Hodgkin-Huxley biophysical models can do all of this. Most biophysical
model parameters will soon be assessed in a systematic manner utilizing an
appropriate mix of immunostaining methods to estimate ion-channel dis-
tribution, calibrated measurements of ion-channel kinetics, and expression
investigations to identify tens of ion channels in individual cells. Along
those same lines, automatic model construction is feasible.
Furthermore, complex nonlinear spatiotemporal effects on the interac-
tion of back-propagating action potentials (those that move into a dendrite)
with shunting inhibition, or local spikes in intracellular calcium concen-
tration that are triggered by numerous, geographically distributed synaptic
inputs, are outside the purview of threshold models. Although traditional
experimental methods have made it difficult to quantify these nonlinear
spatiotemporal aspects, modern imaging techniques that quantify the
voltage phase time course from across dendritic tree at a high spatial res-
olution in tandem with a governed multisite sensory input by glutamate
uncaging or optogenetic methods will introduce a fresh era of statistically
predictive biophysical models (Gerstner & Naud, 2009).

2.1.3 Objective
We now turn to briefly consider how models of neural networks are being
employed in medical computing. Current methods can be divided into two
major categories. Neural models are employed as computational tools in
one class of applications to execute certain information processing tasks.
Neural models are employed as modeling tools in the second category of
applications to replicate diverse neurobiological or psychological events.
Aside from its roots in brain modelling and cognitive research, neural
models offer a general computational framework with potential applica-
tions (Reggia, 1993) in many areas of medical informatics. To the degree
that an issue can be described as a neural model, vast computer capacity
may be utilized in parallel processing to solve that problem. Furthermore,
because neural models have the potential to learn, the knowledge acquisi-
tion difficulty experienced when deploying classic AI systems may
22 Computational Techniques in Neuroscience

occasionally be fine-tuned by training an evolving system with realistic test

cases. In medical informatics, neural models have been used to solve a range
of problems. Image processing is one of these tasks, in which a neural
network is utilized to categorize pictures into categories representing ther-
apeutically relevant items. Neural networks, for example, have been taught
to identify hepatic ultrasound pictures, as well as to distinguish tumor
stages in infrared images and avascular necrosis of the femoral head in
magnetic resonance imaging. Another topic that has garnered attention is
signal processing. Neural models, for example, have been used to eliminate
noise from EKG readings, distinguish evoked potentials, and categorize
EMG and electroencephalogram (EEG) waveforms. In the treatment of
hypertension, neural models have also been utilized to assist therapeutic
decisions such as medication selection and dosage. (Poli et al., 1991) has
several further examples of medical uses. In the next section, we will look in
depth at one specific application: the use of neural networks for diagnostic
issue solving. In a diagnostic problem, one is given specific manifestations
(symptoms, signs, and abnormal laboratory test results) and must diagnose
the illnesses producing those findings. This type of diagnostic problem
solving has been the focus of a great amount of effort in medical informatics
over the last 15 years; thus, it provides an excellent field for comparing
neural modelling approaches with more classic AI methods.

2.2 DYNAMIC AND ARCHITECTURE FOR NEURAL

COMPUTATION

Hopfield in 1982 proposed that the collective features of physical systems

may be exploited to directly execute computational tasks, which sparked a
lot of interest in brain computation. This new computation paradigm
promises to provide new section of computing devices in which the physics
of the machine and algorithms of the computation are inextricably linked.
The model-independent behavior of dynamical systems will be explored
first, and it will be demonstrated that how the systems’ phenomenology
may be isolated into two fundamental architectural components that con-
duct the operations of continuous non-linear transformation and auto-
associative recall. These filters are primal in the sense that they are essential
building pieces from which hierarchical designs can be constructed.
Backpropagation approaches for programming filters will be developed for
a basic model, although the concepts are applicable to a wide range of neu-
rodynamical models. The recurrent backpropagation algorithms will be given
in a formalism suitable for practical nonlinear dynamical system implemen-
tation. One advantage of this formalism is that it employs continuous time
and so does not display certain types of oscillations that are linked with
backpropagation in models with discrete time (Katiyar et al., 2022).
 Neural modeling and neural computation in a medical approach 23

Among them, one of the model-independent investigation’s findings

will be used to explain why the backpropagation method cannot store
numerous patterns in a basic associative memory model. The problem will
be solved by constraining the system during learning. The resultant algo-
rithm not only moves fixed points, but also produces new ones. As a result,
autoassociative memory exhibits discontinuous learning behavior.
Two rudimentary filters will be integrated to form an elementary pattern
recognition machine as an example of a simple hierarchical design. If
the modifications are tiny enough, this machine can recognize patterns
that have been corrupted by arbitrary transformations. In other words,
the machine can only recognize a limited number of invariant patterns. The
two filters in the machine can train separately since error signals do not
spread across filter borders. Thus, the two-filter system is a simple illus-
tration of Ballard’s modular learning approach (1987)(Pineda, 1988).

2.2.1 Overview of dynamic model

The field of neural network models that can be trained through recurrent
backpropagation is very extensive, which is why it’s important to focus
on a specific system. In this case, we’re looking at a neural network
algorithm that relies on a set of differential equations.

dai
a = ai + j
wij f (ai ) + Ki (2.1)
dt

These equations involve a state vector called a, an external device vector

called K, and an array of weights called w that determine the strength of
connections between neurons. The equation also includes a relaxation time
scale represented by -a, and a differentiable function called f(ai) that helps
determine the system’s dynamic properties. Physiologically inspired choices
for this function include the logistic and hyperbolic tangent functions.
When the weight matrix w is symmetric with zero diagonals, the resulting
system is known as a Hopfield model with graded neurons, as described
by (Pineda, 1989).
The solutions of equation (2.1) can exhibit oscillations, chaotic behavior,
or converge towards isolated fixed points. Since we want to use the fixed
point value as the system’s output, convergence towards isolated fixed
points is the preferred behavior for our purposes. Therefore, when the
network is initialized, the weights are adjusted so that the network’s output
matches the desired output result.
There are several methods available to ensure the convergence of the
network, and one such method is to impose a specific structure on the
network’s connectivity, such as requiring the weight matrix to be lower
triangular or symmetric. While symmetry is mathematically elegant, it can
24 Computational Techniques in Neuroscience

be quite strict as it restricts microscopic connectivity by necessitating

symmetric coupling between pairs of neurons.
Alternatively, the Jacobian matrix for equation (2.1) can serve as a less
stringent criterion for ensuring convergence. The Jacobian matrix can be
made bilaterally dominant, which means that the absolute value of the
diagonal elements is greater than the sum of the absolute values of the off-
diagonal elements in each row. By ensuring the Jacobian matrix is bilat-
erally dominant, the network can be made stable and converge towards
fixed points, even in the absence of strict symmetry constraints.

Lij = ij wij f (aj ) (2.2)

where ij are the elements of the identity matrix and f (aj ) is the derivative
of f (ai ).
Gradient descent dynamics can often result in a neural network that
violates stability criteria such as feedforward, symmetric, or diagonal
dominance. However, it has been observed that even if the initial network
violates these criteria, it does not become unstable during the learning
process. This suggests that the stability assumptions underlying recurrent
backpropagation are sufficient and that a dynamical system that allows
only stable behavior is not necessary.
In gradient descent learning, the objective is to optimize an objective
function with the weights as independent parameters. The number of
weights, N, is proportional to n2 if the fan-in/fan-out ratio of the processing
units is proportional to R. Relaxing the network and generating a target
formula based on the steady-state a0 takes O(mN) or O(mn2) procedures.
However, computing the gradient of the objective function numerically
requires O(mN2) or O(mn^4) calculations, which becomes impractical for
large problems. Moreover, the number of gradient evaluations required for
solution convergence may diverge for certain problems.
Backpropagation adaptive dynamics, which is based on gradient descent,
uses two methods to reduce computation. The first method represents the
gradient of the objective function as an outer-product for equations of
the type (2.1), that is,

wE = b0f (a0)T (2.3)

where a0 is the solution to equation (2.1) and b0 is the “error vector,”

b0 = (LT ) 1K (2.4)

The transpose of the n matrix stated in equation (2.2) is denoted by LT, and
K is just an external error signal that is dependent on the objective function
and a0. Because L−1 can be determined in O(n3) operations from L and x2 in
 Neural modeling and neural computation in a medical approach 25

Table 2.1 Results summarization of numerical algorithm complexity

Complexity of Numerical Algorithm
Worst case (e.g., numerical differentiation) O(mN2)
Matrix inversion (e.g., gaussian elimination) O(mN3/2)
Matrix inversion by relaxation (e.g., recurrent backpropagation) O(mN)
Recursion (e.g., classical feedforward backpropogation O(N)

only O(mn2) operations, this technique reduces the computational cost of the
gradient estimated by a factor of n. As an outcome, the whole computation has
scales such as O(mn3) or O(mN3/2). The second approach takes use of the fact
that b0 can be determined via relaxation, or that it is the (stable) fixed point of
the linear differential equation.

dbi
b = bi + f (ai ) j
wij bj + Ki (2.5)
dt

Pineda developed a variant of this equation. Almeida separately

developed a discrete time version (1987). It takes O(n2) operations every
time step to relax b (that integrate equation (2.5) until b achieves its
steady state). As a result, if the system does not wander chaotically,
the amount of computing required scales like O(mn2) or O(mn). The
approach is computationally efficient if the network is big and sparse
enough, and the fixed points are not marginally stable. Table 2.1 sum-
marizes these findings. It is worth noting that the two backpropagation
algorithms have decreased computation by a factor of N. Because the
classical feedforward approach is more efficient since it does not require
relaxation to a steady state. (Hong et al., 2007).

2.3 NEURAL MODELING IN FUNCTIONING BRAIN IMAGING

In the last decade, there has been significant advancement in the study of
brain function utilizing functioning neuroimaging. It has been notably correct
in the realm of human behavior studies in regional cerebral images. The
numerous forms of functional neuroimaging approaches are founded on two
distinct modalities: (Abeles et al., 1995) (1) hemodynamic-metabolic – this
domain includes positron emission tomography (PET), single photon emis-
sion tomography (SPECT), functional magnetic resonance imaging (fMRI),
all predominantly used in humans, and autoradiographic de-oxyglucose and
the optical imaging and method, both primarily used in nonhuman animals
(Ackermann et al., 1984); and (2) electric-magnetic – this realm includes EEG
and magnetoencephalography (MEG), both primarily used with human
subjects. These two basic forms of imaging have fundamentally distinct
26 Computational Techniques in Neuroscience

properties, the most notable of which are associated with temporal resolution
and the amount and type of spatial information provided by each.

2.3.1 Hemodynamic-metabolic methods of functional

neuroimaging signal
The study initiated the measurement of brain metabolism and blood flow in
humans (Oxide & Of, 1947), whose approach allowed for the measure-
ment of the mean flow of cerebral blood or metabolism for the entire brain.
Lassen and Ingvar improved the approach by allowing them to calculate
regional cerebral blood flow (rCBF). Several procedures, some less intrusive
than others, that were modifications of the Lassen-Ingvar method, were
developed throughout the years. (Sokoloff et al., 1977) devised a method
for estimating the zonal cerebral metabolic rate for glucose based on the
absorption of radioactive deoxyglucose (rCMRglc). Quantitative auto-
radiography is used in subhuman animals to quantify radioactive tracer
concentrations in the brain. This approach was adapted for use with PET in
humans, where radio-labeled fluorodeoxyglucose was used (Horwitz &
Sporns, 1994).

2.3.1.1 Functional MRI

Over the last five years, fMRI has become the most widely utilized tech-
nology for functional brain imaging (Adey et al., 1961). The most often
studied signal is the change in concentration in blood oxygenation and
blood volume caused by changing brain activity, which is referred to as
BOLD (blood oxygenation level-dependent contrast) (Kwong et al., 1992).
Endogenous paramagnetic contrast is provided by deoxygenated hemo-
globin. Increase in blood flow lowers the local concentration of deox-
ygenated hemoglobin, resulting in an increase in the MRI signal on a T2p-
weighted picture (Ogawa et al., 1993) These signals, which do not require
contrast fluid infusions, can be identified using standard MRI scanners;
however, specific gear (e.g., rapid gradient coils) is required. fMRI has a
spatiotemporal component.

2.3.1.2 Electric–magnetic methods

Both electric or magnetic fields associated with neuronal activity are mea-
sured in the second main kind of functional brain imaging (for reviews, see
Gevins et al., 1999). The first functional neuroimaging technologies used on
humans captured electrical activity from the scalp. EEGs, which are con-
tinual recordings spanning tens to hundreds of seconds, and event-related
potentials (ERPs), which are electrical reactions to specific cognitive stimuli;
ERPs, which generally correspond to roughly a second’s worth of brain
activity, are included in this category. Researchers have also been able to
 Neural modeling and neural computation in a medical approach 27

track the magnetic fields created by electric current flows associated with
cerebral activity paving the way for the application of MEG to study cog-
nitive functions of the brain (Horwitz et al., 2000).

2.3.2 A brief review of neural modeling in functional

brain imaging
2.3.2.1 Neuromodeling and PET/fMRI
Neuromodeling has been utilized in three ways in combination with PET
and fMRI data (for overviews, see (Horwitz et al., 1999). The first is an
investigation into local variations in brain activity, which are translated into
alterations in blood flow and metabolism (Motion, 1990).
Secondly, the approach has been used in modelling to identify the
systems-level networks that mediate certain cognitive activities (McIntosh
& Gonzalez-Lima, 1991) Finally, numerous organizations have begun to
build multiple neurobiologically realistic models for simulating PET and
fMRI experiments, allowing one to compare system-level results to neu-
ronal and neural ensemble-level results (Goodisman & Asmussen, 1997).

2.3.2.2 EEG/MEG and neuromodeling

The first neural modelling experiments utilizing functional brain imaging
data appeared to be concentrated on EEG data. Given that EEG was the
initial neuroimaging modality, this is not surprising (Brasil-Neto et al.,
1992). The neurophysiological underpinning for the EEG/ERP signals was a
major issue. The fundamental mechanics are now rather well known. The
extracellular environment is affected by ionic currents, which flow through
the neuronal membrane that are sinks as well as sources. When several
comparable parts of an anatomically structured ensemble of neurons are
engaged at the same time, these sinks and sources (dipoles) can reach
macroscopic size. Neuronal modelling of such ensembles led to the con-
clusion in the cortex, which is the activity of postsynaptic potentials linked
with pyramidal neurons that is important, rather than the activity of
pyramidal neurons themselves.

2.3.3 Conclusion
The field of functional brain imaging is a significant source of complicated
data in neuroscience study. Temporal as well as spatial domains are com-
plicated. Therefore, these data enable researchers to look into the neuro-
logical underpinnings of human sensory, motor, affective, and cognitive
functions. The secondary argument is that this complexity limits tranquil
comprehension and necessitates a similarly rich computational approach to
data analysis and, more importantly, data interpretation. The tertiary part
28 Computational Techniques in Neuroscience

is that computational modeling of functional imaging data can occur at

different levels (microscopic, mesoscopic, and macroscopic), which also
leads to the fourth point that bridging models will be needed to connect all
of these approaches into unified and accordant accounts of brain function
(Horwitz et al., 2000).

2.4 LITERATURE REVIEW

2.4.1 Type of neural model

There are three types of neural models that are described below:

2.4.1.1 Single cell level models

Dendritic tree modelling at single nerve cells is an essential method that is
extensively utilized for examining all the features of nerve cells (Horwitz et al.,
1998) Such investigations often highlight the interaction of several biophysical
factors that are related to giving out synaptic inputs also accompanied by
dendritic shape for the development of a response in the modelled neuron. For
example, Traub (1982) explored the development of electro-physiologically
visible exploding patterns within CA3 hippocampal pyramidal neurons using
compartmental modelling and discovered essential variables. A new model of
neurons in the piriform cortex demonstrated in what manner neuro-
pharmacological action (acetylcholine release) might alter pattern storage. A
comprehensive model about the cortical neuropil that included dendritic
shape along with fundamental interrelationship in the middle of neurons
marked a variety of issues starting with neural evolution to biochemical trace
generation associated with neural action (Gally et al., 1990). Nitric oxide is
utilized as a free diffusing extracellular messenger in the model. This type of
“tissue modelling” lies down between the ensemble and single cell model.

2.4.1.2 Ensemble-level models

Models comprising tens to thousands of nerve cells linked to create net-
works address a distinct set of problems. These models may show how
inhibitory & excitatory effects are combined in aggregative forms about
neuronal action. For vertebrate & invertebrate central motif generators,
detailed and realistic models have been created.
Freeman’s investigations about the olfactory bulb of rats in higher ver-
tebrate brains have provided crucial awareness about electroencephalo-
graphic signals collected above the bulb’s plane are formed by the bulb’s
constituent neurons (Jindra, 1976). This and comparable olfactory bulb
computational models have been highly effective in describing a wide range
of actual data. They extended their effort at the basis of single cells by
 Neural modeling and neural computation in a medical approach 29

creating anatomically and physiologically accurate model of groupings of

hippocampal neurons. They are able to demonstrate how harmonic action
is created among the hippocampus, as well as why their anatomical &
physiological characteristics at the level of neurons are important. Their
computer investigations were closely related to actual experimental data.
Neocortical network models have demonstrated in what way a bunch of
neurons function broadly & how the action of a collective single neuron
converts within a macroscopically visible quantity like local field capacity
(Seemanthini & Manjunath, 2021).

2.4.1.3 Systems-level models

Neurological model of entire brain areas or complete nervous systems try to
reveal the nerve foundation for effective processes like behavioral or cog-
nitive tasks. Cognitive models and neural models differ in their approach to
studying the brain and its functions. While neural models focus on
describing the underlying neural mechanisms that give rise to behavior,
cognitive models attempt to explain behavior in terms of abstract mental
processes or representations (Cowan, 1998).
Systems-level neural models are often connected to simulated or actual
behaviors, which may be used to objectively analyze such a model’s per-
formance in the context of a specific task. There have been few attempts to
formally model whole cortical regions or groupings of areas. The relative
youth of this sort of study is due, in part, to the astonishing computational
needs of a genuine neural model at about this scale, and, in part, to the
conceptual difficulties of identifying important structural and dynamic
constraints to also be incorporated in such a model. Existing models often
employ crude approximations of single neural units, typically representing
complete neuronal circuits or populations rather than single cells.

2.4.2 Machine learning

Machine learning enables machines to learn without being explicitly pro-
grammed. What does “without being expressly programmed” mean? To
understand this, consider the typical programming technique, in which we
explicitly program everything and provide it to the machine, and the
machine acts or delivers output based on the program. Traditional pro-
gramming examples include calculator programs, finding factorials, de-
termining if a value is odd or even, and so on. If we give the machine an
addition logic, it will constantly execute addition; it will never learn to
conduct subtraction until expressly programmed to do so. The machine is
not learning in this case; it just produces the output that has been pro-
grammed (Srivastava et al., 2021).
We need two things to properly apply machine learning: data and an
algorithm. These are the absolutely necessary conditions for implementing
30 Computational Techniques in Neuroscience

machine learning. Appropriate data aids in the generation of more accurate

models. What does “suitable” imply in this context? For example, if our
goal is to categorize students based on grades, the relevant data would
include academic data from students, which will aid in better categorizing
students based on grades. However, if we offer students’ personal infor-
mation data instead of academic data, the model developed will not be
useful in achieving our goal.

2.4.2.1 Artificial intelligence vs machine learning vs deep learning

2.4.2.1.1 Artificial intelligence
While AI is a field of science that involves creating machines that can
perform tasks that would normally require human intelligence, it is not
necessarily focused solely on creating robots. In fact, AI can be used to
develop a wide range of intelligent systems, including software programs
and digital assistants, as well as physical robots.
There are two stages of AI:

a. General Artificial Intelligence: This can execute any intellectual work

as accurately as a person.
b. Narrow Artificial Intelligence: This can outperform humans in a
certain activity.

2.4.2.1.2 Machine learning

Machine learning is a subsection of Al in which we attempt to equip machines
with the capacity to learn on their own without being explicitly programmed.
To forecast/predict the output in machine learning, we apply statistical al-
gorithms including decision tree, support vector machine), and KNN.
There are three types of machine learning categories:

2.4.2.1.2.1 SUPERVISED LEARNING

“Machine learns under supervision,” as the name indicates, and “train me”
to make it easier for you. Consider a student studying for test questions and
having access to the answer key to all those questions; in this scenario, the
student is our model, the exam questions are the input, and the solution key
is the intended output. This form of data is referred to as “labeled data,”
and it is used in supervised learning. We train the model using labeled data,
and the model learns the link between both the output and input variables/
features during training. As the training phase concludes, the model is
tested and therefore enters the testing phase, in which the test input char-
acteristics are supplied to the model, and the model will now categorize the
output/predicted output. Now that the anticipated and intended outputs
 Neural modeling and neural computation in a medical approach 31

have been matched, we can claim that the model is more appropriate and
the error margin is low; however, if the gap between the two is more, we
can say that the error margin is greater and the model must be trained more/
better. So in supervised learning:

• Labeled data should be used to train the model.

• Feedback is provided for the model to improve.
• Predicting/classifying the output.

2.4.2.1.2.2 UNSUPERVISED LEARNING

“Machine learns on its own without supervision,” as the name implies. The
output variables in unsupervised learning are unlabeled. There are no known
combos of input and output variables. Unsupervised learning is concerned
with examining correlations between input variables and identifying hidden
patterns that may be used to generate new labels for potential outputs. For
example, a student is given 50 shapes, but the kids have no clue what they are
(they are still learning), and we do not define any labels or names of the forms.
This is an example of unsupervised learning with unlabeled data. Now the
student will try to understand the patterns; for example, the student will
make a group of shapes that has four comers, another group of shapes with
three corners, and one more group of shapes with two corners. So, here the
student tried to make clusters of similar input elements, and that’s what we
do in unsupervised learning. Further to those clusters made by students, new
labels could be given. You are right if you think that the shape labels/names
are quadrilateral, triangle, and circle. So, in unsupervised learning:

• Unlabeled data is provided.

• No feedback is given.
• Finds hidden patterns in data.

2.4.2.1.2.3 REINFORCEMENT LEARNING

In contrast to supervised and unsupervised learning, reinforcement learning

constructs a prediction model by gathering input from random “trial and
error” as well as learning from previous iterations. We can also say that
reinforcement learning is a prize; for example, in the case of self-driving
cars, avoiding a crash results in a positive score, while crashing results in a
negative score; from this, the model learns which actions should be exe-
cuted under what conditions in the environment.

2.4.2.1.3 Deep learning

Deep learning is a subdivision of machine learning. It is the next footstep in
progression of machine learning. It has a layered architecture and applies an
32 Computational Techniques in Neuroscience

artificial neural network (ANN), which was motivated more by biological

neural networks. The human brain often evaluates and processes the
information it gets before attempting to recognize it from earlier stored
knowledge. Deep learning algorithms are similarly trained to recognize
patterns and categorize various forms of information in order to produce
the required output when given an input. In machine learning, we must
manually give the features. However, deep learning automatically extracts
features for categorization, which necessitates a large quantity of data for
training the deep learning algorithm. As a result, the accuracy of deep
learning output is dependent on the quantity of data. We employed con-
volutional neural networks (CNN), ANN, and recurrent neural networks
(RNN) in deep learning.

2.4.3 Application of machine learning

Machine learning is an interdisciplinary discipline that may be utilized in a
variety of fields such as science and business.
Machine learning is applied in medical science to detect and diagnose
disorders. As a result, many diseases are treated and medical technology is
speedily evolving. We have the ability to create three-dimensional images
that can anticipate the exact location of a lesion inside the brain. It makes
possible detection of brain cancers as well as other brain-related illnesses,
image recognition, traffic prediction, speech recognition, product recom-
mendations, email spam and malware filtering, self-driving cars, virtual
personal assistant, stock market trading, online fraud detection, automatic
language translation (Srivastava et al., 2021).
Machine learning implementations are not restricted to the examples
given here; there are several more areas where machine learning has proven
its worth, such as picture identification, speech recognition, medical diag-
nosis, and learning connections.
It is a phenomenal breakthrough in the area of AI. These machine
learning applications are just a few examples of how this technology may
enhance our lives.

2.4.3.1 Machine learning in healthcare

Machine learning is critical in health care. The following are some data on
this subject:

• Disease Identification
• Drug Discovery
• Personalized Treatment
• Clinical Trial Results
• Smart Electronic Health Records
• Treatment Queries and Suggestions
 Neural modeling and neural computation in a medical approach 33

2.4.4 Types of algorithms being used

2.4.4.1 Logistic regression
Logistic regression is also another approach adopted from statistics by
machine learning. It is the preferred strategy for binary classification issues
(those with only two class values) (By, 2018). There are two sorts of
variables in logistic regression: dependent variables and independent vari-
ables. The dependent variables are binary, which means they are either 1
(true, success) or 0 (false, failure). The purpose of this technique is to dis-
cover the model that best fits the connection between the independent and
dependent variables. Independent variables can be either continuous or
binary in nature. It is also known as logistic regression. This can deal with
the likelihood of measuring the relationship between the dependent and
independent variables.

2.4.2.2 Convolutional neural network

CNN is among the most widely used deep technique based on the animal’s
visual cortex (Levitt et al., 1994). CNN is now widely utilized in object
identification and tracking (Fan et al., 2010), text recognition and detec-
tion, visual detection, posture estimation (Toshev & Szegedy, 2014), scene
labelling (Couprie et al., 2013), and a variety of other applications (Nithin
& Sivakumar, 2015). CNNs are very close to ANNs, which may be seen as
an acyclic graph with a well-organized collection of neurons. Unlike neural
networks, CNNs only link the buried layers of neurons to the preceding
layer, which contains a subset of neurons. This form of sparse connection
allows the system to learn characteristics directly. Each portion of the CNN
layer has two or more dimensional filters that are convolved with the layer’s
input. Deep convolutional networks yield hierarchical feature extraction
(Aloysius & Geetha, 2018). CNN‘s architecture is made up of convolu-
tional layers, pooling layers, and fully linked layers.

2.4.2.3 Artificial neural networks

ANN is a part of deep learning that is an advanced algorithm that is formed
on the basis of the human brain. It is a blueprint to mimic the human brain.
As our brain trains itself using previous data, the ANN may train through
data and deliver answers in the sort of forecasts or classifications. ANN
may also be known as a neural network that is made up of many nodes that
are equivalent to the neurons as neurons work in the nervous system by
exactly the same mechanism as nodes.
An ANN node is equivalent to neurons. The node has two parts. The first
one is the summation part, and the second one is the activation function
part. The summation part calculates the weighted sum of all the in-
puts (x1 w1 + x 2 w 2 + + xn wn = xi wi ).
34 Computational Techniques in Neuroscience

Once the weighted sum is calculated, it is sent to the activation func-

tion. The activation function generates a particular output for a given
node based on the input that is getting provided. Logistic regression, the
statistical approach with which they have the most similarities, is a novel
option for ANNs. Neural networks are algorithms that mimic the struc-
ture of the human brain (How Neurl Networks Len Rom Eperience,
1992). They are made up of a set of mathematical equations that are
employed to replicate biological processes like learning and memory. For
this purpose, neural networks have been constructed. Among other uses,
they are used to diagnose acute myocardial infarction and acute pulmo-
nary embolism and to forecast intensive care unit (ICU) resource utili-
zation following cardiac surgery (Tu & Guerriere, 1992). The purpose of
a neural network is identical to logistic regression modelling: to predict a
result using the values of some predictor variables. The technique taken in
constructing the model, on the other hand, is entirely different. Although
there are several forms of neural networks, each node in the input layer is
normally associated with each or every node in the hidden layer, so each
node inside the hidden layer is generally linked to every node in the
output layer (Tu, 1996). ANN is divided into three parts, as described in
Figure 2.1

• Input layer
• Hidden layer
• Output layer

Figure 2.1 Architecture of artificial neural network.

 Neural modeling and neural computation in a medical approach 35

Figure 2.2 Correlation learning mechanism.

2.4.5 Considered learning algorithms

The suggested correlation learning mechanism (CLM) model is made up of
CNNs that collaborate with traditional neural networks throughout the
training stage (ANN). Both neuronal designs are part of a structure. It is
learning how to analyze CT brain scans while sharing data in the form of
filter palettes for CNN and numeric characterizing the evaluated picture for
ANN. Figure 2.2 depicts the overall concept of CLM.

2.5 BEST PERFORMING ALGORITHM

Medical examinations in modern medical clinics are supported by computer

technologies that employ computational intelligence to detect potential
health concerns more efficiently. One of the most significant applications is
the examination of CT brain images, where deep learning algorithms pro-
duce the most exact findings. In this paper, we present a unique CLM of
DNN topologies that blends CNNs with the traditional design. The
36 Computational Techniques in Neuroscience

supporting neural network assists CNN in determining the best filers for
pooling for convolution layers. As a result, the primary neural classifier
learns more quickly and efficiently. The results reveal that our CLM model
can achieve 96% accuracy, 95% precision, and 95% recall.

2.6 NORMALIZATION AND NEURAL CODING

The brain is built in modules. Engineers are well aware of the benefits of
modularity as strong technologies are built on modules that could be
copied and trumbled, such as transistors and web servers. This idea ap-
pears to be used by the brain in the following two ways (i) modular
circuits and using (ii) modular calculations. Anatomic evidence shows the
presence of canonical microcircuits, which are repeated across brain
areas, such as the cerebral cortex (Rodney et al., 1991). Physiological or
behavioral evidence points to the presence of canonical brain computa-
tions, which are typical computational modules that conduct the same
core processes in diverse contexts. A canonical neural computation can
use a variety of circuits and processes. Various brain areas or species may
use different accessible components to execute it.
Exponentiation and linear filtering are two well-known paradigms of
canonical neural computations. Exponentiation, a kind of thresholding,
acts at the neuronal and network levels, for example, the mechanism
through which eye and limb motions are produced. (Saito et al., 2018). This
procedure serves several important functions, including preserving sensory
selectivity, decorrelating signals, and establishing perceptual choice. A
common computational method in sensory mechanism, which is linear fil-
tering (weighted summing by linear receptive fields), is carried out, at least
roughly, at several levels of vision, hearing, and somatosensation. It aids in
the explanation of a wide range of perceptual events and may possibly be
implicated in sensory and motor systems (Prasetyoputri, 2021).
In several neural networks, a third type of computation has been
observed: divisive normalization. Normalization evaluates a ratio between
the response of a single neuron and the overall activity of a clump of
neurons. Normalization was proposed in the early 1990s to explain the
non-linear characteristics of neurons in the primal visual cortex. Similar
computations have previously been suggested to explain light adaptation of
the retina, size invariance of the fly visual system, and associative memory
in the hippocampus. Evidence gathered since then shows that normalization
is involved in a wide range of modalities, brain areas, and animals.
Theorists have proposed numerous (not mutually incompatible) rationales
for normalizing, the majority of which are connected to code efficiency and
enhancing sensitivity. Normalization modifies the procurement of neural
responses to make better use of the dynamic range offered, enhancing sen-
sitivity to changes as input (Prasetyoputri, 2021). Light adaptation of the
 Neural modeling and neural computation in a medical approach 37

retina allows for great sensitivity to tiny changes in visual characteristics over
a broad range of intensities. Normalizing reward values produces depiction
capable of distinguishing between dollars and million dollars, hence ex-
panding the reward system’s effective dynamic range and the invariance with
regard to some stimulus dimensions. Normalization in the fly’s antennal lobe
is hypothesized for allowing odorant identification and discrimination inde-
pendent of concentration. In the retina where normalization discards infor-
mation about the mean light level in order to retain invariant representations
of other visual properties (e.g., contrast). Normalization in V1 disposes
contrast information used to encode picture patterns (e.g.,orientation),
maximizing discriminability independent of contrast.
It is believed that MT encodes velocity independently of structural
pattern. Normalization inside the ventral visual pathway might aid in the
creation of object representations that are immune to changes (size, loca-
tion, lighting, and occlusion) during the process of decoding a distributed
neural representation. The responses of a group of neurons adjusted for
distinct speeds and directions are assumed to represent visual motion in
visual region MT. These responses may be seen as discrete samples of a
likelihood density function for which the firing rate of each neuron is
proportionate to a probability, the means of the distribution anticipates
stimulus velocity and the variance of the distribution indicates the uncer-
tainty in that prediction. If the firing rates are normalized to sum a constant,
the mean & variance may be determined simply as weighted sums of the
firing rates, the same constant for each stimulus. Differentiating between
stimuli, normalization can help a linear classifier distinguish between neural
representations of various inputs. A point in n-dimensional space represents
the response of ˆ neurons to a stimulus. The points associated with com-
parable stimuli group together. A linear classifier distinguishes between
stimulus categories by connecting them using hyperplanes. This is chal-
lenging if certain stimuli elicit indestructible reactions while others elicit
weak responses; the hyperplane that establishes the boundary far from the
origin might fail close to the origin, and vice versa. This difficulty is avoided
via normalization.
Max-pooling (winner-take-all) (Katiyar, 2022) a neuronal population is
normalized. It can work in two modes, i.e., when the inputs are almost
equal, averaging them, and determining a winner, when one input is much
higher than the rest, the competition is winner-take-all (max-pooling,
choosing the maximum in inputs). Max-pooling is hypothesized to work
across different brain systems and to underpin perceptual judgments by
picking the neuronal subpopulation (or psychophysical channel) with the
most responses. Object recognition models suggest numerous phases of
max-pooling. Similar to the “biased competition” concept, attention may
depend on normalization to change the calculation from average to max-
pooling, thereby picking the sub-population with the most replies and
suppressing the others.
38 Computational Techniques in Neuroscience

When the outputs of a neural representation are duplicated, it is con-

sidered inefficient. Normalization can help to decrease redundancy.
Normalization helps to ensure that V1 reactions to natural visuals are
statistically independent. As a result, the response V1 summation field
may be calculated using the responses of surrounding V1 summation fields.
Normalization removes the dependency, resulting in a more efficient rep-
resentation. Similarly, in the fly olfactory bulb, reactions of the community
of antennal lobe neurons (post-normalization) are more differentiable than
olfactory system neurons (pre-normalization) (Carandini & Heeger, 2012).

2.7 CONCLUSION

There has been a significant renewed interest in neural modelling

throughout the last decade. Work in this area has resulted in powerful new
communication mechanisms that are being used for a growing number of
medical conditions. The suggested CLM model learns quickly from data.
We can see from the data that everyone has the potential to learn quickly
and efficiently. The approach provides a fresh and simple concept of CNN
composition. The palette might be made up of the number of filters to
change the image and grids to extract the most significant aspects available.
We picked those that produced the greatest results in brain tumor identi-
fication in our study experiments; however, the CLM can also be employed
for other applications. The CLM may analyze numerous incoming palettes
(parameter of the CNN architecture) from each cycle due to concurrent
implementation.

REFERENCES

Abeles, M., Bergman, H., Gat, I., Meilijson, I., Seidemann, E., Tishby, N., &
Vaadia, E. (1995). Cortical activity flips among quasi-stationary states.
Proceedings of the National Academy of Sciences of the United States of
America, 92(19), 8616–8620. 10.1073/pnas.92.19.8616
Ackermann, R. F., Finch, D. M., Babb, T. L., & Engel, J. (1984). Increased glucose
metabolism during long-duration recurrent inhibition of hippocampal pyram-
idal cells. Journal of Neuroscience, 4(1), 251–264. 10.1523/jneurosci.04-01-
00251.1984
Adey, W. R., Walter, D. O., & Hendrix, C. E. (1961). Computer Techniques in
Correlation and Spectral Analyses of Cerebral Slow Waves during Discriminative
Behavior. 524, 501–524.
Aloysius, N., & Geetha, M. (2018). A review on deep convolutional neural net-
works. Proceedings of the 2017 IEEE International Conference on
Communication and Signal Processing, ICCSP 2017, 2018-Janua (November
2020), 588–592. 10.1109/ICCSP.2017.8286426
 Neural modeling and neural computation in a medical approach 39

Brasil-Neto, J. P., McShane, L. M., Fuhr, P., Hallett, M., & Cohen, L. G. (1992).
Topographic mapping of the human motor cortex with magnetic stimulation:
factors affecting accuracy and reproducibility. Electroencephalography and
Clinical Neurophysiology/ Evoked Potentials, 85(1), 9–16. 10.1016/0168-
5597(92)90095-S
By, S. (2018). Prediction of Asthma as Side Effect After Vaccination. May.
Carandini, M., & Heeger, D. J. (2012). Normalization as a canonical neural
computation. Nature Reviews Neuroscience, 13(1), 51–62. 10.1038/nrn3136
Couprie, C., Najman, L., & Lecun, Y. (2013). For Scene Labeling. Pattern Analysis
and Machine Intelligence, IEEE Transactions On, 35(8), 1915–1929. 10.1109/
TPAMI.2012.231
Cowan, N. (1998). Visual and auditory WM capacity. In Trends in Cognitive
Sciences (Vol. 2, Issue 3, pp. 77–78).
Fan, J., Xu, W., Wu, Y., & Gong, Y. (2010). Human tracking using convolutional
neural networks. IEEE Transactions on Neural Networks, 21(10), 1610–1623.
10.1109/TNN.2010.2066286
Gally, J. A., Montague, P. R., Reeke, G. N., & Edelman, G. M. (1990). The NO
hypothesis: Possible effects of a short-lived, rapidly diffusible signal in the
development and function of the nervous system. Proceedings of the National
Academy of Sciences of the United States of America, 87(9), 3547–3551.
10.1073/pnas.87.9.3547
Gerstner, W., & Naud, R. (2009). How good are neuron models? Science,
326(5951), 379–380. 10.1126/science.1181936
Gevins, A., Smith, M. E., McEvoy, L. K., Leong, H., & Le, J. (1999).
Electroencephalographic imaging of higher brain function. Philosophical
Transactions of the Royal Society B: Biological Sciences, 354(1387),
1125–1134. 10.1098/rstb.1999.0468
Goodisman, M. A. D., & Asmussen, M. A. (1997). Zyxwvutsrqp Zyxwvu Zyxwv
Zyx Zyxwvuts Zyxwv Zyxwvutsrqponm Zyxwvutsr Zyxw Zyxwvutsrqp
Zyxwvu Z Zyxwvutsrq Zyxwvu Zyxwvutsrq Zyxwvu. 338, 321–338.
Herz, A. V. M., Gollisch, T., Machens, C. K., & Jaeger, D. (2006). Modeling single-
neuron dynamics and computations: A balance of detail and abstraction.
Science, 314(5796), 80–85. 10.1126/science.1127240
Hong, S., Ag, B., & Fairhall, A. L. (2007). Single Neuron Computation: From
Dynamical System to Feature Detector. 3172, 3133–3172.
Horwitz, B., Friston, K. J., & Taylor, J. G. (2000). Neural modeling and functional
brain imaging: An overview. Neural Networks, 13(8–9), 829–846. 10.1016/
S0893-6080(00)00062-9
Horwitz, B., Rumsey, J. M., & Donohue, B. C. (1998). Functional connectivity of
the angular gyrus in normal reading and dyslexia. Proceedings of the National
Academy of Sciences of the United States of America, 95(15), 8939–8944.
10.1073/pnas.95.15.8939
Horwitz, B., & Sporns, O. (1994). Neural modeling and functional neuroimaging.
Human Brain Mapping, 1(4), 269–283. 10.1002/hbm.460010405
Horwitz, B., Tagamets, M. A., & McIntosh, A. R. (1999). Neural modeling,
functional brain imaging, and cognition. Trends in Cognitive Sciences, 3(3),
91–98. 10.1016/S1364-6613(99)01282-6
How Neural Networks Learn from Experience. (1992).
40 Computational Techniques in Neuroscience

Jindra, R. H. (1976). Mass action in the nervous system. Neuroscience, 1(5), 423.
10.1016/0306-4522(76)90135-4
Katiyar, K. (2022). AI-Based Predictive Analytics for Patients’ Psychological
Disorder BT - Predictive Analytics of Psychological Disorders in Healthcare:
Data Analytics on Psychological Disorders (M. Mittal & L. M. Goyal (Eds.);
pp. 37–53). Springer Nature Singapore. 10.1007/978-981-19-1724-0_3
Katiyar, K., Kumari, P., & Srivastava, A. (2022). Interpretation of Biosignals and
Application in Healthcare. In M. Mittal & G. Battineni (Eds.), Information and
Communication Technology (ICT) Frameworks in Telehealth (pp. 209–229).
Springer International Publishing. 10.1007/978-3-031-05049-7_13
Kwong, K. K., Belliveau, J. W., Chesler, D. A., Goldberg, I. E., Weisskoff, R. M.,
Poncelet, B. P., Kennedy, D. N., Hoppel, B. E., Cohen, M. S., Turner, R.,
Cheng -, H. M., Brady, T. J., & Rosen, B. R. (1992). Dynamic magnetic
resonance imaging of human brain activity during primary sensory stimula-
tion. Proceedings of the National Academy of Sciences of the United States of
America, 89(12), 5675–5679. 10.1073/pnas.89.12.5675
Levitt, J. B., Kiper, D. C., & Movshon, J. A. (1994). Receptive fields and functional
architecture of macaque V2. Journal of Neurophysiology, 71(6), 2517–2542.
10.1152/jn.1994.71.6.2517
McIntosh, A. R., & Gonzalez-Lima, F. (1991). Structural modeling of functional
neural pathways mapped with 2-deoxyglucose: effects of acoustic startle
habituation on the auditory system. Brain Research, 547(2), 295–302. 10.1016/
0006-8993(91)90974-Z
Motion, O. F. (1990). Differential Equations of Motion. March, 66.
Nithin, D. K., & Sivakumar, P. B. (2015). Generic Feature Learning in Computer
Vision. Procedia Computer Science, 58, 202–209. 10.1016/j.procs.2015.
08.054
Ogawa, S., Menon, R. S., Tank, D. W., Kim, S. G., Merkle, H., Ellermann, J. M., &
Ugurbil, K. (1993). Functional brain mapping by blood oxygenation level-
dependent contrast magnetic resonance imaging. A comparison of signal
characteristics with a biophysical model. Biophysical Journal, 64(3),
803–812. 10.1016/S0006-3495(93)81441-3
Oxide, T. H. E. N., & Of, D. (1947). The Nitrous Oxide Method for the
Quantitative. 1, 476–483.
Pineda, F. J. (1988). Dynamics and architecture for neural computation. Journal of
Complexity, 4(3), 216–245. 10.1016/0885-064X(88)90021-0
Pineda, F. J. (1989). Recurrent Backpropagation and the Dynamical Approach to
Adaptive Neural Computation. 172(1986), 161–172.
Poli, R., Cagnoni, S., Coppini, G., & Valli, G. (1991). A Neural Network Expert
System for Diagnosing and Treating Hypertension. Computer, 24(3), 64–71.
10.1109/2.73514
Prasetyoputri, A. (2021). Detection of Bacterial Coinfection in COVID-19 Patients
Is a Missing Piece of the Puzzle in the COVID-19 Management in Indonesia.
ACS Infectious Diseases, 7(2), 203–205. 10.1021/acsinfecdis.1c00006
Reggia, J. A. (1993). Neural Computation in Medicine. 5, 143–157.
Rodney, B. Y., Douglas, J., & Martin, K. A. C. (1991). Physiology, 440, 735–769.
Saito, B., Nakashima, H., Abe, M., Murai, S., Baba, Y., Arai, N., Kawaguchi, Y.,
Fujiwara, S., Kabasawa, N., Tsukamoto, H., Uto, Y., Ariizumi, H.,
Yanagisawa, K., Hattori, N., Harada, H., & Nakamaki, T. (2018). Efficacy of
 Neural modeling and neural computation in a medical approach 41

palonosetron to prevent delayed nausea and vomiting in non-Hodgkin’s

lymphoma patients undergoing repeated cycles of the CHOP regimen.
Supportive Care in Cancer, 26(1), 269–274. 10.1007/s00520-017-3845-y
Seemanthini, K., & Manjunath, S. S. (2021). Recognition of trivial humanoid group
event using clustering and higher order local auto-correlation techniques. In
Cognitive Computing for Human-Robot Interaction: Principles and Practices
(Issue August). 10.1016/B978-0-323-85769-7.00001-X
Sincich, L. C., Horton, J. C., & Sharpee, T. O. (2009). Preserving information in
neural transmission. Journal of Neuroscience, 29(19), 6207–6216. 10.1523/
JNEUROSCI.3701-08.2009
Sokoloff, L., Reivich, M., Kennedy, C., Rosiers, M. H. D., Patlak, C. S.,
Pettigrew, K. D., Sakurada, O., & Shinohara, M. (1977). the [14C]
Deoxyglucose Method for the Measurement of Local Cerebral Glucose
Utilization: Theory, Procedure, and Normal Values in the Conscious and
Anesthetized Albino Rat. Journal of Neurochemistry, 28(5), 897–916.
10.1111/j.1471-4159.1977.tb10649.x
Srivastava, A., Seth, A., & Katiyar, K. (2021). Microrobots and Nanorobots
in the Refinement of Modern Healthcare Practices. Robotic Technologies
in Biomedical and Healthcare Engineering, May 2021, 13–37. 10.1201/
9781003112273-2
Toshev, A., & Szegedy, C. (2014). DeepPose_Human_Pose_2014_CVPR_paper.pdf.
2014 IEEE Conference on Computer Vision and Pattern Recognition,
1653–1660. https://www.cv-foundation.org/openaccess/content_cvpr_2014/
html/Toshev_DeepPose_Human_Pose_2014_CVPR_paper.html
Tu, J. V., & Guerriere, M. R. (1992). Use of a neural network as a predictive
instrument for length of stay in the intensive care unit following cardiac
surgery. Proceedings / the … ... Annual Symposium on Computer Application
[Sic] in Medical Care. Symposium on Computer Applications in Medical Care,
666–672.
Tu, J. V. (1996). Advantages and Disadvantages of Using Artificial Neural
Networks versus Logistic Regression for Predicting Medical Outcomes.
49(11), 1225–1231.
Chapter 3

Neural networks and neurodiversity

The key foundation for neuroscience
Hera Fatma, Harshit Mishra, and Kalpana Katiyar
Department of Biotechnology, Dr. Ambedkar Institute of Technology
for Handicapped, Kanpur, Uttar Pradesh, India

3.1 INTRODUCTION

Neuroscience, which includes neurology, psychology, and biology, is a

relatively recent field of study. Over the past century, significant progress
has been made in the understanding that covers the anatomical features,
physiology, biochemistry, and pharmacology of the mammalian brain
(Parra, Hruby, and Hruby, n.d.). Fundamentally, brains are organic com-
puter devices. They enable the capacity of an organism to see and com-
prehend its environment, combine and decide from several data streams,
and adjust to a changing situation by turning a flood of complicated and
confusing sensory input into coherent cognition and action. In light of this,
it is probably not surprising that biology has long served as an inspiration
for computer science, the field of creating artificial computing systems (Cox
and Dean 2014a). Although there are many prospects for collaboration
between computer science and neuroscience, the path to developing neu-
rologically algorithms has been difficult and winding. Here, we look back
at the historical relationships between computer science and neuroscience
and ahead to a new era of potential cooperation enabled by recent rapid
developments in experimental neuroscience and biochemically based com-
puter vision approaches. Our attention is on the applications of neuro-
scientific algorithms, areas that have been successful, where they are
currently ineffective, and where deeper links are expected to be beneficial.
Autistic and other neurodivergent advocates and activists have histori-
cally led and made up the majority of the neurodiversity movement, with
limited engagement from neurotypical stakeholders. We are starting to
notice a favorable change in neurotypical stakeholders’ views about autism
as the neurodiversity movement gets support within the larger autistic
community. Treatment objectives are increasingly centered on problems of
critical concern for the autistic community rather than the normalization
of autistic persons, and strengths-based methods to support and interven-
tion are widely acknowledged as best practices (den Houting 2019).
Anatomical, electrophysiological, and computational restrictions are taken

42 DOI: 10.1201/9781003398066-3
 Neural networks and neurodiversity 43

into account when using recurrent neural networks (RNNs), a family of

computer models, to explain neurobiological events (Barak 2017). RNNs
may be taught using input-output examples or they can be created to
implement a specific dynamical principle. Making use of trained RNNs for
computing jobs and as neural phenomenon elucidations have advanced
significantly in recent years. I’ll go through how using trained RNNs
in conjunction with the use of reverse engineering might offer an alternate
structure for modelling in neurological science and perhaps act as a potent
tool for developing hypotheses. Despite recent developments and potential
advantages, there are still several basic holes in the theory of these networks
(Srivastava and Jha 2022). We will talk about these difficulties and
potential solutions.

3.2 WHAT IS NEUROSCIENCE?

Neuroscience refers to a relatively recent field that integrates biology,

psychology, and neurology. The understanding of biochemistry, psy-
chology, pharmacy, and structural features of the mammalian brain has
advanced significantly during the past century. Considering that cognitive
neurosciences have developed, which concentrate explicitly on compre-
hending greater-level cognitive functions using imaging technologies,
comprehension of some of the basic emotional, mnemonic, attentional,
perceptual, and cognitive processes has also improved (Parra, Hruby, and
Hruby, n.d.). We now have a better grasp of the extremely complicated
processes that underlie arithmetic, reading, reading comprehension, and
speaking and articulation thanks to neuroimaging. Thereafter, it seems
appropriate thinking about how we may apply improved knowledge of our
mammalian brain growth and activity to research pedagogic issues.
According to its broad definition, neuroscience studies how the brain
develops and remembers at all levels of organization, from molecules and
cells to whole brain systems (like the network of neural regions and path-
ways that underlies our capacity for language comprehension and speech,
for example). This emphasis on memory and learning can occur at different
extents. Studying synaptic processes and cell signalling – where one neuron
joins another neuron through synapse – is critical for comprehending
knowledge, but is looking closely at the activities of particular areas in the
brain, such as the hippocampus, using invasive or non-invasive techniques
(Material 2002).
Our vocabulary for describing the brain and our grasp of modern man-
made technology have been crucial to our understanding of the brain
throughout history. Descartes used hydraulic analogy and the flow of fluids
to illustrate how the mind works (Cox and Dean 2014b). More and more,
“channels” and “frequencies” were used to describe how the brain func-
tions, throughout the radio era (Srivastava, Seth, and Katiyar 2021).
44 Computational Techniques in Neuroscience

Perhaps unsurprisingly, we speak in the language of contemporary tech-

nology today. Neuroscientists talk more and more about the “circuits” and
“computations” that occur within neurons and how different parts of the
brain interact to create “networks” of activity. It would be easy to dismiss
the “computational” perspective on neurological science as just another
fad, but it is an analogy with deeper roots: beyond the adage “the brain is
a computer,” machine learning science offers a strict systematic approach
and methodologies for reasoning about information systems, isolating what
gets tabulated (the “algorithm”) from the direction it gets tabulated (the
“implementation”) (Ogden and Miller 1966). Today’s world has access
to a tremendous amount of computing power. Since the invention of the
Internet, we have become accustomed to interacting with enormous com-
puter networks and have developed the technology to take advantage of
their combined capacity. Several organizations have started massive,
international projects to model individual brain cells or whole brains in
silico (McCalpin 1995). General perception and object identification in
specific, however, presents a fascinating case study for the confluence of
computers and neurological science, even if a thorough examination of all
relationships between computers and neurological science is outside the
purview of the current chapter. Here, we examine this interface’s past and
present and offer potential channels for further cross-pollination (Cox and
Dean 2014b).

3.3 ARTIFICIAL NEURAL NETWORK: A BRIEF CHRONOLOGY

Unexpectedly far back in the timeline of computers lies the origin of bio-
logically inspired algorithms. In 1943, McCulloch and Pitts established
the concept of a “integrate and fire” neuron, and Hebb was the first to put
out the notion of assimilation and accommodation in the brain cells in
the late 1940s: “What fires together, wires together” (Graben and Wright
2011). In contrast, the transistor was not created until 1947, usable inte-
grated circuits did not arise until the end of late 1950s, “minicomputers”
and mainframes didn’t become widely used until the late 1960s, and per-
sonalized computer systems didn’t come into being until the late 1980s
and 1990s. The fact that the proposal would come to pass before actual
application by such a long time is evidence of these early pioneers’ vision
(Cox and Dean 2014b). Rosenblatt’s “perceptron,” which presented a
straightforward configuration of neurons with output and input that may
judge based on input vectors, was one of first examples of a neural network
that can learn. Since the original perceptron could only learn linear func-
tions of the inputs, it was found to be fundamentally limited. As a result,
neural network research temporarily suffered by the opposing “symbolic
artificial intelligence” faction, which sought to mimic intelligence through
processes that used abstract representations rather than deriving inspiration
 Neural networks and neurodiversity 45

directly from the mammalian brain’s neural network architecture (White

and Rosenblatt 1963). The conceptual limitations of the perceptron were
overcome by the inclusion of non-linear between the output and input of
the networks and a layer of units with activation functions, leading to the
emergence of numerous types of artificial neural networks over the fol-
lowing two decades (Thibault and Grandjean 1991).
Through the 1980s, artificial neural networks proliferated, and hope was
high. The study of many early neural network types with the goal of
addressing a variety of issues, from vision to language, became known as
“connectionism,” and the name quickly gained popularity. The condition in
neural networks was significantly advanced by a number of researchers
(including LeCun, Bengio, Hinton, and Schmidthuber, to mention a few),
and they were progressively used to address a range of problems in the
real world. A potent tool for image analysis is the convolutional neural
network, which made significant contributions to the developing science
of computer vision by doing very well on the challenge of hand-typed digit
identification, a practical use of neural networks. In the meantime, the
creation of neural networks has been guided by neurology, which has
inspired the architectural aspects of neural networks (e.g., pooling in
Fukushima’s noncognition water from basic to complicated) (Cox and
Dean 2014a).

3.3.1 Do deep learning and neuroscience still

need each other?
Autistic and some other neurodivergent advocates and activists have
historically been at the forefront of the neurodiversity movement, with
neurotypical stakeholders playing a far smaller role. Now that the neuro-
diversity movement is gaining support throughout the larger autism
community, we are starting to notice a favorable change in neurotypical
stakeholders’ views about autism (den Houting 2019). The fact that there
were difficult or unavailable ways to teach a machine using numerous
levels has been a main drawback of classic artificial neural networks.
However, it is clear from a visual neuroscience standpoint why building
multilayer networks is appealing. The ventral visual pathway in primates
is structured as a hierarchical network of interconnected visual regions
and is hypothesized to provide visual shape and object vision (Cox and
Dean 2014b). “Is neuroscience still relevant to machine learning with the
current excitement around contemporary deep learning techniques?” is a
natural question to pose. There is no doubt that deep learning techniques
have made significant progress, and a lot of optimism that deep learning
techniques may still be continued to solve a number of the machine
learning issues that are now plaguing the field.
The transfer of concepts from computational sciences to neurological
sciences, meanwhile, has been irregular and hasn’t resulted in the best
46 Computational Techniques in Neuroscience

advancement. In light of this, it could seem simple to conclude that

neuroscience is no longer necessary for machine learning. In fact, A.I.
Spring’s major player Yann LeCun was recently quoted as noting that
while biology might serve as inspiration, we shouldn’t let it lead us
astray. Out-of-set generalization is a concern for contemporary computer
vision. Performance evaluation is a key difficulty in computational
perspective. Performance in computational perspective is frequently
measured in comparison to data sets (K. Katiyar 2022). Torralba and
Efros, however, have neatly demonstrated the majority of the system
learned on one data set outperform those trained; a prejudice in such data
sets may be seen when one is compared to another that has the same item
categories (Hyvärinen 2010). Deep learning literature also lists additional,
less obvious warning indicators of problems. Szegendy and colleagues,
for instance, demonstrated how adding purposefully created “noise” to
photos might result in images being randomly misclassified by a con-
temporary deep learning system (Szegedy et al. 2014).

3.4 NEURO-IMAGING METHODS FOR COGNITIVE

DEVELOPMENTAL NEUROSCIENCES

Researches on neuroimaging are predicated on the idea that every intel-

lectual endeavor places certain demands on the brain, and that these
demands will be fulfilled by alterations to brain activity. Through PET
(positron emission tomography), local blood flow may be measured
explicitly or implicitly (by these changes in activity) (fMRI). ERPs can
monitor dynamic interactions between mental processes (Parra, Hruby, and
Hruby, n.d.). Children should not undergo PET, which requires infusion of
radioactive tracing elements (tracers). The tracers are present in greater
quantities in brain regions with increased blood flow, allowing images
of the radiation distribution to be formed and enabling the localization of
various neuronal activities. The localization of brain activity is also made
possible by fMRI (functional magnetic resonance imaging). The participant
must be placed within a sizable magnet (e.g., a huge tube) for this approach
to operate. It measures the computed tomography signal produced by the
proton of water molecules in brain cells (Geake and Cooper 2003).
Most fMRI studies use the changes in BOLD response as their primary
outcome indicator. Participants receive headphones to protect their ears
from the loud noise inside the magnet as well as a panic alarm switch, i.e.,
the magnet is claustrophobic. Considerations have made it difficult to
modify fMRI for usage with minors (who are frequently moving as well,
reducing imaging precision). Moreover, the number of such investigations
is increasing due to the development of specifically designed coils and less
confining head scanners (Parra, Hruby, and Hruby, n.d.). The first few
years of life see a decline in ERP latencies (concurrent with myelinization),
 Neural networks and neurodiversity 47

and late in childhood they reach adult levels. The temporal course of brain
processing has been extensively documented by ERP investigations, which
are also sensitive to millisecond changes (S. Katiyar and Katiyar 2021). To
comprehend the underlying cognitive processes, one uses the sequence of
recorded potentials, together with their magnitude and duration.

3.5 NEUROMYTHS

The fascinating phrase “neuromyths,” which was used in the OECD

study on studying the mammalian brain (OECD, 2002), illustrates how
quickly as well as easily scientific material may be misrepresented as
having educational value. The three myths that the OECD report focuses
on the most are the widespread idea that hemispheres differ (e.g., learning
using the left or right brain), the idea that certain types of education must
take place during “critical periods” when the brain is most plastic, and
the notion that the best educational interventions should coincide with
synaptogenesis (Berns, Cohen, and Mintun, n.d.) With regard to neuro-
myth, the statements about the left and right brains are likely supported
by the fact that various talents are localized in distinct hemispheres,
which is a form of hemispheric specialization. Many parts of analyzing
language, for instance, are left-lateralized (nevertheless, given what we
have shown, in blind persons or among those who go abroad to different
language group later in infancy). The right hemisphere is lateralized in
several elements of facial recognition, in contrast. However, the normal
brain has a significant number of cross-hemispheric connections, and in
every cognitive activity that has been studied thus far using neuroimaging,
including language and facial recognition tests, both hemispheres col-
laborate (Caspi et al. 2002).

3.6 NEURAL NETWORKS

A highly parallel distributed processor called a neural network is described

as having a natural predisposition to retain experience information and
make it accessible to users. Neural networks are made up of basic proces-
sing units. In a wide variety of applications, including robotics, speech
recognition, face identification, medical applications, manufacturing, and
economics, many researchers employ neural networks (Abdel-Nasser
Sharkawy 2020). Deep network researchers started to amass a constant
stream of useful accomplishments. Fully-connected designs, which examine
a picture with a number of filters at each level of a high hierarchy, have
shown to be effective at very successful at this time in the field of vision. The
spatial stationarity in natural images – where a group of comparable visual
elements tends to exist at various spatial places in an image – is naturally
48 Computational Techniques in Neuroscience

Figure 3.1 Non-linear neuron model.

captured by these systems, which require less weight in training, and per-
form effectively even if there are no weights learned (Cox and Dean 2014a).
Numerous medical uses and diagnoses employ neural networks (Abdel-
Nasser Sharkawy 2020).

3.6.1 Neuron models

The information-processing unit, or neuron, is essential to the functioning
of a neural network. Figure 3.1: Non-linear Neuron Model presents the
block diagram illustrating the neuron concept. The design of a family of
massive neural networks is based on this figure (Barak 2017).

3.6.2 General properties of neural networks

The numerous benefits that neural networks offer make them popular in a
wide range of applications. The following is a summary of these benefits.

• A neural network is an effective method for identifying non-linear

systems (Sharkawy, Koustoumpardis, and Aspragathos 2020).
• The massively parallel distributed architecture of the neural network
and its capacity for learning and generalization give it tremendous
computational power. Given the right training data, it was demon-
strated that such a neural network might approximate any compli-
cated (large-scale) linear or non-linear function (“Fundamentals of
Machine Learning and Softcomputing” 2006).
 Neural networks and neurodiversity 49

• The neural network is essential for identifying dynamic systems

and detecting faults since it not only may be used to identify when
a problem occurs but also gives a post-fault model of the robotic
manipulator. If practicle, the failure can be accommodated using
this post-fault model, which can also be utilized to isolate and
diagnose the issue (Wang, Czerminski, and Jamieson 2021).
• The estimate of seamless batch data comprising the input, output,
or perhaps gradients data of a function, as well as the approxima-
tion of a function’s derivatives, are two examples of how neural
network has some universality (Abdelhameed and Tolbah 2002).

3.6.3 Neural network classification

Here, we focus on the three primary neural networks —MLFNN, SLFFNN
and RNN.

3.6.3.1 Multilayer feedforward neural network (MLFFNN)

In artificial feedforward neural networks, there isn’t a recurrence in the
links in between units. The first artificial neural networks were referred
to as feedforward neural networks. They are also less complicated than
RNNs, their counterpart. They were referred to as feedforward since infor-
mation can only advance via the network’s input nodes, hidden nodes
(if any are present), and output nodes in that order (no loops) (Abdel-
Nasser Sharkawy 2020).

3.6.3.2 Single-layer feedforward neural network (SLFFNN)

The network output layer of the computing nodes is referred to as the single
layer (neurons). Source nodes’ input layer is not counted since no compu-
tation takes place there, as depited in Figure 3.2: Single-Layer Feedforward
Neural Network.

3.6.3.3 Recurrent neural network (RNN)

The neural models known as RNNs are those that can incorporate context
into their decision-making process. A feedforward neural network differs
from an RNN in that it contains at least one feedback loop (Wang,
Czerminski, and Jamieson 2021). The Elman and Jordan RNN models are
the two primary variations of the RNN, sometimes known as “simple”
RNNs, that have been suggested in the literature. An input layer, a hidden
layer, a delay layer, and an output layer are layers that make up both the
Elman and Jordan neural networks. While the Jordan neural networks’ delay
neurons are supplied from the output layer, the Elman neural networks’ delay
neurons are fed from the hidden layer (Abdel-Nasser Sharkawy 2020).
50 Computational Techniques in Neuroscience

Figure 3.2 Single-layer feedforward neural network.

3.7 RNNS AS A TOOL OF NEUROLOGICAL SCIENCE

RESEARCH

Continuous arrangements of enhancement issues are frequently required in

logical and design issues, including signal handling, system verification,
channel configuration, capability estimation, and regression analysis, and
neural networks have been generally explored for this reason. The quan-
tities of decision variables and limitations are typically extremely enormous,
and wide scope advancement methodology are considerably more chal-
lenging when they must be finished progressively to upgrade the output of a
dynamical framework. For such applications, classical enhancement strat-
egies may not be sufficient due to the issue of dimensionality and tough
prerequisites on computational time (K. Katiyar, Kumari, and Srivastava
2022). The neural network approach can tackle enhancement issues in
running times significant degrees quicker than the most famous enhance-
ment calculations executed on universally useful advanced computer
systems.
RNNs are a gathering of computational models that are generally
carried out as a device to portray neurobiological cycles taking into
consideration the electronic, physiological, and computational entangle-
ments (Barak 2017). The excellent spotlight on innovative work on these
 Neural networks and neurodiversity 51

Figure 3.3 Types of RNNs.

neural networks has been used during the 1990s. They have been
intended to comprehend time shifting or sequential patterns. A recurrent
net is a kind of neural network which includes feedback (or closed loop
connections). A few examples of RNN incorporate Hopfield, Boltzmann
machine, BAM, and so forth. RNN strategies have been generally exe-
cuted to a wide scope of issues. Basic to some degree, RNNs were created
in the 1980s with the intent to learn strings of characters. RNNs have
additionally resolved the issues that incorporate dynamical frameworks
with time sequences of events. The two primary variations of RNN,
likewise called “simple” RNNs, are, the Elman and the Jordan RNN
models. Both the Elman and Jordan neural networks comprise a delay
layer along with an input, hidden, and output layer. The postpone neu-
rons of an Elman neural network are taken care of from the hidden layer,
while the defer neurons of a Jordan network are taken care of from the
output layer (Abdel-Nasser Sharkawy 2020). Figure 3.3: Types of RNNs
below depicts the different types of RNNs.

3.7.1 RNNs as an important model for computations

RNNs are a type of computer models that are employed for problem
solving as well as for the explanation of neurobiological phenomena and
52 Computational Techniques in Neuroscience

dealing issues in machine learning (Sussillo 2014). These networks lack a

clear definition of upstream or downstream and allow a neuron to provide
information to any other neuron in the network. As a result, the network’s
present state as well as the current stimulus has an impact on the activity of
its neurons. Such networks are perfectly suited for computations that take
place over time, including keeping information in working memory or
gathering evidence to support a choice, because of this characteristic. The
justification for adopting RNNs as model comes from both electro-
physiological and anatomical considerations. A substantial portion of the
output from practically all cortical regions is directed toward the region of
origin. The normal cortical network is hence continually linked to itself.
Additionally, an RNN has the capacity to produce complex intrinsic
activity patterns that are similar to the continuing activity seen in the brain
(Barak 2017).

3.7.2 RNNs designing

Designing the connections based on intuition is one strategy that pro-
duced numerous significant results. Take the Romo lab’s delayed dis-
crimination task as an illustration. In this task, a monkey must decide
whether the frequency of the first vibrotactile stimulus is higher or lower
after receiving two stimuli separated by a short period of time greater in
contrast to the second one. As a first stage in the modelling process, we
idealize the task by stating that the behavior is essentially an input-
output transformation. The output transiently increases only if the first
input amplitude is greater than the second one, and decreases otherwise.
The input is made up of two pulses whose amplitudes correspond to the
vibration frequency in the experiment. The realization that a represen-
tation of the first frequency that does not alter over time is a useful
element for completing this job is now the insightful stage (observing the
neural activity can also yield insights). The two stimuli could be com-
pared using such a representation: Figure 3.4: RNN with Hidden
Neurons.
The network should represent an abstract variable called f1 that, in the
terminology of dynamical systems, follows the following low-dimensional
dynamics during the delay: df1/dt = 0. Using intuition or computational
techniques, these dynamics can be converted to the connectivity of an
RNN (Machens, Romo, and Brody 2005). In the N-dimensional phase
space, every value of the variable f1 corresponds to a certain point x.
These points are gathered into a line attractor, which was designed into
the network dynamics as a one-dimensional manifold of fixed points.
Different tasks result in various dynamical objects, such as saddle points
that mediate judgments, line attractors that implement the accumulation
of evidence, point attractors that represent associative memories, and
many more (Hopfield 1982). A low-dimensions dynamical system is used
 Neural networks and neurodiversity 53

Figure 3.4 RNN with hidden neurons.

in each of these instances to generate a hypothesis about the underlying

mechanism, which is then translated into a high-dimensional RNN.
Because it makes all assumptions clear, the implementation serves as a
consistency check for the hypothesis (Abbott 2008). Additionally, pre-
dictions may be made by comparing the activity of model neurons to that
observed experimentally.

3.7.3 Functionality and optimization

Various utilitarian roles have been proposed for recurrent associations, by
both the neuroscience and computer vision networks. One normal
thought is that repetition empowers relevant data to be consolidated to
improve generally equivocal sources of information. One can perceive
profoundly debased images when an outsourced setting furnishes addi-
tional hints. Consolidating such a setting, for example in a Bayesian
system, is a famous idea, however, one that still needs to be figured out.
Likewise, various models place explicit jobs for hierarchical criticism
associations in distributing consideration regarding various pieces of a
scene. New apparatuses in neuroscience progressively give test access to
straightforwardly concentrate on these connections. For example, infec-
tions presently exist that can bounce across a solitary neural connection,
conveying hereditarily encoded markers and opsins that empower the
action of neurons that give a contribution to a given objective to be es-
timated and controlled. RNN approaches give a clear idea of these con-
nections, and their functions will be one region where neuroscience and
computer vision could appreciate exceptional cooperative energy (Cox
and Dean 2014b).
54 Computational Techniques in Neuroscience

When a network is optimized, it is told what it should achieve but is

not given many specific instructions on how to do it. As a result, optimizing
and evaluating RNNs may be used to generate hypotheses for the next
experiments and data analysis if the processes used by RNNs after opti-
mization can be understood. For instance, the network could find a totally
unexpected solution to the issue. It’s possible that the best (or locally
optimal) solution produced by optimization will take into account
insignificant elements of the issue but nonetheless have an impact and alter
the underlying answer. Modelling gain fields in the parietal cortex, as well
as pattern production and neural dynamics in the motor system, are early
examples of approaches that use this strategy (Sussillo 2014).

3.8 RNNS CAN BE TRAINED WITHOUT INTUITION

A different approach to building functioning RNNs is training, which is

widely utilized in the machine learning community. The computational
strength of RNNs comes from the fact that a neuron’s activity is influenced
by more than just the current flow, not only by the network’s input but also
by the network’s state, which keeps track of previous inputs. This benefit
comes with a cost because RNNs are challenging to train. However, sub-
stantial advancements in training techniques recently have made it possible
to use RNNs for a range of tasks (Jaeger and Haas 2004). Consider the
above-mentioned vibrotactile discrimination task. By providing the net-
work with a large number of example trials and adjusting connectivity in
accordance with a learning rule, the connectivity can be taught.
The outcome will be a high-dimensional neural network that can solve
the problem, and it will be possible to compare the activity of the model
neurons to that observed experimentally (Enel et al. 2016). In some in-
stances, the trained example compared to designed networks has a higher
correspondence to genuine neurons (Barak et al. 2013).
The final trained network, in contrast to the intended networks, re-
sembles a mysterious black box. The fundamental method through which
the network can complete the task is unknown to us. It appears that by
skipping the “intelligence hypothesis” stage, we have replaced the poorly
known brain with yet another poorly understood complex system (RNN).
But as we’ll see below, this substitute is a lot more easily accessible for study
(Gao and Ganguli 2015).

3.9 HYPOTHESIS AND THEORY GENERATION

If the learned artificial network can be successfully reverse engineered, we

will have inadvertently uncovered a hypothesis rather than deliberately
 Neural networks and neurodiversity 55

considering it (Sussillo 2014). This was the situation in the context-

dependent accumulation of evidence, which led to the discovery of a pair
of line attractors with a specific relationship between their left- and right-
eigenvectors (Mante et al. 2013). Although the finding of this mechanism
was partially an automation of the scientific method, it was not hypoth-
esized. This is because determining how much reverse engineering is
actually interpretable is a judgment call. The mechanism discovered for
the vibrotactile discrimination example was both anticipated and
unexpected. On one hand, as was predicted for such a task, requiring
varied delay times resulted in the creation of a line attractor. On the other
hand, it was discovered that the decision itself was mediated on some
networks by a saddle point and on other networks by a progressive dis-
tortion of the output pulse.
The most recent findings and understandings from training RNNs for
different tasks are encouraging. However, there is a significant knowledge
gap between what we know about designed or random networks and what
we know about such networks (Hopfield 1982). What are the restrictions
of this strategy? Where will the networks fail, and which tasks can be
trained? The solutions discovered by training algorithms are how invariant?
What aspects of the single neuron model being utilized or the learning
algorithm itself affect the solutions? Modern theories for trained RNNs
must be developed in order to address these and numerous other unresolved
concerns. Such theories have a variety of directions. In terms of mathe-
matics, it is feasible to train several networks on the same job while sys-
tematically changing particular network architecture, for instance. This
strategy was refined in a recent study by training a network on a task with
many phases (Enel et al. 2016). The researchers demonstrated that creating
an explicit representation of the task phase, as opposed to letting one
emerge implicitly during training, enhanced performance and offered a
superior fit to electrophysiological data from monkeys executing this task.
Utilizing straightforward tasks that lend themselves to more in-depth study
and serve as building blocks for more complicated environments is another
strategy. Given that many trained networks have been found to contain
fixed points, examining networks that have been trained to contain a
number of specified fixed points is a logical first step in this direction. Tools
from mean field theory and systems theory were recently used to produce a
better match to the monkeys’ electrophysiological data collected during this
task. Utilizing straight forward tasks that lend themselves to more in-depth
study and serve as building blocks for more complicated environments is
another strategy. Given that many trained networks have been found to
contain fixed points, examining networks that have been trained to contain
a number of specified fixed points is a logical first step in this direction. The
dynamics of such networks were recently analyzed using a mix of mean
field and systems theory methods (Rivkind and Barak 2017).
56 Computational Techniques in Neuroscience

3.10 INTRODUCTION TO NEURODIVERSITY

Unfortunately, it is difficult to define “neurodiversity.” The neurodiversity

can be segregated into three distinct meanings. First and foremost, “neu-
rodiversity” might simply refer to the fact that people have a variety of
minds and brains, just as “biodiversity” can refer to the fact that living
things are various. In this sense, since no two people have exactly the same
mind or brain, even groups of neurotypical people are neurodiverse. It is far
more difficult to describe “neurodiversity” in its various contexts. Although
some (Kapp 2020) prefer the term “neurodiversity framework”. The
“neurodiversity movement,” refers to an activist group that works to en-
hance the rights and welfare of persons with neurologically abnormal dis-
abilities, or “neurodivergent” people. This concept is very different from
the term “neurodiversity approaches”.
In many aspects, the neurodiversity methods do actually resemble a
paradigm. Both concepts are hard to describe; therefore, it makes sense to
think that individuals rely mainly on their everyday assessments of whether
something adheres to a neurodiversity approach on how closely it resembles
other examples that have been shown to do so.
The neurodiversity approaches aim to prescribe a proper course of action
in relation to human neurocognitive diversity, much like paradigms attempt
to prescribe a proper course of action for conducting science. Furthermore,
they are largely action-oriented and prescriptive, again similar to para-
digms. However, compared to a scientific paradigm, the neurodiversity
methods seem far more varied. They are not just difficult to describe; as will
be covered in more detail below, different people seem to have funda-
mentally differing opinions on key aspects of what a proper neurodiversity
strategy should entail. It could be more accurate to refer to different
“neurodiversity approaches” rather than a single “neurodiversity para-
digm” (Dwyer 2022).

3.11 NEURODIVERSITY: THE SITUATION OF INCLUDING

AUTISTIC EMPLOYEES AT WORK

Rapid progress of technology over the past few years has resulted in sub-
stantial changes in workplace employment practices due to greater com-
puting capacity, the massive growth of technical data and massive
algorithmic advancements. The previous technological revolutions have
been primarily brought about by improvements in popularly used tech-
nologies, such as steam power, electricity, and computer-based technolo-
gies. However, the current Industrial Revolution includes a paradigm shift
across all fields of study, industries and economies because it is raising
significant political and philosophical issues in the process (Last 2017).
Artificial intelligence (AI) algorithms, for instance, “trained” on historical
 Neural networks and neurodiversity 57

huge data and reflect the ideals and preconceptions of their creators as well
as programmers (Montes and Goertzel 2019).
With the algorithms expanding globally, they amplify prejudices and
perpetuate stereotypes against the most marginalized people. It becomes
essential to include ethical and human values in technologies. In fact, the
separation between technological and social advancement is no longer
viable in a setting of widening wealth disparities among workers. New
paradigms and stories must be developed immediately for inclusive pros-
perity (Last 2017). The term “neurodiversity” in psychology refers to the
amalgamation of benefits and drawbacks brought on by a person’s unique
brain makeup. These variations include disorders, including attention def-
icit hyperactivity disorder (ADHD), autism spectrum disorder, dyslexia,
and dyspraxia, among others. As autistic workers are presently in the
spotlight for the majority of programs pertaining to employment that favor
neurodiversity, the word “neurodiversity” is used in this research to more
precisely refer to those individuals. Autism is a neurological psychiatric
illness that lasts a lifetime and affects people differently in terms of per-
ception and cognition. The neurodiversity paradigm takes into account the
intangible distinctions between individual brains and intelligences, while
the majority of literature based on diversity, incorporation, or inequality
highlights overt characteristics like gender, age, or race.
This study offers a novel conceptual framework for examining the dif-
ferent connections between neurodiversity and the digital transition, which
researchers view as a complementary pair. Despite their close connections,
neurodiversity management and the digital transition have not yet been
thoroughly researched.

3.11.1 The links between technology, organization,

and skills
According to the skill-bias technological change theory (SBTC) (Acemoglu
and Autor 2011), based upon the extent of complementarity among various
inserts, the labor division between employees and machines, defined by
their skill sets, is altered as a result of technological development. Since it
disregards the organizational spectrum of the job, SBTC theory is deter-
ministic. According to technological determinism, workstation organiza-
tion and the distribution of work load between employees and technologies
may be dictated by technology. In order to assert that developments,
training, and institutional changes are necessary to create gains in terms of
productivity with the involvement of new technologies, scientists are also
taking note of the institutional aspect of technology-based progress
(Camiña, Díaz-Chao, and Torrent-Sellens 2020). In other words, the
implementation of technologies, expertise based on skill set, and supervi-
sory design practices all work well together. In terms of microeconomics,
when the reciprocal variables pertaining to a function escalate at the same
58 Computational Techniques in Neuroscience

time, the value of that particular function escalates higher than the sum-
mation of the changes that are brought about by the escalation in every
variable when considered separately. Also, this strategy is very acceptable.
Economic acceptance takes the place of technological determinism. The
coordination of technical and organizational decisions is motivated by the
pursuit of performance (Lambrecht and Tucker 2019). Reversing the loss to
examine the influence of human acceptance in the technology framework is
another method for highlighting biases in technologies. Instead of focusing
on how technologies affect the workforce, let’s examine how the makeup of
the workforce affects technologies. Biases in the values and traits contained
in technologies are created by technology designers. While the obvious
gender imbalance in the IT sector is receiving a lot of attention, algorithmic
biases related to impairments also raise additional concerns about social
norms and stereotypes as well as job impediments. Initiatives to promote
neurodiversity address the issue of preconceptions in the workplace while
aiming for a more inclusive hiring process (Lambrecht and Tucker 2019).

3.11.2 Problem analysis

Researchers have devised two important and connected ideas, generally
known as skill bias technology and organizational change (SBTC and
SBOC) and productive interrelations, to explain the developing disparities
at work associated with technological change.
These ideas suggest that a worker’s abilities or degree of competence
serve as the basis for comparison. Homogeneous workers possess hetero-
geneous talents or abilities. Obviously, additional factors distinguish
workers when they are hired or promoted. On the job market and at work,
stereotypes and discrimination against workers are prevalent. It is essential
to take into account the workforce diversity, or the identity of the em-
ployees. The observational approach is applied by taking into account the
particular traits, propensities, and stereotypes associated with autism. A
field experiment demonstrated that appeals for considering jobs that said an
employee had Asperger syndrome and autism obtained 26% lesser ex-
pressions of significance for a similar standard of skill (Ameri et al. 2018).
The SBOC, SBTC, and fruitful reciprocal studies are used to frame the
connections between digital revolution and the workforce’s neurodiversity
in ensuing subsections.

3.11.3 Neurodiversity at the workplace at different levels

Neurodiversity recognizes that each person’s uniqueness as a human being
is marked by the diversity of their perceptual and mental abilities.
Understanding human variety is essential for comprehending neurodi-
versity. Consider the example of the intellectual quotient (IQ) Gaussian
distribution, which indicates that 95% of the population has an IQ between
 Neural networks and neurodiversity 59

70 and 130. Individuals in the tails of the distribution differ from the mean
or median because they deviate statistically from it. Each person is clearly
defined by a variety of cognitive, emotional, and perceptual characteristics.
But for the majority of people, the differences between two people pale in
comparison to what they have in common. These differences are greater for
autistic people, which beg the query of how these individuals specifically
add up to the organization.
Both at the macro- and microeconomic levels, neurodiversity practices are
acknowledged to have good social and inclusionary effects (Krzeminska et al.
2019). Apart from reputational advantages, transnational as well as major
companies from various organizations have pioneered a change in how these
individuals take charge of their staff and have created programs to utilize
hitherto untapped autistic abilities. As a result, these organizations are actu-
ally acquiring a competitive benefit because of these creative projects (Austin
and Pisano 2017). More businesses of all sizes and in a variety of industries
are adopting neurodiversity hiring policies (Austin and Sonne 2014).
At the level of the individual, characteristics defining exceptional autistic
employees involve their capacity for concentration, pattern recognition,
ability to accomplish monotonous tasks, remarkable attention to detail, and
participation. This may be the reason why many businesses that specialize
in the placement of autistic individuals provide software-testing positions or
coding jobs. According to a study, certain autistic people may be well suited
for these positions since they have good rule-based system-building skills
and a low mistake probability on activities requiring close attention to
detail. The American Psychiatric Association (2013) noted that reoccurring
motif hobbies or pursuits may also be a strength for autistic professionals
who complete repetitive tasks in their field of interest. However, digitali-
zation could have a negative effect on these jobs due to the automation of
repetitive operations and the replacement of skilled jobs based on rules by
artificial intelligence. This observation calls into question whether autistic
workers’ skills can be replaced by machinery. In fact, repetitive tasks and/or
those requiring pattern recognition may be more susceptible to automation
(substitution).
The establishment of an organization that accepts employees on the
spectrum of autism is necessary for their employment and engagement at
the organizational level, creating a workforce of neurodiverse employees
(S.Markel & Elia, 2016). In a study, a variety of concerns and strategies
are highlighted that have been used in various national, organizational,
and institutional settings to promote the employment of autistic em-
ployees. According to (Austin and Pisano 2017), there are seven essential
steps in the process of identifying neurodiversity in the worksite: the
creation of non-stereotypical hiring practices; training of employees and
employers; customization of the supervisory environment; specification of
techniques for career management; guidance and counselling from experts
having deep knowledge of neurodiversity in order to gain a clear
60 Computational Techniques in Neuroscience

understanding relating to the specifications of the employees; creations

related to scaling strategies; and mainstreaming the program.
It is not a surprise that the IT sector receives the majority of attention in
the articles pertaining to the effective handling of neurodiversity at the
corporate level (Austin and Pisano 2017). In a field that moves quickly and
where talents are continuously changing, the issue of a skill shortage is one
that is frequently raised. Our study advances knowledge related to signifi-
cance of gap in digital skills, the dynamism in IT job dismantling, and their
possible connections to the adoption of neurodiversity initiatives.
The lack of digital skills is cited as a major motivator for neurodiversity
initiatives in academia management research (Austin and Pisano 2017),
although there is no clear evidence of a scarcity of digital and technical
skills. According to Cappelli (2015), employers are primarily responsible
for creating mismatches involving the ability gap between demand, supply,
and shortage of skill sets, as a result of inadequate vocational guidance at
the organization. This is an indication that the scarcity of digitally skilled
workers is endogenous, resulting from corporate behavior, rather than
exogenous to businesses. In fact, unfair hiring practices and discrimination
restrict the pool of viable candidates and could lead to a scarcity of digital
skills. As a result, encouraging neurodiversity at the corporate level is
crucial for finding untapped sources of creative genius.

3.11.4 Methodology
This exploratory study seeks to determine the linkage between digital
transition and neurodiversity management. The research was shaped by a
phenomenological approach. A qualitative research method known as
phenomenology describes participants’ actual experiences in order to better
comprehend their nature or significance. To accomplish the goals of the
study, researchers employed a purposive sample strategy in accordance with
the phenomenological approach. To find suitable interview candidates,
researchers looked at two factors: leadership or competence in neurodi-
versity efforts, and solid IT industry knowledge (Walkowiak 2021).
In order to gather information with a suitable level of complexity and
diversity, sixteen candidates were interviewed in 2018 and 2019. As per
the participant expertise and contact intensity, this number in phenome-
nology typically ranges from 5 to 30 people. Sixteen participants offered the
amount of saturation that was seeked for, given the research’s targeting of
specialists, and the nascent approach of workplace initiatives in neurodi-
versity. Participants are listed in Table 3.1: Participants.
Three sets of questions served as the framework for the quasi-interviews.
Identifying autistic workers’ skills and how they relate to the digital transition
was the goal of the first set of questions. Second series of queries centered on
highlighting advantages as well as disadvantages pertaining to autistic em-
ployees related to innovation, artificial intelligence, and the shortage of
 Neural networks and neurodiversity 61

Table 3.1 Participants

P No. Field Gender Country
1. Financial F Canada
2. Information Technology M Australia
3. Social Enterprise M America
4. Software F Canada
5. Specially-abled Services F Canada
6. Information Technology M Britain
7. Information Technology F Australia
8. Academic M Australia
9. Charity Trust Institutes M Australia
10. Accountancy M America
11. Information Technology M Australia
12. Software F Britain
13. Social Enterprise M Canada
14. Specially-abled Services M Canada
15. Academic F Australia
16. Information Technology M America

programmed skills. The final array of queries focused on the organizational

architecture pertaining to neurodiversity’s digital component.
An integrated method was implemented to code the content. It is less
restricting than a start list method and less biased than an entirely induction-
based method evolved from grounded theory. The broad code types were
started, and further data-based sub-codes were created. The initial types
of broad code covered the compatibility between autistic workers’ abilities
and programmed technologies, connections between neurodiversity-aware
Human Resource practices as well as technology utility, and ultimately
the dynamics belonging to hiring especially abled employees in conjunction
with mechanization, AI, and program technologies (Figure 3.5: Proposed
Research Framework). Researchers found various productivity factors that
control the interactions between digital changeover and the employee’s
neurodiversity (Walkowiak 2021).

3.11.5 Result
While discussing a range of interests and abilities they saw in a neurodiverse
workforce, participants also explicitly referred to noteworthy “performa-
tive” qualities that were connected with the analyzed output of an employee
when he/she was carrying out an IT oriented task. Resilience, a different
mode of innovative thinking as well as query resolving, which promotes
creativity in the workplace and aids in the digital transformation, was one
of these expressive capabilities.
62 Computational Techniques in Neuroscience

Figure 3.5 Proposed research framework.

In addition to technical qualifications, STEM, math, and computer

science qualifications, which are connected to the programmed transition,
all candidates cited a wide spectrum of capabilities and qualifications
displayed by autistic employees. Additionally, they mentioned expertise in
games, languages, the arts, history, literature, human resources, com-
munication, budgeting, medical, and film. The diverse skill set displayed
by autistic employees implies that they may be suitable for a range of
occupations. Participants talked about a range of interests and abilities
they saw in autistic employees, but they also methodically identified a few
standout skills. Participants mentioned a variety of abilities, but they also
noted that some of these skills—or how they are combined—might be
very different from person to person. These abilities included emphasis to
detail, a constant capacity for focus as well as discipline, firm involve-
ment, the capability to recognize patterns, knowledge-grasping caliber,
and innovative brain storming. Five participants felt that highlighting
these particular skills possessed by autistic people was undesirable
because it would help to reinforce negative preconceptions about autistic
workers. These preconceptions, according to participant 16, are a result
of the lack of leaders from the autistic community and social variations on
squad in charge of neurodiversity projects. The creation of “ergonomic”
environments that reduce the sensory challenges faced by autistic em-
ployees is a crucial component of neurodiversity projects’ success.
Workspace placement in locations with little stimuli promotes concen-
tration. Additionally, relatively basic technology that enhances the sen-
sory experience for autistic workers includes headsets and various light
colors (Walkowiak 2021).
 Neural networks and neurodiversity 63

3.12 SCOPE AND CONCLUSION

The combined scope of artificial intelligence, neuroscience, and computa-

tional systems biology is significantly broad. The motivation of this unique
difficulty is to create a drone view on areas and demanding situations in
which the three fields overlap with their defining aims and where these fields
might also benefit from a synergetic mutual exchange of ideas. The purpose
in the back on this special issue is that a multidisciplinary technique in
present-day artificial intelligence, neuroscience, and systems biology is vital
and that progress in these fields requires a mess of perspectives and con-
tributions from a huge spectrum of members. This special difficulty,
therefore, objectives to create a center of gravity pulling together instruc-
tional researchers and industry practitioners from a selection of regions and
backgrounds to percentage outcomes of modern-day studies and improve-
ment and to talk about existing and rising theoretical and realistic troubles
in synthetic intelligence, neuroscience, and structural biology, transporting
them beyond the event horizon in their individual domain names (Berrar,
Sato, and Schuster 2010).
Bioscience and neuroscience are both devoted to growing our infor-
mation of the human frame and its different aspects, with a mutual
interest in the use of expertise to enhance human existence. These fields
significantly diverge, however, due to their focus on observation.
Questions that arise in biomedicine (and finally bioethics) are commonly
constrained to the area of clinical research and clinical motion. While
neuroscience has its analogues to those questions, its scope extends nicely
beyond the confines of clinical exercise: the mind’s precise function in
behavior engenders questions in a spectrum of fields with significant ef-
fects on society, such as neuro-politics, neuro-economics, and neuro-
regulation (McCoy et al. 2020).

REFERENCES

Abbott, L. F. 2008. “Theoretical Neuroscience Rising.” Neuron 60 (3): 489–495.

10.1016/j.neuron.2008.10.019.
Abdel-Nasser Sharkawy. 2020. “Principle of Neural Network and Its Main Types:
Review.” Journal of Advances in Applied & Computational Mathematics 7:
8–19. 10.15377/2409-5761.2020.07.2.
Abdelhameed, Magdy M, and Farid A Tolbah. 2002. “Design and Implementation
of a Flexible Manufacturing Control System Using Neural Network.”
International Journal of Flexible Manufacturing Systems 14 (3): 263–279.
10.1023/A:1015883728142.
Acemoglu, Daron, and David Autor. 2011. Skills, Tasks and Technologies:
Implications for Employment and Earnings. Handbook of Labor Economics.
Vol. 4. Elsevier B.V. 10.1016/S0169-7218(11)02410-5.
64 Computational Techniques in Neuroscience

Ameri, Mason, Lisa Schur, Meera Adya, F. Scott Bentley, Patrick McKay, and
Douglas Kruse. 2018. “The Disability Employment Puzzle: A Field Experi-
ment on Employer Hiring Behavior.” ILR Review 71 (2): 329–364. 10.1177/
0019793917717474.
Austin, Robert D., and Gary P. Pisano. 2017. “Neurodiversity as a Competitive
Advantage.” Harvard Business Review 2017 (May-June): 9.
Austin, Robert D., and Thorkil Sonne. 2014. “The Dandelion Principle: Redesigning
Work for the Innovation Economy.” MIT Sloan Management Review 55 (4):
67–72.
Barak, Omri. 2017. “Recurrent Neural Networks as Versatile Tools of Neuro-
science Research.” Current Opinion in Neurobiology 46: 1–6. 10.1016/
j.conb.2017.06.003.
Barak, Omri, David Sussillo, Ranulfo Romo, Misha Tsodyks, and L. F. Abbott. 2013.
“From Fixed Points to Chaos: Three Models of Delayed Discrimination.”
Progress in Neurobiology 103: 214–222. 10.1016/j.pneurobio.2013.02.002.
Berns, Gregory S, Jonathan D Cohen, and Mark A Mintun. n.d. “Absence of
Awareness.”
Berrar, Daniel, Naoyuki Sato, and Alfons Schuster. 2010. “Artificial Intelligence in
Neuroscience and Systems Biology: Lessons Learnt, Open Problems, and the
Road Ahead.” Advances in Artificial Intelligence 2010: 1–2. 10.1155/2010/
578309.
Camiña, Ester, Ángel Díaz-Chao, and Joan Torrent-Sellens. 2020. “Automation
Technologies: Long-Term Effects for Spanish Industrial Firms.” Technological
Forecasting and Social Change 151 (November 2019): 119828. 10.1016/
j.techfore.2019.119828.
Caspi, Avshalom, Joseph McCray, Terrie E. Moffitt, Jonathan Mill, Judy Martin,
Ian W. Craig, Alan Taylor, and Richie Poulton. 2002. “Role of Genotype in
the Cycle of Violence in Maltreated Children.” Science 297 (5582): 851–854.
10.1126/science.1072290.
Cox, David Daniel, and Thomas Dean. 2014a. “Neural Networks and Neuroscience-
Inspired Computer Vision.” Current Biology 24 (18): R921–R929. 10.1016/
j.cub.2014.08.026.
Cox, David Daniel, and Thomas Dean. 2014b. “Neural Networks and Neuroscience-
Inspired Computer Vision.” Current Biology 24 (18): R921–R929. 10.1016/
j.cub.2014.08.026.
Dwyer, Patrick. 2022. “The Neurodiversity Approach(Es): What Are They and
What Do They Mean for Researchers?” Human Development 66 (2): 73–92.
10.1159/000523723.
Enel, Pierre, Emmanuel Procyk, René Quilodran, and Peter Ford Dominey. 2016.
“Reservoir Computing Properties of Neural Dynamics in Prefrontal Cortex.”
PLoS Computational Biology 12 (6): 1–35. 10.1371/journal.pcbi.1004967.
“Fundamentals of Machine Learning and Softcomputing.” 2006. Neural Networks
in a Softcomputing Framework, no. Vc: 27–56. 10.1007/1-84628-303-5_2.
Gao, Peiran, and Surya Ganguli. 2015. “On Simplicity and Complexity in the Brave
New World of Large-Scale Neuroscience.” Current Opinion in Neurobiology
32: 148–155. 10.1016/j.conb.2015.04.003.
Geake, John, and Paul Cooper. 2003. “Cognitive Neuroscience: Implications
for Education?” Westminster Studies in Education 26 (1): 7–20. 10.1080/
0140672030260102.
 Neural networks and neurodiversity 65

Graben, Peter beim, and James Wright. 2011. “From McCulloch-Pitts Neurons
Toward Biology.” Bulletin of Mathematical Biology 73 (2): 261–265. 10.1007/
s11538-011-9629-5.
Hopfield, J. J. 1982. “Neural Networks and Physical Systems with Emergent
Collective Computational Abilities.” Proceedings of the National Academy
of Sciences of the United States of America 79 (8): 2554–2558. 10.1073/
pnas.79.8.2554.
Houting, Jacquiline den. 2019. “Neurodiversity: An Insider’s Perspective.” Autism
23 (2): 271–273. 10.1177/1362361318820762.
Hyvärinen, Aapo. 2010. “Statistical Models of Natural Images and Cortical Visual
Representation.” Topics in Cognitive Science 2 (2): 251–264. 10.1111/j.1756-
8765.2009.01057.x.
Jaeger, Herbert, and Harald Haas. 2004. “Harnessing Nonlinearity: Predicting
Chaotic Systems and Saving Energy in Wireless Communication.” Science 304
(5667): 78–80. 10.1126/science.1091277.
Kapp, Steven K. 2020. Conclusion - Autistic Community and the Neurodiversity
Movement: Stories from the Frontline. Autistic Community and the Neuro-
diversity Movement: Stories from the Frontline. 10.1007/978-981-13-8437-
0_22.
Katiyar, Kalpana. 2022. “AI-Based Predictive Analytics for Patients’ Psychological
Disorder.” In Predictive Analytics of Psychological Disorders in Healthcare:
Data Analytics on Psychological Disorders, 37–53. Springer.
Katiyar, Kalpana, Pooja Kumari, and Aditya Srivastava. 2022. “Interpretation of
Biosignals and Application in Healthcare.” In Information and Communication
Technology (ICT) Frameworks in Telehealth, 209–229. Springer.
Katiyar, Sarthak, and Kalpana Katiyar. 2021. “Recent Trends Towards Cognitive
Science: From Robots to Humanoids.” In Cognitive Computing for Human-
Robot Interaction, 19–49. Elsevier.
Krzeminska, Anna, Robert D. Austin, Susanne M. Bruyère, and Darren Hedley.
2019. “The Advantages and Challenges of Neurodiversity Employment in
Organizations.” Journal of Management and Organization 25 (4): 453–463.
10.1017/jmo.2019.58.
Lambrecht, Anja, and Catherine Tucker. 2019. “Algorithmic Bias? An Empirical
Study of Apparent Gender-Based Discrimination in the Display of Stem Career
Ads.” Management Science 65 (7): 2966–2981. 10.1287/mnsc.2018.3093.
Last, Cadell. 2017. “Global Commons in the Global Brain.” Technological
Forecasting and Social Change 114 (2016): 48–64. 10.1016/j.techfore.2016.
06.013.
Machens, Christian K., Ranulfo Romo, and Carlos D. Brody. 2005. “Flexible Control
of Mutual Inhibition: A Neural Model of Two-Interval Discrimination.” Science
307 (5712): 1121–1124. 10.1126/science.1104171.
Mante, Valerio, David Sussillo, Krishna V. Shenoy, and William T. Newsome. 2013.
“Context-Dependent Computation by Recurrent Dynamics in Prefrontal
Cortex.” Nature 503 (7474): 78–84. 10.1038/nature12742.
Material, Supplementary. 2002. “Caspi_2002_MAOA_AggressionMsat_SIMethods,”
1–7. papers2://publication/uuid/D97CED7A-D777-4E60-A883-A49E4CE1B625.
McCalpin, John D. 1995. “Memory Bandwidth and Machine Balance in Current
High Performance Computers.” IEEE Computer Society Technical Committee
on Computer Architecture (TCCA) Newsletter, no. May: 19–25.
66 Computational Techniques in Neuroscience

McCoy, Liam G., Connor Brenna, Felipe Morgado, Stacy Chen, and Sunit Das. 2020.
“Neuroethics, Neuroscience, and the Project of Human Self-Understanding.”
AJOB Neuroscience 11 (3): 207–209. 10.1080/21507740.2020.1778127.
Montes, Gabriel Axel, and Ben Goertzel. 2019. “Distributed, Decentralized, and
Democratized Artificial Intelligence.” Technological Forecasting and Social
Change 141 (February): 354–358. 10.1016/j.techfore.2018.11.010.
Ogden, Thomas E., and Robert F. Miller. 1966. “Studies of the Optic Nerve of
the Rhesus Monkey: Nerve Fiber Spectrum and Physiological Properties.”
Vision Research 6 (10). 10.1016/0042-6989(66)90108-8.
Parra, Miguel, George Hruby, and George Hruby. n.d. “Neuroscience and Education
Related Papers.”
Rivkind, Alexander, and Omri Barak. 2017. “Local Dynamics in Trained Recurrent
Neural Networks.” Physical Review Letters 118 (25): 1–5. 10.1103/PhysRevLett.
118.258101.
S. Markel, Karen, and Brittany Elia. 2016. “How Human Resource Management
Can Best Support Employees with Autism: Future Directions for Research and
Practice.” Journal of Business and Management 22 (1): 71–86.
Sharkawy, Abdel-Nasser, Panagiotis N Koustoumpardis, and Nikos Aspragathos.
2020. “A Neural Network-Based Approach for Variable Admittance Control
in Human-Robot Cooperation: Online Adjustment of the Virtual Inertia.”
Intelligent Service Robotics 13 (4): 495–519. 10.1007/s11370-020-00337-4.
Srivastava, Aditya, and Shashank Jha. 2022. “Data-Driven Machine Learning:
A New Approach to Process and Utilize Biomedical Data.” In Predictive
Modeling in Biomedical Data Mining and Analysis, 225–252. Elsevier.
Srivastava, Aditya, Aparna Seth, and Kalpna Katiyar. 2021. “Microrobots and
Nanorobots in the Refinement of Modern Healthcare Practices.” In Robotic
Technologies in Biomedical and Healthcare Engineering, 13–37. CRC Press.
Sussillo, David. 2014. “Neural Circuits as Computational Dynamical Systems.”
Current Opinion in Neurobiology 25: 156–163. 10.1016/j.conb.2014.01.008.
Szegedy, Christian, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan,
Ian Goodfellow, and Rob Fergus. 2014. “Intriguing Properties of Neural
Networks.” 2nd International Conference on Learning Representations,
ICLR 2014 - Conference Track Proceedings, 1–10.
Thibault, Jules, and Bernard P.A. Grandjean. 1991. “A Neural Network
Methodology for Heat Transfer Data Analysis.” International Journal of Heat
and Mass Transfer 34 (8): 2063–2070. 10.1016/0017-9310(91)90217-3.
Walkowiak, Emmanuelle. 2021. “Neurodiversity of the Workforce and Digital
Transformation: The Case of Inclusion of Autistic Workers at the Workplace.”
Technological Forecasting and Social Change 168 (April): 120739. 10.1016/
j.techfore.2021.120739.
Wang, Hongming, Ryszard Czerminski, and Andrew C. Jamieson. 2021. “Neural
Networks and Deep Learning.” The Machine Age of Customer Insight, 91–101.
10.1108/978-1-83909-694-520211010.
White, B. W., and Frank Rosenblatt. 1963. “Principles of Neurodynamics: Perceptrons
and the Theory of Brain Mechanisms.” The American Journal of Psychology
76 (4): 705. 10.2307/1419730.
Chapter 4

Brain waves, neuroimaging

(fMRI, EEG, MEG, PET, NIR)
Surbhi Kumari and Amit Kumar Dutta
Amity Institute of Biotechnology, Amity University Jharkhand Ranchi, India

4.1 INTRODUCTION

Neuroscience is the integrative science dealing with the study of the nervous
system (both central and peripheral nervous system) as well as its functions
and underlying diseases. The brain is the vital organ of the central nervous
system controlling all the other activities in the body. Computational
neuroscience is the study of the mechanism of brain functioning by various
tools and techniques using computer science (Sejnowski et al., 1988).
Our complex brain consists of fundamental units called neurons. The
functional neuron in the brain transmits the electrical signals. There are on
an average 90 billion neurons present in the human brain (Goriely et al.,
2015). The electrochemical signals are carried by these neurons and result
in passing of the iconic currents via the synapse. The coordination of the
electrical activities of the neurons results in the repeated rhythmic altera-
tions all over the regions of brain and are termed as brain waves (Buskila,
2019). (Figure 4.1)
The five brain waves that have been recognized to date are: gamma (g),
beta (b), alpha (a), theta (θ), and delta (d) waves. All these waves possess
different frequencies, for e.g., gamma waves have the highest frequency range
(<30 Hz), whereas delta waves have the lowest frequency range (0–4 Hz).
They all are responsible for different states of mind in the human brain
(Abhang et al., 2016). These brain waves are measured by a technique known
as electroencephalography, or EEG, that uses electrodes placed on the scalp
for recording the waves by non-invasive approach (Teplan, 2002). The brain
structure and function can be easily analyzed by studying the images of the
brain by the process of neuroimaging. It has been broadly classified into
structural as well as functional neuroimaging techniques. The functional
techniques like fMRI, PET, MEG, and NIR are covered in the chapter
(Noggle & Davis, 2021). Numerous methods of non-invasive neuroimaging
have fostered an enormous impact in upgrading our knowledge about the
brain functioning and neurological disorders and provided a novel insight
into the treatment of these diseases (Supek & Aine, 2016).

DOI: 10.1201/9781003398066-4 67
68 Computational Techniques in Neuroscience

Figure 4.1 The different lobes of the brain viz. frontal, parietal, occipital and temporal
lobe, that possess diverse functions like controlling emotions, processing
sensory data and visual information, and conducting auditory functions,
respectively.
( https://imotions.com/wp-content/uploads/2022/10/brain-lobes-iMotions.png).

4.2 BRAIN WAVES

The synchronized wave patterns generated by the flow of electric current

through the neurons are called brain waves (Ismail et al., 2016). The
voltage variations in the neurons occur due to the ionic flow within the
brain. These waves come about naturally while in mobile as well as in a
relaxed phase. They are determined using an EEG. These electrical activities
of the brain are recorded using the placement of metal electrodes on the
scalp to detect the distinct frequencies of the brain waves viz. gamma, beta,
alpha, theta, and delta (Jeong et al., 2011) (Figure 4.2).

a. Gamma (γ) Brain waves – The gamma brain waves are the fastest waves,
having the largest frequency range above 30 Hz (i.e., more than 30 cycles
per second). They govern while the brain is involved in analytical
problem solving (Jeong et al., 2011), deep learning, as well as creative
processing of language (Ismail et al., 2016). These waves have emerged
from the thalamic region of the brain. They are initiated to adjust the
activity and systems of the neurons. People with some injury in the
thalamus lack consciousness along with cognition, and as a result may
slip into a coma (Desai et al., 2015). There should be balance in gamma
waves for the right perception, cognitive focus, inspired learning, and
processing of information. Lowered levels result in impairment in
learning, depression, as well as ADHD. On the other hand, increased
 Brain waves, neuroimaging (fMRI, EEG, MEG, PET, NIR) 69

Figure 4.2 A non-invasive process of electroencephalography (EEG) for the analysis and
computation of brain waves.
( https://assets.nhs.uk/nhsuk-cms/images/E5RF5X.width-1534.jpg).

levels lead to stress, hyperactivity, and anxiousness (Dudeja, 2017). In

research, it was found that people practicing meditation had increased
gamma activation frequencies. The brain corresponds to the occurring
stream of consciousness and attention, among other cognitive functions,
which have all been related to gamma waves (Braboszcz et al., 2017).
They are regulated via interior processes like attention as well as working
memory. Its level elevates with the cognitive and noncognitive events
along with the sensory drive. In disorders associated with the central
nervous system like Alzheimer’s, epilepsy and Parkinson’s disease, the
gamma activities have been seen as irregular (Jia & Kohn, 2011).
b. Beta (β) Brain Waves – Beta waves lie between the frequency range
14–30 Hz (i.e., 14–30 cycles per second). These waves act through-
out the waking consciousness (Jeong et al., 2011). They are highly
active during problem solving like logical or analytical reasoning
(Koudelková et al., 2018), initiating new ideas, and generating solu-
tions. They promote elevated concentration levels and alertness of our
brain towards tasks (Ismail et al., 2016). Optimal amount of beta
waves leads to better focus, memory and ability to solve a problem.
When the frequency is low, it leads to a state of depression, deficit
cognition, daydreaming and disorders like attention-deficit hyper-
activity disorder, or ADHD. On the other hand, too much of it causes
increased stress, anxiety and even hyperactivity with a distracted mind
(Dudeja, 2017). These waves are present all over the motor region of
the cortex during the muscular isotonic contractions. The triggering of
the beta waves corresponds to better performance in academics due
to improved cognition as well as concentration (Desai et al., 2015).
70 Computational Techniques in Neuroscience

The alert mind is the result of beta waves that in turn result in effectual
functioning (Ashtaputre-Sisode, 2016).
c. Alpha (α) Brain Waves – Alpha waves are also called Berger’s wave as
they were manifested by Hans Berger in the 1930s (Stinson & Arthur,
2013). The conscious thinking in addition to the subconscious mind are
interconnected by the alpha waves. Their frequency lies between 8–13 Hz
(i.e., 8–13 cycles per second). They assist the feeling of relaxation and
calm down the body (Koudelková et al., 2018). They are initiated inside
the cortex, near the thalamus and the occipital lobe (Desai et al., 2015).
The alpha waves govern the state of light meditation or daydreaming
(Jeong et al., 2011), when the mind is relaxed, and thoughts are passing
over. The requirement of a stabilized amount of alpha waves is necessary
because if the quantity is less it results to insomnia, stress, and obsessive
compulsive disorder. If the quantity is more it leads to less concentration
and complete relaxed mind (Dudeja, 2017). These waves intensify the
learning process and the physical as well as mental health of a person. It
also aids in the therapy of depression, anxiety, and sleep dysfunctions by
introducing the deep relaxation approach along with mindfulness
(Stinson & Arthur, 2013). In older adults these waves ameliorate the
identification of words and enable precise memory performance.
The person behaves in a calm and composed manner (Desai et al.,
2015). A study revealed that the person suffering from anxiety
disorder had an increased frequency of alpha waves on the frontal
lobe (both sides) as well as on the parietal lobe (front side). The
disorders of anxiety are basically related to the dysfunction of the
forebrain (Cho et al., 2011) (Figure 4.3).
d. Theta (θ) Brain Waves – The frequency of the Theta waves ranges
between 4–8 Hz (i.e., 4 to 8 cycles per second). They are dominant
throughout normal sleeping hours, while meditating (Jeong et al.,
2011) and during states of fatigue (Ashtaputre-Sisode, 2016). Walter
and Dovey first described the presence of theta waves in 1944 in cases
associated with tumors in the sub-cortex (Schacter, 1977). They result
in profound meditation, relaxation, and enhanced memory in in-
dividuals. During light sleep or conscious dreaming in daytime, they get
activated and correlate with relieving stress (Ismail et al., 2016). Theta
rhythm is the other name given to these waves, which eventuate as a
repeated function turns to be self-governed. These waves originate in
the region of the cortex as well as hippocampus. It was suggested in a
study that theta waves are associated with constructing memories via
action within the hippocampus (Desai et al., 2015). While we are
dreaming or in a state of intuition, unconsciousness, or imagination,
these waves are governing our mind. Their excessive number results in
disorders like ADHD, distracted mind, impulsiveness, depression, and
hyperactivity disorders. Decreased amount of these waves causes stress,
anxiousness, and deficit emotional balance. Adequate balance in waves
 Brain waves, neuroimaging (fMRI, EEG, MEG, PET, NIR) 71

Figure 4.3 The different types of brain waves associated with their dominant states.
( https://ars.els-cdn.com/content/image/3-s2.0-B9780128044902000026-f02-01-
9780128044902.jpg).

leads to proper flow of thoughts, emotional stability, being intuitive

and relaxed (Dudeja, 2017). During hypnosis state, theta waves are
active, and our brain is filled with thoughts of our inner self. In some
research, it has been found that these waves are associated with psy-
chological occurrences in humans (Schacter, 1977).
e. Delta (δ) Brain Waves – The slowest brain waves are the delta waves,
with frequency ranging between 0–4 Hz (i.e., less than 4 cycles per
second). During deep sleep or profound meditation, these waves are
dominant as this deep sleep state is an essential and vital process that aids
in healing and the process of regeneration. Excessive number of delta
waves leads to brain trauma, disoriented mind, cognition disability and
disorders like ADHD. On the contrary, fewer delta waves causes prob-
lems in sleep cycles, thinking, learning, and inability of the body to
regenerate as well as revitalize itself (Dudeja, 2017). The Human Growth
Hormone is released along with the delta waves, which aids in the healing
process. If it is generated in a state of waking consciousness, it gives the
72 Computational Techniques in Neuroscience

chance to approach the subconscious state (Ismail et al., 2016). People

with impairment associated with learning processes or brain injuries
have unusual activity of the delta waves (Koudelková et al., 2018).

4.3 NEUROIMAGING

Neuroimaging is a non-invasive approach to get in touch with different

structures and activities of our brain (Gui et al., 2010). It has been an
emerging and indispensable procedure for recognizing the brain dynamics
on a structural level (Bandettini, 2009). The functional dynamics of the brain
can be encapsulated by neuroimaging by capturing the disparate timespans
of the brain activities (Kringelbach & Deco, 2020). The different functional
neuroimaging techniques, namely PET (positron emission tomography),
MEG (magnetoencephalography), fMRI (functional magnetic resonance
imaging), NIR (near infrared), provide remarkable probability in neurolog-
ical disease detection and examination (Drevets, 2000). The fMRI is a non-
invasive process that determines accurate brain activities with elevated
geometric resolution. The PET is the significant technique for depicting the
overall physiological features of the brain (Savoy, 2001) (Figure 4.4).

Figure 4.4 A magnetic resonance imagining (MRI) machine.

( https://ccnmtl.columbia.edu/projects/neuroethics/module1/foundationtext/
index.html).
 Brain waves, neuroimaging (fMRI, EEG, MEG, PET, NIR) 73

a. Functional Magnetic Resonance Imaging (fMRI) – The fMRI is a

non-invasive neuroimaging technique that has evolved to reveal the
time-dependent variations of the metabolisms occurring in the brain
(Glover, 2011). It has become a governing technique due to its least
exposure to radiation, less invasive nature, and greater availability
(Gui et al., 2010). From its establishment in 1992, it has enhanced
cognitive studies for the healthy as well as the defective brain. It uses
the MRI to represent the vital changes in the tissues underlying the
brain generated by the variations in the neurological metabolisms
(Chen & Glover, 2015). The fMRI is constructed on the principle of
MRI that uses magnetic fields with their gradients and the radio waves
for producing the images. It is basically associated with the learning of
the cognitive behavior of the brain, diagnosis of certain neurological
diseases, keeping track of different therapies, as well as depicting the
efficacy of certain drugs (Figure 4.5).

The neural activities of the brain result in various metabolic actions such as
rise in oxygen supply along with elongated blood flow, and these activities
are detected using various techniques namely BOLD, perfusion, and con-
trast fMRI. Among these the BOLD (blood oxygenation level dependent)
fMRI method is widely used (Gui et al., 2010). It maps the neurological

Figure 4.5 A fMRI scan of human brain depicting certain regions in a person suffering from
severe traumatic brain injuries.
( https://imotions.com/blog/learning/research-fundamentals/eeg-vs-mri-vs-
fmri-differences/ https://imotions.com/wp-content/uploads/2022/10/fMRI-
explained.jpg).
74 Computational Techniques in Neuroscience

activities by determining the variation in blood flow in the brain and de-
termines the alteration in corresponding signal strength related to the var-
ious cognitive conditions amidst the imaging process (Matthews & Jezzard,
2004). There lies a limitation for the same as the detection of brain activities
is not carried directly as it determines the variation in blood oxygenation
level instead of directly measuring the neuron activity (Chen et al., 2020).

b. Electroencephalography (EEG) – The discovery of the brain pos-

sessing electrical currents was carried out by Richard Caton in
1875, whereas discovering these currents can be measured at the
scalp surface was demonstrated by Hans Berger, who was a German
neurologist, in 1924 (Teplan, 2002). EEG is a neuroimaging tech-
nique involving a non-invasive method to determine the electrical
signals of the brain by making use of a cap containing metal elec-
trodes and sensors that are laid on different positions on the scalp.
EEG was first employed by Hans Berger in 1929, and in 1957 the
use of electrodes was discovered in recording the activities of the
brain by Gray Walter (Abhang et al., 2016). During the activation
of neurons, the flow of current is generated that is measured by an
EEG. The current perforates through various layers of the skin into
the skull and then to the neurons as a result recording their electrical
activities (Teplan, 2002). It records the brain waves of different
frequencies in accordance with the variations of mental state
(Abhang et al., 2016). Different types of electrodes are employed
such as the gel-based electrodes, water-based electrodes, and dry
electrodes (Figure 4.6).

They are positioned on a specific location. According to the 10–20 system,

they are laid down at 10% and 20% from the left, right, nasion and inion

Figure 4.6 Different metal electrodes incorporated into the cap that is placed on the scalp
to record the brain oscillations with contrasting frequencies.
( https://info.tmsi.com/blog/types-of-eeg-electrodes).
 Brain waves, neuroimaging (fMRI, EEG, MEG, PET, NIR) 75

points, although to attain a spatial resolution, an additional number of

electrodes are placed in this system (Müller-Putz, 2020). These metal
electrodes scrutinize the signals at the surface of head, then they are
amplified and are converted to digital pattern that is recorded on a com-
puter (Teplan, 2002), and the voltage potentials generated by the flowing
current in the neurons are recorded by the electrodes (Biasiucci et al., 2019).
The EEG readings are provided to gain comprehensive information about
the cognitive aspects, mental state such as stress, relaxation, etc. (Jebelli
et al., 2018). In clinical diagnosis and treatment of neural rehabilitation it
has its diverse range (Biasiucci et al., 2019). This neuroimaging method can
detect the disorders of the brain such as epilepsy, tumors, brain strokes,
injury, and dysfunction (Abhang et al., 2016).

c. Magnetencephalography (MEG) – MEG is a non-invasive functional

neuroimaging technique that includes measuring the magnetic field
produced by the electrical currents flowing through the neurons
(Singh, 2014). The activities of neurons as recorded by the MEG
technique are the result of both excited and inhibited postsynaptic
potential of the dendrites. This flow of current produces a magnetic
field projecting in a radial direction (Proudfoot et al., 2014).
According to the right-hand rule in Ampere’s law, electric current is
related to the magnetic field that is perpendicular to the direction of
current. The activated neurons produce electric current, which gen-
erates magnetic fields that get recorded with the help of MEG (Singh,
2014) (Figure 4.7).

Figure 4.7 Magnetoencephalography neuroimaging process determining the magnetic field

brought about by the electronic activity of the neurons in the brain.
( https://www.york.ac.uk/psychology/research/york-neuroimaging-centre/
research/magnetoencephalography/).
76 Computational Techniques in Neuroscience

Currently, MEG is the sole non-invasive imaging with higher-resolution

technique that is independent from vascular responses. Furthermore, it is
the only functional neuroimaging providing both elevated spatial as well as
temporal resolution (Wilson et al., 2016). The MEG readings are capable of
providing an insight regarding the processes and the functional framework
of the brain. The magnetic fields determined by the MEG imaging are very
minute; therefore, extremely sensitive sensors and detectors like SQUIDs or
superconducting quantum interference devices are employed. Current MEG
set-up comprises 100–300 sensors that envelope the head (Supek & Aine,
2016). The SQUIDs allow detection of very small magnetic field gradients
with unit femtotesla or fT that are produced at the surface of the skull with
the help of currents flowing in the cortex (Savoy, 2001). In the detection
and study of several neurological disorders, MEG imaging has wide ap-
plications (Wilson et al., 2016). In patients with epilepsy, MEG aids in the
identification of eloquent brain areas that helps in determining the affected
region during surgery (Singh, 2014).

d. Positron Emission Tomography (PET) – The PET is a functional

neuroimaging tool that aids in visualizing the physical as well as
atomic activities in a specific region of the brain or different body
parts (Raichle, 1983). This non-invasive technique is helpful in
ameliorating knowledge about the pathophysiological aspects of
a disease, diagnosis, and treatment results (Zimmer, 2009). It assists
in the diagnosis of neurological disorders and underlying injuries
of brain parts. The radioactive decay induces a huge amount of
radiation, which is employed in PET (Raichle, 1983). This technique
uses the radiotracers (radioactive particles) that cohere with the
target molecule with greater bond strength and selectivity. These
radiotracer molecules consist of an atom having immoderate
quantity of nuclear energy generated because of its contrasting
neutron count when compared to its stable state (Walker & Bilgel,
2021) (Figure 4.8).

Firstly, the radionuclides that emit positrons are generated and get
included in molecules to give the radiotracers. These radionuclides pro-
duce PET radioisotopes. The radiotracers are administered in patients,
which reaches to the intended organs, and positrons move and get united
with the electrons, producing the photons that get captured by the camera
and recorded via detectors. The images with temporal and spatial reso-
lutions are generated and examined for diagnosis as well as treatment of
diseases (Zimmer, 2009). Assessment of patients suffering from brain
stroke is really complicated as a distinct region of the brain examined to
check for permanent or partial damage caused, but with PET imaging it
has been quite simple to picture a distinguished area for suitable therapy
to be conducted. The same is true with a person with epilepsy, where this
 Brain waves, neuroimaging (fMRI, EEG, MEG, PET, NIR) 77

Figure 4.8 An amyloid PET scan (both positive and negative) depicting deposition of
proteins such as tau and amyloid that forms plaques in the brain of a person
suffering from Alzheimer’s disease.
( https://radiology.ucsf.edu/patient-care/services/specialty-imaging/alzheimer).

technique is widely used to study the symptoms of seizures associated

with epilepsy by detecting the affected functioning area of the brain
through implementation of PET (Raichle, 1983).
Alzheimer’s disease is a continuous degrading neurological ailment
accompanied by weakened memory, inability to think, and inability to fulfil
daily activities. For effective diagnosis of this disease, any of the two
amyloid or FDG PET scannings are acceptable as biomarkers. In amyloid
PET imaging the plagues causing damage to the nerve cells are visualized in
the patient (Rice & Bisdas, 2017). PET neuroimaging also aids in scruti-
nizing the synthesis of proteins, process of Krebs cycle, glucose metabolism
as well as DNA replications. The neurological activities of the brain and its
hemodynamic variations are demonstrated using PET for e.g., the volume
and flow of blood along with the use of oxygen (Savoy, 2001).

e. Near-Infrared Spectroscopy (NIRS) – NIRS is a non-invasive, functional

neuroimaging procedure that includes close supervision of the altera-
tions in the blood volume along with blood oxygenation associated with
brain functioning. It makes use of certain light wavelengths that are
exposed to the scalp surface to record the fluctuations in respective
ratios of both oxygenated as well as deoxygenated haemoglobin
throughout complete brain activity (Izzetoglu et al., 2005) (Figure 4.9).

However, the functional activities of the brain or any tissue can dominate
its optical characteristics. In the same way, the human brain retaliates
to the external stimuli and experiences certain physiological variations
such as fluctuations in the level of blood along with the electrical activi-
ties, resulting in changes in optical characters, too. The oxygenated
haemoglobin, as well as the deoxyhaemoglobin (oxy-Hb and deoxy-Hb)
78 Computational Techniques in Neuroscience

Figure 4.9 NIRS imaging with head cap containing sensors, light detectors, and light
source.
( https://www.pnas.org/doi/10.1073/pnas.2208729119).

acting as chromophores, comprises optical characters in both the near-

infrared range and the visible range. The difference in concentration of
both is determined using Beer-Lambert’s Law (Irani et al., 2007). The
deoxy-Hb chromophore absorbs below 790 nm, whereas the oxy-Hb
absorbs above 790 nm. The NIRS signals respond to the variations in
the neural activities that are entwined with the hemodynamic feedback
functions (Chen et al., 2020).
Several NIRS studies have built strong insight in detection of neurological
disorders in children as well as in adults like ADHD, Alzheimer’s disease,
epilepsy, migraine, etc. (Ehlis et al., 2014). Variations such as decrease in
oxygenated Hb and total Hb in the frontal lobe were found in Alzheimer’s
patients. This suggests that there was gradual decrease in the oxygen supply
in the degenerated regions of the brain, resulting in poor cognition in
them. In recent studies, it has been observed that in epileptic patients the
concentration of oxy-Hb and the blood volume increases profoundly during
seizures (Irani et al., 2007).
 Brain waves, neuroimaging (fMRI, EEG, MEG, PET, NIR) 79

4.4 CONCLUSION

The brain oscillations or brain waves are synchronized, and continuous

patterns are generated due to the electrical activities of the neurons in the
brain. They can be recorded easily via the process of EEG, which is a non-
invasive technique. It includes several sensors and electrodes that are placed
directly onto the scalp for recording the neuronal activities and mapping
them on a computer system. These waves are categorized into five major
types based on their frequencies and characteristics viz. gamma, beta,
alpha, theta, and delta. Functional neuroimaging is a speedily evolving field
in neuroscience. Diverse approaches and advancements have been made
so far to the prevailing techniques. A few recent examples comprise of the
establishment of fMRI, NIRS having high-intensity data recording proto-
cols, greater analysis of the subjects, and better treatment of the con-
valescents. Due to its speedy development and extensive implementation,
neuroimaging has provided a completely different scenario in clinical neu-
roscience as well as in research associated with better cognition and brain
dynamics.

REFERENCES

Abhang, P. A., Gawali, B., & Mehrotra, S. C. (2016). Introduction to EEG-and

speech-based emotion recognition. Academic Press. https://books.google.com/
books?hl=en&lr=&id=o4t4CgAAQBAJ&oi=fnd&pg=PP1&dq=introduction
+to+eeg+and+speech+based&ots=pDinWkxU5l&sig=Exyz8LZpITWcqBEZz
IFCzO7jUMc
Ashtaputre-Sisode, A. (2016). Emotions and brain waves. The International Journal of
Indian Psychology, 3(2), 14–18. https://books.google.com/books?hl=en&lr=&
id=XniVCwAAQBAJ&oi=fnd&pg=PA14&dq=EMOTIONS+AND+BRAIN
+WAVES&ots=NjAQTRCOTY&sig=UoOC2RNvKKuoEW3FBEZp5YhdtdM
Bandettini, P. A. (2009). What’s new in neuroimaging methods? Annals of the
New York Academy of Sciences, 1156(1), 260–293. 10.1111/j.1749-6632.
2009.04420.x
Biasiucci, A., Franceschiello, B., & Murray, M. M. (2019). Electroencephalography.
Current Biology, 29(3), R80–R85. 10.1016/j.cub.2018.11.052
Braboszcz, C., Cahn, B. R., Levy, J., Fernandez, M., & Delorme, A. (2017).
Increased gamma brainwave amplitude compared to control in three different
meditation traditions. PloS one, 12(1), e0170647. 10.1371/journal.pone.
0170647
Buskila, Y., Bellot-Saez, A., & Morley, J. W. (2019). Generating brain waves,
the power of astrocytes. Frontiers in neuroscience, 13, 1125. 10.3389/fnins.
2019.01125
Chen, J. E., & Glover, G. H. (2015). Functional magnetic resonance imaging
methods. Neuropsychology Review, 25(3), 289–313. 10.1007/s11065-015-
9294-9
80 Computational Techniques in Neuroscience

Chen, W. L., Wagner, J., Heugel, N., Sugar, J., Lee, Y. W., Conant, L., … … &
Whelan, H. T. (2020). Functional near-infrared spectroscopy and its clinical
application in the field of neuroscience: advances and future directions.
Frontiers in Neuroscience, 14, 724. 10.3389/fnins.2020.00724
Cho, J. H., Lee, H. K., Dong, K. R., Kim, H. J., Kim, Y. S., Cho, M. S., &
Chung, W. K. (2011). A study of alpha brain wave characteristics from
MRI scanning in patients with anxiety disorder. Journal of the Korean
Physical Society, 59(4), 2861–2868. https://www.researchgate.net/profile/
Woon-Kwan-Chung/publication/270110634_A_Study_of_Alpha_Brain_
Wave_Characteristics_from_MRI_Scanning_in_Patients_with_Anxiety_
Disorder/links/570df7f808aed31341cf87f0/A-Study-of-Alpha-Brain-Wave-
Characteristics-from-MRI-Scanning-in-Patients-with-Anxiety-Disorder.pdf
Desai, R., Tailor, A., & Bhatt, T. (2015). Effects of yoga on brain waves and
structural activation: A review. Complementary therapies in clinical practice,
21(2), 112–118. 10.1016/j.ctcp.2015.02.002
Drevets, W. C. (2000). Neuroimaging studies of mood disorders. Biological
Psychiatry, 48(8), 813–829. 10.1016/S0006-3223(00)01020-9
Dudeja, J. P. (2017). Scientific analysis of mantra-based meditation and its bene-
ficial effects: An overview. International Journal of Advanced Scientific
Technologies in Engineering and Management Sciences, 3(6), 21–26. https://
www.researchgate.net/profile/Jai-Dudeja/publication/318395933_Scientific_
Analysis_of_Mantra-Based_Meditation_and_its_Beneficial_Effects_An_
Overview/links/5baa003192851ca9ed23aabd/Scientific-Analysis-of-Mantra-
Based-Meditation-and-its-Beneficial-Effects-An-Overview.pdf
Ehlis, A. C., Schneider, S., Dresler, T., & Fallgatter, A. J. (2014). Application of
functional near-infrared spectroscopy in psychiatry. Neuroimage, 85, 478–488.
10.1016/j.neuroimage.2013.03.067
Glover, G. H. (2011). Overview of functional magnetic resonance imaging.
Neurosurgery Clinics, 22(2), 133–139. 10.1016/j.nec.2010.11.001
Goriely, A., Budday, S., & Kuhl, E. (2015). Neuromechanics: from neurons
to brain. Advances in Applied Mechanics, 48, 79–139. 10.1016/bs.aams.
2015.10.002
Gui, X. U. E., Chuansheng, C. H. E. N., Zhong-Lin, L. U., & Qi, D. O. N. G.
(2010). Brain imaging techniques and their applications in decision-
making research. Xin li Xue Bao. Acta Psychologica Sinica, 42(1), 120.
10.3724%2FSP.J.1041.2010.00120
Irani, F., Platek, S. M., Bunce, S., Ruocco, A. C., & Chute, D. (2007).
Functional near infrared spectroscopy (fNIRS): an emerging neuroimaging
technology with important applications for the study of brain disorders.
The Clinical Neuropsychologist, 21(1), 9–37. 10.1080/1385404060091
0018
Ismail, W. W., Hanif, M., Mohamed, S. B., Hamzah, N., & Rizman, Z. I. (2016).
Human emotion detection via brain waves study by using electroencephalogram
(EEG). International Journal on Advanced Science, Engineering and Infor-
mation Technology, 6(6), 1005–1011. https://www.researchgate.net/profile/
Wan-Woaswi/publication/329584955_Human_Emotion_Detection_Via_
Brain_Waves_Study_by_Using_Electroencephalogram_EEG/links/5c10b7a645
85157ac1bbb648/Human-Emotion-Detection-Via-Brain-Waves-Study-by-
Using-Electroencephalogram-EEG.pdf
 Brain waves, neuroimaging (fMRI, EEG, MEG, PET, NIR) 81

Izzetoglu, M., Izzetoglu, K., Bunce, S., Ayaz, H., Devaraj, A., Onaral, B., &
Pourrezaei, K. (2005). Functional near-infrared neuroimaging. IEEE Trans-
actions on Neural Systems and Rehabilitation Engineering, 13(2), 153–159.
10.1109/TNSRE.2005.847377
Jebelli, H., Hwang, S., & Lee, S. (2018). EEG signal-processing framework to obtain
high-quality brain waves from an off-the-shelf wearable EEG device. J. Comput.
Civ. Eng, 32(1), 04017070. https://www.researchgate.net/profile/Houtan-
Jebelli/publication/319550134_An_EEG_Signal_Processing_Framework_to_
Obtain_High-Quality_Brain_Waves_from_an_Off-the-Shelf_Wearable_EEG_
Device/links/5c85a1cd92851c69506b1fa8/An-EEG-Signal-Processing-
Framework-to-Obtain-High-Quality-Brain-Waves-from-an-Off-the-Shelf-
Wearable-EEG-Device.pdf
Jeong, E. G., Moon, B., & Lee, Y. H. (2011). A platform for real time brain-
waves analysis system. In International Conference on Grid and Distributed
Computing (pp. 431–437). Springer, Berlin, Heidelberg. 10.1007/978-3-
642-27180-9_53
Jia, X., & Kohn, A. (2011). Gamma rhythms in the brain. PLoS Biology, 9(4),
e1001045. 10.1371/journal.pbio.1001045
Koudelková, Z., Strmiska, M., & Jašek, R. (2018). Analysis of brain waves ac-
cording to their frequency. Int. J. Of Biol. And Biomed. Eng., 12, 202–207.
https://www.researchgate.net/profile/Zuzana-Koudelkova-4/publication/
334805116_Analysis_of_brain_waves_according_to_their_frequency/links/
5d4194e84585153e59312c60/Analysis-of-brain-waves-according-to-their-
frequency.pdf
Kringelbach, M. L., & Deco, G. (2020). Brain states and transitions: insights from
computational neuroscience. Cell Reports, 32(10), 108128. 10.1016/j.celrep.
2020.108128
Matthews, P. M., & Jezzard, P. (2004). Functional magnetic resonance imaging.
Journal of Neurology, Neurosurgery & Psychiatry, 75(1), 6–12. https://jnnp.
bmj.com/content/75/1/6.short
Müller-Putz, G. R. (2020). Electroencephalography. Handbook of Clinical Neurology,
168, 249–262. 10.1016/B978-0-444-63934-9.00018-4
Noggle, C. A., & Davis, A. S. (2021). Advances in neuroimaging. In Understanding
the Biological Basis of Behavior (pp. 107–137). Springer, Cham. 10.1007/
978-3-030-59162-5_5
Proudfoot, M., Woolrich, M. W., Nobre, A. C., & Turner, M. R. (2014).
Magnetoencephalography. Practical neurology, 14(5), 336–343. 10.1136/
practneurol-2013-000768
Raichle, M. E. (1983). Positron emission tomography. Annual Review of Neuro-
science, 6(1), 249–267. https://www.annualreviews.org/doi/pdf/10.1146/
annurev.ne.06.030183.001341
Rice, L., & Bisdas, S. (2017). The diagnostic value of FDG and amyloid PET in
Alzheimer’s disease—A systematic review. European Journal of Radiology,
94, 16–24. 10.1016/j.ejrad.2017.07.014
Savoy, R. L. (2001). History and future directions of human brain mapping and
functional neuroimaging. Acta Psychologica, 107(1-3), 9–42. 10.1016/S0001-
6918(01)00018-X
Schacter, D. L. (1977). EEG theta waves and psychological phenomena: A review and
analysis. Biological Psychology, 5(1), 47–82. 10.1016/0301-0511(77)90028-X
82 Computational Techniques in Neuroscience

Sejnowski, T. J., Koch, C., & Churchland, P. S. (1988). Computational neuro-

science. Science, 241(4871), 1299–1306. 10.1126/science.3045969
Singh, S. P. (2014). Magnetoencephalography: basic principles. Annals of Indian
Academy of Neurology, 17(Suppl 1), S107. 10.4103%2F0972-2327.128676
Stinson, B., & Arthur, D. (2013). A novel EEG for alpha brain state training,
neurobiofeedback and behavior change. Complementary Therapies in Clinical
Practice, 19(3), 114–118. 10.1016/j.ctcp.2013.03.003
Supek, S., & Aine, C. J. (2016). Magnetoencephalography. Springer-Verlag Berlin
An. https://link.springer.com/content/pdf/10.1007/978-3-642-33045-2.pdf
Teplan, M. (2002). Fundamentals of EEG measurement. Measurement Science
Review, 2(2), 1–11. http://www.edumed.org.br/cursos/neurociencia/Methods
EEGMeasurement.pdf
Walker, K. A., & Bilgel, M. (2021). PET Neuroimaging of Neurologic Disease:
Methods, Clinical and Research Applications. 10.31234/osf.io/u72e8
Wilson, T. W., Heinrichs-Graham, E., Proskovec, A. L., & McDermott, T. J. (2016).
Neuroimaging with magnetoencephalography: A dynamic view of brain path-
ophysiology. Translational Research, 175, 17–36. 10.1016/j.trsl.2016.01.007
Zimmer, L. (2009). Positron emission tomography neuroimaging for a better un-
derstanding of the biology of ADHD. Neuropharmacology, 57(7-8), 601–607.
10.1016/j.neuropharm.2009.08.001

WEB SOURCE

https://imotions.com/wp-content/uploads/2022/10/brain-lobes-iMotions.png
https://assets.nhs.uk/nhsuk-cms/images/E5RF5X.width-1534.jpg
https://ars.els-cdn.com/content/image/3-s2.0-B9780128044902000026-f02-01-
9780128044902.jpg
https://ccnmtl.columbia.edu/projects/neuroethics/module1/foundationtext/index.html
https://imotions.com/blog/learning/research-fundamentals/eeg-vs-mri-vs-fmri-
differences/
https://imotions.com/wp-content/uploads/2022/10/fMRI-explained.jpg
https://info.tmsi.com/blog/types-of-eeg-electrodes
https://www.york.ac.uk/psychology/research/york-neuroimaging-centre/research/
magnetoencephalography/
https://radiology.ucsf.edu/patient-care/services/specialty-imaging/alzheimer
https://www.pnas.org/doi/10.1073/pnas.2208729119
Chapter 5

EEG
Concepts, research-based analytics,
and applications
Rashmi Gupta, Sonu Purohit, and Jeetendra Kumar
Atal Bihari Vajpayee University, Bilaspur, Chhattisgarh, India

5.1 INTRODUCTION

EEG is an abbreviation for electroencephalogram. It is a medical test that

measures brain electrical activity. The electrodes are attached to the scalp
and pick up the electrical signals produced by the brain, which are then
transmitted to a machine, which displays the results in the form of a graph.
The EEG is used to diagnose a wide range of neurological conditions,
including epilepsy, sleep disorders, and brain damage. EEG devices record
electrical signals generated from the brain. “Electroencephalography (EEG)
has been instrumental in making discoveries about cognition, brain func-
tion, and dysfunction” [1]. It is already known that the human brain is
always generating electrical signals, even when the person is sleeping.
During different types of activities, the brain generates many waves. The
signals are also known as neuro signals. In EEG, metal discs (electrodes)
pick up the electrical activity that is generated by the neurons inside the
brain cells. In the recording process of EEG, tiny metal discs (electrodes)
are connected to the scalp to capture the electrical activity of the brain. The
sensors carry out the signals received by a machine and display them on
the computer monitor. The signals are in the form of waves. The recording
procedure is done by a specialist, who is called a neurologist, in the hospital
for a diagnostic test or any other medical purpose research center for
research purposes. For the computerized processing of these brain signals
generated from EEG devices, at first, these signals are amplified and then
transferred to the computer. After preprocessing, machine learning and
deep learning based algorithms are used for data analysis and making
predictions. This process is shown in Figure 5.1.
Designing technology of EEG: EEG technology involves the use of
electrodes attached to the scalp to pick up electrical signals produced
by the brain, amplifiers to increase the strength of the signals, filters to
remove any unwanted noise, and an analog-to-digital converter to convert
the signals into a digital format that can be analyzed and displayed. The
EEG machine is typically made up of a computer and specialized software

DOI: 10.1201/9781003398066-5 83
84 Computational Techniques in Neuroscience

Figure 5.1 EEG signal processing.

that processes the data and displays the results in a visual format, such as
a graph or image. Over the years, advances in materials, electronics, and
software have resulted in smaller, more portable, and more efficient
machines with higher resolution and accuracy. Furthermore, advances
in brain-computer interface technology have enabled the development of
EEG-based systems that can be used for a variety of applications other
than traditional medical diagnosis, such as controlling prosthetic devices,
monitoring brain activity for cognitive assessment and research, and
more. EEG is a subpart of the brain-computer interface(BCI). BCI devices
can be of the following types:

A. Invasive: In these types of devices, microelectrodes are directly placed

into the cortex. These types of EEG devices are generally inserted
into the internal area of the brain using surgical procedures. These
invasive BCI devices effectively read brain signals, because they are
placed inside the brain.
B. Semi-invasive: In these types of devices, electrodes are placed on the
upper surface of the brain using a surgical procedure. These types of
devices are known as ECoG (electrocorticography).
C. Non-invasive: In these types of devices, sensors are placed on the
scalp of the head. This type of device does not very effectively read
the signals, but it is used mostly because it does not require a surgical
procedure to read brain waves as well as there is no chance of
 EEG 85

physical harm to the brain as compared with invasive and semi-

invasive devices. These types of devices are known as EEG or
MEG devices.

EEG devices can also be wired or wireless. The wired brain-wear devices
offer strong signals and more stability than the wireless ones. There is no
chance of dysconnectivity during the recording of electrical activity. During
the utility of wired connected devices, we face a lack of movement but
it is easy to record without any external interruption. In the wireless
headset, the recording can be affected by connectivity. Due to the mobility
of the device, it is easy to handle and can be used from a distance. The
primary drawback of the wireless device is that sometimes the device can
be disconnected for external reasons. Battery backup of the device can be
one of the reasons for an interruption during recording.

5.2 PREPROCESSING TECHNIQUES OF EEG SIGNALS

Electric signals inside the brain may be affected by different types of reasons
like body movements, eye blinking, opening the mouth, moving the leg as
well as the imagination. Many variations can be seen in recorded signals
when a person moves her/his leg, hand, head, or another part of the body.
This kind of variation can be called noise or artifacts. The removal of noise
from the datasets is called the preprocessing of datasets. Two types of ar-
tifacts can be present in the dataset – the first one is an internal artifact,
and the second one is a system artifact. The internal artifact can arise due
to body movements of the subject like eye blinking, hand, foot, or head
movements, and system artifacts can arise due to fluctuations in the power
supply in the device, movements in the electrodes of the device, etc.
There are a few artifact removal techniques to remove artifacts from the
raw EEG dataset that are called filtering. The filtering includes three
techniques, which are the high pass filter, low pass filter, and notch filter.
The high pass filters out the slow frequencies that are less than 0.1 hz or
1.0 hz. High-pass filtering is used for removing drift and trends. The low
pass filter is used for removing unwanted noise from data and is to make
the dataset clear and improve its quality. The low pass filter removes high-
frequency interferences greater than 50-70hz. EEG signals are often
affected by power line noise, and such noise is sometimes avoided from
a source with a carefully equipped system, but this work cannot always
be successful. Moreover, it cannot work in the already gathered dataset.
Therefore, notch filters are used for the removal of such noises from the
dataset. The notch filter removes 50/60 Hz interference in the dataset.
Apart from all these basic techniques, some new EEG data processing
techniques are also proposed in some research works. Following are some
of these techniques.
86 Computational Techniques in Neuroscience

5.3 MACHINE LEARNING AND DEEP LEARNING BASED

EEG DATA ANALYSIS TECHNIQUES

The EEG data consists of the activity of billions of neurons. Most of

the time, we want to classify or analyze the EEG data according to
some activity. For example, we can classify the EEG data according to
the attention state of students during learning for mental disease classi-
fication, sleep stage classification, controlling the robotic action, etc. For
the classification task, many artificial intelligence (AI) methods are
used. In broad categories, these AI methods can be machine learning
methods and deep learning methods. Machine learning techniques can be
defined as computer-based models and algorithms that can learn from
the datasets. There are so many machine learning techniques for data
analysis.
Nowadays deep learning techniques are widely used in various areas.
In the field of EEG analysis, deep learning techniques are very useful.
In the last few years, some of the deep learning techniques have been
very successful in the analysis of complex datasets such as EEG, audio,
and image signals. Deep learning, a subfield of machine learning tech-
nology, helps to find out the feature extraction of the dataset. It can
produce a higher level of performance for the different analysis tasks.
Many deep learning techniques like RNN (recurrent neural network),
LSTM (long short-term memory), DBN (deep belief network), and CNN
(convolution neural network) are being used for EEG data analysis
and classification, and some of the recent research work is shown in
Table 5.1.

5.4 APPLICATIONS OF EEG

EEG devices are able to grasp the internal state of mind using the elec-
trical signals generated from the mind. EEG is used to measure the
electrical activity of the brain in a non-invasive manner. Because this
activity is related to a variety of functions such as movement, sensation,
thought, and emotion, EEG is a useful tool for understanding brain
function. Furthermore, EEG can be used to diagnose various neurolog-
ical disorders as well as monitor brain activity during surgery, which is
critical for patient safety. EEG is also useful for researching sleep and
assessing the efficacy of brain stimulation techniques. Furthermore,
EEG’s ability to detect changes in brain activity during drug trials is
critical for the development of effective medicines for a variety of dis-
eases. Due to the non-invasive nature of EEG, it can also be used by a
non-medical person. Even people related to computers can also record
brain waves and analyze them. In this chapter, EEG applications in some
areas like neuro-marketing, behavioral science, cognitive neuroscience,
Table 5.1 Deep learning-based EEG data analysis techniques
Research Objective Year Dataset used No. of AI technique used Model used Accuracy
work volunteers
[ 2] To classify 2023 Own dataset 31 Machine learning KNN Highest 89% accuracy
schizophrenia during LR with SVM
resting state DT
RF
SVM
[ 3] To classify emotional 2023 DEAP dataset [ 4] 32 Machine learning SVM Classifier 97.42% accuracy
valence
[ 5] To detect drowsiness 2023 Public dataset by 12 Machine learning NB 100% highest accuracy
Jianliang Min et al [ 6] SVM
KNN
RFA
[ 7] To classify bipolar and 2023 Own collected dataset 57 Machine learning SVM 89.3% accuracy
other depressive KNN
disorders RF
[ 8] To diagnose brain 2019 The TUH EEG 13500 Deep learning HMM (Hidden 90%
disorders Corpus [ 9] Markov Model)
[ 10] To find the efficient 2021 Andrzejak et al. [ 11] 24 Deep learning CNN-E CNN-E model
classification method performed well with
for different same frequency
frequency data sampling and different
frequency sampling
[ 12] To classify eye close 2022 Own dataset 27 Deep learning DLVQ (deep 91% F-Score
and eye-opening learning vector
states using deep quantizer)
EEG

learning
87

(Continued)
 88

Table 5.1 (Continued) Deep learning-based EEG data analysis techniques

Research Objective Year Dataset used No. of AI technique used Model used Accuracy
work volunteers
[ 13] Epileptic patient 2022 Bonn EEG Database 5 Deep learning Attention-based 98.89%
classification Bern-Barcelona 10 wavelet CNN
Database
[ 14] To classify motor 2022 Own collected dataset 57 Deep learning CSP+LDA Average 69.07%
imagery task 2D CNN F- score
[ 15] To classify emotions 2022 DEAP dataset [ 4] 32 Deep learning Bi-LSTM Accuracy 99.45%,
96.67%, and 99.68%
for valence, arousal,
and liking,
respectively
[ 16] To classify raw EEG 2022 PhysioNet EEG motor 109 Deep learning Transformer Highest 83.31%
movement/imagery based approach accuracy
Computational Techniques in Neuroscience

dataset [ 17]
[ 18] To detect 2022 One public dataset 128 Deep learning CNN 78% accuracy
schizophrenia using and self-collected
sound perception dataset
[ 19] To classify normal 2022 UCI dataset 122 Deep learning DWT+ CNN+ 99.32%
and alcoholic people Bi-LSTM
[ 20] To predict tinnitus 2023 Own collected dataset 9 Deep learning TFI score + CNN Accuracy ranging
treatment outcomes from 98%-100%
[ 21] To predict awakening 2023 Own collected dataset 145 Deep learning CNN Positive prediction -
from the coma 0.83 ± 0.03 and
Negative prediction -
0.57 ± 0.04
 EEG 89

Figure 5.2 Applications of EEG.

education, security, etc. have been discussed. Figure 5.2 illustrates the
applications of EEG in many areas.

5.4.1 Cognitive neuroscience

Cognitive neuroscience is the study of the brain’s relationship to behavior.
It combines psychology, neuroscience, and computer science to better
understand how the brain processes information and generates behavior.
Cognitive neuroscientists study brain activity using techniques such as
EEG, MRI, and functional MRI (fMRI) to better understand the underlying
mechanisms of cognitive processes such as attention, memory, language,
decision making, and perception. Cognitive neuroscience is an inter-
disciplinary field that draws knowledge from many disciplines, including
biology, physics, and mathematics. The ultimate goal of cognitive neuro-
science is to shed light on the neural basis of mental processes and to
develop new treatments for neurological and psychiatric conditions. By
understanding the relationship between brain activity and behavior, cog-
nitive neuroscientists can improve our understanding of the human mind
and develop more effective treatments for conditions such as Alzheimer’s
disease, depression, and schizophrenia. The field of cognitive neuroscience
is highly interdisciplinary and draws on knowledge from a wide range of
fields, including biology, physics, mathematics, and computer science. This
interdisciplinary approach allows cognitive neuroscientists to tackle com-
plex questions about the brain and behavior with a variety of tools and
techniques, leading to discoveries and advancements in the field.
90 Computational Techniques in Neuroscience

5.4.2 Behavioral neuroscience

The study of the biological and physiological basis of behavior is known
as behavioral neuroscience. It combines psychology, neuroscience, and
biology to better understand how the nervous system controls behavior
and how genetics, development, and the environment influence behavior.
Behavioral neuroscience is also known as bio-psychology. This is a study
of behavior development in humans as well as animals. In this study, the
psychological events are examined through the neurons by the biological
activity. Animal models, brain imaging (e.g., EEG, MRI, and fMRI),
and genetic and molecular methods are used by behavioral neuroscientists
to study behavior. Behavioral neuroscience seeks to identify the neural
circuits and mechanisms that underpin various behaviors such as learning
and memory, emotion, sensation, movement, and perception. Behavioral
neuroscientists can improve our understanding of the brain and its func-
tions, as well as develop new treatments for neurological and psychiatric
conditions, by understanding the biological basis of behavior. The field
of behavioral neuroscience also investigates the effects of drugs and other
substances on behavior, as well as the neural mechanisms underlying
these effects.

5.4.3 Neuro-marketing
The application of neuroscience techniques to the study of consumer
behavior and decision making is known as neuro-marketing. It seeks to
comprehend how people make purchasing decisions and what factors
influence them. Neuro-marketing measures brain activity and responses to
marketing stimuli using tools such as EEG, fMRI, and eye-tracking. Neuro-
marketing researchers can gain insights into consumer preferences, atti-
tudes, and motivations by studying brain activity. These insights can then
be used to develop more effective marketing strategies. Because it is still a
young field, neuro-marketing’s methods and conclusions can be debatable.
However, by offering a deeper understanding of consumer behavior and the
underlying neural mechanisms that drive it, it has the potential to com-
pletely alter how businesses approach marketing and consumer research.
Neuro-marketing is a highly interdisciplinary field that draws on knowledge
from marketing, psychology, and neuroscience. It is also a field that is
constantly changing as new strategies and tactics are created. For neuro-
marketing, EEG is an easy-to-use device.

Benefits of EEG in Neuro-marketing:

i. Analysis of consumer’s decision-making process: With the help of
EEG signals, the decision-making process that is running in the brain
of the consumer can be analyzed. Moreover, the buying behavior
of the consumer can be found.
 EEG 91

ii. Analysis of marketing behavior: In this method, the analysis process

can be done using EEG signals. Analyzing consumer behavior in
marketing strategies and analysis of brain functioning of liking/dis-
liking any product can be done. This process can be a little investi-
gation to find out the interest of the customer in buying any product
to improve marketing goals.
iii. Setup the marketing goals: The main objective of this process is the
brainstorming of the consumer without any inquiries. Spread a com-
mercial message to the consumer and develop their interest to purchase
the product by hidden extracting patterns with the help of EEG.

5.4.4 Sports and meditation

Meditation sessions can change the way of thinking. The stress can be
decreased by meditation or sports. The changes can be seen in neurons
inside the human brain in the form of several types of frequencies like
delta, beta, theta, and alpha by the EEG technique. The amplification
of alpha frequency, even in the non-meditative closed eyes state, is the
EEG sign of meditation. EEG is used for a variety of reasons in both
sports and meditation. To enhance performance and lower the risk of
injury in sports, EEG can be used to monitor brain activity while an
athlete is engaged in physical activity. An athlete’s cognitive state can
be tracked using EEG, which can also be used to spot signs of fatigue,
stress, and changes in attention that could affect performance. Then,
with this knowledge, training programs can be improved, recovery can
be accelerated, and injuries can be avoided. EEG is used in meditation
to track changes in brain activity as practitioners progress. EEG can be
used to study the effects of meditation, which have been shown to have
a variety of advantages, including lowered stress, enhanced mood, and
increased focus. EEG can also be used for comparing different medita-
tion practices.

5.4.5 Educational purpose

EEG can be applied in academic settings to examine and enhance the
learning process. A student’s brain activity during academic tasks like
reading, problem solving, and memory exercises can be studied using EEG,
which measures the electrical activity of the brain. By monitoring these
brainwave signals in real time, it’s easy to find out how the students are
feeling during the study in their minds. If a teacher gives them a task, then
he can see how they respond along with the feelings that are running inside
their brain. The teachers can monitor their students’ engagement level
during learning by analyzing the brain waves with the EEG process. So,
they can know the state of the student’s brain, how the student received
92 Computational Techniques in Neuroscience

the task, whether the task was easy or too hard to understand, or whether
the mind of the student was going through stress.

5.4.6 Security
User authentication system plays a very important role in security. There
are many types of user authentication systems like pin, password, fin-
gerprint, face recognition systems, etc. A new authentication system is
going to be introduced named EEG authentication system. EEG contains
a large number of psychological activities running in the human brain.
There are many differences between an individual’s brain structure and
cognitive activities. The EEG signals of different people are different. But
the same individual’s brain performs the same activities, which are even
repeatable. The same frequency of the brain can be used as authentication
by EEG signals for security purposes. The EEG-based authentication
system is best in cyber security or cryptography as it is unique always and
not replicable as well. The EEG-based biometric system needs the BCI
application to develop security systems.

5.4.7 Brain control robotics

Electrodes are affixed to the scalp to record the electrical activity of the brain
in brain control technology. The signals from this activity are then processed
and interpreted by a computer, which employs algorithms to convert them
into control instructions for the robot. Numerous potential uses for brain-
controlled robots include assistive technology for the disabled, rehabilitation
technology for those with motor impairments, and research tools for ex-
amining brain-machine interactions. Additionally, brain-controlled robots
have the potential to completely change the field of robotics by allowing
humans to control robots directly with their brains. An EEG-based brain
control robotic is a gadget or a robot that receives a command from a human
operator and performs different activities according to that. The performance
of brain control devices depends on different BCIs, which may have different
speeds and accuracies as well. Table 5.2 shows recent research work done
in many application areas of EEG.

5.5 CHALLENGES ASSOCIATED WITH EEG

EEG is a device that measures the electrical activity of the brain during
any mental task. From the first viewpoint, anyone can say that it is
technology to access the brain state of humans. New research doors are
already open for conducting research in neuroscience using EEG devices,
but as with other technologies, many challenges are also associated with
EEG. These challenges can be categorized as follows.
Table 5.2 Latest research work done on different application areas of EEG
EEG application Research work & Year Dataset used Device used No of. Methodology used Findings
area objective volunteer
Cognitive [ 22] Diagnosis of 2023 Publically Walter EEGPL- 59 SVM Accuracy 97.22%
Neuroscience Alzheimer’s disease available 2311
dataset [ 23]
[ 24] To predict the 2021 Own HydroCel 74 Statistical analysis EEG signals can
decision-making collected Geodesic discriminate decision
process dataset Sensory Net variables.
[ 25] To predict 2020 Own Brain Products 14 COH Highest accuracy
decision making collected GmbH) 0.90 ± 0.10
dataset
Behavior [ 26] To analyze the 2022 Own BioSemi Inc., 81 Statistical analysis Way-finding cognition
Neuroscience effect of colors, collected was significantly
graphics, and design dataset improved with color
and graphic
enhancement.
[ 27] To detect driver 2022 Own - 6 Bi-LSTM Highest accuracy 92.48%
distraction collected
dataset
[ 28] To analyze brain 2021 Own BioSemi B.V. 186 Statistical analysis 72% of sessions were
behavior and collected completed.
synchrony during dataset
parent–child
interactions
EEG

(Continued)
93
 94

Table 5.2 (Continued) Latest research work done on different application areas of EEG
EEG application Research work & Year Dataset used Device used No of. Methodology used Findings
area objective volunteer
Neuro- [ 29] To classify 2022 DEAP Biosemi Active 32 DNN Highest accuracy 94%
marketing preferences dataset [ 4] Two system RF
SVM
KNN
[ 30] To predict 2022 Publicly Emotiv Epoc+ 25 Ensemble Accuracy 96.89%
consumer’s emotion available classification
dataset [ 31] using GA
[ 32] To classify 2022 Own Neurowerk 45 K means clustering Highest accuracy 92.4%
consumer preference collected EEG Sigma
dataset
Sports and [ 33] To access mental 2021 Own Muse EEG 10 VAS ruler Game positively affected
Meditation fatigue and stress collected Device stress and
dataset concentration
Computational Techniques in Neuroscience

[ 34] To access the 2019 Own Mindwave and 14 Calculated average of Shooting score and
mental status of rifle collected neurosky shooting score, meditation were higher
shooter dataset attention, and heart on the expert, and
rate before 5 s of attention and heart rate
shooting were higher on the
novice.
Education [ 35] To analyze split 2022 Own g.Nautilus, g.tec 40 Statistical test The split-attention group
attention effects collected in the beta brain wave
dataset and the focused
attention group in the
alpha brain wave both
showed a significant
difference.
 [ 36] To analyze 2020 Own Mindwave 13 Comparison of Different media of study
attention collected attention values differently affects
dataset attention value.
Security [ 37] To develop EEG- 2022 Own EGI GES 300 15 Auto Weka 95.6%
based personal collected
authentication system dataset
[ 38] To analyze EEG- 2019 Own Truscan EEG 20 KNN Accuracy 98.04%
based authentication collected device LDA
dataset
Brain Control [ 39] To develop brain 2020 Own − 15 MDCBN Success rate 0.60
Robotics control robotic arm collected
dataset
[ 40] To decode hand 2020 Own EEG with 57 15 LDA Average 48% score
movements collected electrodes
dataset
EEG
95
96 Computational Techniques in Neuroscience

5.5.1 Technical challenges

• The placement of electrodes on the scalp and their stability during
recording can significantly impact the quality of the EEG signal. Any
movement or displacement of electrodes can cause artifacts in the
EEG data.
• The EEG signal’s quality can be impacted by the electrical impedance
of electrodes, which can vary depending on several factors like skin
preparation, sweat, and oils.
• Every time, a special gel is required to be applied to electrodes during
data collection. In the case of dry electrodes, data contain more noise
and artifacts than wet electrodes.
• Poor fitting of the EEG cap can occur on different head sizes.
• High synchronization between the amplifier and EEG device is
required.
• The choice of reference electrodes and their placement can signifi-
cantly impact the interpretation of the EEG data.
• The EEG signals can also be contaminated by electrical activity gen-
erated by the muscles, especially in the vicinity of electrodes. This is
known as an electromyographic (EMG) artifact and can make it dif-
ficult to distinguish between brain and muscle activity.

5.5.2 Social and ethical challenges

• Some people are afraid of participating in research conducted
with EEG.
• Wearing an EEG headset for a long time results in a headache.
• EEG data can be highly sensitive and personal, as it provides infor-
mation about an individual’s thoughts and mental processes. This
raises privacy concerns, as EEG data could be used to profile in-
dividuals or access their personal information.
• The use of EEG data in research or clinical settings requires informed
consent from the participants, which can be difficult to obtain in some
cases, such as in children or individuals with cognitive or neurological
disorders.
• EEG data can be easily manipulated or misinterpreted, leading to
incorrect conclusions about an individual’s brain activity or mental
state. This can have serious consequences for individuals, such as false
diagnoses or inappropriate treatment.

5.5.3 Environmental challenges

• Ambient light and noise can affect the quality of the EEG signal, as
both can generate electrical noise that can interfere with the EEG
signal.
 EEG 97

• Changes in temperature and humidity can affect the impedance of the

electrodes and the stability of the EEG signal.
• The electrode gel used to prepare the scalp for EEG recording can
dry over time, leading to increased impedance and decreased signal
quality.
• EEG signals can be contaminated by other electrical devices in the
environment, such as cell phones, computers, or medical equipment,
which can generate electrical noise that interferes with the EEG signal.
• Movement during EEG recording can cause significant artifacts in
the EEG signal, making it difficult to accurately interpret the data.

5.6 CONCLUSION

EEG is a non-invasive method used to measure the electrical activity of

the brain. It involves placing electrodes on the scalp to record brainwaves,
which can be analyzed to understand brain function and diagnose neuro-
logical disorders. EEG research has led to the development of advanced
analytics techniques for analyzing brain signals, such as time-frequency
analysis and machine learning algorithms. Numerous research is going on
to develop efficient algorithms for preprocessing and classification of EEG
signals. Highly efficient deep learning algorithms have been introduced by
researchers to EEG analytics. In this chapter, we have discussed recent
research done in the area of algorithm development for EEG analytics.
We have also discussed recent research done in many application areas of
EEG like cognitive neuroscience, behavioral neuroscience, sport, medita-
tion, education, neuromarketing, etc.

REFERENCES

[1] M. X. Cohen, “Where Does EEG Come From and What Does It Mean?,”
Trends Neurosci., vol. 40, no. 4, pp. 208–218, Apr. 2017.
[2] J. Ruiz de Miras, A. J. Ibáñez-Molina, M. F. Soriano, and S. Iglesias-Parro,
“Schizophrenia Classification Using Machine Learning on Resting State
EEG Signal,” Biomed. Signal Process. Control, vol. 79, p. 104233, Jan.
2023.
[3] L. Abdel-Hamid, “An Efficient Machine Learning-Based Emotional Valence
Recognition Approach Towards Wearable EEG,” Sensors 2023, Vol. 23,
Page 1255, vol. 23, no. 3, p. 1255, Jan. 2023.
[4] S. Koelstra et al., “DEAP: A Database for Emotion Analysis; Using
Physiological Signals,” IEEE Trans. Affect. Comput., vol. 3, no. 1, pp. 18–31,
Jan. 2012.
[5] I. A. Fouad, “A Robust and Efficient EEG-based Drowsiness Detection
System Using Different Machine Learning Algorithms,” Ain Shams Eng. J.,
vol. 14, no. 3, p. 101895, Apr. 2023.
98 Computational Techniques in Neuroscience

[6] J. Min, P. Wang, and J. Hu, “Driver Fatigue Detection through Multiple
Entropy Fusion Analysis in an EEG-based System,” PLoS One, vol. 12,
no. 12, Dec. 2017.
[7] M. Ravan et al., “Discriminating between Bipolar and Major Depressive
Disorder Using a Machine Learning Approach and Resting-state EEG Data,”
Clin. Neurophysiol., vol. 146, pp. 30–39, Feb. 2023.
[8] M. Golmohammadi, A. H. Harati Nejad Torbati, S. Lopez de Diego,
I. Obeid, and J. Picone, “Automatic Analysis of EEGs Using Big Data and
Hybrid Deep Learning Architectures,” Front. Hum. Neurosci., vol. 13, p. 76,
Feb. 2019.
[9] “Temple University EEG Corpus - Downloads.” [Online]. Available: https://
isip.piconepress.com/projects/tuh_eeg/html/downloads.shtml. [Accessed: 10-
Feb-2023].
[10] T. Wen, Y. Du, T. Pan, C. Huang, and Z. Zhang, “A Deep Learning-Based
Classification Method for Different Frequency EEG Data,” Comput. Math.
Methods Med., vol. 2021, 2021.
[11] R. G. Andrzejak, K. Lehnertz, F. Mormann, C. Rieke, P. David, and
C. E. Elger, “Indications of Nonlinear Deterministic and Finite-Dimensional
Structures in Time Series of Brain Electrical Activity: Dependence on
Recording Region and Brain State,” Phys. Rev. E, vol. 64, no. 6, p. 061907,
Nov. 2001.
[12] F. Husham Almukhtar, A. Abbas Ajwad, A. S. Kamil, R. A. Jaleel, R. Adil
Kamil, and S. Jalal Mosa, “Deep Learning Techniques for Pattern Recognition
in EEG Audio Signal-Processing-Based Eye-Closed and Eye-Open Cases,”
Electron. 2022, Vol. 11, Page 4029, vol. 11, no. 23, p. 4029, Dec. 2022.
[13] Q. Xin, S. Hu, S. Liu, L. Zhao, and Y. D. Zhang, “An Attention-Based
Wavelet Convolution Neural Network for Epilepsy EEG Classification,”
IEEE Trans. Neural Syst. Rehabil. Eng., vol. 30, pp. 957–966, 2022.
[14] N. Tibrewal, N. Leeuwis, and M. Alimardani, “Classification of Motor
Imagery EEG Using Deep Learning Increases Performance in Inefficient
BCI Users,” PLoS One, vol. 17, no. 7, p. e0268880, Jul. 2022.
[15] M. Algarni, F. Saeed, T. Al-Hadhrami, F. Ghabban, and M. Al-Sarem, “Deep
Learning-Based Approach for Emotion Recognition Using Electroencephalo-
graphy (EEG) Signals Using Bi-Directional Long Short-Term Memory (Bi-
LSTM),” Sensors 2022, Vol. 22, Page 2976, vol. 22, no. 8, p. 2976, Apr. 2022.
[16] J. Xie et al., “A Transformer-Based Approach Combining Deep Learning
Network and Spatial-Temporal Information for Raw EEG Classification,”
IEEE Trans. Neural Syst. Rehabil. Eng., vol. 30, pp. 2126–2136, 2022.
[17] “EEG Motor Movement/Imagery Dataset v1.0.0.” [Online]. Available:
https://physionet.org/content/eegmmidb/1.0.0/. [Accessed: 10-Feb-2023].
[18] C. Barros, B. Roach, J. M. Ford, A. P. Pinheiro, and C. A. Silva, “From
Sound Perception to Automatic Detection of Schizophrenia: An EEG-Based
Deep Learning Approach,” Front. Psychiatry, vol. 12, p. 2659, Feb. 2022.
[19] E. Tamilia, L. Ricci, H. Li, and L. Wu, “EEG Classification of Normal
and Alcoholic by Deep Learning,” Brain Sci. 2022, Vol. 12, Page 778,
vol. 12, no. 6, p. 778, Jun. 2022.
[20] M. Doborjeh et al., “Prediction of Tinnitus Treatment Outcomes Based
on EEG Sensors and TFI Score Using Deep Learning,” Sensors 2023, Vol. 23,
Page 902, vol. 23, no. 2, p. 902, Jan. 2023.
 EEG 99

[21] F. M. Aellen et al., “Auditory Stimulation and Deep Learning Predict

Awakening From Coma after Cardiac Arrest,” Brain, Jan. 2023.
[22] A. An Approach et al., “An Approach toward Artificial Intelligence
Alzheimer’s Disease Diagnosis Using Brain Signals,” Diagnostics 2023,
Vol. 13, Page 477, vol. 13, no. 3, p. 477, Jan. 2023.
[23] M. Cejnek, O. Vysata, M. Valis, and I. Bukovsky, “Novelty Detection-based
Approach for Alzheimer’s Disease and Mild Cognitive Impairment Diagnosis
from EEG,” Med. Biol. Eng. Comput., vol. 59, no. 11–12, pp. 2287–2296,
Nov. 2021.
[24] Y. Yau, T. Hinault, M. Taylor, P. Cisek, L. K. Fellows, and A. Dagher,
“Evidence and Urgency Related EEG Signals during Dynamic Decision-
Making in Humans,” J. Neurosci., vol. 41, no. 26, pp. 5711–5722, Jun. 2021.
[25] Y. Si et al., “Predicting Individual Decision-Making Responses Based
on Single-Trial EEG,” Neuroimage, vol. 206, p. 116333, Feb. 2020.
[26] S. Kalantari et al., “Evaluating the Impacts of Color, Graphics, and
Architectural Features on Wayfinding in Healthcare Settings Using EEG Data
and Virtual Response Testing,” J. Environ. Psychol., vol. 79, p. 101744,
Feb. 2022.
[27] X. Zuo, C. Zhang, F. Cong, J. Zhao, and T. Hamalainen, “Driver
Distraction Detection Using Bidirectional Long Short-Term Network Based
on Multiscale Entropy of EEG,” IEEE Trans. Intell. Transp. Syst., vol. 23,
no. 10, pp. 19309–19322, Oct. 2022.
[28] E. S. Norton et al., “Social EEG: A Novel Neurodevelopmental Approach to
Studying Brain-Behavior Links and Brain-to-Brain Synchrony During
Naturalistic Toddler–Parent Interactions,” Dev. Psychobiol., vol. 64, no. 3,
p. e22240, Mar. 2022.
[29] M. Aldayel, M. Ykhlef, and A. Al-Nafjan, “Deep Learning for EEG-Based
Preference Classification in Neuromarketing,” Appl. Sci. 2020, Vol. 10,
Page 1525, vol. 10, no. 4, p. 1525, Feb. 2020.
[30] S. M. A. Shah et al., “An Ensemble Model for Consumer Emotion Prediction
Using EEG Signals for Neuromarketing Applications,” Sensors 2022,
Vol. 22, Page 9744, vol. 22, no. 24, p. 9744, Dec. 2022.
[31] M. Yadava, P. Kumar, R. Saini, P. P. Roy, and D. Prosad Dogra, “Analysis
of EEG signals and its application to neuromarketing,” Multimed. Tools
Appl., vol. 76, no. 18, pp. 19087–19111, Sep. 2017.
[32] S. Raiesdana and M. Mousakhani, “An EEG-Based Neuromarketing
Approach for Analyzing the Preference of an Electric Car,” Comput. Intell.
Neurosci., vol. 2022, 2022.
[33] S. Gündoğdu, Ö. H. Çolak, E. A. Doğan, E. Gülbetekin, and Ö. Polat,
“Assessment of Mental Fatigue and Stress on Electronic Sport Players with
Data Fusion,” Med. Biol. Eng. Comput., vol. 59, no. 9, pp. 1691–1707, Sep.
2021.
[34] “The psychophysiological differences between expert and novice rifle shooters
during the aiming period.” [Online]. Available: https://cyberleninka.ru/article/
n/the-psychophysiological-differences-between-expert-and-novice-rifle-
shooters-during-the-aiming-period/viewer. [Accessed: 09-Feb-2023].
[35] D. Mutlu-Bayraktar, P. Ozel, F. Altindis, and B. Yilmaz, “Split-Attention Effects
in Multimedia Learning Environments: Eye-Tracking and EEG Analysis,”
Multimed. Tools Appl., vol. 81, no. 6, pp. 8259–8282, Mar. 2022.
100 Computational Techniques in Neuroscience

[36] D. Ni, S. Wang, and G. Liu, “The EEG-Based Attention Analysis in

Multimedia m-Learning,” Comput. Math. Methods Med., vol. 2020, 2020.
[37] C. Stergiadis, V. D. Kostaridou, S. Veloudis, D. Kazis, and M. A. Klados,
“A Personalized User Authentication System Based on EEG Signals,” Sensors
2022, Vol. 22, Page 6929, vol. 22, no. 18, p. 6929, Sep. 2022.
[38] T. Zhi Chin, A. Saidatul, and Z. Ibrahim, “Exploring EEG Based
Authentication for Imaginary and Non-Imaginary Tasks Using Power
Spectral Density Method,” IOP Conf. Ser. Mater. Sci. Eng., vol. 557, no. 1,
p. 012031, Jun. 2019.
[39] J. H. Jeong, K. H. Shim, D. J. Kim, and S. W. Lee, “Brain-Controlled Robotic
Arm System Based on Multi-Directional CNN-BiLSTM Network Using
EEG Signals,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 28, no. 5,
pp. 1226–1238, May 2020.
[40] A. Schwarz, M. K. Höller, J. Pereira, P. Ofner, P. Ofner, and G. R. Müller-
Putz, “Decoding Hand Movements from Human EEG to Control a Robotic
Arm in a Simulation Environment,” J. Neural Eng., vol. 17, no. 3,
p. 036010, May 2020.
Chapter 6

Classification of gait signals for

detection of neurodegenerative
diseases using log energy entropy
and ANN classifier
Prasanna J, S. Thomas George, and M.S.P Subathra
Dept. of Robotics Engineering, Karunya Institute of Technology and Sciences,
Coimbatore, Tamil Nadu, India

6.1 INTRODUCTION

The prevalence of neurodegenerative disorders (NDDs) has increased

worldwide, and they have a considerable negative impact on economical,
developmental, and health fronts [1]. They may result in severe movement
issues, such as tremors in the limbs, jaw, or face; stiffness; or slowness of
movement [2]. Gait analysis is a crucial method for evaluating NDD, even
though the gait anomaly as a deviation from walking may indicate distinct
problem patterns [3]. Neurodegenerative illnesses may manifest when
neurons are harmed or begin to deteriorate. A number of serious NDDs,
such as Parkinson’s disease (PD), Huntington’s disease (HD), Alzheimer’s
disease (AD), and amyotrophic lateral sclerosis (ALS), are brought on by
changes in these cells that cause abnormal functioning and progressive
damage of the arrangement of the neurons, including death of neurons [4].
Gait fluctuations analysis is a crucial method for evaluating NDD, even
if gait abnormalities as a divergence from walking gait fluctuation may
reflect many problem patterns [5,6]. Stride length and cadence, two fun-
damental temporal-spatial gait metrics, are used as input features in [7] by
Kamruzzaman to assess the gait of people with cerebral palsy. The gait
information explains the various gait patterns that humans exhibit while
walking [6]. For the examination of PD gait, the authors of [8] revealed
various regression normalization techniques that took physical character-
istics and self-selected speed into account. In [9], Wu used a statistical
analysis of gait rhythm with the nonparametric Paren-window approach to
evaluate the probability density functions of intervals of stride, stance, and
swing periods with its sub-phases.
Each gait cycle consists of a series of ordered gait events that take place at
particular gait parameters indicated with temporal locations. The metrics
of stride, posture, and swing are used to analyze human gait patterns to
identify aberrant neurodegeneration classes. As the earliest study based on
gait fluctuations-based illness analysis, Hausdorff et al. [10–12] examined

DOI: 10.1201/9781003398066-6 101

102 Computational Techniques in Neuroscience

the stride interval time series of the gait in participants with HD as well as
the healthy older subjects in comparison to control subjects. Their findings
demonstrated that compared to controls, subjects with HD and older
subjects have greater random stride interval fluctuations. [5,13,14] are
more studies for consideration on gait oscillations analysis intended for
illness state analysis. Driven by their findings, Kamruzzaman applied the
support vector machine method and two fundamental temporal-spatial
gait characteristics (stride length and cadence) as input features to examine
the gait of people with cerebral palsy [15]. A study that uses fluctuation
analysis and frequency range distribution to get a fresh perspective on
gait rhythm is found in [16]. A tensor decomposition model for higher-
dimensional analysis in PD was put forth in [17] by the author [18]
described numerous regression normalization techniques for PD gait
analysis that took into account the patient’s physical characteristics and
self-selected speed. With the statistical investigation of gait rhythm, Wu
employed a nonparametric Paren-window approach in [19] to evaluate the
gait intervals. For the gait dynamics analysis for classifying NDDs, frequency
range distribution [20], tensor decomposition [21], and texture-based images
with fuzzy recurrence plots [22] were proposed. The improvement of fall
prediction, treatment, and rehabilitation procedures may result from research
on the dynamics of gait patterns in neurodegenerative illness to determine the
severity. Thus, we proposed non-linear entropy for extracting optimum
features for the detection of diseases. The proposed research could aid in the
classification of NDDs and the analysis of gait variations.

Contributions:
• An entropy-based feature extraction technique is proposed for effec-
tive feature extraction.
• Statistical-based features such as minimum, maximum, mean, energy,
and normalized energy were used for the classification.
• An artificial neural network (ANN) classifier is utilized to classify
the task, such as HC vs PD, HC vs. HD, HC vs. ALS, and HC
vs. NDD.
• The classification performance shows better results with the proposed
approach. The framework of the proposed method is shown in
Figure 6.1.

Outline of the Paper:

• Section 6.2 provides the framework of the proposed methodology and
dataset information and classification.
• In Section 6.3, we illustrate the performance of the classification of a
different task, and we discuss the achieved results. The comparison
between the existing and proposed method is reviewed based on the
classification performance.
• Finally, Section 6.4 concludes the article with future directions.
 Classification of gait signals for detection of NDDs 103

Figure 6.1 The framework of the proposed method.

6.2 METHOD AND MATERIALS

6.2.1 Dataset used

In this investigation, we use the Physionet [23] database’s Gait Dynamics
in Neuro-Degenerative Disease Dataset [24]. The dataset was suggested in order
to comprehend the pathophysiology of NDDs better. There are 16 healthy
controls, 15 PD patients, 20 HD patients, and 13 people with ALS. We abbre-
viate using HC, ALS, HD, and PD. The dataset includes a detailed description of
the patients’ clinical characteristics, including their age, gender, height, weight,
walking speed, PD and HD illness severity, and ALS length. Examples of the
HC, PD, HD, and ALS groups are shown in Tables 6.1–6.4, respectively.
Force-sensitive resistors, with an output roughly proportionate to the
force under the foot, were used to collect the database’s raw data. The
signals were used to obtain the stride-to-stride measures of footfall contact
periods, which include the left and right stride intervals, left and right swing
intervals, left and right stance intervals, and double support intervals.
Additionally, the percentages of stride for the left/right swing interval, left/
right stance interval, and double support period were extracted. You can
refer to [24,25] for a thorough description of the dataset’s experiment
conditions. Figure 6.2 displays examples for each category in the dataset.

Table 6.1 The classification performance using all five statistical features
Classification performance
Classification task Accuracy (%) Sensitivity (%) Specificity (%) PPV (%) NPV (%)
PD vs. HC 93.54 99.28 88.82 88.12 99.33
HD vs. HC 92.41 89.20 97.62 98.12 85.34
ALS vs. HC 91.10 84.97 97.89 97.50 86.00
NDD vs.HC 92.21 92.28 91.31 92.25 93.20
104 Computational Techniques in Neuroscience

Table 6.2 Classification result (PD vs. HC)

Classification performance (PD vs. HC)
Features Accuracy (%) Sensitivity (%) Specificity (%) PPV (%) NPV (%)
Minimum 90.17 89.18 90.71 91.22 89.22
Maximum 89.29 89.28 87.56 89.90 90
Mean 90.38 87.78 90.78 93.22 90.27
Energy 92.29 89.28 91.55 92.90 93.2
N-Energy 91.25 92.45 91.86 90.76 91.29

Table 6.3 Classification performance (HD vs. HC)

Features Accuracy (%) Sensitivity (%) Specificity (%) PPV (%) NPV (%)
Minimum 91.72 90.68 90.97 91.92 87.22
Maximum 91.29 91.28 92.56 90.90 92.23
Mean 90.78 94.78 90.58 93.82 91.47
Energy 91.29 89 93.30 92.90 92
N-Energy 91.87 92.56 92.63 91.86 91

Table 6.4 Classification performance (ALS vs. HC)

Features Accuracy (%) Sensitivity (%) Specificity (%) PPV (%) NPV (%)
Minimum 92.29 91.28 89.56 90.90 91.34
Maximum 90.38 87.78 90.78 93.22 90.27
Mean 91.72 90.68 90.97 91.92 87.22
Energy 91.25 92.45 91.86 90.76 91.29
N-Energy 90.17 89.18 90.71 91.22 89.22

6.2.2 Feature extraction

To effectively classify the gait patterns of NDDs (e.g., ALS, HD, and PD)
and healthy people employed in this work, the standardized dataset is
processed for the discovery of distinguishing feature sets. Statistical features
have been computed by applying log energy entropy.
Log energy entropy is used to calculate the level of involvement in the
non-stationary signals. Log energy entropy is defined mathematically as

N
Log energy entropy = i =1
log(yi2) (6.1)

where N denotes the total length of the signal yi and denotes the signal’s
ith sample. The log energy entropy feature was used by the authors of [26]
and [27] to achieve high classification accuracy.
 Classification of gait signals for detection of NDDs 105

Parkinson signal
1000
0
-1000
-2000
0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
4
10
Huntington signal
1000
0
-1000
Amplitude

-2000
0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
4
10
ALS signal
0

-1000

-2000
0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
4
10
Healthy signal
1000
0
-1000
-2000
0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8
4
10
No. of samples

Figure 6.2 The visual representation of gait patterns of the different NDD classes.

The five optimum features – minimum, maximum, mean, energy, and

normalized energy features – have been computed on both right and left foot.

6.2.3 Classification
Classification is a significant area of study in data mining, and neural net-
works are one of the most widely used techniques for classification. NDD
classification employing gait features is typically challenged by three
parameters that can affect performance: 1) a small number of clinical
samples; 2) a large number of noisy or redundant features; and 3) the
requirement to meet real-time requirements [28]. To eliminate these prob-
lems ANN classification is used. ANN is a complex adaptive system that
can change its internal structure in response to the information it receives. It
is accomplished by varying the weight of the connection. Each link carries a
certain amount of weight.
A weight is a numerical value that governs the signal between two neu-
rons. The output layer, input layer, and one or more hidden layers (middle
layer) with a larger number of processing nodes or artificial neurons are all
connected in an ANN [29]. The connection between two neurons is con-
trolled by weights that can be adjusted to improve system accuracy, shown
in Figure 6.3. Supervised learning is illustrated by an ANN. The knowledge
was acquired by the ANN in the form of connected network units. Humans
have a difficult time extracting this knowledge. This factor has motivated
the extraction of classification rules in data mining.
ANN learning is achieved through the use of various training algorithms
that are based on training guidelines or functions. Various training algo-
rithms are used to recognize the training algorithm, which is based on
106 Computational Techniques in Neuroscience

Figure 6.3 Functional unit of ANN classifier.

training rules or functions. The neural network is trained to perform its

function by the training algorithm. In this work, the input layer is given five
entropy features per channel, for a total of ten nonlinear features, to dis-
criminate between normal healthy control and NDD gait signals. Physical
modelling is commonly performed using a feed-forward multilayered neural
network. The few hidden neurons used for classification are inefficient. As a
result, the total number of hidden neurons studied in this study is ten. The
neural network is trained using the Levenberg-Marquardt backpropagation
(LMBP) training algorithm, which is a type of quasi-Newton algorithm.

6.2.3.1 Classification performance

The classification performance is computed for individual healthy control,
PD, HD, and ALS subjects based on the classification results, such as
classification accuracy, sensitivity, specificity, positive predictive value, and
negative predictive value, which are given by the following formulas:

True (+) + True ( )

Classification Accuracy = 100 (6.2)
True (+) + True ( ) + False (+) + False ( )

True (+)
Sensitivity = 100 (6.3)
True (+) + False ( )

True ( )
Specificity = 100 (6.4)
True ( ) + False (+)

True (+)
Positive predictive value (PPV) = 100 (6.5)
True (+) + False (+)
 Classification of gait signals for detection of NDDs 107

True ( )
Negative Predictive Value(NPV) = 100 (6.6)
True ( ) + False ( )

Where, True(+) indicates True positive, True(−) refers to True negative,

False(+) corresponds to False positive and False(−) implies False negative.

6.3 RESULTS AND DISCUSSION

The gait signals taken from the pressure sensors from gait dynamics for
the NDDs dataset is utilized to investigate the proposed technique. In the
presented work, the non-linear characteristics of the gait signals are ana-
lyzed by applying log energy entropy computation on gait signals. It cap-
tures effective non-linear properties of the given signal where the optimum
statistical-based features – minimum, maximum, mean, and energy and
normalized energy – were captured from the left and right feet of a healthy
control subject, an ALS patient, an HD patient, and a PD patient, respec-
tively. The extracted features were given to the ANN classifier for further
classification of the healthy control and individual with NDD. The total
subjects taken for the experiment are 15 PD, 20 HD, 13 ALS, and 16
healthy controls. Since each recording has a 5-minute time duration with
sampling of 300 samples for a single second, total number of samples
for each foot is 90,000, respectively. The classification performance of the
proposed work is shown in Table 6.1 and Figure 6.4.
The proposed method achieved the classification accuracy of 93.54% for
the classification between PD vs. HC, 92.41% for HD vs. HC, 91.10% for
ALS vs. HC, and 92.21% for NDD vs. HC. The sensitivity of PD vs. HC
classification achieved 99.28% maximum sensitivity than all other classi-
fication tasks, but minimum specificity than all other classification tasks.
Considering single statistical features for classification also performed

100

95

90

85

80

75
PD vs. HD vs. ALS vs. NDD vs.
HC HC HC HC

Accuracy (%) Sensi!vity (%) Specificity (%)

PPV (%) NPV (%)

Figure 6.4 Classification results of proposed technique.

108 Computational Techniques in Neuroscience

94 Accuracy (%)
92
Sensi!vity (%)
90
Specificity (%)
88
86 PPV (%)

84 NPV (%)

Figure 6.5 Classification performance of PD vs. HC.

better classification results. Table 6.2 and Figure 6.5 show the classification
results of the classification between PD and HC by considering individual
statistical features.
For the classification of PD vs. HC, while considering single features,
energy features achieved the maximum classification performance with an
accuracy of 92.29%. In the classification between HD vs. HC, normalized
energy feature achieved the highest accuracy of 91.87% than all other
features that are shown in Table 6.3 and Figure 6.6.
The classification result of ALS vs. HC is shown in Table 6.4 and
Figure 6.7. Minimum feature outperforms the classification with an accu-
racy of 92.29%, sensitivity of 91.28%, specificity of 89.56%, PPV of
90.90%, and NPV of 91.34%, comparatively.
For the classification between the NDD and HC, all features have similar
classification results with minor variations, shown in Table 6.5 and
Figure 6.8. Hence, the proposed method produces promising results.
In comparison to some traditional variables for comparing multiple groups
of subjects, the indices and HC appear to be far more useful. For example,
if the p value is minimum of all Kruskal-Wallis tests for these subjects with
disease, that are related to the statistical-like average value, standard devia-
tion, fractal scaling index, decay time of autocorrelation, and nonstationarity

96 Accuracy (%)
94
92 Sensi!vity (%)
90
88 Specificity (%)
86
84 PPV (%)
82
NPV (%)

Figure 6.6 HD vs. HC results.

 Classification of gait signals for detection of NDDs 109

94
Accuracy (%)
92
90 Sensivity (%)

88 Specificity (%)
86
PPV (%)
84
NPV (%)

Figure 6.7 ALS vs. HC results.

Table 6.5 Classification performance (NDD vs. HC)

Features Accuracy (%) Sensitivity (%) Specificity (%) PPV (%) NPV (%)
Minimum 91.92 90.56 90.27 91.62 89.22
Maximum 91.25 92.45 91.86 91.96 92.29
Mean 91.29 91.38 93.70 92.50 92.08
Energy 91.87 92.56 92.63 91.86 91.18
N-Energy 91.13 92.78 91.71 91.57 92.44

96
Accuracy (%)
94
92 Sensi!vity (%)
90
88 Specificity (%)
86
PPV (%)

NPV (%)

Figure 6.8 NDD vs. HC results.

index [20,24,30,31]. These findings clearly demonstrate the dependability of

the proposed technique which may be significant for automatic diagnosing
and continuous monitoring of all disease progression, that can be involved in
the evaluation of potential therapy and treatment regimes. The proposed
method may also assist physicians in making reliable decisions, that would
eventually use individuals, families, society, and healthcare.

6.4 CONCLUSION

A new automated framework based on gait fluctuation measures was

proposed in this study as a step forward into clinical and cost-effective
110 Computational Techniques in Neuroscience

detection of ALS, PD, and HD. To investigate a better metric for mon-
itoring the progression of NDD and the effects of intervention therapies, the
proposed entropy-based feature extraction using log energy entropy with
statistical feature measures has produced a better classification of the entire
task with promising results with consideration of single features and mul-
tiple features. This strongly demonstrates the dependability of classification
results, which may be beneficial for automated intensive care of disease
development as well as for early diagnosing. Furthermore, the proposed
method may decrease clinical diagnosis imprecision. In the future, this
research work will concentrate on improving the performance by incor-
porating deep learning techniques for classification.

REFERENCES

[1] Alaskar, Haya, Abir Jaafar Hussain, Wasiq Khan, Hissam Tawfik,
Pip Trevorrow, Panos Liatsis, and Zohra Sbaï. “A data science approach
for reliable classification of neuro-degenerative diseases using gait patterns.”
Journal of Reliable Intelligent Environments 6 (2020): 233–247.
[2] Jankovic, Joseph. “Parkinson’s disease: clinical features and diagnosis.”
Journal of Neurology, Neurosurgery & Psychiatry 79, no. 4 (2008):
368–376.
[3] Hausdorff, Jeffrey M., Yosef Ashkenazy, Chang-K. Peng, Plamen Ch Ivanov,
H. Eugene Stanley, and Ary L. Goldberger. “When human walking becomes
random walking: fractal analysis and modeling of gait rhythm fluctuations.”
Physica A: Statistical Mechanics and its Applications 302, no. 1–4 (2001):
138–147.
[4] Lahmiri, Salim. “Image characterization by fractal descriptors in variational
mode decomposition domain: application to brain magnetic resonance.”
Physica A: Statistical Mechanics and its Applications 456 (2016): 235–243.
[5] Shahbakhi, Mohammad, Danial Taheri Far, and Ehsan Tahami. “Speech
analysis for diagnosis of Parkinson’s disease using genetic algorithm and
support vector machine.” Journal of Biomedical Science and Engineering
2014 (2014).
[6] Salvatore, Christian, Antonio Cerasa, Isabella Castiglioni, F. Gallivanone,
A. Augimeri, M. Lopez, G. Arabia, M. Morelli, M.C. Gilardi, and
A. Quattrone. “Machine learning on brain MRI data for differential diag-
nosis of Parkinson’s disease and Progressive Supranuclear Palsy.” Journal
of Neuroscience Methods 222 (2014): 230–237.
[7] Kamruzzaman, Joarder, and Rezaul K. Begg. “Support vector machines
and other pattern recognition approaches to the diagnosis of cerebral palsy
gait.” IEEE Transactions on Biomedical Engineering 53, no. 12 (2006):
2479–2490.
[8] Wahid, Ferdous, Rezaul K. Begg, Chris J. Hass, Saman Halgamuge, and
David C. Ackland. “Classification of Parkinson’s disease gait using spatial-
temporal gait features.” IEEE Journal of Biomedical and Health Informatics
19, no. 6 (2015): 1794–1802.
 Classification of gait signals for detection of NDDs 111

[9] Wu, Yunfeng, and Sridhar Krishnan. “Statistical analysis of gait rhythm
in patients with Parkinson’s disease.” IEEE Transactions on Neural Systems
and Rehabilitation Engineering 18, no. 2 (2009): 150–158.
[10] Hass, Chris J., Thomas A. Buckley, Chris Pitsikoulis, and Ernest J. Barthelemy.
“Progressive resistance training improves gait initiation in individuals
with Parkinson’s disease.” Gait & Posture 35, no. 4 (2012): 669–673.
[11] McNeely, Marie E., and Gammon M. Earhart. “Medication and subthalamic
nucleus deep brain stimulation similarly improve balance and complex gait
in Parkinson disease.” Parkinsonism & Related Disorders 19, no. 1 (2013):
86–91.
[12] Picelli, Alessandro, Camilla Melotti, Francesca Origano, Andreas Waldner,
Antonio Fiaschi, Valter Santilli, and Nicola Smania. “Robot-assisted gait
training in patients with Parkinson disease: a randomized controlled trial.”
Neurorehabilitation and Neural Repair 26, no. 4 (2012): 353–361.
[13] Eskofier, Bjoern M., Sunghoon Ivan Lee, Manuela Baron, André Simon,
Christine F. Martindale, Heiko Gaßner, and Jochen Klucken. “An overview
of smart shoes in the internet of health things: gait and mobility assessment
in health promotion and disease monitoring.” Applied Sciences 7, no. 10
(2017): 986.
[14] Genetics Home, Huntington disease. National Library of Medicine [Online].
https://ghr.nlm.nih.gov/condition/huntingtondisease(2018). Accessed 03 Jan
2018
[15] Kremer, H.P.H., and Hungtington Study Group. “Unified Huntington’s
disease rating scale: Reliability and consistency.” Movement disorders 11
(1996): 136–142.
[16] Ren, Peng, Shanjiang Tang, Fang Fang, Lizhu Luo, Lei Xu, Maria L. Bringas-
Vega, Dezhong Yao, Keith M. Kendrick, and Pedro A. Valdes-Sosa. “Gait
rhythm fluctuation analysis for neurodegenerative diseases by empirical
mode decomposition.” IEEE Transactions on Biomedical Engineering 64,
no. 1 (2016): 52–60.
[17] Pham, Tuan D., and Hong Yan. “Tensor decomposition of gait dynamics
in Parkinson’s disease.” IEEE Transactions on Biomedical Engineering 65,
no. 8 (2017): 1820–1827.
[18] Long, Jeffery D., Jane S. Paulsen, Karen Marder, Ying Zhang, Ji‐In Kim,
James A. Mills, and Researchers of the PREDICT‐HD Huntington’s Study
Group. “Tracking motor impairments in the progression of Huntington’s
disease.” Movement Disorders 29, no. 3 (2014): 311–319.
[19] Mannini, Andrea, Diana Trojaniello, Andrea Cereatti, and Angelo M.
Sabatini. “A machine learning framework for gait classification using inertial
sensors: Application to elderly, post-stroke and huntington’s disease pa-
tients.” Sensors 16, no. 1 (2016): 134.
[20] Hausdorff, Jeffrey M., Merit E. Cudkowicz, Renée Firtion, Jeanne Y. Wei,
and Ary L. Goldberger. “Gait variability and basal ganglia disorders:
Stride‐to‐stride variations of gait cycle timing in Parkinson’s disease and
Huntington’s disease.” Movement disorders 13, no. 3 (1998): 428–437.
[21] Barnéoud, Pascal, and Olivier Curet. “Beneficial effects of lysine acetylsa-
licylate, a soluble salt of aspirin, on motor performance in a transgenic
model of amyotrophic lateral sclerosis.” Experimental Neurology 155, no. 2
(1999): 243–251.
112 Computational Techniques in Neuroscience

[22] Cho, Chien-Wen, Wen-Hung Chao, Sheng-Huang Lin, and You-Yin Chen.
“A vision-based analysis system for gait recognition in patients with
Parkinson’s disease.” Expert Systems with Applications 36, no. 3 (2009):
7033–7039.
[23] PhysioBank, PhysioToolkit. “Physionet: Components of a new research
resource for complex physiologic signals.” Circulation 101, no. 23 (2000):
e215–e220.
[24] Hausdorff, Jeffrey M., Apinya Lertratanakul, Merit E. Cudkowicz, Amie L.
Peterson, David Kaliton, and Ary L. Goldberger. “Dynamic markers of
altered gait rhythm in amyotrophic lateral sclerosis.” Journal of Applied
Physiology (2000).
[25] Hausdorff, Jeffrey M., Susan L. Mitchell, Renee Firtion, Chung-Kang Peng,
Merit E. Cudkowicz, Jeanne Y. Wei, and Ary L. Goldberger. “Altered fractal
dynamics of gait: Reduced stride-interval correlations with aging and
Huntington’s disease.” Journal of Ppplied Physiology 82, no. 1 (1997):
262–269.
[26] Saka, Kübra, Önder Aydemir, and Mehmet Öztürk. “Classification of EEG
signals recorded during right/left hand movement imagery using Fast Walsh
Hadamard Transform based features.” In 2016 39th International Conference
on Telecommunications and Signal Processing (TSP), pp. 413–416. IEEE,
2016.
[27] Gupta, Vipin, Tanvi Priya, Abhishek Kumar Yadav, Ram Bilas Pachori, and
U. Rajendra Acharya. “Automated detection of focal EEG signals using fea-
tures extracted from flexible analytic wavelet transform.” Pattern Recognition
Letters 94 (2017): 180–188.
[28] Sharma, Bhavna, and K.J.I.J.C.E. Venugopalan. “Comparison of neural
network training functions for hematoma classification in brain CT images.”
IOSR Journal of Computer Engineering 16, no. 1 (2014): 31–35.
[29] Subasi, Abdulhamit, M. Kemal Kiymik, Ahmet Alkan, and Etem Koklukaya.
“Neural network classification of EEG signals by using AR with MLE pre-
processing for epileptic seizure detection.” Mathematical and Computational
Applications 10, no. 1 (2005): 57–70.
[30] Goetz, Christopher G., Werner Poewe, Olivier Rascol, Cristina Sampaio,
Glenn T. Stebbins, Carl Counsell, Nir Giladi et al. “Movement Disorder
Society Task Force report on the Hoehn and Yahr staging scale: Status
and recommendations the Movement Disorder Society Task Force on
rating scales for Parkinson’s disease.” Movement Disorders 19, no. 9 (2004):
1020–1028.
[31] Hausdorff, Jeffrey M. “Gait dynamics in Parkinson’s disease: Common
and distinct behavior among stride length, gait variability, and fractal-like
scaling.” Chaos: An Interdisciplinary Journal of Nonlinear Science 19, no. 2
(2009): 026113.
Chapter 7

An optimized text summarization

for healthcare analytics using
swarm intelligence
Rekha Jain1, Pratistha Mathur2, and Manisha3
1
Department of Computer Applications, Manipal University Jaipur, India
2
Department of Information Technology, Manipal University Jaipur, India
3
Dept of Computer Science, Manipal University Jaipur, India

7.1 INTRODUCTION

Healthcare providers and persons involved in healthcare need accurate and

updated information to make better decisions. As in healthcare, proper
decision making can be life changing for patients and their relatives. Data
analytics in healthcare, i.e., healthcare analytics, can play a vital role in
this. Making decisions about surgery or therapy, predicting the course of
major health events, and making long-term plans all depend on the decision
makers’ ability to promptly collect and analyze complete, reliable data.
With digitalization, healthcare-related data and medical research docu-
ments are available in bulk. Therefore, users of the Internet do not wish
to waste their valuable time on any material, whether it be a document,
research paper, or book, without first understanding the significance of such
material to their needs. Before spending their time reading the source pa-
pers, users like to have a summary or overview of such publications.
Automatic text summarization (ATS) can play a vital role in this. Efficient
summary generation has become challenging in this scenario. Swarm
intelligence optimization methods can be used to generate an efficient
summary. The following section discusses in detail text summarization,
approaches of text summarization, the role of text summarization in
healthcare, and the role of swarm intelligence in optimizing the summary(s).
Section 7.2 focuses on a literature review of work done in healthcare
analytics. Sections 7.3 and 7.4 describe the TF-IDF algorithm and PSO
algorithm, respectively, that are used in the generation of an optimized
summary. The proposed methodology is explained in section 7.5. Results
are discussed in section 7.6. Finally, section 7.7 concludes the chapter with
future work.

7.1.1 Text summarization

ATS techniques are becoming important as the amount of online text
materials is increasing day by day. “A summary is a text produced
DOI: 10.1201/9781003398066-7 113
114 Computational Techniques in Neuroscience

from one or more texts that contain a significant portion of the infor-
mation in the original text(s) and is no longer than half of the original
text(s)”. Summaries can be categorized as indicative, informative, ex-
tractive, and abstractive. Indicative summaries give an idea of content,
while informative summaries give a brief of the content. Content can
be shortened by creating extracts or abstracts. Extracts are generated by
reusing a portion of the text, i.e., sentences or words of input text,
while abstracts are regenerated by using new phrases. Extractive text
summarization and abstractive text summarization are the two primary
methods of ATS.

7.1.2 Text summarization approaches

ATS systems look for the most important content contained in the original
document(s). One can use the following two approaches.

7.1.2.1 Extractive text summarization

Extractive summarization techniques use scoring methods to assign an
importance score or rank to each fragment of the text and return the highest
scoring fragments to generate the summary.

7.1.2.2 Abstractive text summarization

Abstractive summarization compression or reformation methods are applied
over extracted text. These summaries look like original texts and are
more cohesive than extractive summaries.
Additionally, classification of summaries can be done as indicative or
informative depending on the information they provide. The characteristics
of indicative summaries are as follows:

• Without containing the substance of the original document, it pro-

vides hints regarding its contents.
• Its major goal is to indicate the content of the original document
without divulging the specifics of the original text.
• It only provides a bare suggestion about the domain or primary
subject of the input document.

However, an informative summary contains the following characteristics:

• It gives the document’s key information.

• It presents the key information or concepts present in the original
work, which means it must include all pertinent material from the
original document and exclude additional details.
 An optimized text summarization for healthcare analytics 115

ATS may have a three-stage procedure. In the first stage, topic selection or
identification, focus is on what portion of text to include in the summary.
This step generally assigns scores to different portions of the text, which
helps in the selection of the text. The second stage is topic interpretation,
which performs fusion or compression and helps in briefing the content.
The last or third stage is a summary generation, which produces the final
summary in the desired form, and it uses mainly the text generation method
to reformulate the text. Most of the existing systems use the first stage only
and produce a summary using pure extracts.
Topic identification is based on assigning a score to each unit (e.g., word,
clause, or sentence) of the input content and then producing the top scoring
n-units as per required length of the summary. In literature, various
methods are used for scoring the fragments of input text. Text summari-
zation systems use an independent scoring method to score each unit
of text.
To compute the score various criterion can be used and experimented in
literature. The first criteria is positional criteria, which says that certain
locations like headings, titles, first paragraphs, etc., contain important
information, so these phrases are given a higher score. Second criterion is
based on cue phrases in which sentences containing these phrases are given
a higher score. A popular method of text identification or scoring is based
on word and phrase frequently, which says that if a text has higher fre-
quency of some words, then sentences containing those words are of higher
importance and are given higher scores. If query-based summarization is
needed, in that case, query and title overlap methods score the sentences
high that contain desired words. The other criterion can be connectedness
or discourse structure, which are finding the connectedness in the sentences.
Summaries generated by different approaches can be evaluated using
different evaluation methods. Lin and Hovey [2003] introduced ROUGE to
evaluate text summaries. ROUGE compares a system summary to a human
summary using different methods like unigram, bigram, skip bigram, etc.
ROUGE measures recall, i.e., number of system units included in the
gold standard.

7.1.3 Text summarization in healthcare

The importance of ATS has been recognized in different verticals and
domains. Healthcare and medicine is one among them. It helps medical
practitioners and researchers to get more and precise information in a short
span of time. The important sources of health information are electronic
health records, clinical reports, and medical research publications [Rohil
2022]. These documents help healthcare professionals to learn about pa-
tients and their health conditions, and research articles help them to learn
about new findings, therapies, and other advancements in healthcare.
Abstractive summarization is generally used in literature to get concise
116 Computational Techniques in Neuroscience

information, whereas to get an in-depth overview, extractive summariza-

tion approaches are popularly implemented.
[Sarkar K 2009] worked in text summarization in the medical domain.
In this work, they used domain knowledge to select important information
from the input text of the medical domain. [Pivovarov R 2015] worked
on the summarization of electronic health and clinical records. [Gayathri
2015] proposed a medical document summarization method using the
extractive text summarization process, and topic selection was done using
the cue phrase method. Medical Subject Heading (MeSH) was used to find
cue phrases. [Milad et al.] and [Paul et al.] proposed deep learning methods
also for text summarization in healthcare.

7.2 LITERATURE REVIEW

There are different types of medical records that can be useful for medical
professionals for decision making such as electronic health records, clinical
reports, medical research publications, etc. The summaries may help in
better and quick understanding of the medical text data. In recent years,
many researchers have done work related to summarization of health data
(Table 7.1).

Table 7.1 Current research in text summarization for the health domain
S. No. Author and Technique used Dataset
publication
1. Kanwal N and Used multi-head attention- MIMIC-III discharge notes
Rizzo G 2022 based mechanism for
performing extractive
summarization
2. Jesus M et al. 2018 Multi Objective Artificial Document understanding
Bee Colony (MOABC) conference’s dataset
3. Gayathri P and On the basis of cue word For the identification of
Jaisankar N 2015 occurrences, important domain-specific terms, MeSH
sentences were extracted vocabulary thesaurus was
from documents to get used.
summarized text
4. Moradi M and For biomedical text, Retrieved articles from
Mattias S 2019 summarization BioMed Central to construct
embeddings learned by development and evaluation
BERT model were used corpora
5. Deepika S et al. Pretrained models BERT, The COVID-19 Open
2021 GPT-2 and Text Rank Research Dataset: CORD-19
6. Kieuvongngam V Pretrained models, BERT CORD-19
et al. 2020 and OpenAI GPT-2
(Continued)
 An optimized text summarization for healthcare analytics 117

Table 7.1 (Continued) Current research in text summarization for the health domain
S. No. Author and Technique used Dataset
publication
7. Gigioli P et al. 2018 Novel reinforcement Biomedical literature from
learning reward metrics MEDLINE from NLM’s
used for learning which PubMed citation database
is based on biomedical
expert tools like UMLS
Metathesaurus and MeSH
8. Chen YP et al. Bidirectional Encoder A dataset of discharge
2020 Representations using diagnoses from NTUH-iMD
Transformers (BERT)
based structure with a
two-stage training method
9. Moradi M 2018 Clustering and Itemset Self created text corpora
mining-based Biomedical
Summarizer
10. Reddy SM and A multi-objective Data from Mendeley datasets
Mirivala S 2020 optimization approach for summarization of clinical
with the use of similarity trial descriptions
and position of the
sentences. The value of
similarity is calculated
using TF-IDF and WMD.

[Kanwal N and Rizzo G 2022] presented an architecture that gave better

results than frequency, graph-oriented, and centroid-based methods on
MIMIC-III dataset. They suggested that if a concoction of abstractive and
extractive approaches is used, it may give more reliable results.
[Jesus M et al. 2018] proposed an architecture using a multi-objective
artificial bee colony method for multi-document summarization. They
claimed it was the first-time implementation of the MOABC algorithm for
text summarization. Results obtained showed improved ROUGE-2 score
with other parallel methods.
[Gayathri P and Jaisankar N 2015] performed research and presented an
approach for single medical document summarization. For this, domain-
specific terms as cue words were used by model to fetch the important
sentences from the dataset of documents.
[Moradi M and Mattias S 2019] used embeddings learned by BERT
model for the summarization of biomedical text. For this, they randomly
retrieved articles from BioMed Central. For the future, researchers sug-
gested the use of contextual representation to address some problems,
such as biomedical-named entity recognition, information extraction, etc.
[Deepika S et al. 2021] focused on cleaning the text as much as possible.
After that, pretrained models GPT-2, Text Rank, and BERT were applied
118 Computational Techniques in Neuroscience

on the CORD-19 dataset. Out of these models, GPT-2 showed the best
results for summarization.
[Kieuvongngam V et al. 2020] worked on the CORD-19 dataset and
showed that the text-to-text multi-loss approach for training can be used
for the fine tuning of a pre-trained model like GPT-2 to perform abstractive
summarization. Results obtained were reasonable and interpretable.
[Gigioli P et al. 2018] proposed a deep reinforced summarization
model that is capable of generating domain-based summaries of biomedical
documents. They used reward metrics to boost the quality of generated
summaries using a NLM PubMed citation dataset.
[Chen YP et al. 2020] presented a model named AlphaBERT model,
which uses bidirectional encoder representation from a transformers-based
structure with two-stage training. This approach decreases the size of the
model without affecting the performance for summarization. This model
helps staff of the hospital by giving quick summarized information.
[Moradi M. 2018] proposed clustering and itemset mining to develop a
biomedical summarizer using self-created corpora. This model extracts
concepts from the inputted data and uses itemset mining to extract main
topics, and a clustering algorithm is used to place sentences in a cluster,
which shows similar topics. After that, it generates summaries.
[Reddy SM and Miriyala S 2020] presented a multi-feature based opti-
mization model for summarization, in which various features such as
cosine similarity, position, and word-mover distance were used. Clinical
trial description dataset was taken from Mendeley.

7.3 TF-IDF ALGORITHM

It is a statistical measure that finds how relevant a word is to a text doc-

ument [Kim and Gil 2019]. The algorithm weighs keywords in the input
text and assigns importance to each of them on the basis of the number of
times they are present in the document. This is calculated with the help
of two values. The first is term frequency, which means how many times
a word is present in a document. This can be computed by counting the
instances of the word in a document. The second is inverse document fre-
quency of the keyword across a set of documents. It finds how rare or how
common a word is in a set of documents [Manning et al. 2008]. If the value
is closer to 0, the word is very common. The TF-IDF score is calculated by
the formula given in Equation 7.1; here t is the keyword or term, d is the
document, D is the complete document set, and N is the total number of
documents present in the document set.

tf _idf (term : t , document : d ,) = TF (t , d) IDF (t) (7.1)

 An optimized text summarization for healthcare analytics 119

Where:

Occurances of t in d
TF (t , d) = (7.2)
Number of terms in d

number of documents
IDF (t) = loge (7.3)
Number of documents with term t in it

This algorithm is used very commonly in machine learning, information

retrieval, and text summarization/keyword extraction.

7.4 SWARM INTELLIGENCE USING PARTICLE

SWARM OPTIMIZATION

Particle Swarm Optimization (PSO) is a simple algorithm for solving

optimization problems. This algorithm is a simulation of social behaviors.
[Reynolds] and [Heppner and Grenader] initially presented the simulation
of bird flocking. PSO has roots in theory of bird flocking, fish schooling,
and swarm theory. A sociobiologist E. O. Wilson wrote in reference to fish
schooling: “In theory at least, individual members of the school can profit
from discoveries and previous experiences of all other members of the
school during the search of food.” Kennedy and Eberhart proposed a
swarm-based stochastic algorithm “Particle Swarm Optimization”.
In PSO, each potential solution is assumed as one particle with some
fixed velocity flying through the space of the problem. It assumes that
knowledge is shared socially, not only between generations but also among
members of the same generation. PSO optimizes a problem iteratively,
starting with a set of solution candidates (population). These solution
candidates are swarms of particles. Each particle knows the global best
position within the swarm and its own best position found so far during the
solution space search process. At every iteration, the velocity and the
position of each particle in the swarm are influenced by individual and
collective knowledge, which decides the next value in search of the optimal
value. It tries to move towards a promising area to get the global optimal.
The process is repeated until some stopping criterion is satisfied. The pro-
cess can be summarized as follows:
In PSO, based on its prior flying experiences and those of its group-
mates, each particle dynamically modifies the speed at which it travels.
The best outcome for each particle is what is known as their personal
best, or pbest. The global best, or gbest, is the best value for any particle,
and each particle alters its position in accordance with its current posi-
tion, current velocity, and the distance between its current position and
p-best, its current location’s separation from g-best.
120 Computational Techniques in Neuroscience

7.4.1 Particle swarm optimization algorithm

Suppose F: objective function, A: population of agents, Vi: velocity of
particle or agent, C1: cognitive constant, U1, U2: random numbers, C2:
social constant, W: inertial weight Xi: position of particle or agent, p-best:
Personal Best, g-best: Global Best
The PSO algorithm is as follows:

1. Generate the initial population of particles (agents) evenly distributed

over X.
2. Evaluate the position of each particle taking the objective function
Z = F(x, y)
3. Update p-best, i.e., If a particle’s current position is better than its
previous best position, update it.
4. Update g-best, i.e., Find the best particle (according to the last best
places of particles).
5. Update the velocity of each particle using the formula given below.

(
Vit +1 = W . Vit + c1 U1t Pbt1 )
Pit + c2 U2t (gbt Pit) (7.4)

6. Update each particle position using the following formula:

Pit +1 = Pit + vit +1 (7.5)

7. Check the stopping criterion; if false, go to step 2.

The most impressive aspect of PSO is its stable topology, which enables
particles to communicate with one another and pick up new information
more quickly in order to reach the global optimum. Because it optimizes
a problem by constantly attempting to improve a potential solution, the
metaheuristic aspect of this optimization algorithm gives us a lot of alter-
natives. Its application will grow as Ensemble Learning is studied more and
more. Candidate summaries are viewed as particles in text summarizing. By
applying the PSO, one concludes the summary(s), which can be considered
as best as per the given criterion.

7.5 PROPOSED METHODOLOGY

7.5.1 Input text

A medical report of neurology is treated as input text, and it is considered
for health analytics purposes. The dataset is taken from mtsamples.com,
where various medical transcripts are available. The proposed methodology
is shown below as a figure 7.1.
 An optimized text summarization for healthcare analytics 121

Figure 7.1 Proposed methodology.

7.5.2 Preprocessing
It is one of the most important steps because correctly preprocessed data
give good results. If data are preprocessed in an incorrect manner, they will
yield incorrect results. Finally, in preprocessing, we need to covert textual
data into numeric form because machines cannot understand textual forms
of data. They can only work with numeric data. Following are the steps
involved in preprocessing of data.

7.5.2.1 Sentence tokenization

In this step, the entire text is broken down into number of sentences. Text
is separated into sentences based on punctuation marks and space (.).
Then each sentence is considered for further processing.
122 Computational Techniques in Neuroscience

7.5.2.2 Stop word removal

Some words are there that do not carry meaning but are required to form
a proper sentence. Examples of these words are “is,” “am,” “are,” “has,”
“have,” etc. These words are of no use for machines. They only reduce
the importance of analysis by increasing the weight of text. Nltk toolkit
is used to remove the stop words.

7.5.2.3 Stemming
In this step, all the words are reduced to their root forms. Porter’s stemmer
algorithm is used to find the root word.

7.5.3 Applying TF-IDF algorithm

Now the text is ready to feed to the TF-IDF algorithm. This algorithm is based
on term frequency and inverse document frequency. It counts the frequency
of words in each sentence and calculates the term frequency matrix. Then,
it calculates the IDF matrix and finally provides the score to sentences. Based
on some threshold value, only candidate sentences that are part of the sum-
mary are filtered. This algorithm consists of the following steps:

a. Creation of a frequency matrix

b. Calculate term frequency
c. Creation of an IDF matrix
d. Calculate & generate a TF-IDF matrix
e. Scoring of sentences
f. Setting threshold values
g. Summary generation

7.5.4 Generation of different versions of summary

By iteratively running the algorithm and setting different threshold values,
various versions of the summary are produced. Next, all possible versions
are passed to the PSO algorithm.

7.5.5 Applying PSO algorithm

The PSO algorithm takes the initial score of generated summaries (particles)
and their bounds. With the help of the fitness function, inertia (w), cognitive
constant (c1), and social constant (c2), position and velocity of each particle
are updated iterative until convergence is reached. The steps are as follows:

7.5.5.1 Find all possible sets of summaries

All the possible subsets of summaries generated by the TF-IDF algorithm
are treated as particles.
 An optimized text summarization for healthcare analytics 123

7.5.5.2 Initialization of PSO parameters

Initialization is performed in this step. w, c1, c2, r1, r2 are initialized
with some values. Bounds are also specified to ensure resultant values
must fall within range.

7.5.5.3 Update the parameters until they get optimized or until

some condition is reached
The PSO algorithm works in an iterative manner. The fitness value, posi-
tion, and velocity of particles are updated iteratively until the values
converge.

7.5.5.4 Get the p-best value for all versions

For each version, the p-best value is calculated by comparing the new
p-best with the previous iteration’s p-best value.

7.5.6 Evaluate the summaries and provide the best

optimized summary as a result
The p-best values for all the particles are evaluated, and finally, the
g-best value (best p-best value) is calculated. The summary that is asso-
ciated with the g-best value is considered as the most optimized sum-
marized result.

7.6 RESULTS AND DISCUSSIONS

The dataset is taken from Mtsamples.com. It belongs to medical specialty:

Neurology. Sample Name is “Adult Hydrocephalus”. This report contains
information about a 74-year-old woman who needs a neurological con-
sultation for possible adult hydrocephalus. She has mild gait impairment
and mild cognitive slowing. For experimental purposes, the following
medical transcription sample report is considered.

Original Text: “The patient is a lovely 74-year-old woman who presents

with possible adult hydrocephalus. Danish is her native language, but she
has been in the United States for many many years and speaks fluent
English, as does her husband. With respect to her walking and balance, she
states, “I think I walk funny”. Her husband has noticed over the last six
months or so that she has broadened her base and become more stooped
in her pasture. Her balance has also gradually declined such that she
124 Computational Techniques in Neuroscience

frequently touches walls and furniture to stabilize herself. She has difficulty
stepping up onto things like a scale because of this imbalance. She does not
festinate. Her husband has noticed some slowing of her speed. She does
not need to use an assistive device. She has occasional difficulty getting in
and out of a car. Recently she has had more frequent falls. In March of
2007, she fell when she was walking to the bedroom and broke her wrist.
Since that time, she has not had any emergency department trips, but she
has had other falls. With respect to her bowel and bladder, she has no
issues and no trouble with frequency or urgency. The patient does not
have headaches. With respect to thinking and memory, she states she is
still able to pay the bills, but over the last few months she states, “I do
not feel as smart as I used to be”. She feels that her thinking has slowed
down. Her husband states that he has noticed, she will occasionally start
a sentence and then not know what words to use as she is continuing.
The patient has not had trouble with syncope. She has had past episodes
of vertigo, but not recently. Significant for hypertension diagnosed in 2006,
reflux in 2000, insomnia, but no snoring or apnea. She has been on
Ambien, which has no longer been helpful. She has had arthritis since
the year 2000, thyroid abnormalities diagnosed in 1968, a hysterectomy
in 1986, and a right wrist operation after her fall in 2007 with a titanium
plate and eight screws. Her father died of heart disease in his 60s, and
her mother died of colon cancer. She has a sister who she believes is
probably healthy. She has had two sons, one who died of a blood clot
after having been a heavy smoker and another who is healthy. She had two
normal vaginal deliveries. She lives with her husband. She is a nonsmoker
and has no history of drug or alcohol abuse. She does drink two to three
drinks daily. She completed 12th grade.”

The mentioned summary is given as input to the proposed system, and

various versions of the summary are produced. Each summary is treated as
a particle, and the value of each particle is equivalent to the sum of sentence
scores of the summary. These values of particles are passed to the PSO
algorithm. This algorithm works in an iterative manner, and values of
velocity, positions, and p-best are calculated until the convergence condi-
tion is reached. Finally, the g-best value is calculated among all the p-best
values (Table 7.2).

Table 7.2 Summary and their p-best scores

Summary 30% 40% 45% 50% 55% 60% 65% 70%
p-best 2.48 3.24 2.95 3.92 3.07 4.08 4.64 4.73
 An optimized text summarization for healthcare analytics 125

p-best score
5 4.64 4.73
4.5 4.08
3.92
4
3.5 3.24
2.95 3.07
3
2.48
2.5
2
1.5
1
0.5
0
30% 40% 45% 50% 55% 60% 65% 70%
Summary Percentage

Figure 7.2 p-best score for various version of summary.

Values in the graph depict that the 70% summary is the best optimized
summary. The g-best value is 4.73 as it is the best among all the p-best values.

7.7 CONCLUSION AND FUTURE WORK

In this study, the authors implemented extractive text summarization using

the PSO approach for summarizing medical transcript reports that are
available online. Initially, the TF-IDF algorithm is used to generate the sen-
tence scores; then the PSO algorithm iteratively calculates the p-best values
for each particle based on the distance from the summarized result. Finally, it
calculates the g-best value that exhibits the optimal solution. In the future,
the authors would work on abstractive text summarization and would try to
generate more optimal solutions from abstractive summaries.

REFERENCES

Chen, Y.P., Chen, Y.Y., Lin, J.J., Huang, C.H., and Lai, F. 2020. Modified
Bidirectional Encoder Representations From Transformers Extractive
Summarization Model for Hospital Information Systems Based on Character-
Level Tokens (AlphaBERT): Development and Performance Evaluation. JMIR
Med Inform. doi: 10.2196/17787. PMID:30445218.
Deepika, S., Lakshmi Krishna, N., and Shridevi, S. 2021. Extractive Text
Summarization for COVID-19 Medical Records. Innovations in Power and
Advanced Computing Technologies (i-PACT). Kuala Lumpur. Malaysia. 1–5.
doi: 10.1109/i-PACT52855.2021.9697019.
126 Computational Techniques in Neuroscience

Gayathri, P., and Jaisankar, N. 2015. Towards an Efficient Approach for Automatic
Medical Document Summarization. Cybernetics and Information Technologies.
vol.15. no.4. 2015, 78–91. 10.1515/cait-2015-0056.
Gigioli, P., Sagar, N., Rao, A., and Voyles, J. 2018. Domain-Aware Abstractive
Text Summarization for Medical Documents. 2018 IEEE International
Conference on Bioinformatics and Biomedicine. Doi: 10.1109/bibm.2018.
8621539.
Jesus, M., Sanchez-Gomez, M., Vega-Rodríguez, A., and Pérez, C.J. 2018.
Extractive multi-document text summarization using a multi-objective artifi-
cial bee colony optimization approach. Knowledge-Based Systems. vol.159.
1–8. ISSN 0950-7051. 10.1016/j.knosys.2017.11.029.
Kanwal, N., and Rizzo, G. 2022. Attention-based clinical note summarization,
SAC ’22: Proceedings of the 37th ACM/SIGAPP Symposium on Applied
Computing. (April): 813–820. 10.1145/3477314.3507256.
Kieuvongngam, V., Tan, B., and Niu, Y. 2020. Automatic text summarization of
COVID-19 medical research articles using BERT and GPT-2. CoRR. Volume
abs/2006.01997. http://dblp.org/rec/journals/corr/abs-2006-01977.bib.
Kim, S.W., and Gil, J.M. 2019. Research paper classification systems based on
TF-IDF and LDA schemes. Human-centric Computing and Information Sciences.
vol. 9. 30. 10.1186/s13673-019-0192-7
Manning C., Raghavan, P., and Schütze, H. 2008. Introduction to Information
Retrieval. England: Cambridge University Press.
Moradi, M., and Samwald, M. 2019. Clustering of Deep Contextualized. for
Summarization of Biomedical Texts. arXiv. doi: 10.48550/ARXIV.1908.02286.
Pivovarov, R., and Elhadad, N. 2015. Automated methods for the summarization
of electronic health records. J. Am. Med. Inf. Assoc. vol.22. no. 5. 938–947.
Rohil, M.K., and Magotra, V. 2022. An exploratory study of automatic text
summarization in biomedical and healthcare domain. Healthcare Analytics.
vol. 2. 100058, ISSN 2772-4425, 10.1016/j.health.2022.100058.
Reddy, S.M., and Miriyala, S. 2020. Exploring Multi Feature Optimization for
Summarizing Clinical Trial Descriptions. IEEE Sixth International Conference
on Multimedia Big Data (BigMM). New Delhi, Indi. 341–345. Doi: 101109/
BigMM50055.2020.00059.
Sarkar, K. 2009. Using domain knowledge for text summarization in medical
domain. Int. J. Recent Trends Eng. vol.1. no. 1. 200–205.
“Transcribed Medical Transcription Sample Reports and Examples.” https://www.
mtsamples.com/site/pages/sample.asp?Type=42-Neurology&Sample=1695-
Adult%20Hydrocephalus.
Chapter 8

Computer aided diagnosis of

neurodegenerative diseases using
discrete wavelet transform and
neural network for classification
Prasanna J1, S. Thomas George1, and M.S.P Subathra2
1
Dept. of Biomedical Engineering, Karunya Institute of technology and
sciences, Coimbatore, India
2
Dept. of Robotics Engineering, Karunya Institute of technology and sciences,
Coimbatore, Tamil Nadu, India

8.1 INTRODUCTION

Neurodegenerative diseases (NDDs) are defined by neurophysiological con-

trol as the enlightened damage of pattern or function of neurons, including
neurodegeneration. Amyotrophic lateral sclerosis (ALS), Parkinson’s disease
(PD), and Huntington’s disease (HD) are examples of common NDDs.
Clinical symptoms of various NDDs may differ [1]. NDDs are a group of
warning signs characterized by a decline in neural networks, cognitive abil-
ities, and human motor systems. These disorders cause unusual changes in
neuromuscular control [2]. Individuals with NDDs are at risk of sustaining
serious physical injuries that will have an impact on their daily lives. This
suggests the need for a precise diagnostic method capable of improving
prognostication skills and assisting in the development of new endovascular
effective treatments [3]. PD causes bradykinesia, rigidity, a resting tremor,
and posture instability. The foremost clinical symptoms of HD are chorea
and behavioral and cognitive changes, whereas the most visible symptom of
ALS is muscular degenerative disorder and atrophy. Meanwhile, as a result
of a malfunction in the neural central controller, the common external feature
of NDD is neurological dysfunction, which has been investigated and con-
firmed according to several studies. PD and HD are both disorders of
the basal ganglia that are characterized by changes in gait rhythm. ALS is
an NDD that primarily affects the moto neurons of the prefrontal cortex,
limbic system, and spinal cord [4].
Standard experimental procedures used to diagnose NDDs include bio-
logical blood tests, electromyography analysis [5], structural imaging
practices, spinal cord imaging [6], and nerve biopsy [7]. These are, how-
ever, time-consuming, intrusive, or costly procedures, and some of these
strategies may be less effective in the early stages of the diseases [8]. For
decades, researchers have been studying the effects of these pathogens on

DOI: 10.1201/9781003398066-8 127

128 Computational Techniques in Neuroscience

human locomotion [9]. There has been a lot of interest in automating

the diagnosis of NDDs using gait dynamics [10], because it allows for a
non-invasive and low-cost observation of the neuromuscular motor control
development. Gait recordings are a complex combination of strength [11],
sensation, and coordination. These physiognomies of the gait time series
recommend that it may be sensitive to vascular dementia in mobility
functioning and thus aid in the diagnosis of such illness [12]. The human
gait is a series of cyclic spatial and temporal motions of the left and right
feet that are directed through the nervous system [2]. Morris et al. [13]
illustrated that PD had a significant impact on the subjects’ gait by reducing
their speed, stride frequency, and total range of walking motion. The
reviewed studies found that stride interval or stride-to-stride fluctuation gait
subtleties differed significantly between NDD patients and control subjects.
Currently, the most frequently used method for diagnosing and evaluating
advancement in NDD patients is static based on various types. It is well
recognized that using questionnaires can result in independent results.
As a result, an impartial evaluation of patients’ physically efficient routine
is critical in daily clinical practice, and it helps clinical doctors improve
a more satisfactory treatment design and a more scientific analysis of
therapy outcome [14]. Duta et al. [15] used Elman’s recurrent neural net-
work to identify healthy and unhealthy gait based on stance, swing, and
double support intervals. Cross-correlograms of those gait signals with
corresponding signals from a reference subject provided the relevant fea-
tures in their work. In the context of the previous research, gait parameters
related to the locomotor system’s sensitivity to local perturbations, which
is important for investigating muscular strength of balance. Detrended
fluctuation analysis (DFA) was used by Hausdorff et al [16]. and Jordan
et al. to show that gait dissimilarities are not casual but exhibit long-
range correlation in healthy subjects. Furthermore, their research found
that the scaling exponent values reflect locomotor control, adaptability, and
versatility of walking in humans. The swing-interval turns count (SWITC)
was used by Wu et al. to represent a significant variation between healthy
control subjects and ALS patients [17]. Daliri et al. [18] extracted features
such as the minimum, maximum, average, and standard deviation (SD)
from each gait stride sequence and fed this relevant data to a support
vector classifier to make a distinction of the healthy group from the NDD
group. Wu and Krishnan [19] used the signal turns count procedure to
assess gait variability with diseased patients. They discovered a significant
difference in the swing interval turns count parameter between healthy
control subjects and PD patients. This study [20] used the nonparametric
frequency analysis based Parzen-Window utilized to determine the proba-
bility density functions (PDFs) of movement patterns from the stride
interval, those can be realized the categorization of gait in subjects and
normal subjects by calculating the average of the left-foot stride intermis-
sion and the altered Kullback-Leibler divergence from the PDF’s projection.
 Computer aided diagnosis of neurodegenerative diseases 129

In addition to significant work towards automatic gait analysis, the past

few decades have seen the rapid growth of gait recognition, which has
procedures similar to gait detection and classification. Moustakidis et al.
[21] used wavelet packet (WP) degradation to measure ground reaction
forces while walking, and then used a fuzzy complementary criterion to
obtain a compact set of robust and corresponding features from WP coef-
ficients using only one camera. Hu [22] anticipated a two-stage based
Gaussian mixture model to depict the dual-tree complex wavelet trans-
form (DTCWT) coefficients after applying gait images. A simple nearest-
neighbor classifier was then used to measure the viewpoint of a gait
sequence. Zeng and Wang [23] used the deterministic learning theory to
extract features from a series of subject aesthetic and then discovered gait
of DTCWT with different dimensions and positioning to gait energy cog-
nition for the realization of inter gait recognition. A feature representation
and a more appropriate classifier are required for improved performance.
Difficulty stems from the following two fundamental facts: Due to the
nonlinear dynamics of the human system, gait cadence exhibits complex
and nonlinear behavior in both NDD and CO subjects [24]. The first
20 seconds of each and every recording were not considered in order to
eliminate start-up consequences, and a median filter was used to remove
sets of data that were three standard deviations above or below the median
feature value. The cracks at the expiration of the hallway are also most
likely responsible for these outliers; Hausdorff et al. proposed all of the pre-
processing steps in their previous study [25]. In [25,26] NDD patients
can be helped by using gait rhythm signals, which are easy to collect using
various sensors. This study also focused on the detection of gait time series
in order to describe their suitability in symmetry assessment for NDD
detection. Classification method based on a machine learning approach was
used to distinguish gait time intervals of NDD patients. Considering small
number of training samples or noisy data, the proposed method is projected
to attain higher classification performance for all classification tasks.

Contributions:
• Time-frequency domain based feature extraction technique is pro-
posed for the effective feature extraction.
• Discrete wavelet transform (DWT) technique has been applied on
the gait time series to decompose the signals to coefficients.
• Statistical and entropy-based features were extracted from the de-
composed coefficients taken for the classification.
• Artificial neural network (ANN) classifier is utilized to classify the
task such as HC vs PD, HC vs. HD, HC vs. ALS, and HC vs. NDD.
• The classification performance shows the better results with the pro-
posed approach. The framework of the proposed method is shown
in Figure 8.1.
130 Computational Techniques in Neuroscience

Figure 8.1 The pictorial framework of the proposed technique.

Outline of the Paper:

• Section 8.2 provides the framework of the proposed methodology and
dataset description and classification technique.
• In Section 8.3, we illustrate the performance of the classification of
different tasks, and we discuss the achieved results. The comparison
between the existing and proposed method is reviewed based on the
classification performance.
• Finally, Section 8.4 concludes the article with future directions.

8.2 METHODS AND MATERIALS

8.2.1 Dataset used

The gait dataset defined in [25] is used in this study. It is freely accessible at
www.physionet.org. The whole gait recordings in the dataset include gait re-
cordings from 20 HD subjects, 13 ALS subjects, 15 PD subjects, and 16 healthy
control (HC) subjects. All patients with HD, PD, and ALS were evaluated
for indicators that could impair gait signals. A comparison of HC and NDD
subjects revealed no significant dissimilarities in heights and weights [26].
All participants were instructed to walk 77 meters in a straight line at their
average speed. The gait signals were recorded using force-sensitive resistors
located beneath each subject’s feet for 5 minutes while walking [27,28].
The visual representation of gait signals is shown in Figure 8.2. A 12-bit
analog-to-digital converter with a sampling rate of 300 Hz was used
to collect the data. The data contains various time series of strike intervals
defined as left and right stride intervals (Sec), swing intervals (Sec), stance
intervals (Sec), and left and right stance intervals (% of stride).
 Computer aided diagnosis of neurodegenerative diseases 131

Figure 8.2 Gait time series representation.

8.2.2 Discrete wavelet transforms

Wavelet transforms are extensively used in many disciplines to solve
a variety of actual issues. A signal’s Fourier transform includes only
the spectral spectrum of the gait over the estimation process and thus
privations any time domain location information. In order to obtain
localization information in time, [29] the time window must be short,
compromising localization in frequency. On the contrary, achieving fre-
quency localization necessitates a large time analysis window, which
compromises time localization. Therein lays the dilemma, from time
to time denoted as the “uncertainty principle”. The short-time Fourier
transform (STFT) is a kind of negotiation of time and frequency-based
perspectives of a signal, and it consists of both time and frequency
information. The size of the prediction interval determines the frequency
resolution of STFT. This sampling rate is constant across the entire fre-
quency range.
DWT is a characterization of the signal, which is an infinite set of a
wavelet on a orthonormal basis. DWT is able to decompose non-linear
features of the signal.
In DWT, the dilation and translation measures are discretized, as shown
in equation (8.1)

a1 = a1oj
a10 < 1, b0 0, j Z, k Z (8.1)
b = kb0 a1oj

The base wavelet has discretized measures, which can be written as,
132 Computational Techniques in Neuroscience

1 t ka10j b0
j , k (t ) = (8.2)
a0j a10j

Therefore, DWT of the signal can be denoted as,

+
D (j, k) = u1(t) j , k (t) dt (8.3)

The signal u1(t) can be duplicated using the inverse DWT, as given below:

J
1
u1(t) = wt (j , k) j , k (t ) a1 R+ (8.4)
a1 j= k=

The signal is instantaneously carried through LP and HP filters during the

initial step of the DWT, also with cut-off frequency being one-fourth of
the sampling frequency [30]. The low and high pass filter outcomes are
mentioned as the approximation (A) and detail (D) coefficients of the first
level, respectively.
According to the Nyquist rule, output signals with half the frequency
bandwidth of the input signal can be down sampled by two. To obtain the
second-level coefficients, repeat the steps used for the first-level approxi-
mation and the detail coefficients. The frequency resolution is doubled
through filtering at each step of this decomposition process, and the time
resolution is halved through down sampling. Figure 8.3 depicts a signal’s
nth-level wavelet decomposition.

8.2.3 Feature extraction

Feature extraction is a critical process in the medical field for event cate-
gorization of signals and images for diagnostic techniques [31]. In this

Figure 8.3 Nth level of wavelet decomposition of the gait time series.
 Computer aided diagnosis of neurodegenerative diseases 133

current work, entropy and statistical-based feature extraction has been

performed for the classification of gait signals for the identification of
NDDs. Nine statistical measures – minimum, maximum, mean, standard
deviation, energy, normalized energy, kurtosis, skewness, and normalized
standard deviation features – have been computed from the wavelet de-
composed coefficients. Five entropy measures – log energy entropy, sample
entropy, approximate entropy, permutation entropy, and fuzzy entropy
features – have been computed for the classification of gait signals. The
level of involvement in non-stationary signals is calculated using log energy
entropy [32]. That can be mathematically illustrated as

N
Log energy entropy = log(y1i2 ) (8.5)
i =1

Approximate Entropy assesses the randomness and uniformity of a gait

signal across multiple dimensions [33]. It is a scale invariant that describes
the similarity of the samples. Sample Entropy is an updated form of
Approximate Entropy that is beneficial for complexity reduction in gait
signals [34]. It is used to compute the uniformity in the signal more
efficiently than Approximate Entropy. It is independent of the length of
the time cycles. Permutation Entropy [35] was used to approximate time
series complexity by comparing neighboring time series signals. It is a non-
stationary time series method that yields efficient, reliable, and quick
results. The relationship between the distinguishable number of equally
spaced values for a past and present esteem is labelled by mapping the
continuous time series record onto a symbolic sequence. Fuzzy Entropy [36]
signifies the information of consistency, and it is implemented from the
fuzzy sets, whose measure on the gait recordings has a specific function,
which is known as membership function with a value remains from 0 to 1.
Fuzzy Entropy is a continuous function that is used to estimate resemblance
in the time sequence of the signal.

8.2.4 Artificial neural network classifier

Classification is an important area of research in big data, and neural
networks are one of the most widely used classification techniques. Three
parameters that can make a significant difference are typically used to
challenge NDD classification using gait features: 1) a limited number of
clinical samples; 2) a large number of noise level or redundant features;
and 3) the need to meet real-time deadlines [37]. ANN is a sophisticated
adaptive system that could also change its inner core in responding to
input. It’s done by altering the weight of the correlation. Each link has
a specific amount of weight. Weights handle the connection between
two neurons, which can be adapted to improve system accurateness.
134 Computational Techniques in Neuroscience

Figure 8.4 Artificial neural network classifier.

The signal between two nerve cells is governed by a weight, which is a

scalar. An ANN connects the output layer, input layer, and one or more
hidden layers (middle layer) with [38] a larger number of processing
nodes or artificial neurons. The function of ANN classifier is given in
Figure 8.4.
In the current study, scaled conjugate gradient back propagation is uti-
lized for training the samples. The classification performance of the ANN
classifier computed by the following parameters.

True (+) + True ( )

Classification Accuracy = 100 (8.6)
True (+) + True ( ) + False (+) + False ( )

True (+)
Sensitivity = 100 (8.7)
True (+) + False ( )

True ( )
Specificity = 100 (8.8)
True ( ) + False (+)

True (+)
Positive predictive value(PPV) = 100 (8.9)
True (+) + False (+)

True ( )
Negative Predictive Value(NPV) = 100 (8.10)
True ( ) + False ( )

Where, True (+) indicates True positive, True(-) refers to True negative,
False (+) corresponds to False positive and False(-) implies False negative.
 Computer aided diagnosis of neurodegenerative diseases 135

8.3 RESULTS AND DISCUSSION

To investigate the proposed technique, gait signals from pressure sensors

from the gait dynamics for NDDs dataset are used. The non-linear char-
acteristics and statistical information of gait signals are investigated in the
presented work by using wavelet analysis based on time-frequency com-
putation on gait signals. It captures the effective non-linear properties and
statistical features from the wavelet decomposed coefficients of a given
signal where the optimum statistical-based features such as minimum,
maximum, mean, standard deviation, normalized standard deviation, kur-
tosis, skewness energy, and normalized energy were captured from the left
and right feet of a healthy control subject, an ALS patient, an HD patient,
and a PD patient, respectively.
Table 8.1 shows the classification performance of the task PD vs. HC,
individual statistical measures, and in Table 8.2, entropy measures are
considered. When we consider all the features, minimum feature achieved

Table 8.1 Classification results achieved using statistical features for the classification of
PD vs. HC
Features Accuracy (%) Sensitivity (%) Specificity (%) PPV (%) NPV (%)
Minimum 91.94 87.50 96.67 95.57 89.27
Maximum 61.61 64.38 58.67 62.19 63.13
Mean 91.94 87.50 96.67 95.57 89.27
Standard deviation 64.84 65.00 64.67 69.09 65.70
Normalized 75.48 70.63 80.67 79.09 72.52
standard deviation
Kurtosis 76.77 78.75 74.67 77.10 77.03
Skewness 73.87 77.50 70.00 73.49 74.64
Energy 67.42 77.50 56.67 66.29 71.07
Normalized energy 62.26 72.50 51.33 62.16 66.85
All features 99.38 99.23 99.41 99.29 98.65

Table 8.2 Classification results achieved using entropy features for the classification of
PD vs. HC
Features Accuracy (%) Sensitivity (%) Specificity (%) PPV(%) NPV(%)
Log- Energy 85.36 87.03 71.26 84.73 84.37
Approximate 98.13 96.75 94.88 97.48 97.43
Sample 91.94 87.50 96.67 95.57 89.27
Permutation 85.16 86.25 84.00 85.36 87.03
fuzzy 97.42 96.88 98.00 98.13 96.75
All features 99.78 99.46 99.70 99.75 99.72
136 Computational Techniques in Neuroscience

Table 8.3 Classification results achieved using statistical features for the classification of
HD vs. HC
Features Accuracy (%) Sensitivity (%) Specificity (%) PPV(%) NPV(%)
Minimum 58.61 26.25 84.50 57.17 58.94
Maximum 61.67 39.38 79.50 60.03 62.34
Mean 63.61 42.50 80.50 64.63 63.57
Standard deviation 63.33 56.25 69.00 60.37 66.02
Normalized standard 54.72 33.75 71.50 50.51 57.17
deviation
Kurtosis 57.78 40.00 72.00 52.97 60.46
Skewness 70.56 69.38 71.50 67.00 74.25
Energy 70.28 60.63 78.00 69.10 71.32
Normalized energy 71.94 79.38 66.00 65.11 80.34
All features 97.17 96.25 98.61 98 98.94

Table 8.4 Classification results achieved using entropy features for the classification of
HD vs. HC
Features Accuracy (%) Sensitivity (%) Specificity (%) PPV(%) NPV(%)
Log- Energy 72.34 90.53 96.84 94.93 81.67
Approximate 83.57 85.44 50.94 58.28 63.61
Sample 66.02 25.81 57.89 61.99 63.33
Permutation 57.17 76.33 39.72 78.53 54.72
fuzzy 95.46 92.68 94.84 92.70 97.78
All features 99.98 99.22 99.72 99.73 99.71

the maximum classification performance by having accuracy of 91.94%,

sensitivity of 87.50%, specificity of 96.67% and PPV of 95.57% with NPV
of 89.27%. When all the 9 statistical features fed in the ANN classifier,
that achieved the accuracy of 99.38%.
All entropies considered for the classification attained the classification
accuracy of 99.78%. Tables 8.3 and 8.4 list the classification performance
of HD vs. HC.
Normalized energy from the statistical features and fuzzy entropy from
the entropy features attained the maximum classification performance with
respect to all the classification parameters considered. Consideration of
all statistical and entropy features attained the accuracy of 97.17% and
99.98%, respectively.
Tables 8.5 and 8.6 show the classification performance of the ALS vs.
HC, thereby concluding that the proposed method outperforms by using
the normalized energy feature from the statistical and fuzzy & approximate
entropy measure attained the maximum accuracy of 88.28%, 86.85%, and
84.86%, respectively.
 Computer aided diagnosis of neurodegenerative diseases 137

Table 8.5 Classification results achieved using statistical features for the classification of
ALS vs. HC
Features Accuracy (%) Sensitivity (%) Specificity (%) PPV(%) NPV(%)
Minimum 75.86 86.25 63.08 75.57 76.72
Maximum 85.17 88.13 81.54 85.54 84.86
Mean 65.52 67.50 63.08 69.65 60.81
Standard deviation 84.48 91.88 75.38 82.16 88.45
Normalized standard 76.21 93.75 54.62 71.97 86.85
deviation
Kurtosis 72.41 87.50 53.85 70.00 77.78
Skewness 64.83 74.38 53.08 60.16 67.07
Energy 78.28 79.38 76.92 82.30 73.03
Normalized energy 88.28 91.25 84.62 87.93 88.95
All features 98.97 93.75 93.08 97.31 91.50

Table 8.6 Classification results achieved using entropy features for the classification of
ALS vs. HC
Features Accuracy (%) Sensitivity (%) Specificity (%) PPV(%) NPV(%)
Log- Energy 76.72 50.74 80.24 72.72 75.86
Approximate 84.86 70.03 86.78 84.73 85.17
Sample 60.81 30.52 68.47 65.08 65.52
Permutation 88.45 68.91 86.72 83.19 84.48
fuzzy 86.85 53.31 81.40 71.30 76.21
All features 97.78 94.45 97.78 98.64 92.41

From Tables 8.7 and 8.8 the classification performance of NDD vs.
HC is considered with the individual features from both, such as statis-
tical and entropy attained, showing the better classification results. As
shown in Figure 8.5, the proposed method outperforms in all the classi-
fication tasks.
The three patient groups with NDDs are at various stages in the current
database. They are assessed by the severity or length of the illness. This is
the Hohn and Yahr score for PD patients. This is the total functional
storage assessment for subjects with HD.
The number here represents the number of months since the subject’s
diagnosis, demonstrated in the experiments that patients with NDDs, even
those in the early stages, can be categorized. This validated the proposed
method’s efficacy in early detection. Nonetheless, more clinical research and
data analysis with larger sample sizes are needed to verify the method’s
efficacy and appropriateness in the early stages.
138 Computational Techniques in Neuroscience

Table 8.7 Classification results achieved using statistical features for the classification of
NDD vs. HC
Features Accuracy (%) Sensitivity (%) Specificity (%) PPV(%) NPV(%)
Minimum 58.94 63.13 35.79 46.19 58.61
Maximum 62.34 70.53 46.84 54.93 61.67
Mean 63.57 75.44 50.94 58.28 63.61
Standard deviation 66.02 65.81 57.89 61.99 63.33
Normalized standard 57.17 76.33 39.72 48.53 54.72
deviation
Kurtosis 60.46 76.68 44.84 52.70 57.78
Skewness 74.25 41.06 67.91 70.21 70.56
Energy 71.32 39.50 64.38 68.61 70.28
Normalized energy 80.34 45.41 71.46 72.29 71.94
All features 92.39 92.22 92.97 93.37 94.70

Table 8.8 Classification results achieved using entropy features for the classification of
NDD vs. HC
Features Accuracy (%) Sensitivity (%) Specificity (%) PPV(%) NPV(%)
Log- Energy 96.67 95.57 89.27 84.50 91.04
Approximate 64.67 69.09 65.70 31.18 62.58
Sample 80.67 79.09 72.52 51.45 74.26
Permutation 74.67 77.10 77.03 53.77 77.67
fuzzy 93.00 93.49 94.64 96.81 94.35
All features 97.00 95.36 97.03 91.26 94.73

100
98
96
94
Accuracy
92
Sensi vity
90
88 Specificity

86 PPV
NPV

Figure 8.5 The summary of classification performance considering all features for the
classification.
 Computer aided diagnosis of neurodegenerative diseases 139

8.4 CONCLUSION

The central nervous system is crucial in the regulation of human gait. PD is

a common NDD that can cause neurophysiologic changes in the central
nervous system, altering the duration of the gait cycle (stride interval). The
time-frequency based wavelet analysis performed on the gait signals by
applying discrete wavelet transform with the decomposed detail and
approximate coefficients. The nine statistical and five entropy measures
from the decomposed coefficients computed for the automated detection
of NDDs. These optimum features are classified using ANN classifier with
the 10-fold cross validation with the better training and testing. The clas-
sification performances of the four different tasks are validated by incor-
porating every single statistical and entropy feature, also computed by
considering all the entropy measures and statistical features.

REFERENCES

[1] Hausdorff, Jeffrey M., Apinya Lertratanakul, Merit E. Cudkowicz, Amie L.

Peterson, David Kaliton, and Ary L. Goldberger. “Dynamic markers of
altered gait rhythm in amyotrophic lateral sclerosis.” Journal of Applied
Physiology (2000).
[2] Viteckova, Slavka, Patrik Kutilek, Zdenek Svoboda, Radim Krupicka, Jan
Kauler, and Zoltan Szabo. “Gait symmetry measures: A review of current
and prospective methods.” Biomedical Signal Processing and Control 42
(2018): 89–100.
[3] Moon, Yaejin, JongHun Sung, Ruopeng An, Manuel E. Hernandez, and
Jacob J. Sosnoff. “Gait variability in people with neurological disorders: A
systematic review and meta-analysis.” Human movement science 47 (2016):
197–208.
[4] Aziz, Wajid, and Muhammad Arif. “Complexity analysis of stride interval
time series by threshold dependent symbolic entropy.” European Journal of
Applied Physiology 98 (2006): 30–40.
[5] Joshi, Deepak, Aayushi Khajuria, and Pradeep Joshi. “An automatic non-
invasive method for Parkinson’s disease classification.” Computer Methods
and Programs in Biomedicine 145 (2017): 135–145.
[6] Chen, J. Jean. “Functional MRI of brain physiology in aging and neuro-
degenerative diseases.” Neuroimage 187 (2019): 209–225.
[7] Blijham, Paul J., H. Jurgen Schelhaas, Henk J. Ter Laak, Baziel G.M. van
Engelen, and Machiel J. Zwarts. “Early diagnosis of ALS: The search for
signs of denervation in clinically normal muscles.” Journal of the
Neurological Sciences 263, no. 1-2 (2007): 154–157.
[8] Baratin, E., L. Sugavaneswaran, K. Umapathy, C. Ioana, and S. Krishnan.
“Wavelet-based characterization of gait signal for neurological abnormali-
ties.” Gait & Posture 41, no. 2 (2015): 634–639.
[9] Bilgin, Suleyman. “The impact of feature extraction for the classification
of amyotrophic lateral sclerosis among neurodegenerative diseases and
140 Computational Techniques in Neuroscience

healthy subjects.” Biomedical Signal Processing and Control 31 (2017):

288–294.
[10] Hausdorff, Jeffrey M., Susan L. Mitchell, Renee Firtion, Chung-Kang Peng,
Merit E. Cudkowicz, Jeanne Y. Wei, and Ary L. Goldberger. “Altered fractal
dynamics of gait: reduced stride-interval correlations with aging and
Huntington’s disease.” Journal of Applied Physiology 82, no. 1 (1997):
262–269.
[11] Pham, Tuan D. “Texture classification and visualization of time series of gait
dynamics in patients with neuro-degenerative diseases.” IEEE Transactions
on Neural Systems and Rehabilitation Engineering 26, no. 1 (2017):
188–196.
[12] Prabhu, Pooja, A. Kotegar Karunakar, H. Anitha, and N. Pradhan.
“Classification of gait signals into different neurodegenerative diseases
using statistical analysis and recurrence quantification analysis.” Pattern
Recognition Letters 139 (2020): 10–16.
[13] Morris, Meg E., Jennifer McGinley, Frances Huxham, Janice Collier, and
Robert Iansek. “Constraints on the kinetic, kinematic and spatiotemporal
parameters of gait in Parkinson’s disease.” Human Movement Science 18,
no. 2-3 (1999): 461–483.
[14] Ghaderyan, Peyvand, and Seyede Marziyeh Ghoreshi Beyrami. “Neuro-
degenerative diseases detection using distance metrics and sparse coding: A
new perspective on gait symmetric features.” Computers in Biology and
Medicine 120 (2020): 103736.
[15] Dutta, Saibal, Amitava Chatterjee, and Sugata Munshi. “An automated
hierarchical gait pattern identification tool employing cross‐correlation‐
based feature extraction and recurrent neural network based classification.”
Expert Systems 26, no. 2 (2009): 202–217.
[16] Hausdorff, Jeffrey M. “Gait variability: methods, modeling and meaning.”
Journal of Neuroengineering and Rehabilitation 2, no. 1 (2005): 1–9.
[17] Wu, Yunfeng, and Sridhar Krishnan. “Computer-aided analysis of gait
rhythm fluctuations in amyotrophic lateral sclerosis.” Medical & Biological
Engineering & Computing 47 (2009): 1165–1171.
[18] Daliri, Mohammad Reza. “Automatic diagnosis of neuro-degenerative dis-
eases using gait dynamics.” Measurement 45, no. 7 (2012): 1729–1734.
[19] Wu, Yunfeng, and Sridhar Krishnan. “Statistical analysis of gait rhythm in
patients with Parkinson’s disease.” IEEE Transactions on Neural Systems
and Rehabilitation Engineering 18, no. 2 (2009): 150–158.
[20] Wu, Yunfeng, and Lei Shi. “Analysis of altered gait cycle duration in
amyotrophic lateral sclerosis based on nonparametric probability density
function estimation.” Medical Engineering & Physics 33, no. 3 (2011):
347–355.
[21] Moustakidis, S.P., J.B. Theocharis, and G. Giakas. “Feature selection based
on a fuzzy complementary criterion: application to gait recognition using
ground reaction forces.” Computer Methods in Biomechanics and Biomedical
Engineering 15, no. 6 (2012): 627–644.
[22] Hu, Haifeng. “Multiview gait recognition based on patch distribution fea-
tures and uncorrelated multilinear sparse local discriminant canonical cor-
relation analysis.” IEEE Transactions on Circuits and Systems for Video
Technology 24, no. 4 (2013): 617–630.
 Computer aided diagnosis of neurodegenerative diseases 141

[23] Zeng, Wei, Cong Wang, and Feifei Yang. “Silhouette-based gait recognition
via deterministic learning.” Pattern Recognition 47, no. 11 (2014):
3568–3584.
[24] Wagenaar, Robert C., and Richard E.A. van Emmerik. “Dynamics of move-
ment disorders.” Human Movement Science 15, no. 2 (1996): 161–175.
[25] Hausdorff, Jeffrey M., Merit E. Cudkowicz, Renée Firtion, Jeanne Y. Wei,
and Ary L. Goldberger. “Gait variability and basal ganglia disorders:
stride‐to‐stride variations of gait cycle timing in Parkinson’s disease
and Huntington’s disease.” Movement Disorders 13, no. 3 (1998):
428–437.
[26] Xia, Yi, Qingwei Gao, and Qiang Ye. “Classification of gait rhythm
signals between patients with neuro-degenerative diseases and normal
subjects: Experiments with statistical features and different classification
models.” Biomedical Signal Processing and Control 18 (2015): 254–262.
[27] Zeng, Wei, and Cong Wang. “Classification of neurodegenerative diseases
using gait dynamics via deterministic learning.” Information Sciences 317
(2015): 246–258.
[28] Hausdorff, Jeffrey M., Zvi Ladin, and Jeanne Y. Wei. “Footswitch system for
measurement of the temporal parameters of gait.” Journal of Biomechanics
28, no. 3 (1995): 347–351.
[29] Ocak, Hasan. “Automatic detection of epileptic seizures in EEG using dis-
crete wavelet transform and approximate entropy.” Expert Systems with
Applications 36, no. 2 (2009): 2027–2036.
[30] Prasanna, J., M.S.P. Subathra, Mazin Abed Mohammed, Robertas Damaševičius,
Nanjappan Jothiraj Sairamya, and S. Thomas George. “Automated epileptic
seizure detection in pediatric subjects of CHB-MIT EEG database—a survey.”
Journal of Personalized Medicine 11, no. 10 (2021): 1028.
[31] Jothiraj, Sairamya Nanjappan, Thomas George Selvaraj, Balakrishnan
Ramasamy, Narain Ponraj Deivendran, and Subathra M.S.P. “Classification
of EEG signals for detection of epileptic seizure activities based on feature
extraction from brain maps using image processing algorithms.” IET Image
Processing 12, no. 12 (2018): 2153–2162.
[32] Subathra, M.S.P., Mazin Abed Mohammed, Mashael S. Maashi,
Begonya Garcia-Zapirain, N.J. Sairamya, and S. Thomas George. “Detection
of focal and non-focal electroencephalogram signals using fast Walsh-
Hadamard transform and artificial neural network.” Sensors 20, no. 17
(2020): 4952.
[33] Acharya, U. Rajendra, Hamido Fujita, Vidya K. Sudarshan, Shreya Bhat,
and Joel E.W. Koh. “Application of entropies for automated diagnosis of
epilepsy using EEG signals: A review.” Knowledge-based Systems 88
(2015): 85–96.
[34] Cuesta–Frau, David, Pau Miró–Martínez, Jorge Jordán Núñez, Sandra
Oltra–Crespo, and Antonio Molina Picó. “Noisy EEG signals classification
based on entropy metrics. Performance assessment using first and second
generation statistics.” Computers in Biology and Medicine 87 (2017):
141–151.
[35] Cao, Yinhe, Wen-wen Tung, J.B. Gao, Vladimir A. Protopopescu, and Lee
M. Hively. “Detecting dynamical changes in time series using the permuta-
tion entropy.” Physical Review E 70, no. 4 (2004): 046217.
142 Computational Techniques in Neuroscience

[36] Chen, Weiting, Zhizhong Wang, Hongbo Xie, and Wangxin Yu.
“Characterization of surface EMG signal based on fuzzy entropy.” IEEE
Transactions on Neural Systems and Rehabilitation Engineering 15, no. 2
(2007): 266–272.
[37] Sharma, Bhavna, and K.J.I.J.C.E. Venugopalan. “Comparison of neural
network training functions for hematoma classification in brain CT images.”
IOSR Journal of Computer Engineering 16, no. 1 (2014): 31–35.
[38] Subasi, Abdulhamit, M. Kemal Kiymik, Ahmet Alkan, and Etem Koklukaya.
“Neural network classification of EEG signals by using AR with MLE pre-
processing for epileptic seizure detection.” Mathematical and Computational
Applications 10, no. 1 (2005): 57–70.
Chapter 9

EEG artifact detection and

removal techniques
A brief review
Sandhyalati Behera and Mihir Narayan Mohanty
Department of Electronics and Communication Engineering, ITER,
Sikhsha ‘O’ Anusandhan (Deemed to be University), Bhubaneswar, India

9.1 INTRODUCTION

By attaching metal electrodes to a patient’s scalp, doctors can record the

brain’s electrical activity in real time during a diagnostic procedure known
as an electroencephalogram (EEG). When the nervous system is at rest,
the brain neurons are still actively engaged in their normal, unprompted
process of exchanging information via the production of electrical currents.
EEG is an effective tool for brain imaging due to its low cost, high flexi-
bility, high temporal resolution, non-invasiveness, ease of use, portability,
and safety in comparison to other functional neuroimaging techniques such
as positron emission tomography (PET), magnetoencephalography (MEG),
functional magnetic resonance imaging (fMRI), and transcranial magnetic
stimulation (TMS). Brain signals, unlike other bodily signals, exhibit peri-
odic patterns in their electrical changes, which can be used to infer the
occurrence of some activity, such as the change from sleep to wakefulness,
which is characterized by slow, very low frequency, high-amplitude waves
(fast, high-frequency, low-amplitude waves). Furthermore, the impact of
anesthesia and epilepsy are investigated on EEGs.
The brain’s electrical activity, as measured by an electroencephalograph,
is a non-linear and non-stationary signal. We can infer the occurrence of
some neuronal activity from the slight shifts in the voltage fluctuations
measured by EEG. Consequently, the visual examination of these signals
varies with the expertise level. Furthermore, manual review of long EEG
recordings takes a significant amount of time, and results may be inaccurate
due to artifacts in the signals. Thus, computer-aided technologies can be
used to process and analyze these signals, yielding quick and precise results.
Particularly in the diagnosis of epilepsy [1,2], major depressive disorder
(MDD) [3,4], alcohol use disorder (AUD) [5,6], and dementias [7,8] like
Alzheimer’s, mild cognitive impairment (MCI), Parkinson’s, and dementia
with Lewy bodies (DLB), the use of computer-aided technologies with EEG
signals has gained widespread popularity. The integration of motor imagery
into neuro prostheses [9,10] through EEG has led to exciting new

DOI: 10.1201/9781003398066-9 143

144 Computational Techniques in Neuroscience

possibilities. In addition, the use of physiological data like EEG is yielding

significant results in other research domains, such as identity authentication
[11,12], sleep stage classification, emotion recognition, eye state detection,
and drowsiness monitoring.
However, artifacts [13,14] are an inherent problem with EEG recordings.
As the original EEG activity is distorted, analyzing it becomes more chal-
lenging due to artifacts, which are unwanted signals that originate from
sources other than neurons. Eye movements, eye blinks, muscle activity, and
cardiac activity are common sources of contamination in the EEG, despite
the fact that only neuronal activity should be present [15,16]. Because arti-
fact contamination modifies the genuine EEG signal, it also impacts the
outcomes of the applied analysis. For instance, it is well established that
artifacts lower classification accuracy and controllability. Therefore, it is
essential for clinical or applied research to address these artifacts before
analyzing EEG signals. This requires a technique that not only effectively
eliminates artifacts but also protects the authentic, undistorted neuronal
activity present in EEG signals.
To achieve this goal, a number of techniques, both manual and automated,
have been created and put into use. This paper provides a comprehensive
overview of the difficulties encountered by artifact removal algorithms when
processing EEG data. This paper is the first of its kind to elaborate on the
algorithm-specific and general difficulties of EEG artifact removal algorithms,
to the best of our knowledge. Recent reviews [17–21] have only focused on
the specifics of implementing various methods. Artifact removal algorithms are
discussed in Jiang et al. [22] and Islam et al. [19,23] with respect to only four
algorithm-specific challenges: additional reference channel, automatic, online,
and applicability to a single EEG channel. Similarly, Mannan et al. [24] dis-
cussed various difficulties associated with artifact removal algorithms in the
discussion section. The challenges of creating and evaluating EEG artifact
removal algorithms are becoming increasingly apparent, and researchers are
beginning to realize they need more information about them. The purpose of
this paper is to highlight the challenges inherent in existing methods for
cleaning EEG data of artifacts and to propose alternatives for addressing those
challenges. The writing of this paper is broken up into a few different parts.
In Section 9.2, we delve deeper into the common artifacts found in EEG data.
Artifact removal and storage methods are briefly covered in Section 9.3.
Sections 9.4 and 9.5 discuss the specific and general challenges of EEG artifact
removal algorithms, respectively. In Section 9.6, we offer some suggestions for
dealing with these problems.

9.2 DIFFERENT TYPES OF EEG ARTIFACTS

When designing or choosing algorithms to remove artifacts from EEG

signals, it is important to have a foundational understanding of the various
 EEG artifact detection and removal techniques 145

types of artifacts that can be present. Specifically, artifacts are defined as

“unwanted signals that may introduce changes in the measurements and
affect the signal of interest” [25]. The subject’s physiological activities
and movement, as well as environmental interferences (e.g., the movement
of electrodes and cables) can introduce both frequency and time domain
artifacts into an EEG recording [26]. Cables, which act as antennas when
in use, can be shielded and grounded to reduce external artifacts. They
can also be isolated and relocated away from recording locations.
Contrarily, internal or physiological artifacts present difficulties for scien-
tists. Artifacts from the eyes (called electrooculograms, or EOGs), artifacts
from the muscles (EMG), cardiac artifacts (ECG artifacts), and motion
artifacts are the most noticeable in a standard EEG recording.

9.2.1 Ocular (EOG) artifact

The most prevalent artifact encountered when measuring EEG signals is the
ocular artifact. Because of the potential disparity between the cornea and
the retina, an artifact is produced. Blinking and other eye movements alter
the potentials in the eye. In particular, ocular conductance artifacts are caused
by changes in contact between the cornea and eyelid during blinking [27].
Artifacts caused by eye movement occur when the retina and cornea dipole
are no longer aligned. Because of the volume conduction effect, the electrodes
were able to pick up both the ocular artifact and the EEG activity occurring
on the surface of the subject’s head. The electrooculogram is a tool for
recording these eye waves (EOG). While the frequencies of EOG and EEG are
similar, the amplitude of EOG is typically much higher [27–29]. It’s impor-
tant that both EEG and EOG have the potential to contaminate the other’s
data. As a result, we will encounter removal error when eliminating EOG
artifacts due to bidirectional interference [30].

9.2.2 Muscular artifact

Muscle or electromyographic (EMG) artifact is another source of noise in
EEG signals. The action of the frontalis and temporalis muscles is to blame
for these. Non-stationary characteristics [31] are inherent to EMG because
it is produced by a collection of muscles that are both physically and
functionally separated from one another. Muscle artifact waveforms and
amplitudes vary by muscle type, degree of contraction, and participant sex
[32,33]. It is generally agreed that the frequency range of EMG activity is
much larger than that of traditional EEG rhythms. According to [25], the
power density of contracting striated muscle is highest at the lower end of
the frequency spectrum, which spans from 20 to 300 Hz. Muscle artifacts
are most likely to occur in the gamma band, which spans 30 to 40 Hz, given
that the peak frequencies of the temporal, masseter, and frontalis muscles
are 40 to 80 Hz, 50 to 60 Hz, and 30 to 40 Hz, respectively [31,34].
146 Computational Techniques in Neuroscience

The beta band overlaps with the frequencies reported in [31] for the frontal
muscles, which are around 20–30 Hz. Furthermore, 2 Hz has been reported
as the lowest frequency of muscle activity. Muscle artifacts also disrupt the
delta, theta, and alpha frequency bands. Therefore, using only the most
fundamental spectral signatures, it is challenging to tell EMG apart
from EEG.

9.2.3 Cardiac artifact

The heart potential over the scalp causes artifacts in an EEG called
electro-cardiogram (ECG). This type of artifact is seen more in cases of a
short person with a wide neck. Placing electrodes on or near a blood
vessel [31], which is expanding and contracting in response to the beating
of the heart, introduces cardiac artifacts. Similar to cardiac artifact, pulse
artifacts, with a frequency of about 1.2 Hz, can appear within the EEG
as a similar waveform and are therefore challenging to remove [29,35].
In contrast to pulse artifacts, the cardiac artifact can be measured with
a characteristic regular pattern [36] and recorded independently of cere-
bral activity, suggesting that removing such artifacts may be as simple
as using a reference waveform. Because of cardiac activity, the EEG sig-
nals are distorted in a static magnetic field, causing BCG artifacts.
Electrode motion is typically brought on by cardiac-related activities [37]
and in 3.0T, the magnitude of the BCG artifacts can reach 400 V (roughly
6–8 times that of EEG) [38].

9.2.4 Motion artifact

Removal of motion artifact (MA) recently gained interest among researchers
in the past two years. When the subject’s head rotates in relation to the
magnetic field, MAs are introduced into the EEG data that were collected at
the same time as the fMRI [39]. Because they have the same frequency
spectrum as the EEG signal (up to 50 Hz) but a larger amplitude than the
brain signals, MAs have a relatively greater impact on ambulatory EEG [40].
These MAs shared a characteristically low frequency with the EEG’s Delta
rhythm. It’s also closely linked to how the subjects and equipment in an
experiment are moving around. Since MAs introduce significant variation
into the EEG data and distort the original shape of the signal, they must be
eliminated [41]. Due to this, incorrect diagnoses of diseases, false alarms, and
other problems can result from analyzing EEG data. Voltage changes occur
when the electrodes shift from their normal position or the cables sway while
the patient is in motion. Various factors, such as ground reaction forces,
cyclic motion, head movements, etc., can cause MAs. Therefore, raw EEG
data, which can range from high to low power spectral density, is severely
corrupted by MAs if recordings are made while the subject is engaged in
everyday activities.
 EEG artifact detection and removal techniques 147

9.3 ARTIFACT REMOVAL TECHNIQUES

Artifacts in EEG data are typically eliminated using EEG artifact removal
strategies. Due to the effectiveness of artifact removal techniques for EEG
data, the techniques can now be used in a wide variety of clinical and
industrial settings. There have been a number of obstacles in the way of
EEG artifact removal techniques. These difficulties may arise from the
nonlinearities of the unwanted signal being added to the EEG signal or from
the complexity of the methods themselves. The “nonlinear” nature of the
artifacts, for instance, makes it challenging to extract only the artifacts
without also losing actual neuronal data. In this literature, various artifact
removal algorithms are discussed in the following subsections.

9.3.1 Regression technique

Since the middle of the 1990s, a number of methods have been proposed to
remove different artifacts from EEG data, but the linear regression method
has gained popularity due to its ease of use. Each EEG channel is thought to
record a linear sum of natural brain activity and artifact signals. The signals
from these artifacts can be retrieved with the help of reference channels or
artifact templates. If the contaminated EEG signal is subtracted from a
regressed portion of the reference signal, the artifacts will be eliminated. As
such, the goal of regression algorithms is to provide an accurate estimate of
the optimal value that characterizes the extent to which the reference
channel (i.e., the artifact) affects the EEG signals [42]. In order to recover
EEG data from ocular noise, linear regression has been used extensively.
A common method for detecting artifactual samples and subsequently
removing them from the model is regression analysis [28,43], which em-
ploys a multi-modal linear model between the observed and a reference
signal. Ocular artifacts can be eliminated with the help of time-domain
regression [44]. Frequency domain regression [45] was also introduced, but
neither of these methods could deal with the inherent problem of
bidirectional contamination in EEG data [28]. This issue can be fixed,
however, by employing the filtering method first, followed by the regression
[46–49]. This strategy is justified on the grounds that filtering out the high-
frequency components of recorded EOG will drastically cut down on the
bidirectional contamination effect, as most of that content belongs to the
neuronal activity [48]. While methods based on regression and blind source
separation (BSS) [43] are able to eliminate the issue of bidirectional con-
tamination, they are severely constrained.
The regression technique was used to remove the ocular and cardiac
artifacts. But when there is no reference channel, such regression analysis
frequently fails. Regression methods have been replaced by more advanced
methodologies [25,50,51] due to their reliance on a reference channel,
which restricts their use to EOG and ECG primarily. In addition, the
148 Computational Techniques in Neuroscience

method assumes incorrectly that the neuronal activity in EEG and EOG
signals are uncorrelated, while attempting to remove artifacts using EOG
signals as reference [25,51]. Thus, the shared neuronal activity between
EEG and EOG can be removed from EEG signals using regression analysis.
There is currently no agreement in the scientific literature regarding the best
approach for low-pass filtering EOG signals. On the other hand, some
authors contend that neural activity pollutes every frequency range [52].
However, regression methods continue to serve as the benchmark against
which the efficacy of all other newly developed methods must be judged.

9.3.2 Filtering technique

Filtering technique is one of the popular artifact removal methods from the
EEG signal. Generally, two types of filtering technique are used in this case.
Classical filters can only be applied when the frequency bands of the noise
and the signal of interest do not substantially overlap. However, the fre-
quency bands of muscle artifact and brain signals often do overlap, which is
the main reason for failure of classical filtering in muscle artifact removal.
To overcome this limitation, there are a number of new filtering techniques
applied for artifact elimination, such as adaptive filtering [53] and Kalman
filtering [54]. In what follows, we briefly introduce the main two filtering
techniques.

9.3.2.1 Adaptive filtering

The adaptive filtering presupposes that the target signal and the artifact are
independent of one another. Based on a reference signal, the filter taps a
delay line to produce a signal that is correlated with the true artifact signal.
The estimated artifact is then subtracted from the recorded signal, and the
resulting residual is a proxy for the true signal [55]. When an artifact is
detected in the output and it is found to be correlated with the reference, the
filter coefficients will continue to adjust until the artifact is eliminated.
Successful application of adaptive filtering relies heavily on selecting an
appropriate reference signal [50] and also on the choice of an appropriate
algorithm to update the filter weights. The algorithms used to update the
weights are least mean squares (LMS) and variants of LMS, recursive least
squares (RLS) and variants of RLS, and state space RLS.
In the case of adaptive filtering, to update the filter weights, reference
signal is used as one of the inputs. Based on reference input, the adaptive
filter produces a signal, which is the correlated version of reference input.
The output of the filter is subtracted from the contaminated EEG signal to
produce an appropriate signal of interest. The electrooculogram (EOG)
signal is used as a reference to filter out ocular artifacts in EEG data [56],
and/or reference signals can be measured using an electrocardiogram (ECG)
for filtering out cardiac artifacts [57]. The artifacts in EEG signals can be
 EEG artifact detection and removal techniques 149

estimated most accurately with the help of an optimization algorithm. The

most widely used adaptive algorithm for modifying a weight vector is the
LMS algorithm [58]. Recursive least squares (RLS)-based adaptive filtering
is another popular algorithm [56,59].
The second-order nature of the RLS algorithm results in very quick
convergence. There are a number of papers that detail the benefits of both
families of algorithms, including least mean absolute value and sign (SIGN)
[60]. The RLS family of algorithms consistently outperforms the competi-
tion in terms of accuracy, but at the expense of a higher computational cost
[60,61]. Adaptive filters have a few advantages, such as online implemen-
tation, no preprocessing/calibration, and ease of use, but one of their lim-
itations is that they need a reference signal using additional sensors.
Other filters, such as Kalman, Wiener, and Bayes filters, can also be used
for artifact removal; however, these methods have not been extensively
explored in the literature of EEG artifact removal [25,62–65]. When
unwanted artifacts are present in the measured signals, Wiener filtering
can be used as a parametric technique to eliminate them [66]. Unlike
other algorithms, the Wiener algorithm can function without any kind of
external reference signal because it is based on statistical principles. It
is assumed that the signal and the (additive) artifact are both stationary
linear stochastic processes with well-characterized spectra or well-
established patterns of autocorrelation and cross-correlation. One dis-
advantage of the Wiener filter over the adaptive filter is that it requires
calibration before use and can’t be used in real-time applications. The
adaptive filter, on the other hand, does not necessitate any supplementary
hardware on the recording device.
Using a predictor-corrector strategy, Bayes filters can be built. The pre-
dictor describes the connection between states from one time sample to the
next using a time update model. Afterwards, a measurement model is used
in the corrector phase to characterize the connection between the external
data and the internal state. The Bayes filter technique, like Wiener filtering,
can eliminate the embedded artifact without the use of a reference signal. In
contrast to the Wiener algorithm, Bayes filtering can function in real time.
Several approaches exist for realizing the various approximations of the
Bayes filter technique used in modern signal processing. Since the algorithm
for the Bayes filter is computationally intractable, the filtering method itself
is not implemented. Bayes filter is implemented by Kalman filter [67–70]
and Particle filter [71,72].
To estimate a process, the Kalman filter employs feedback control by first
making an estimate of the state of the process at a given time and then
receiving information about the process in the form of (noisy) measure-
ments [67]. Kalman filters can be used in a variety of ways to remove ar-
tifacts [73–75]. The first step is to create a model of both the target signal
and the noise. The recorded signal can be understood in terms of the sum
of these two model signals, allowing one to infer both the process and
150 Computational Techniques in Neuroscience

measurement models. Kalman filters are a great alternative to adaptive

filters if process and measurement models for the necessary system are
already available.
The use of the particle filter to eliminate artifacts in physiological signals
[76] is an area that has seen surprisingly little research. The particle filter,
on the other hand, can remove artifacts in a way that’s analogous to the
Kalman filter’s success. For optimal performance, Kalman filters necessitate
sensors with both high accuracy and a high rate of update. When compared
to other Bayes’ method-based filtering techniques, they are one of the most
time- and space-efficient. However, if precise sensors are unavailable or
if a comprehensive model of the process and measurements is lacking,
particle filtering emerges as the preferred method. Particle filters are a
versatile tool with low implementation overhead because they don’t need
a detailed model of the systems [77].

9.3.3 Decomposition technique

Separating the EEG and artifact sources into distinct components and
then eradicating them during reconstruction is another denoising method.
Techniques for decomposing EEG data include BSS, wavelet transform
(WT), empirical mode decomposition (EMD), and variational mode
decomposition (VMD).

9.3.3.1 Techniques of blind source separation (BSS)

BSS is based on a wide variety of unsupervised learning algorithms [78]
with the intention of estimating sources (which are not necessarily inde-
pendent) and mixing system parameters. Because it can decouple the source
signals of neuronal activity from the artifacts, it is one of the most well-
known and widely used methods for extracting EEG data [25,26,50]. When
it comes to blending multiple sources together, BSS shines because it doesn’t
necessitate any prior knowledge (or, in some cases, uses very limited
knowledge). Independent component analysis (ICA), canonical correlation
analysis (CCA), morphological component analysis (MCA), and indepen-
dent vector analysis (IVA) are just some of the algorithms available for BSS.

9.3.3.1.1 Independent component analysis (ICA)

ICA separates multichannel EEG data into its constituent parts called
independent components (ICs). ICA is used on the premise that the signals
coming from various sources are linearly mixed. Because it avoids the
problems that plague parametric methods like adaptive filtering, ICA has
recently emerged as a useful tool for cleaning up EEG data. EEG artifact
removal studies employ ICA more frequently than any other BSS technique
[79–84]. JADE [16], fast ICA, SOBI, InfoMax [85], constrained ICA [86],
 EEG artifact detection and removal techniques 151

AMICA [87], and AMUSE [88] are just a few of the ICA variants proposed
by researchers for EEG artifact removal.
The necessity of using non-Gaussian data for ICA is a significant limi-
tation of the method. If the sources are not Gaussian, ICA can be used to
estimate them. ICA can only account for a single Gaussian component,
which can be estimated as the residual after all other independent compo-
nents have been extracted. The Gaussian or non-Gaussian nature of a given
component is rarely known in advance. Normalizing the data is a standard
practice before ICA is calculated. Another drawback of ICA is that it
requires a number of channels that is at least as large as the number of
sources, so it can’t be used with just one or a small number of channels.

9.3.3.1.2 Canonical correlation analysis (CCA)

In order to discover the nature of the connection between two datasets,
CCA was developed as a statistical technique. Second-order statistics (SOS)
are utilized by the CCA method to generate components that are intrinsi-
cally uncorrelated. CCA eliminates the BSS issue by making all sources
maximally autocorrelated and uncorrelated with one another [89].
The recorded EEG data and its time-delayed version are used as the first
data set and second data set, respectively, for CCA‘s application to the BSS
problem. CCA is a solution to the BSS problem because it ensures that the
canonical variates are maximally correlated across both data sets while
remaining independently correlated within each [90]. Since muscle artifacts
have a lower autocorrelation than other types of artifacts, they can be re-
moved from a reconstructed image by setting the autocorrelation of the
final several source components to zero.

9.3.3.1.3 Morphological component analysis (MCA)

MCA functions by breaking down the recorded signal into sub-signals with
distinct morphological properties. A collection of waveforms (called atoms)
is used to describe the various underlying signals, and these are sparsely
represented in an overcomplete dictionary. Consequently, the total signal is
simply the product of these atoms multiplied by their respective coefficient
vectors in a linear fashion. The MCA algorithm was used to clean up a
single-channel EEG recording from artifacts like brow furrowing, jaw
clenching, swallowing, and blinking eyes by the authors of [91].

9.3.3.1.4 Principal component analysis (PCA)

Principal component analysis (PCA) is an often-used BSS technique whose
algorithm relies on the eigen values of a covariance matrix [22]. The first step
in this process is to use orthogonal transformation to transform any corre-
lated variables into independent ones. These sets of independent variables are
152 Computational Techniques in Neuroscience

referred to as principal components (PCs). EEG signal PCs will be im-

plemented via single value decomposition (SVD). According to [92], PCA of
variance is used to extract the main components standing for blinks and eye
movements. After that, we used an inverse operation to eliminate the asso-
ciated components and obtain the clean EEG data. It was shown in [93] that
PCA is a more computationally efficient alternative to linear regression. On
the other hand, it can be challenging to ensure that artifact components are
independent of EEG readings. PCA also fails to discriminate interferences
when the potential of drifts and EEG data are similar. The subsequent
studies, therefore, favor more adaptable approaches like ICA.

9.3.3.2 Wavelet transform (WT)

WT is one of the decomposition techniques that acts as a sub-band filter.
Use of this transform eliminates high-frequency signal to restore the low-
frequency signal and is applied to remove the artifacts from EEG signal.
The artifacts, such as ocular artefacts in EEG [94–96] and EMG artifacts in
fNIRS. Decomposing the signal into shifted and scaled versions of the
wavelet basis [97] is the first step in the method for eradicating artifacts in
EEG, after which the remaining signal components are cleaned up and re-
constructed. The time-frequency analysis of EEG signals commonly em-
ploys the complex Morlet wavelet as the wavelet basis [98]. Although the
discrete wavelet transform (DWT) is a fast and effective method, its major
drawback is its time-variance, which is very important for EEG data [99].
This is because DWT uses iterative high and low pass filters to represent
signals into its discrete wavelets. Although slower than DWT [100], sta-
tionary wavelet transform (SWT) gets around DWT’s weakness in trans-
lation invariance. The key distinction between the two is that DWT uses a
sampling factor of 2j 1 to reconstruct the signal while SWT uses a sam-
pling factor of 2j to down-sample the filter output before reconstructing it.
When the measured signal’s spectral properties overlap with the spectral
properties of the artefacts, wavelet-based methods are unable to remove the
artifacts completely [25,26].

9.3.3.3 Empirical mode decomposition (EMD)

As similar to time-frequency WT technique, EMD is another decomposition
method for this application. This method is an adaptive and flexible data-
driven method that splits a time-domain signal into a collection of intrinsic
mode functions (IMFs) [101,102]. Artifacts in EEG data can be removed
using EMD alone [103] or in conjunction with other techniques [101,104].
Due to its sensitivity to background noise, EMD has undergone refinements
to address the challenges posed by mode mixing. A noise-assisted data
analysis method, enhanced EMD (EEMD) is developed with the average
number of IMFs from EMD as the optimal IMFs [105,106].
 EEG artifact detection and removal techniques 153

9.3.3.4 Variational mode decomposition

Modification of EMD is termed variational mode decomposition (VMD),
where variation of center frequency needed to generate IMFs. It is a relatively
new non-recursive signal decomposition method that can be used to infer the
structure of a non-stationary signal [107,108]. VMD is a method for
breaking down a multi-component EEG signal into smaller, more manage-
able pieces, or “modes,” each of which has its own unique spectral density.
Despite being effective at reducing single-channel OAs, most methods for
eliminating OAs also distort important clinical aspects of the EEG signal in
the process. In [109] the authors propose a reliable framework for the
detection and removal of OAs using VMD and turning point count to address
these concerns. The VMD procedure consisted of two phases, VMD-I and
VMD-II. Rejecting low-frequency baseline components from the raw EEG
signal using VMD-I; dividing the processed EEG signal into three modes
using VMD-II; and finally, rejecting the mode that contains OAs using a
threshold based on the number of turning points in the signal [109].
To process signals in an adaptive and non-recursive way, VMD is a
popular choice. To a large extent, the outcome of signal decomposition is
controlled by two parameters in VMD. Parameter-related problems are
reduced thanks to a revised VMD based on the squirrel search algorithm
(SSA). The method is used to fine-tune VMD’s critical settings. To filter out
eye movements and other visual noise from EEG recordings, the GOSSA-
VMD model is developed [110].
Although the amplitude of an eye blink is greater than that of a typical
EEG signal, a standard peak-detection algorithm applied to the raw EEG
signal would likely generate numerous false positives. Blink regions in IMFs
were sought after by researchers using an extension of VMD [107], called
multivariate variational mode decomposition (MVMD) [111,112].

9.3.4 Machine learning technique

Algorithms based on machine learning can also be used to clean up EEG
data that has been contaminated by artifacts. In order to detect artifacts in
EEG data using traditional methods, trained observers are required. Using
machine learning based algorithms, the artifactual patterns can be detected
and eliminated mechanically. This has the potential to greatly improve the
integrity of the original EEG data while simultaneously enhancing artifact
removal accuracy. Artificial neural networks (ANN) and support vector
machines (SVMs) are the most common machine learning based algorithms.
Automatic artifact removal using linear SVMs is a computationally cheap,
robust, and extensible method [113]. In addition to ECG artifacts, ocular
and muscular artifacts can be eliminated with ANN [114]. Recently the
deep learning methods used for removal of artifact from the EEG signal
[115,116].
154 Computational Techniques in Neuroscience

9.3.5 Combined approach for artifact removal

To exploit the advantage of each method, recently, several researchers have
opted to use a hybrid strategy, which is a combination of two or more
methods. This categorization is based on how many individual algorithms
are involved in the pre-processing pipeline. Popular single methods for EEG
artifact removal include linear regression, adaptive filtering, WT, BSS, and
empirical mode decomposition (EMD). On the contrary, the hybrid
methods involve a combination of these single methods. Some major hybrid
methods are discussed below.

9.3.5.1 Blind source separation and adaptive filtering

This hybrid technique is a combination of the BSS (BSS: ICA) and adaptive
filtering. When it comes to eliminating artifacts, adaptive filtering is used to
find them while BSS is responsible for locating them. Decomposing EEG
signals into ICs is a task for ICA. The neuronal information contained in
artificial ICs is preserved through further processing by an adaptive filter. In
order to effectively remove artifacts from EEG data, combined adaptive
filtering with ICA to create a hybrid method [43]. Cardiac artifact was
removed using adaptive noise canceller and ICA [117]. ICA and adaptive
noise cancellation (ANC), or ICA-ANC, are the foundation of this tech-
nique. As a means of determining the reference signal for ANC, ICA is
applied to a small number of EEG signals. As the method only requires a
small number of EEG channels and no synchronous ECG channel, it is well
suited for use in portable BCIs [118]. An unsupervised and fully automatic
method of detecting and eliminating visual artifacts using a combination of
canonical correlation analysis and a multi-channel Wiener filter
(ACCAMWF). Segments of artifacts can be automatically annotated using
the spatial distribution entropy (SDE) and the spectral entropy (SE). After
that, the clean EEG data are supplemented with neural signal extracted
from artifact-contaminated data using the CCA algorithm [119].

9.3.5.2 Adaptive filtering and wavelet transform

To get rid of these ocular artifacts, [120] proposes a method that combines
DWT and ANC, where the OA reference is derived from DWT decompo-
sition and then used in the adaptive filter as reference. Ocular artefact
removal using Multiresolution Analysis and Adaptive Filtering [MRAF]
was used in [121]. To begin, discrete wavelet transform (DWT) is applied to
the EEG signal in order to effectively help localize the epileptic region.
These decomposed wavelet components are then subjected to multi-
resolutional soft thresholding to smooth out any peaks and valleys. Low-
frequency components representing physiological artifacts are also filtered
out using adaptive filtering.
 EEG artifact detection and removal techniques 155

9.3.5.3 Technique of BSS and WT

Since the number of sources in ICA must be equal to the number of sources in
the measurement, and the WT fails when the artifacts overlap in the spectral
domain, a wavelet-ICA technique has been proposed in [81,122–124] to
combine the good parts and avoid the bad ones. When processing
EEG data, WICA is employed to filter out extraneous signals like EMG
and to clean up single-channel recordings of the brain’s electrical activity.
After the EEG data has been recorded, the WT is used to decompose it, and
the resulting resolution, which contains likely artifactual components, is then
fed into an ICA algorithm of choice. Finally, using the preserved wavelet
components and the disposed components, artifact-free reconstruction of the
EEG signals is carried out.
An Automatic Wavelet Independent Component Analysis (AWICA) for
an automatic artifact rejection from multichannel scalp EEG [125].
Firstly, the input EEG will be decomposed using the DWT to partition
each channel of the original data set into four major bands of the brain
activity. Each rhythm of each channel is represented by a Wavelet
Component (WC). Each of the WCs will be channeled to ICA analysis
to concentrate the artifacts into a few ICs. The resulting artifactual
Wavelet Independent Components (WICs) are automatically detected and
removed. Subsequently, the reconstruction of the clean signal involves
two steps, which are inverse ICA and inverse DWT.
Wavelet is frequently used in conjunction with ICA for artifact removal
in EEG [126]. Artifacts are identified using ICA to find the ICs. In order to
obtain an EEG free of artifacts, we will flag the offending ICs as critical,
denoise the signal using DWTs, and then rebuild the ICs using the corrected
data. When trying to get usable neural signals out of artifact ICs, a lifting
wavelet transform (LWT) can help [127]. Since ICA-LWT does not neces-
sitate intricate computation, it can be used in real-time settings. An
unsupervised eye blink artifact removal with modified multiscale sample
entropy (mMSE), kurtosis, and wavelet-ICA is proposed by authors in
[128,129]. After the decomposition of ICA, mMSE is computed for all the
ICs. Normal distribution is used to detect the artifact. ICs with mMSE
lower than the threshold will be marked for wavelet correction. The ICs
computed will undergo the same process for Kurtosis. Finally, the marked
ICs will be denoised using biorthogonal wavelets.

9.3.5.4 Technique of EMD and BSS

The EEG signals are first subjected to EMD in order to obtain IMFs; then,
the IMFs are subjected to the BSS method, which detects and removes
artifactual components [103]. By iteratively summing the remaining IMFs,
the original signal can be reconstructed from scratch without any artifacts
[130]. This is accomplished by multiplying the mixing matrix by the
156 Computational Techniques in Neuroscience

extracted IC components. Removing EMG and ocular artifact from EEG

using EEMD in conjunction with ICA was first investigated in [131].
EEMD-ICA, EMD-CCA [132], EEMD-CCA [133–136], BSS-EMD,
EEMD-SCICA [137], CEEMDAN-ICA-WTD [138], and multivariate em-
pirical mode decomposition (MEMD) and CCA (MEMD-CCA) [139] used
for removal of artifact from the EEG signal.

9.3.5.5 Adaptive filtering and EMD

The adaptive filtering and EMD are the two main components of this new
method. Adaptive filter (using RLS algorithm) and empirical mode decom-
position (EMD) are used to filter out ECG artifacts in EEG data [140].

9.3.5.6 Technique of BSS and SVM

BSS-SVM is a hybrid approach that can be used to further BSS’s applica-
tion. To begin, BSS algorithms are used to decompose the recorded EEG
data into multiple components. The next step is to extract a variety of
component features, including temporal, spatial, and statistical character-
istics. The features are then fed into a linear SVM classifier to determine
what kinds of artifact parts are present. Signals free of artifacts are re-
constructed using the remaining components [22,25,141,142].

9.3.6 Summary of earlier methods used for

EEG artifact removal
As number of artifact removal methods are discussed in previous section,
for ease of observing all the above methods in terms of their methods,
Performance parameters taken into consideration, way of acquiring EEG
data are given in tabular form.
A summary of artifact removal techniques for filtering techniques is
provided in Table 9.1. Decomposition techniques such as EMD and VMD
and combined techniques that include EMD and VMD are provided in
Table 9.2. Source separation algorithms and transformed methods such as
WT, ST, and TWT are provided in Table 9.3. And recent machine learning
techniques applied for artifact removal from EEG signals are provided in
Table 9.4. It is observed that most of the authors collected EEG data from
openly available databases for verification of their artifact removal algo-
rithm. Some authors collect EEG data from the hospital, and some record
EEG data using the recording machine available to them. The main problem
with collecting EEG data is that appropriate data may not be available for
removing particular types of artifacts. For recording appropriate EEG data
to verify the efficacy of the artifact removal algorithm, a proper set of EEG
recording machines is required, which is not cost-effective. In this regard,
researchers depend on other sources for collecting EEG data. For the
Table 9.1 Filtering techniques utilized in EEG artifact removal process
Author Method Type of artifact Performance measure Data Base
C. S. Kim et al. [ 118] Independent component analysis Ocular RMSE and CC BCI Competition IV and
(ICA) and adaptive filter (AF) Dataset IIIa of BCI
Competition III
Bo Hua [ 143] LMS (least mean square) based Eye movement SNR, MI, Coherence Recorded
algorithm
S. Blum et al. [ 144] Artifact Subspace Eye blink SNR, Sensitivity, Data recorded using
Reconstruction (ASR) Computation time smart phone
EGLE BUTKEVIČI ˙ Baseline estimation and denoising Movement artifact Pearson CC Data recorded using
UT¯ E et al. [ 145] with sparsity (BEADS) filter smart phone
algorithm
S. Kohli and A. J. Moving averages and adaptive transcranial Direct Current SNR Recorded
Casson [ 146] filtering Stimulation (tDCS) and
transcranial alternating
current artifact
A. J. Mohammad ICA and ANC ECG artifact RRMSE and frequency Mitsar amplifier and
Ali B. [ 117] correlation WinEEG software
M. Chavez et al. [ 147] Surrogate-based artifact removal ocular and muscular SNR and RRMSE Recorded
(SuBAR) artifacts
M. Miao et al. [ 119] CCA-MWF (ACCAMWF) EOG RMSE and MI Semi-simulated EEG/EOG
dataset(Mendelay)
S. Sharma and DFT and ACMD Ocular CCN, SNR, AEM, AEMAX, Mendeley database, MIT-
U. Satija [ 148] MAXN, RDN, RDP, BIH Polysomnographic
RMSE, SNRW, CCW, η database and EEGMAT
EEG artifact detection and removal techniques

database
(Continued)
157
 158

Table 9.1 (Continued) Filtering techniques utilized in EEG artifact removal process
Author Method Type of artifact Performance measure Data Base
G. S. Spencer RLAS Motion artifact RMSE Recorded
et al. [ 39]
A. Kilicarslan Adaptive de-noising framework Motion artifact PSD Recorded
et al. [ 149] with H∞ adaptation rule
R. Ranjan et al. [ 150] Modified-EMD and LoG filter Motion artifact ∆SNR, γ, mean absolute Semi-simulated EEG data,
error in PSD of δ-band mobile brain-body
(MAEδ PSD), MI, imaging (MoBI) real-time
percentage improvement EEG dataset with BCI
in correlation [corr (%)], task and synthetic
percentage improvement dataset
in coherence [coh(%)],
power spectral distortion
[PSDdis (%)] and
execution time
Computational Techniques in Neuroscience
Table 9.2 Summery of decomposition techniques utilized for EEG artifact removal process
Author Method Type of artifact Performance measure Data Base
Shivam Sharma and DFT and ACMD Ocular CC, SNR, AE, AE MAX, Mendeley database, MIT-BIH
Udit Satija [ 148] MAXN, RDN, RDP, RMSE, polysomnographic database,
SNRW, CCW, percentage EEGMAT database
reduction in coefficient of
correlation (η), M I
Md S Hossain, VMD, VMD-PCA, VMD-CCA Motion artifact SNR and percentage PhysioNet
et al. [ 151] reduction in motion artifact
R. Ranjan, et al. [ 150] Modified EMD and optimized Motion artifact SNR, signal to artifact gain Semi-simulated EEG data
LoG filter coefficient, PSNR, SAR, contaminated with motion
MAE, MI, improvement in artifacts, mobile brain-body
correlation, improvement in imaging (MoBI) real-time
coherence, power spectral EEG data
destruction and execution
time
Miao Shi, et al. [ 110] GOSSA-VMD Ocular SNR, RMSE and CC Semi-simulated EEG dataset
C. Kaur, et al. [ 152] VMD-DWT and VMD-WPT SNR, PSNR, and MSE. Data collected from Hospital
University Sains Malaysia
(HUSM)
L. Chang, et al. [ 153] MVMD-CCA Accuracy and ITR Simulated data using Matlab
M. Saini, et al. [ 109] VMD in two stages denoted as Ocular Sensitivity, NCC, SNR, Mendeley database, MIT-BIH
VMD-I and VMD-II MAE,MAX,NMAX,NRD, polysomnographic database
PRD, and RMSE and EEGMAT database
R. Gavas, et al. [ 112] MVMD Eye blink SER, correlation, variance- Synthetically generated EEG
based metric (V), percentage data, Covert Shift Dataset,
change in band power, Cog Beacon Dataset
EEG artifact detection and removal techniques

classification accuracy
159

(Continued)
 160

Table 9.2 (Continued) Summery of decomposition techniques utilized for EEG artifact removal process
Author Method Type of artifact Performance measure Data Base
C. Dora and Modified VMD ECG artifact SAR, CF MIT/BIH polysomnography
P. K. Biswal [ 154] data
Q. Li, et al. [ 138] CEEMDAN-ICA-WTD Ocular RMSE Recorded
C. Dora and VMD-based algorithm Ocular PSD Capslpdb and sleep-edf of
P. K. Biswal [ 155] Physionet
A. Yadav and EEMD and SCICA Ocular MI, CC and Coherence physionet.org
M. S. Choudhry [ 137]
Yan Liu, et al. [ 156] NALSMEMD Motion artifact Percentage change in Artifact physionet.org
X. Chen, et al. [ 133] EEMD-CCA Muscle artifact SNR, RMS, RRMSE Recorded and simulated data
A. Egambaram, EMD and CCA Eye blink Accuracy (VMI), Error (VMI), Recorded
et al. [ 132] CC, RMSE, Time of
execution
MD E. Alam, EMD Eye blink and SNR BCI2000
Computational Techniques in Neuroscience

et al. [ 157] power line

noise
K. K. Dutta, et al. [ 158] EEMD Muscle artifact Recorded
K. T Sweeney, EEMD-CCA Motion artifact ∆SNR and percentage Recorded
et al. [ 135] reduction in artifact λ
S. Tavildar and MEMD-CCA Motion artifact Correlation the percent Recorded
A. Ashraf [ 159] reduction in artifacts λ
Table 9.3 Methods of wavelet transform and blind source separation
Author Method Type of artifact Performance measures Source of EEG data
R. Mahajan and B. I. ICA and DWT Eye blink Sensitivity, specificity, agreement Recorded
Morshed [ 128] rate, MI and Cross Power spectral
density
Chi Zhang et al. [ 160] DWT and ICA EOG and EMG Correlation scores and accuracy Recorded
Charvi A. Majmudar DWT Ocular TFA, Magnitude Square Coherence Recorded
et al. [ 161]. (MSC), CC, MI
S. A. Gaikwad ICA(SOBI) and DWT Ocular SDR, Variance and RMSE Recorded
K.P.Paradeshi [ 162]
R. Upadhyay ICA-DOST Ocular By visual Inspection Recorded and simulated
et al. [ 163]
Xiaobai CAI and WT and HT Eye blink Visual inspection Collected from Salk
Junjun CHEN [ 164] computational neurobiology
Laboratory of California
University
B. Somers and A. CCA Eye blink SER, ARR, and SNR Recorded
Bertrand [ 165]
Mst. Jannatul Ferdous Lifting WT Eye blink MSE and SAR BCI Database
et al. [ 166]
Sim Kuan Goh ICA Multiple SNR, CSR Synthesized EEG artifacts
et al. [ 167]
Anwesha Khasnobish KNN -DWT Head movement artifact SD Recorded
et al. [ 168]
EEG artifact detection and removal techniques

(Continued)
161
 162

Table 9.3 (Continued) Methods of wavelet transform and blind source separation
Author Method Type of artifact Performance measures Source of EEG data
Matteo Dora, David WT Different kinds of artifact NMSE, ∆R, ∆SNR Semi-simulated EEG
Holcman [ 169]
Chi-Yuan Chang ICA and ASR Muscle, eye-blink, and RMS Driving simulator EEG data
et al. [ 170] lateral eye-movement
activities
S. R. Sreeja MCAand KSVD Eye blink RMSE, SAR, CC, MI and MSE Recorded
et al. [ 171]
Nitesh S. Malan Shiru DTCWT Ocular RRMSE Recordeded using NI LABVIEW
Sharma [ 172] 2015 Biosignal toolkit
Chong Yeh Sai SVM and WICA Eye blink CC Recorded and simulated
et al. [ 113]
A. J.M. Ali ICA-ANC ECG artifact RRMSE and frequency correlation Recorded
Computational Techniques in Neuroscience

Badamchizadeh
et al. [ 117]
Pranjali Gajbhiye MTV, MWTV Motion artifact Difference in SNR and η physionet
et al. [ 173] and DWT
Md. Kafiul Islam Amir SWT Chewing, swallowing, SNR, ∆SNR, ∆RMSE, ∆PSD, ∆corr, BCI competition-IV Scalp EEG
Rastegarnia [ 174] eye blinks, subject ∆SNDR Database: dataset-1, dataset-2a
movements, talking, and dataset-2b
Soojin Lee et al. [ 175] JBSS and quadrature High-amplitude Root mean square (RMS), relative Recorded and simulated
regression and q-IVA stimulation artifact root-mean-squared error
(RRMSE), correlation coefficient
(CC), power deviation (Pdev)
Mohamed F. Issa and WEICA EOG artifact Signal-to-noise ratio ∆SNR, RMSE, Klados EEG dataset and
Zoltan Juhasz [ 176] MSC, Percentage of artifact Recorded dataset
removal
Abhijit Bhattacharyya TQWT Cortical stimulation (CS) MSE and CCI Simulated data and Data
et al. [ 177] Collected from Nancy
University Hospital (CHU
Nancy), France
Vandana Roy Shailja ICA,CCA,DWT Motion artifact ∆SNR, RMSE Online open source interface
Shukla [ 178] and SWT
Young-Eun Lee cICA with cIOL Movement artifact AUC(Area under the ROC curve) Recorded
et al. [ 179] and SNR
Chi-Yuan Chang ASR Eye blink and Muscle Retained power Recorded
et al. [ 180] artifact
Dhanalekshmi P. DWT and MRAF Ocular Sensitivity, specificity, Accuracy and CHB MIT scalp EEG dataset
Yedurkar, Shilpa P. precision chb01 15
Metkar [ 121]
K. Jindal et al. [ 181] FPIC and GLCT Eye blink, EOG, SNR, RMSE, NMAX and NRD Simulated and recorded
muscular and other
high frequency artifact
Nikesh Bajaja WPD Muscle, motion and MI, CC, PSD Recorded
et al. [ 182] ocular artifact
K. P. Paradeshi and Enhanced WICA Ocular ADP, RMSE, PSNR, CC Recorded
U. D. Kolekar [ 183]
Sayedu Khasim Ov-ASSA -ANC Ocular RRMSE, MAE CHB-MIT
Noorbasha, Gnanou
Florence
Sudha [ 184]
EEG artifact detection and removal techniques

(Continued)
163
 164

Table 9.3 (Continued) Methods of wavelet transform and blind source separation
Author Method Type of artifact Performance measures Source of EEG data
Ian McNulty WT using the SURE Ocular NMSE and SNR Bonn database
et al. [ 185] Shrink algorithm
with the hard
thresholding
S. T. AUNG AND M-mDistEn Motion artifact Accuracy and p-values PhysioNet Database
Y. W.WAT [ 186]
Pranjali Gajbhiye WOSG filtering Motion artifact NMSE, ∆R, ∆SNR Two publicly available databases
et al. [ 187]
Zainab Jamil ICA-DWT Eye movement artifact MI, Sensitivity and specificity Recorded
et al. [ 188]
S. Phadikar WT With Eye blink CC, NMSE, SSIM
et al. [ 189] Heuristically
Optimized threshold
Computational Techniques in Neuroscience

M. Shahbakhti SWT Electrical shift and linear CC, PSNR, NRMSE Mendelay
et al. [ 190] trend artifacts (ESLT)
Sridhar Chintala Mixed step size Ocular MSD Recorded
et al. [ 191] normalized least
mean fourth adaptive
algorithm
Salim Çınar [ 192] ICA-ANC Ocular RE, CC, SAR, SNR, Sensitivity, ERP-based Brain-Computer
Specificity, and AUC Interface (BCI) Records dataset
Christos Stergiadis BSS Ocular Entropy, MIR, CC, Execution time Real recorded data and semi
et al. [ 193] simulated data
Ruisen Huang DSMF Motion artifact SDR, NMSE Recorded
et al. [ 194]
Mary Judith A. MD-SVD -ICA SNR, PSNR, MSE MIT EEG database
et al. [ 195]
Velu Prabhakar LOF and ASR Newborn non FTR, SME, F1 Score Data collected from Neonatal
Kumaravel stereotypical artifacts. Neuroimaging Unit CIMeC,
et al. [ 196] University of Trento) and
simulated data
Yuheng FENG SSA-CCA Muscle artifact PSD and mean time cost Semi simulated data
et al. [ 197]
Sagar S. Motdhare, SSA-MEMD EMG artifact Visual inspection Recorded
Dr. Garima
Mathur [ 198]
A. K. Maddirala and K. SSA-CWT and k- Eye blink RRMSE, SNR, RMS, CC, artifact Fatigue EEG database
C. Veluvolu [ 199] means clustering reduction ratio, MAE, precision
and accuracy
J. Yedukondalu and L. CiSSA-DWT EOG artifact SAR, MAE, RRMSE, and CC Dataset 2a from the BCI
D. Sharma [ 200] Competition IV
EEG artifact detection and removal techniques
165
 166

Table 9.4 Summery of machine learning technique in the field of artifact removal from EEG signal
Author Method Type of artifact Performance measure Data Base
C. Burger et al. [ 201] ICA and WNN Ocular PSD, RMSE and frequency Simulated and recorded data
correlation collected from motor imagery test
W. Suna et al. [ 202] 1D-ResCNN model EMG, ECG, EOG SNR, RMSE CHB-MIT Scalp EEG Database
S. K. Sahoo and EMCD and ODCN Ocular MAE, SNR BCI competition IV database
S. K. Mohapatra [ 203]
R. Ghosh et al. [ 204] kNN classifier and a Eyeblink and muscular CC, SAR, MAE, NMSE, SSIM Recorded
LSTM network
B. Yang et al. [ 116] DLN Ocular PSD, RMSE and EEG “Data sets 1” for BCI Competition IV
classification accuracy
M.H. Quazi et al. [ 205] FLM optimization- EMG, ECG, EOG MSE, SNR PhysioNet
based learning
algorithm for neural
network-enhanced
adaptive filtering
Computational Techniques in Neuroscience

model
X. Li et al. [ 206] Discriminative model Ocular Recorded
for joint OAC and
feature learning
S. Behera and WVFLN Ocular MSE, RMSE, Mendeley database
M. N. Mohanty [ 115]
S. Behera and RVFLN model Ocular and cardiac MSE, NMSE, RE, GSAR, SNR, Mendeley database
M. N. Mohanty artifact IQ, and INPS
 EEG artifact detection and removal techniques 167

removal of cardiac artifacts, a proper data set is not available, so re-

searchers create a synthesized version of an ECG artifact-contaminated
EEG signal to verify the artifact removal algorithm. In [36], the authors
recorded the cardiac artifact-contaminated EEG data from a small animal
(a rat), and it is also practically very difficult to collect real-time EEG data.
The researchers also rely on semi-simulated data. In the case of a semi-
simulated dataset, the artifact is synthetically added to the clean EEG signal
to form contaminated EEG data.
Another problem in the artifact removal process is the validation mea-
sure. It is not possible to verify the performance of an algorithm visually by
observing the output signal obtained. Due to this, the researcher considers
different types of measurement parameters to verify the performance of the
artifact removal algorithm. It is observed from the tables that the per-
formance measurement parameters are calculated by various authors. There
is no fixed number of performance parameters that should be taken as a
standard parameter in the case of artifact removal. It is also observed that in
recent papers the authors considering a greater number of performance
parameters for verifying their artifact removal algorithm.

9.4 PROPOSED TECHNIQUE

As the artifacts are unwanted and need to be removed prior to the infor-
mation extraction, an automatic way of removing artifact is proposed in
this paper. In the literature review section, some of the automatic artifact
removal techniques are discussed. Most of the automatic EEG artifact
removal methods are based on transforming the EEG signal to another
domain. However, when the signal is transformed from the time domain to
another domain it suffers from the problem of time-frequency resolution.
To avoid the problem of time-frequency resolution the Fast Discrete S
Transform (FDST) is proposed in this work.

9.4.1 Fast discrete S transform (FDST)

The general S transform given in equation (9.1) can be written as:

n N 1
j 2 kn
Seeg jT , = k =0
Seeg (kT ). w (kT , ). e N (9.1)
NT

Where Seeg (kT ) is the EEG signal, k = [0, 1, 2, …. N 1] and N is the

sample size of the EEG signal, T is the sampling interval of the signal
considered. For implementing FDST the Fourier transform of EEG signal is
obtained in the first step. Required kernel functions and the window
function are calculated in the second step [207,208]. The bandpass filter is
168 Computational Techniques in Neuroscience

applied to the Fourier transform of the signal, and then inverse Fourier
transform is applied to obtain S eeg . In the last step, FST is calculated for
each kernel function, which are given in equation (9.2) and (9.3).

3n N 1 3n + 3n
Seeg jT , = k =0
S eeg (kT ). w kT T, . kT , (9.2)
4NT 4NT 4NT

3n N 1 3n 3n
Seeg jT , = k =0
S eeg (kT ). w kT T, . kT , (9.3)
4NT 4NT 4NT

From the above equations + and are the kernel functions, is equivalent
to in DOST. S eeg (kT ) is obtained by taking inverse Fourier transform of
the bandpass version of the original signal. The inverse of FDST is obtained
by the equation (9.4)

n+1 1 M 1 n i 2 jl 1
Seeg = Seeg jT , .e M . (9.4)
NT N j =0
NT W ( 1
NT
, )
( ) n+1
Where Seeg NT is the Fourier transform of the EEG signal, and to get back
the original signal we need to apply inverse Fourier transform. FDST and
DOST use DFT for their operation and can be represented as:

k
DOST = i =1
Bi DFT (9.5)

Where direct sum matrix Bi forms a block diagonal matrix. Each block in
direct sum matrix is composed of a phase correction factor coming from ei
and dimensional inverse Fourier transform. The DOST basis functions are
compact in frequency, but not in space. DOST coefficients carry symmetry
property due to higher frequencies that are required in frequency, and in
space DOST suffers from ringing effect. The problems that arise in the case
of DOST and FDST can be overcome by DCST. The mathematical overview
of DCST is given in the following subsection.

9.5 RESULT AND DISCUSSION

The proposed algorithm is tested by considering the clean EEG signal and
two types of contaminated EEGs. The clean EEG is collected from the
Mendeley data base [209], as shown in Figure 9.1.The length of the clean
EEG signal is 6001, and its amplitude is within the range of −30 µV to
+30 µV. The clean EEG signal is considered here as a reference to validate
 EEG artifact detection and removal techniques 169

Figure 9.1 Clean EEG signal taken from Mendeley database.

Figure 9.2 Contaminated EEG signal taken from Mendeley database.

the performance of the transform-based algorithm. One type of contami-

nated EEG collected from the Mendeley data base is the semi simulated
EEG. In the case of semi simulated contaminated EEG, the artifactual EEG
is obtained by adding the vertical and horizontal eye movement artifact by
following the appropriate procedure and is only meant to verify the artifact
removal algorithms. The contaminated signal is shown in Figure 9.2 and
is of 6001 sample size. The amplitude of the contaminated EEG is within
−60 µV to + 110 µV. By visual inspection it is clearly distinguished from
the clean EEG signal. Another artifactual EEG signal collected from the
PhysioNet database [210] is shown in Figure 9.3, and its amplitude is
within the range of −600 µV to + 1600 µV. It is observed that at the time
of occurrence of artifact, the signal amplitude drastically increased from
170 Computational Techniques in Neuroscience

Figure 9.3 Artifactual EEG signal taken from PhysioNet database.

Figure 9.4 Fast DST of clean EEG signal.

its normal amplitude. Due to artifact the EEG signal shape is drastically
changed, and it is removed by the proposed algorithm.
The FDST of clean EEG and contaminated EEG are shown in Figure 9.4
and Figure 9.5, respectively. From the Figure 9.4 and Figure 9.5 it is
observed that at the time of occurrence of artifact, the phase change is more
prominent. More phase changes are indicated by the red and yellow color.
Figure 9.6 shows the clean EEG signal, contaminated signal, and the
signal after removal of artifact. It is observed that the recovered EEG signal
amplitude is lesser than the amplitude of the contaminated EEG signal.
Figure 9.7 shows the artifactual EEG signal collected from PhysioNet
database and the recovered artifact-free EEG signal.
 EEG artifact detection and removal techniques 171

Figure 9.5 Fast DST of contaminated EEG signal.

Figure 9.6 The clean EEG signal, contaminated EEG and the signal after removal of artifact.

Figure 9.7 Artifactual EEG and signal after removal of artifact.

172 Computational Techniques in Neuroscience

9.6 CONCLUSION

In this paper, we have covered some of the most common artifacts in EEG
data as well as some of the methods currently in use to eliminate them. To
this day, there is no universal method that can transform all contaminated
EEG data into usable EEG data. To achieve desirable outcomes, it is
common practice to employ multiple methods in sequence. It is common
practice to use ICA-based algorithms as the next step after filtering in
order to obtain clean EEG data. After that, various artifact removal
strategies can be implemented, each tailored to the specific artifacts
present in the dataset. Both WT and regression can be used to clean up
EEG data that has been contaminated by electrocardiogram (ECG) or
electrooculogram (EOG) artefacts, respectively. EOG and ECG artifact
removal using machine learning techniques has shown remarkable
improvement over the past three years. Muscle and body movement, as
well as other extrinsic artifacts, should be minimized or eliminated
whenever possible during the EEG recording process. Some artifacts, such
as EOG and ECG artifacts, are unavoidable but can be eliminated with
these techniques.

REFERENCES

[1] A. Subasi, “EEG signal classification using wavelet feature extraction and
a mixture of expert model,” Expert Systems with Applications, vol. 32,
no. 4, pp. 1084–1093, 2007.
[2] T. Zhang, W. Chen, and M. Li, “Generalized Stockwell transform and SVD-
based epileptic seizure detection in EEG using random forest,” Biocybernetics
and Biomedical Engineering, vol. 38, no. 3, pp. 519–534, 2018.
[3] W. Mumtaz, S. S. A. Ali, M. A. M. Yasin, and A. S. Malik, “A machine
learning framework involving EEG-based functional connectivity to diagnose
major depressive disorder (MDD),” Medical & Biological Engineering &
Computing, vol. 56, pp. 233–246, 2018.
[4] U. R. Acharya, S. L. Oh, Y. Hagiwara, J. H. Tan, H. Adeli, and D. P. Subha,
“Automated EEG-based screening of depression using deep convolutional
neural network,” Computer Methods and Programs in Biomedicine,
vol. 161, pp. 103–113, 2018.
[5] A. Anuragi and D. S. Sisodia, “Alcohol use disorder detection using EEG
Signal features and flexible analytical wavelet transform,” Biomedical Signal
Processing and Control, vol. 52, pp. 384–393, 2019.
[6] W. Mumtaz, N. Kamel, S. S. A. Ali, and A. S. Malik, “An EEG-based
functional connectivity measure for automatic detection of alcohol use dis-
order,” Artificial Intelligence in Medicine, vol. 84, pp. 79–89, 2018.
[7] R. Yuvaraj, U. Rajendra Acharya, and Y. Hagiwara, “A novel Parkinson’s
Disease Diagnosis Index using higher-order spectra features in EEG sig-
nals,” Neural Computing and Applications, vol. 30, pp. 1225–1235,
2018.
 EEG artifact detection and removal techniques 173

[8] H. Yu, X. Lei, Z. Song, C. Liu, and J. Wang, “Supervised network-based

fuzzy learning of EEG signals for Alzheimer’s disease identification,” IEEE
Transactions on Fuzzy Systems, vol. 28, no. 1, pp. 60–71, 2019.
[9] P. Ofner and G. R. Müller-Putz, “Movement target decoding from EEG and
the corresponding discriminative sources: A preliminary study,” in 2015
37th Annual International Conference of the IEEE Engineering in Medicine
and Biology Society (EMBC), 2015: IEEE, pp. 1468–1471.
[10] G. R. Müller-Putz et al., “Towards non-invasive EEG-based arm/hand-
control in users with spinal cord injury,” in 2017 5th International Winter
Conference on Brain-Computer Interface (BCI), 2017: IEEE, pp. 63–65.
[11] Y. Zeng et al., “EEG-based identity authentication framework using face
rapid serial visual presentation with optimized channels,” Sensors, vol. 19,
no. 1, p. 6, 2018.
[12] J. Chen, Z. Mao, W. Yao, and Y. Huang, “EEG-based biometric identification
with convolutional neural network,” Multimedia Tools and Applications,
vol. 79, pp. 10655–10675, 2020.
[13] M. M. N. Mannan, M. Y. Jeong, and M. A. Kamran, “Hybrid
ICA—Regression: Automatic identification and removal of ocular artifacts
from electroencephalographic signals,” Frontiers in Human Neuroscience,
vol. 10, p. 193, 2016.
[14] M. M. N. Mannan, S. Kim, M. Y. Jeong, and M. A. Kamran, “Hybrid
EEG—Eye tracker: Automatic identification and removal of eye movement
and blink artifacts from electroencephalographic signal,” Sensors, vol. 16,
no. 2, p. 241, 2016.
[15] A. Delorme and S. Makeig, “EEGLAB: an open source toolbox for analysis
of single-trial EEG dynamics including independent component analysis,”
Journal of Neuroscience M, vol. 134, no. 1, pp. 9–21, 2004.
[16] J. Iriarte et al., “Independent component analysis as a tool to eliminate ar-
tifacts in EEG: a quantitative study,” Journal of Clinical Neurophysiology,
vol. 20, no. 4, pp. 249–257, 2003.
[17] S. Kotte and J. K. Dabbakuti, “Methods for removal of artifacts from EEG
signal: A review,” in Journal of Physics: Conference Series, vol. 1706, no. 1:
IOP Publishing, p. 012093, 2020.
[18] A. Bisht, C. Kaur, and P. Singh, “Recent advances in artifact removal tech-
niques for EEG signal processing,” Intelligent Communication, Control and
Devices: Proceedings of ICICCD 2018, pp. 385–392, 2020.
[19] W. Mumtaz, S. Rasheed, and A. Irfan, “Review of challenges associated with
the EEG artifact removal methods,” Biomedical Signal Processing and Control,
vol. 68, p. 102741, 2021.
[20] A. Khosla, P. Khandnor, and T. Chand, “A comparative analysis of signal
processing and classification methods for different applications based on
EEG signals,” Biocybernetics and Biomedical Engineering, vol. 40, no. 2,
pp. 649–690, 2020.
[21] D. Mahmood, H. Nisar, and Y. V. Voon, “Removal of Physiological
Artifacts from Electroencephalogram Signals: A Review and Case Study,” in
2021 IEEE 9th Conference on Systems, Process and Control (ICSPC 2021),
2021: IEEE, pp. 141–146.
[22] X. Jiang, G.-B. Bian, and Z. Tian, “Removal of artifacts from EEG signals: a
review,” Sensors, vol. 19, no. 5, p. 987, 2019.
174 Computational Techniques in Neuroscience

[23] M. K. Islam, A. Rastegarnia, and Z. Yang, “Les méthodes de détection et de

rejet d’artefact de l‘EEG de scalp: revue de littérature,” Neurophysiologie
Clinique, vol. 46, no. 4-5, pp. 287–305, 2016.
[24] M. M. N. Mannan, M. A. Kamran, and M. Y. Jeong, “Identification and
removal of physiological artifacts from electroencephalogram signals: A
review,” Ieee Access, vol. 6, pp. 30630–30652, 2018.
[25] J. A. Urigüen and B. Garcia-Zapirain, “EEG artifact removal—state-of-the-art
and guidelines,” Journal of Neural Engineering, vol. 12, no. 3, p. 031001,
2015.
[26] M. K. Islam, A. Rastegarnia, and Z. Yang, “Methods for artifact detection
and removal from scalp EEG: A review,” Neurophysiologie Clinique/Clinical
Neurophysiology, vol. 46, no. 4-5, pp. 287–305, 2016.
[27] A. Schlögl, C. Keinrath, D. Zimmermann, R. Scherer, R. Leeb, and
G. Pfurtscheller, “A fully automated correction method of EOG artifacts in
EEG recordings,” Clinical Neurophysiology, vol. 118, no. 1, pp. 98–104,
2007.
[28] G. L. Wallstrom, R. E. Kass, A. Miller, J. F. Cohn, and N. A. Fox,
“Automatic correction of ocular artifacts in the EEG: a comparison of
regression-based and component-based methods,” International Journal of
Psychophysiology, vol. 53, no. 2, pp. 105–119, 2004.
[29] A. Q. Hamal and A. W. bin Abdul Rehman, “Artifact processing of epileptic
EEG signals: an overview of different types of artifacts,” in 2013 International
Conference on Advanced Computer Science Applications and Technologies,
2013: IEEE, pp. 358–361.
[30] R. J. Croft and R. J. Barry, “Removal of ocular artifact from the EEG: a
review,” Neurophysiologie Clinique/Clinical Neurophysiology, vol. 30, no.
1, pp. 5–19, 2000.
[31] I. I. Goncharova, D. J. McFarland, T. M. Vaughan, and J. R. Wolpaw,
“EMG contamination of EEG: spectral and topographical characteristics,”
Clinical Neurophysiology, vol. 114, no. 9, pp. 1580–1593, 2003.
[32] S. D. Muthukumaraswamy, “High-frequency brain activity and muscle ar-
tifacts in MEG/EEG: a review and recommendations,” Frontiers in Human
Neuroscience, vol. 7, p. 138, 2013.
[33] X. Chen et al., “Removal of muscle artifacts from the EEG: A review and
recommendations,” IEEE Sensors Journal, vol. 19, no. 14, pp. 5353–5368,
2019.
[34] R. D. O’Donnell, J. Berkhout, and W. R. Adey, “Contamination of scalp EEG
spectrum during contraction of cranio-facial muscles,” Electroencephalography
and Clinical Neurophysiology, vol. 37, no. 2, pp. 145–151, 1974.
[35] K. Wang, W. Li, L. Dong, L. Zou, and C. Wang, “Clustering-constrained
ICA for ballistocardiogram artifacts removal in simultaneous EEG-fMRI,”
Frontiers in Neuroscience, vol. 12, p. 59, 2018.
[36] C. Dai, J. Wang, J. Xie, W. Li, Y. Gong, and Y. Li, “Removal of ECG ar-
tifacts from EEG using an effective recursive least square notch filter,” IEEE
Access, vol. 7, pp. 158872–158880, 2019.
[37] P. J. Allen, G. Polizzi, K. Krakow, D. R. Fish, and L. Lemieux, “Identification
of EEG events in the MR scanner: the problem of pulse artifact and a method
for its subtraction,” Neuroimage, vol. 8, no. 3, pp. 229–239, 1998.
 EEG artifact detection and removal techniques 175

[38] P. J. Allen, O. Josephs, and R. Turner, “A method for removing imaging

artifact from continuous EEG recorded during functional MRI,”
Neuroimage, vol. 12, no. 2, pp. 230–239, 2000.
[39] G. S. Spencer, J. A. Smith, M. E. Chowdhury, R. Bowtell, and K. J. Mullinger,
“Exploring the origins of EEG motion artefacts during simultaneous fMRI
acquisition: Implications for motion artefact correction,” NeuroImage,
vol. 173, pp. 188–198, 2018.
[40] A. Islam, E. J. Esha, S. F. B. Ahmed, and M. K. Islam, “Study and Analysis of
Motion Artifacts for Ambulatory EEG,” Article in International Journal of
Electrical and Computer Engineering, vol. 9, no. 4, 2021.
[41] J. T. Gwin, K. Gramann, S. Makeig, and D. P. Ferris, “Removal of
movement artifact from high-density EEG recorded during walking and
running,” Journal of Neurophysiology, vol. 103, no. 6, pp. 3526–3534,
2010.
[42] M. H. Kutner, C. J. Nachtsheim, J. Neter, and W. Wasserman, Applied linear
regression models. McGraw-Hill/Irwin New York, 2004.
[43] M. A. Klados, C. Papadelis, C. Braun, and P. D. Bamidis, “REG-ICA: a
hybrid methodology combining blind source separation and regression
techniques for the rejection of ocular artifacts,” Biomedical Signal Processing
and Control, vol. 6, no. 3, pp. 291–300, 2011.
[44] S. A. Hillyard and R. Galambos, “Eye movement artifact in the CNV,”
Electroencephalography and Clinical Neurophysiology, vol. 28, no. 2,
pp. 173–182, 1970.
[45] J. L. Whitton, F. Lue, and H. Moldofsky, “A spectral method for removing
eye movement artifacts from the EEG,” Electroencephalography and Clinical
Neurophysiology, vol. 44, no. 6, pp. 735–741, 1978.
[46] G. L. Wallstrom, R. E. Kass, A. Miller, and N. A. Fox, “in the EEG Using
Bayesian Adaptive Regression Splines,” Case Studies in Bayesian Statistics,
vol. 6, p. 351, 2018.
[47] S. Romero, M. Mañanas, and M. J. Barbanoj, “Ocular reduction in EEG
signals based on adaptive filtering, regression and blind source separation,”
Annals of biomedical engineering, vol. 37, pp. 176–191, 2009.
[48] T. Gasser, P. Ziegler, and W. F. Gattaz, “The deleterious effect of ocular
artefacts on the quantitative EEG, and a remedy,” European Archives of
Psychiatry and Clinical Neuroscience, vol. 241, pp. 352–356, 1992.
[49] F. Ghaderi, S. K. Kim, and E. A. Kirchner, “Effects of eye artifact removal
methods on single trial P300 detection, a comparative study,” Journal of
Neuroscience Methods, vol. 221, pp. 41–47, 2014.
[50] K. T. Sweeney, T. E. Ward, and S. F. McLoone, “Artifact removal in
physiological signals—Practices and possibilities,” IEEE Transactions on
Information Technology in Biomedicine, vol. 16, no. 3, pp. 488–500,
2012.
[51] P. Sadasivan and D. N. Dutt, “ANC schemes for the enhancement of EEG
signals in the presence of EOG artifacts,” Computers and Biomedical
Research, vol. 29, no. 1, pp. 27–40, 1996.
[52] D. Hagemann and E. Naumann, “The effects of ocular artifacts on (later-
alized) broadband power in the EEG,” Clinical Neurophysiology, vol. 112,
no. 2, pp. 215–231, 2001.
176 Computational Techniques in Neuroscience

[53] S. Narasimhan and D. N. Dutt, “Application of LMS adaptive predictive

filtering for muscle artifact (noise) cancellation from EEG signals,”
Computers & Electrical Engineering, vol. 22, no. 1, pp. 13–30, 1996.
[54] F. Bartoli and S. Cerutti, “An optimal linear filter for the reduction of noise
superimposed to the EEG signal,” Journal of Biomedical Engineering, vol. 5,
no. 4, pp. 274–280, 1983.
[55] V. Ingle, S. Kogon, and D. Manolakis, Statistical and adaptive signal pro-
cessing. Artech, 2005.
[56] S. Puthusserypady and T. Ratnarajah, “H/sup/spl infin//adaptive filters for
eye blink artifact minimization from electroencephalogram,” IEEE Signal
Processing Letters, vol. 12, no. 12, pp. 816–819, 2005.
[57] A. G. Correa, E. Laciar, H. Patiño, and M. Valentinuzzi, “Artifact removal
from EEG signals using adaptive filters in cascade,” in Journal of Physics:
Conference Series, vol. 90, no. 1: IOP Publishing, p. 012081, 2007.
[58] B. Jervis, M. Thomlinson, C. Mair, J. Lopez, and M. Garcia, “Residual
ocular artefact subsequent to ocular artefact removal from the electro-
encephalogram,” IEE Proceedings-Science, Measurement and Technology,
vol. 146, no. 6, pp. 293–298, 1999.
[59] P. He, G. Wilson, and C. Russell, “Removal of ocular artifacts from electro-
encephalogram by adaptive filtering,” Medical and Biological Engineering
and Computing, vol. 42, pp. 407–412, 2004.
[60] C. Marque, C. Bisch, R. Dantas, S. Elayoubi, V. Brosse, and C. Perot,
“Adaptive filtering for ECG rejection from surface EMG recordings,”
Journal of Electromyography and Kinesiology, vol. 15, no. 3, pp. 310–315,
2005.
[61] M. Milanesi et al., “Multichannel techniques for motion artifacts removal
from electrocardiographic signals,” in 2006 International Conference
of the IEEE Engineering in Medicine and Biology Society, 2006: IEEE,
pp. 3391–3394.
[62] J. Mateo, E. M. Sánchez-Morla, and J. Santos, “A new method for removal
of powerline interference in ECG and EEG recordings,” Computers &
Electrical Engineering, vol. 45, pp. 235–248, 2015.
[63] R. Sameni, M. B. Shamsollahi, and C. Jutten, “Model-based Bayesian fil-
tering of cardiac contaminants from biomedical recordings,” Physiological
Measurement, vol. 29, no. 5, p. 595, 2008.
[64] J. J. Kierkels, J. Riani, J. W. Bergmans, and G. J. Van Boxtel, “Using an eye
tracker for accurate eye movement artifact correction,” IEEE Transactions
on Biomedical Engineering, vol. 54, no. 7, pp. 1256–1267, 2007.
[65] B. Somers, T. Francart, and A. Bertrand, “A generic EEG artifact removal
algorithm based on the multi-channel Wiener filter,” Journal of Neural
Engineering, vol. 15, no. 3, p. 036007, 2018.
[66] P. Diniz, “Adaptive Filtering: Algorithms and Practical Implementation.
Springer,” New York, NY, USA, 2008.
[67] G. Welch, “An Introduction to the Kalman Filter” “Univ. of North Carolina,”
http://www.cs.unc.edu/~welch/media/pdf/kalman_intro.pdf, 2006.
[68] M. Izzetoglu, P. Chitrapu, S. Bunce, and B. Onaral, “Motion artifact can-
cellation in NIR spectroscopy using discrete Kalman filtering,” Biomedical
Engineering Online, vol. 9, pp. 1–10, 2010.
 EEG artifact detection and removal techniques 177

[69] S. Seyedtabaii and L. Seyedtabaii, “Kalman filter based adaptive reduction of

motion artifact from photoplethysmographic signal,” World Acad. Sci. Eng.
Technol, vol. 37, pp. 173–176, 2008.
[70] F. Morbidi, A. Garulli, D. Prattichizzo, C. Rizzo, and S. Rossi, “Application
of Kalman filter to remove TMS-induced artifacts from EEG recordings,”
IEEE Transactions on Control Systems Technology, vol. 16, no. 6,
pp. 1360–1366, 2008.
[71] J. V. Candy, Bayesian signal processing: classical, modern, and particle fil-
tering methods. John Wiley & Sons, 2016.
[72] C.-M. Ting, S.-H. Salleh, Z. M. Zainuddin, and A. Bahar, “Spectral esti-
mation of nonstationary EEG using particle filtering with application to
event-related desynchronization (ERD),” IEEE Transactions on Biomedical
Engineering, vol. 58, no. 2, pp. 321–331, 2010.
[73] S. Nakagome, T. P. Luu, J. A. Brantley, and J. L. Contreras-Vidal, “Prediction
of EMG envelopes of multiple terrains over-ground walking from EEG signals
using an unscented Kalman filter,” in 2017 IEEE International Conference on
Systems, Man, and Cybernetics (SMC), 2017: IEEE, pp. 3175–3178.
[74] T. P. Luu, Y. He, S. Nakagame, J. Gorges, K. Nathan, and J. L. Contreras-
Vidal, “Unscented Kalman filter for neural decoding of human treadmill
walking from non-invasive electroencephalography,” in 2016 38th Annual
International Conference of the IEEE Engineering in Medicine and Biology
Society (EMBC), 2016: IEEE, pp. 1548–1551.
[75] M. Yakoubi, R. Hamdi, and M. B. Salah, “EEG enhancement using extended
Kalman filter to train multi-layer perceptron,” Biomedical Engineering:
Applications, Basis and Communications, vol. 31, no. 01, p. 1950005, 2019.
[76] C.-M. Ting, S.-H. Salleh, Z. M. Zainuddin, and A. Bahar, “Artifact removal
from single-trial ERPs using non-Gaussian stochastic volatility models and
particle filter,” IEEE Signal Processing Letters, vol. 21, no. 8, pp. 923–927,
2014.
[77] V. Fox, J. Hightower, L. Liao, D. Schulz, and G. Borriello, “Bayesian fil-
tering for location estimation,” IEEE Pervasive Computing, vol. 2, no. 3,
pp. 24–33, 2003.
[78] S. Choi, A. Cichocki, H.-M. Park, and S.-Y. Lee, “Blind source separation
and independent component analysis: A review,” Neural Information
Processing-Letters and Reviews, vol. 6, no. 1, pp. 1–57, 2005.
[79] A. Delorme, T. Sejnowski, and S. Makeig, “Enhanced detection of artifacts
in EEG data using higher-order statistics and independent component anal-
ysis,” Neuroimage, vol. 34, no. 4, pp. 1443–1449, 2007.
[80] A. Mognon, J. Jovicich, L. Bruzzone, and M. Buiatti, “ADJUST: An auto-
matic EEG artifact detector based on the joint use of spatial and temporal
features,” Psychophysiology, vol. 48, no. 2, pp. 229–240, 2011.
[81] I. Winkler, S. Brandl, F. Horn, E. Waldburger, C. Allefeld, and M. Tangermann,
“Robust artifactual independent component classification for BCI practi-
tioners,” Journal of Neural Engineering, vol. 11, no. 3, p. 035013, 2014.
[82] T.-P. Jung, S. Makeig, M. Westerfield, J. Townsend, E. Courchesne, and
T. J. Sejnowski, “Removal of eye activity artifacts from visual event-related
potentials in normal and clinical subjects,” Clinical Neurophysiology,
vol. 111, no. 10, pp. 1745–1758, 2000.
178 Computational Techniques in Neuroscience

[83] Y. Li, Z. Ma, W. Lu, and Y. Li, “Automatic removal of the eye blink artifact
from EEG using an ICA-based template matching approach,” Physiological
Measurement, vol. 27, no. 4, p. 425, 2006.
[84] N. Mammone and F. C. Morabito, “Enhanced automatic artifact detection
based on independent component analysis and Renyi’s entropy,” Neural
Networks, vol. 21, no. 7, pp. 1029–1040, 2008.
[85] S. P. Fitzgibbon, D. M. Powers, K. J. Pope, and C. R. Clark, “Removal of
EEG noise and artifact using blind source separation,” Journal of Clinical
Neurophysiology, vol. 24, no. 3, pp. 232–243, 2007.
[86] C. J. James and O. J. Gibson, “Temporally constrained ICA: an application
to artifact rejection in electromagnetic brain signal analysis,” IEEE
Transactions on Biomedical Engineering, vol. 50, no. 9, pp. 1108–1116,
2003.
[87] A. Delorme, J. Palmer, J. Onton, R. Oostenveld, and S. Makeig, “Indepen-
dent EEG sources are dipolar,” PloS One, vol. 7, no. 2, p. e30135,
2012.
[88] K. Ting, P. Fung, C. Chang, and F. Chan, “Automatic correction of artifact
from single-trial event-related potentials by blind source separation using
second order statistics only,” Medical Engineering & Physics, vol. 28,
no. 8, pp. 780–794, 2006.
[89] W. De Clercq, A. Vergult, B. Vanrumste, W. Van Paesschen, and S. Van
Huffel, “Canonical correlation analysis applied to remove muscle artifacts
from the electroencephalogram,” IEEE Transactions on Biomedical
Engineering, vol. 53, no. 12, pp. 2583–2587, 2006.
[90] D. M. Vos et al., “Removal of muscle artifacts from EEG recordings of
spoken language production,” Neuroinformatics, vol. 8, pp. 135–150, 2010.
[91] X. Yong, R. K. Ward, and G. E. Birch, “Artifact removal in EEG
using morphological component analysis,” in 2009 IEEE International
Conference on Acoustics, Speech and Signal Processing, 2009: IEEE,
pp. 345–348.
[92] P. Berg and M. Scherg, “Dipole modelling of eye activity and its application
to the removal of eye artefacts from the EEG and MEG,” Clinical Physics
and Physiological Measurement, vol. 12, no. A, p. 49, 1991.
[93] S. Casarotto, A. M. Bianchi, S. Cerutti, and G. A. Chiarenza, “Principal
component analysis for reduction of ocular artefacts in event-related po-
tentials of normal and dyslexic children,” Clinical Neurophysiology,
vol. 115, no. 3, pp. 609–619, 2004.
[94] P. S. Kumar, R. Arumuganathan, K. Sivakumar, and C. Vimal, “Removal
of ocular artifacts in the EEG through wavelet transform without using
an EOG reference channel,” Int. J. Open Problems Compt. Math, vol. 1,
no. 3, pp. 188–200, 2008.
[95] D. Safieddine et al., “Removal of muscle artifact from EEG data: comparison
between stochastic (ICA and CCA) and deterministic (EMD and wavelet-
based) approaches,” EURASIP Journal on Advances in Signal Processing,
vol. 2012, no. 1, pp. 1–15, 2012.
[96] V. Krishnaveni, S. Jayaraman, L. Anitha, and K. Ramadoss, “Removal of
ocular artifacts from EEG using adaptive thresholding of wavelet coeffi-
cients,” Journal of Neural Engineering, vol. 3, no. 4, p. 338, 2006.
 EEG artifact detection and removal techniques 179

[97] M. Unser and A. Aldroubi, “A review of wavelets in biomedical applica-

tions,” Proceedings of the IEEE, vol. 84, no. 4, pp. 626–638, 1996.
[98] S. Slobounov, M. Hallett, C. Cao, and K. Newell, “Modulation of cortical
activity as a result of voluntary postural sway direction: an EEG study,”
Neuroscience Letters, vol. 442, no. 3, pp. 309–313, 2008.
[99] G. Inuso, F. La Foresta, N. Mammone, and F. C. Morabito, “Wavelet-ICA
methodology for efficient artifact removal from Electroencephalographic
recordings,” in 2007 international joint conference on neural networks,
2007: IEEE, pp. 1524–1529.
[100] P. S. Kumar, R. Arumuganathan, K. Sivakumar, and C. Vimal, “An adaptive
method to remove ocular artifacts from EEG signals using wavelet trans-
form,” J. Appl. Sci. Res, vol. 5, no. 7, pp. 711–745, 2009.
[101] M. K. I. Molla, M. R. Islam, T. Tanaka, and T. M. Rutkowski, “Artifact
suppression from EEG signals using data adaptive time domain filtering,”
Neurocomputing, vol. 97, pp. 297–308, 2012.
[102] Z. Wu and N. E. Huang, “Ensemble empirical mode decomposition: a noise-
assisted data analysis method,” Advances in Adaptive Data Analysis, vol. 1,
no. 01, pp. 1–41, 2009.
[103] B. Mijović, M. De Vos, I. Gligorijević, J. Taelman, and S. Van Huffel, “Source
separation from single-channel recordings by combining empirical-mode
decomposition and independent component analysis,” IEEE Transactions
on Biomedical Engineering, vol. 57, no. 9, pp. 2188–2196, 2010.
[104] M. G. Keshava and K. Z. Ahmed, “Correction of ocular artifacts in EEG
signal using empirical mode decomposition and cross-correlation,” Research
Journal of Biotechnology, vol. 9, no. 12, pp. 21–26, 2014.
[105] M. Guarascio and S. Puthusserypady, “Automatic minimization of ocular
artifacts from electroencephalogram: A novel approach by combining
Complete EEMD with Adaptive Noise and Renyi’s Entropy,” Biomedical
Signal Processing and Control, vol. 36, pp. 63–75, 2017.
[106] B. Yang, T. Zhang, Y. Zhang, W. Liu, J. Wang, and K. Duan, “Removal of
electrooculogram artifacts from electroencephalogram using canonical cor-
relation analysis with ensemble empirical mode decomposition,” Cognitive
Computation, vol. 9, pp. 626–633, 2017.
[107] K. Dragomiretskiy and D. Zosso, “Variational mode decomposition,” IEEE
Transactions on Signal Processing, vol. 62, no. 3, pp. 531–544, 2013.
[108] T. Dutta, U. Satija, B. Ramkumar, and M. S. Manikandan, “A novel method
for automatic modulation classification under non-Gaussian noise based on
variational mode decomposition,” in 2016 twenty second national confer-
ence on communication (NCC), 2016: IEEE, pp. 1–6.
[109] M. Saini and U. Satija, “An effective and robust framework for ocular
artifact removal from single-channel EEG signal based on variational
mode decomposition,” IEEE Sensors Journal, vol. 20, no. 1, pp. 369–376,
2019.
[110] M. Shi, C. Wang, W. Zhao, X. Zhang, Y. Ye, and N. Xie, “Removal of
ocular artifacts from electroencephalo-graph by improving variational mode
decomposition,” China Communications, vol. 19, no. 2, pp. 47–61, 2022.
[111] N. ur Rehman and H. Aftab, “Multivariate variational mode decomposition,”
IEEE Transactions on Signal Processing, vol. 67, no. 23, pp. 6039–6052,
2019.
180 Computational Techniques in Neuroscience

[112] R. Gavas, D. Jaiswal, D. Chatterjee, V. Viraraghavan, and R. K. Ramakrishnan,

“Multivariate variational mode decomposition based approach for blink
removal from EEG signal,” in 2020 IEEE International Conference on Pervasive
Computing and Communications Workshops (PerCom Workshops), 2020:
IEEE, pp. 1–6.
[113] C. Y. Sai, N. Mokhtar, H. Arof, P. Cumming, and M. Iwahashi, “Automated
classification and removal of EEG artifacts with SVM and wavelet-ICA,”
IEEE Journal of Biomedical and Health Informatics, vol. 22, no. 3,
pp. 664–670, 2017.
[114] S. Behera and M. N. Mohanty, “Classification of EEG signal using SVM,”
in Advances in Electrical Control and Signal Systems: Select Proceedings of
AECSS 2019, 2020: Springer, pp. 859–869.
[115] S. Behera and M. N. Mohanty, “Artifact removal using deep WVFLN for
brain signal diagnosis through IoMT,” Measurement: Sensors, vol. 24,
p. 100465, 2022.
[116] B. Yang, K. Duan, C. Fan, C. Hu, and J. Wang, “Automatic ocular artifacts
removal in EEG using deep learning,” Biomedical Signal Processing and
Control, vol. 43, pp. 148–158, 2018.
[117] A. Jafarifarmand and M. A. Badamchizadeh, “Real-time cardiac artifact
removal from EEG using a hybrid approach,” in 2018 International
Conference BIOMDLORE, 2018: IEEE, pp. 1–5.
[118] C. S. Kim, J. Sun, D. Liu, Q. Wang, and S. G. Paek, “Removal of ocular
artifacts using ICA and adaptive filter for motor imagery-based BCI,” IEEE/
CAA journal of automatica sinica, 2017.
[119] M. Miao, W. Hu, B. Xu, J. Zhang, J. J. Rodrigues, and V. H. C. de Albuquerque,
“Automated CCA-MWF algorithm for unsupervised identification and removal
of EOG artifacts from EEG,” IEEE Journal of Biomedical and Health
Informatics, vol. 26, no. 8, pp. 3607–3617, 2021.
[120] H. Peng et al., “Removal of ocular artifacts in EEG—An improved approach
combining DWT and ANC for portable applications,” IEEE Journal of
Biomedical and health informatics, vol. 17, no. 3, pp. 600–607, 2013.
[121] D. P. Yedurkar and S. P. Metkar, “Multiresolution approach for
artifacts removal and localization of seizure onset zone in epileptic EEG
signal,” Biomedical Signal Processing and Control, vol. 57, p. 101794,
2020.
[122] J. Lin and A. Zhang, “Fault feature separation using wavelet-ICA filter,”
NDT & e International, vol. 38, no. 6, pp. 421–427, 2005.
[123] B. Azzerboni, M. Carpentieri, F. La Foresta, and F. Morabito, “Neural-ICA
and wavelet transform for artifacts removal in surface EMG,” in 2004
IEEE International Joint Conference on Neural Networks (IEEE Cat.
No. 04CH37541), 2004, vol. 4: IEEE, pp. 3223–3228.
[124] S. Calcagno, F. La Foresta, and M. Versaci, “Independent component
analysis and discrete wavelet transform for artifact removal in biomedical
signal processing,” American Journal of Applied Sciences, vol. 11, no. 1,
p. 57, 2014.
[125] N. Mammone, F. La Foresta, and F. C. Morabito, “Automatic artifact
rejection from multichannel scalp EEG by wavelet ICA,” IEEE Sensors
Journal, vol. 12, no. 3, pp. 533–542, 2011.
 EEG artifact detection and removal techniques 181

[126] N. K. Al-Qazzaz, S. Hamid Bin Mohd Ali, S. A. Ahmad, M. S. Islam, and

J. Escudero, “Automatic artifact removal in EEG of normal and demented
individuals using ICA–WT during working memory tasks,” Sensors, vol. 17,
no. 6, p. 1326, 2017.
[127] S. Jirayucharoensak and P. Israsena, “Automatic removal of EEG
artifacts using ICA and lifting wavelet transform,” in 2013 International
Computer Science and Engineering Conference (ICSEC), 2013: IEEE,
pp. 136–139.
[128] R. Mahajan and B. I. Morshed, “Unsupervised eye blink artifact denoising of
EEG data with modified multiscale sample entropy, kurtosis, and wavelet-
ICA,” IEEE Journal of Biomedical and Health Informatics, vol. 19, no. 1,
pp. 158–165, 2014.
[129] S. Behera and M. N. Mohanty, “A statistical approach for ocular
artifact removal in brain signals,” in 2018 2nd International Conference
on Data Science and Business Analytics (ICDSBA), 2018: IEEE,
pp. 500–503.
[130] X. Chen, Z. J. Wang, and M. McKeown, “Joint blind source separation
for neurophysiological data analysis: Multiset and multimodal methods,”
IEEE Signal Processing Magazine, vol. 33, no. 3, pp. 86–107, 2016.
[131] P. He, M. Kahle, G. Wilson, and C. Russell, “Removal of ocular artifacts
from EEG: a comparison of adaptive filtering method and regression method
using simulated data,” in 2005 IEEE Engineering in Medicine and Biology
27th Annual Conference, 2006: IEEE, pp. 1110–1113.
[132] A. Egambaram, N. Badruddin, V. S. Asirvadam, E. Fauvet, C. Stolz, and
T. Begum, “Automated and online eye blink artifact removal from electro-
encephalogram,” in 2019 IEEE International Conference on Signal and
Image Processing Applications (ICSIPA), 2019: IEEE, pp. 159–163.
[133] X. Chen, Q. Chen, Y. Zhang, and Z. J. Wang, “A novel EEMD-CCA
approach to removing muscle artifacts for pervasive EEG,” IEEE Sensors
Journal, vol. 19, no. 19, pp. 8420–8431, 2018.
[134] X. Chen, C. He, and H. Peng, “Removal of muscle artifacts from single-channel
EEG based on ensemble empirical mode decomposition and multiset canonical
correlation analysis,” Journal of Applied Mathematics, vol. 2014, 2014.
[135] K. T. Sweeney, S. F. McLoone, and T. E. Ward, “The use of ensemble em-
pirical mode decomposition with canonical correlation analysis as a novel
artifact removal technique,” IEEE Transactions on Biomedical Engineering,
vol. 60, no. 1, pp. 97–105, 2012.
[136] G. Wang, C. Teng, K. Li, Z. Zhang, and X. Yan, “The removal of EOG
artifacts from EEG signals using independent component analysis and mul-
tivariate empirical mode decomposition,” IEEE Journal of Oiomedical and
Health Informatics, vol. 20, no. 5, pp. 1301–1308, 2015.
[137] A. Yadav and M. S. Choudhry, “A new approach for ocular artifact removal
from EEG signal using EEMD and SCICA,” Cogent Engineering, vol. 7,
no. 1, p. 1835146, 2020.
[138] Q. Li, L. Wei, Y. Xu, and B. Yang, “Ocular Artifact Removal Algorithm of
Single Channel EEG Based on CEEMDAN-ICA-WTD,” in 2021 IEEE
6th International Conference on Signal and Image Processing (ICSIP), 2021:
IEEE, pp. 451–455.
182 Computational Techniques in Neuroscience

[139] X. Chen, X. Xu, A. Liu, M. J. McKeown, and Z. J. Wang, “The use of

multivariate EMD and CCA for denoising muscle artifacts from few-channel
EEG recordings,” IEEE Transactions on Instrumentation and Measurement,
vol. 67, no. 2, pp. 359–370, 2017.
[140] X. Navarro, F. Porée, and G. Carrault, “ECG removal in preterm EEG
combining empirical mode decomposition and adaptive filtering,” in 2012
IEEE international conference on acoustics, speech and signal processing
(ICASSP), 2012: IEEE, pp. 661–664.
[141] L. Shoker, S. Sanei, and J. Chambers, “Artifact removal from electro-
encephalograms using a hybrid BSS-SVM algorithm,” IEEE Signal Processing
Letters, vol. 12, no. 10, pp. 721–724, 2005.
[142] S. Halder et al., “Online artifact removal for brain-computer interfaces using
support vector machines and blind source separation,” Computational
Intelligence and Neuroscience, vol. 2007, 2007.
[143] B. Hua, “Improved adaptive filtering based artifact removal from EEG
signals,” in 2020 13th International Congress on Image and Signal Process-
ing, BioMedical Engineering and Informatics (CISP-BMEI), 2020: IEEE,
pp. 424–428.
[144] S. Blum, N. S. Jacobsen, M. G. Bleichner, and S. Debener, “A Riemannian
modification of artifact subspace reconstruction for EEG artifact handling,”
Frontiers in Human Neuroscience, vol. 13, p. 141, 2019.
[145] E. Butkevičiūtė et al., “Removal of movement artefact for mobile EEG
analysis in sports exercises,” IEEE Access, vol. 7, pp. 7206–7217, 2019.
[146] S. Kohli and A. J. Casson, “Removal of gross artifacts of transcranial
alternating current stimulation in simultaneous EEG monitoring,” Sensors,
vol. 19, no. 1, p. 190, 2019.
[147] M. Chavez, F. Grosselin, A. Bussalb, F. D. V. Fallani, and X. Navarro-Sune,
“Surrogate-based artifact removal from single-channel EEG,” IEEE Transactions
on Neural Systems and Rehabilitation Engineering, vol. 26, no. 3, pp. 540–550,
2018.
[148] S. Sharma and U. Satija, “Automated Ocular Artifacts Removal Framework
Based on Adaptive Chirp Mode Decomposition,” IEEE Sensors Journal,
vol. 22, no. 6, pp. 5806–5814, 2022.
[149] A. Kilicarslan and J. L. C. Vidal, “Characterization and real-time removal
of motion artifacts from EEG signals,” Journal of Neural Engineering,
vol. 16, no. 5, p. 056027, 2019.
[150] R. Ranjan, B. C. Sahana, and A. K. Bhandari, “Motion artifacts suppression
from EEG signals using an adaptive signal denoising method,” IEEE
Transactions on Instrumentation and Measurement, vol. 71, pp. 1–10, 2022.
[151] M. S. Hossain et al., “Motion artifacts correction from EEG and fNIRS signals
using novel multiresolution analysis,” IEEE Access, vol. 10, pp. 29760–29777,
2022.
[152] C. Kaur, A. Bisht, P. Singh, and G. Joshi, “EEG Signal denoising using hybrid
approach of Variational Mode Decomposition and wavelets for depression,”
Biomedical Signal Processing and Control, vol. 65, p. 102337, 2021.
[153] L. Chang, R. Wang, and Y. Zhang, “Decoding SSVEP patterns from EEG
via multivariate variational mode decomposition-informed canonical corre-
lation analysis,” Biomedical Signal Processing and Control, vol. 71,
p. 103209, 2022.
 EEG artifact detection and removal techniques 183

[154] C. Dora and P. K. Biswal, “Correlation-based ECG artifact correction from

single channel EEG using modified variational mode decomposition,”
Computer Methods and Programs in Biomedicine, vol. 183, p. 105092,
2020.
[155] C. Dora and P. K. Biswal, “An improved algorithm for efficient ocular artifact
suppression from frontal EEG electrodes using VMD,” Biocybernetics
and Biomedical Engineering, vol. 40, no. 1, pp. 148–161, 2020.
[156] Y. Liu, F. An, X. Lang, and Y. Dai, “Remove motion artifacts from scalp
single channel EEG based on noise assisted least square multivariate em-
pirical mode decomposition,” in 2020 13th International Congress on
Image and Signal Processing, BioMedical Engineering and Informatics
(CISP-BMEI), 2020: IEEE, pp. 568–573.
[157] M. E. Alam and B. Samanta, “Performance evaluation of empirical mode
decomposition for EEG artifact removal,” in ASME International
Mechanical Engineering Congress and Exposition, 2017, vol. 58387:
American Society of Mechanical Engineers, p. V04BT05A024.
[158] K. K. Dutta, K. Venugopal, and S. A. Swamy, “Removal of muscle artifacts
from EEG based on ensemble empirical mode decomposition and classifi-
cation of seizure using machine learning techniques,” in 2017 International
Conference on Inventive Computing and Informatics (ICICI), 2017: IEEE,
pp. 861–866.
[159] S. Tavildar and A. Ashrafi, “Application of multivariate empirical mode
decomposition and canonical correlation analysis for EEG motion artifact
removal,” in 2016 Conference on Advances in Signal Processing (CASP),
2016: IEEE, pp. 150–154.
[160] C. Zhang et al., “Automatic artifact removal from electroencephalogram
data based on a priori artifact information,” BioMed research international,
vol. 2015, 2015.
[161] C. A. Majmudar, R. Mahajan, and B. I. Morshed, “Real-time hybrid
ocular artifact detection and removal for single channel EEG,” in 2015
IEEE International Conference on Electro/Information Technology (EIT),
2015: IEEE, pp. 330–334.
[162] S. Gaikwad and K. Paradeshi, “Design of effective algorithm for removal
of ocular artifact from multichannel EEG Signal using ICA and wavelet
method,” International Journal of Computer Science and Information
Technologies, vol. 7, no. 3, pp. 1531–1535, 2016.
[163] R. Upadhyay, P. Padhy, and P. K. Kankar, “EEG artifact removal and noise
suppression by discrete orthonormal S-transform denoising,” Computers &
Electrical Engineering, vol. 53, pp. 125–142, 2016.
[164] C. Xiaobai and C. Junjun, “A Method for Blink Artifact Detection and
Removal with Wavelet Transform and Hilbert Transform,” in International
Conference on Biological Engineering and Pharmacy 2016 (BEP 2016),
2016: Atlantis Press, pp. 104–109.
[165] B. Somers and A. Bertrand, “Removal of eye blink artifacts in wireless
EEG sensor networks using reduced-bandwidth canonical correlation anal-
ysis,” Journal of neural engineering, vol. 13, no. 6, p. 066008, 2016.
[166] M. J. Ferdous, M. S. Ali, M. E. Hamid, and M. K. I. Molla,
“Eyeblink Artifact Suppression from EEG Signal using Lifting Wavelet
Transform.”
184 Computational Techniques in Neuroscience

[167] S. K. Goh, H. A. Abbass, K. C. Tan, A. Al-Mamun, C. Wang,

and C. Guan, “Automatic EEG artifact removal techniques by detecting
influential independent components,” IEEE Transactions on Emerging
Topics in Computational Intelligence, vol. 1, no. 4, pp. 270–279,
2017.
[168] A. Khasnobish, K. Chakravarty, D. Chatterjee, and A. Sinha, “Wavelet
based head movement artifact removal from electrooculography signals,” in
2017 IEEE International Conference on Acoustics, Speech and Signal
Processing (ICASSP), 2017: IEEE, pp. 984–988.
[169] M. Dora and D. Holcman, “Adaptive single-channel EEG artifact
removal with applications to clinical monitoring,” IEEE Transactions
on Neural Systems and Rehabilitation Engineering, vol. 30, pp. 286–295,
2022.
[170] C.-Y. Chang, S.-H. Hsu, L. Pion-Tonachini, and T.-P. Jung, “Evaluation
of artifact subspace reconstruction for automatic EEG artifact removal,”
in 2018 40th Annual International Conference of the IEEE Engineering in
Medicine and Biology Society (EMBC), 2018: IEEE, pp. 1242–1245.
[171] S. Sreeja, R. R. Sahay, D. Samanta, and P. Mitra, “Removal of eye blink
artifacts from EEG signals using sparsity,” IEEE Journal of Biomedical and
Health Informatics, vol. 22, no. 5, pp. 1362–1372, 2017.
[172] N. S. Malan and S. Sharma, “Removal of Ocular Atrifacts from Single Channel
EEG Signal Using DTCWT with Quantum Inspired Adaptive Threshold,” in
2018 2nd International Conference on Biomedical Engineering (IBIOMED),
2018: IEEE, pp. 94–99.
[173] P. Gajbhiye, R. K. Tripathy, A. Bhattacharyya, and R. B. Pachori, “Novel
approaches for the removal of motion artifact from EEG recordings,” IEEE
Sensors Journal, vol. 19, no. 22, pp. 10600–10608, 2019.
[174] M. K. Islam and A. Rastegarnia, “Probability mapping based artifact
detection and wavelet denoising based artifact removal from scalp EEG for
BCI applications,” in 2019 IEEE 4th International Conference on Computer
and Communication Systems (ICCCS), 2019: IEEE, pp. 243–247.
[175] S. Lee, M. J. McKeown, Z. J. Wang, and X. Chen, “Removal of high-
voltage brain stimulation artifacts from simultaneous EEG recordings,” IEEE
Transactions on Biomedical Engineering, vol. 66, no. 1, pp. 50–60, 2018.
[176] M. F. Issa and Z. Juhasz, “Improved EOG artifact removal using wavelet
enhanced independent component analysis,” Brain Sciences, vol. 9, no. 12,
p. 355, 2019.
[177] A. Bhattacharyya et al., “A multi-channel approach for cortical stimulation
artefact suppression in depth EEG signals using time-frequency and spatial
filtering,” IEEE Transactions on Biomedical Engineering, vol. 66, no. 7,
pp. 1915–1926, 2018.
[178] V. Roy and S. Shukla, “Designing efficient blind source separation methods
for EEG motion artifact removal based on statistical evaluation,” Wireless
Personal Communications, vol. 108, pp. 1311–1327, 2019.
[179] Y.-E. Lee, N.-S. Kwak, and S.-W. Lee, “A real-time movement artifact removal
method for ambulatory brain-computer interfaces,” IEEE Transactions on
Neural Systems and Rehabilitation Engineering, vol. 28, no. 12, pp. 2660–2670,
2020.
 EEG artifact detection and removal techniques 185

[180] C.-Y. Chang, S.-H. Hsu, L. Pion-Tonachini, and T.-P. Jung, “Evaluation
of artifact subspace reconstruction for automatic artifact components
removal in multi-channel EEG recordings,” IEEE Transactions on
Biomedical Engineering, vol. 67, no. 4, pp. 1114–1121, 2019.
[181] K. Jindal, R. Upadhyay, and H. S. Singh, “Application of hybrid GLCT-
PICA de-noising method in automated EEG artifact removal,” Biomedical
Signal Processing and Control, vol. 60, p. 101977, 2020.
[182] N. Bajaj, J. R. Carrión, F. Bellotti, R. Berta, and A. De Gloria,
“Automatic and tunable algorithm for EEG artifact removal using wavelet
decomposition with applications in predictive modeling during auditory
tasks,” Biomedical Signal Processing and Control, vol. 55, p. 101624,
2020.
[183] K. Paradeshi and U. Kolekar, “Ocular artifact suppression in multichannel
EEG using dynamic segmentation and enhanced wICA,” IETE Journal
of Research, vol. 68, no. 4, pp. 2683–2696, 2022.
[184] S. K. Noorbasha and G. F. Sudha, “Removal of EOG artifacts from single
channel EEG–an efficient model combining overlap segmented ASSA
and ANC,” Biomedical Signal Processing and Control, vol. 60, p. 101987,
2020.
[185] I. McNulty et al., “Analysis of Artifacts Removal Techniques in EEG
Signals for Energy-Constrained Devices,” in 2021 IEEE International
Midwest Symposium on Circuits and Systems (MWSCAS), 2021: IEEE,
pp. 515–519.
[186] S. T. Aung and Y. Wongsawat, “Analysis of EEG signals contaminated
with motion artifacts using multiscale modified-distribution entropy,” IEEE
Access, vol. 9, pp. 33911–33921, 2021.
[187] P. Gajbhiye, N. Mingchinda, W. Chen, S. C. Mukhopadhyay,
T. Wilaiprasitporn, and R. K. Tripathy, “Wavelet domain optimized
Savitzky–Golay filter for the removal of motion artifacts from EEG re-
cordings,” IEEE Transactions on Instrumentation and Measurement,
vol. 70, pp. 1–11, 2020.
[188] Z. Jamil, A. Jamil, and M. Majid, “Artifact removal from EEG signals
recorded in non-restricted environment,” Biocybernetics and Biomedical
Engineering, vol. 41, no. 2, pp. 503–515, 2021.
[189] S. Phadikar, N. Sinha, and R. Ghosh, “Automatic eyeblink artifact removal
from EEG signal using wavelet transform with heuristically optimized
threshold,” IEEE Journal of Biomedical and Health Informatics, vol. 25,
no. 2, pp. 475–484, 2020.
[190] M. Shahbakhti et al., “SWT-kurtosis based algorithm for elimination of
electrical shift and linear trend from EEG signals,” Biomedical Signal
Processing and Control, vol. 65, p. 102373, 2021.
[191] S. Chintala, J. Thangaraj, and D. R. Edla, “Mixed step size normalized
least mean fourth adaptive algorithm for artifact elimination from raw
EEG signals,” Biomedical Signal Processing and Control, vol. 65,
p. 102392, 2021.
[192] S. Çınar, “Design of an automatic hybrid system for removal of eye-blink
artifacts from EEG recordings,” Biomedical Signal Processing and Control,
vol. 67, p. 102543, 2021.
186 Computational Techniques in Neuroscience

[193] C. Stergiadis, V.-D. Kostaridou, and M. A. Klados, “Which BSS method

separates better the EEG Signals? A comparison of five different algorithms,”
Biomedical Signal Processing and Control, vol. 72, p. 103292, 2022.
[194] R. Huang, K. Qing, D. Yang, and K.-S. Hong, “Real-time motion artifact
removal using a dual-stage median filter,” Biomedical Signal Processing and
Control, vol. 72, p. 103301, 2022.
[195] A. M. Judith, S. B. Priya, and R. K. Mahendran, “Artifact Removal
from EEG signals using Regenerative Multi-Dimensional Singular Value
Decomposition and Independent Component Analysis,” Biomedical Signal
Processing and Control, vol. 74, p. 103452, 2022.
[196] V. P. Kumaravel, E. Farella, E. Parise, and M. Buiatti, “NEAR: An artifact
removal pipeline for human newborn EEG data,” Developmental Cognitive
Neuroscience, vol. 54, p. 101068, 2022.
[197] Y. Feng, Q. Liu, A. Liu, R. Qian, and X. Chen, “A Novel SSA-
CCA Framework forMuscle Artifact Removal from Ambulatory EEG,”
Virtual Reality & Intelligent Hardware, vol. 4, no. 1, pp. 1–21, 2022.
[198] S. S. Motdhare and G. Mathur, “An Experimental Analysis on EMG
Artifact Removal Methods from EEG Signal Records,” Mathematical
Statistician and Engineering Applications, vol. 71, no. 1, pp. 72–78-72–78,
2022.
[199] A. K. Maddirala and K. C. Veluvolu, “SSA with CWT and k-means for eye-
blink artifact removal from single-channel EEG signals,” Sensors, vol. 22,
no. 3, p. 931, 2022.
[200] J. Yedukondalu and L. D. Sharma, “Circulant Singular Spectrum Analysis
and Discrete Wavelet Transform for Automated Removal of EOG Artifacts
from EEG Signals,” Sensors, vol. 23, no. 3, p. 1235, 2023.
[201] C. Burger and D. J. Van Den Heever, “Removal of EOG artefacts
by combining wavelet neural network and independent component analysis,”
Biomedical Signal Processing and Control, vol. 15, pp. 67–79, 2015.
[202] W. Sun, Y. Su, X. Wu, and X. Wu, “A novel end-to-end 1D-ResCNN
model to remove artifact from EEG signals,” Neurocomputing, vol. 404,
pp. 108–121, 2020.
[203] S. K. Sahoo and S. K. Mohapatra, “Recognition of Ocular Artifacts in
EEG Signal through a Hybrid Optimized Scheme,” BioMed Research
International, vol. 2022, 2022.
[204] R. Ghosh, S. Phadikar, N. Deb, N. Sinha, P. Das, and E. Ghaderpour,
“Automatic Eye-blink and Muscular Artifact Detection and Removal from
EEG Signals using k-Nearest Neighbour Classifier and Long Short-Term
Memory Networks,” IEEE Sensors Journal, 2023.
[205] M. Quazi and S. Kahalekar, “Artifacts removal from EEG signal: FLM
optimization-based learning algorithm for neural network-enhanced adap-
tive filtering,” Biocybernetics and Biomedical Engineering, vol. 37, no. 3,
pp. 401–411, 2017.
[206] X. Li, C. Guan, H. Zhang, and K. K. Ang, “Discriminative ocular artifact
correction for feature learning in EEG analysis,” IEEE Transactions on
biomedical engineering, vol. 64, no. 8, pp. 1906–1913, 2016.
[207] R. A. Brown and R. Frayne, “A fast discrete S-transform for biomedical signal
processing,” in 2008 30th Annual International Conference of the IEEE
Engineering in Medicine and Biology Society, 2008: IEEE, pp. 2586–2589.
 EEG artifact detection and removal techniques 187

[208] W. Yao, Q. Tang, Z. Teng, Y. Gao, and H. Wen, “Fast S-transform for time-
varying voltage flicker analysis,” IEEE Transactions on Instrumentation and
Measurement, vol. 63, no. 1, pp. 72–79, 2013.
[209] M. A. Klados and P. D. Bamidis, “A semi-simulated EEG/EOG dataset
for the comparison of EOG artifact rejection techniques,” Data in brief,
vol. 8, pp. 1004–1006, 2016.
[210] I. Silva and G. B. Moody, “An open-source toolbox for analysing and
processing physionet databases in matlab and octave,” Journal of Open
Research Software, vol. 2, no. 1, 2014.
Chapter 10

Analysis of neural network

and neuromorphic computing
with hardware
A survey
Manish Bhardwaj1, Kailash Nath Tripathi2,
Yogendra Narayan Prajapati3, and Analp Pathak4
1
Department of Computer Science and Information Technology,
KIET Group of Institutions, Delhi-NCR, Ghaziabad, India
2
Department of AIML, ISBM College of Engineering, Pune, India
3
Department of Computer Science and Engineering, Ajay Kumar Engineering,
Ghaziabad, India
4
Department of Information Technology, KIET Group of Institutions,
Delhi-NCR, Ghaziabad, India

10.1 INTRODUCTION

In this study, the author presents a thorough overview of the field of neu-
romorphic computing, including topics such as objectives, neuron/synapse
concepts, procedures and education, deployments, advances in hardware,
and supplementary materials and systems [1]. By providing a comprehen-
sive and historical overview of the topic, we hope to stimulate future
research and serve as a jumping-off point for interested newcomers [2–4].
For decades, the goal of computer scientists has been to construct a
system that can perceive the world at a rate greater than that of a person,
and the von Neumann architecture has emerged as the undisputed gold
standard for this kind of system. While parallels to the human brain are
inescapable, the vastly different organizational structure, power require-
ments, and processing capabilities of the two systems underline the limi-
tations of both architectures [5]. This begs the obvious question of whether
or not artificial neural networks (ANNs) can be designed to perform as well
as the human brain.
Paralleling von Neumann systems, neuromorphic computing has devel-
oped in recent years. In 1990, the term “neuromorphic computing” was
coined by Mead; “neuromorphic” refers to a type of very large scale inte-
gration (VLSI) that uses analog components to simulate organic neural
networks [6]. These days, the phrase also refers to implementations that
make use of or are inspired by neural networks that aren’t necessarily based
on biology.

188 DOI: 10.1201/9781003398066-10

 Analysis of neural network & neuromorphic computing with hardware 189

These neuromorphic designs are unique in that they are highly linked and
parallel, need little energy, and co-localize storing and function. The
impending end of Moore’s law, the rising power requirements of Dennard
scaling, and the poor connectivity between processor and memory, known
as the von Neumann bottleneck, have all drawn more attention to neuro-
morphic designs, which are interesting in their own right. When compared
to conventional von Neumann designs, neuromorphic computers have the
ability to do complicated calculations more quickly, with less power con-
sumption, and in a smaller physical footprint [7]. Taking advantage of
neuromorphic systems in hardware development is strongly recommended
due to these features.
Furthermore, machine learning is a driving force behind the growing
popularity of neuromorphic computing. This method has the potential to
significantly enhance learning efficiency on specific tasks [8–10]. The focus
shifts from the architecture benefits of research in artificial intelligence to
its prospective operational benefits, with the hope that programs capable
of online, real-time learning like that of biological brains can be developed,
shown in Figure 10.1. Modern machine learning technique implementa-
tions may find neuromorphic structures to be the most perfect platform.
Researchers from many disciplines, including materials science, neu-
roscience, electrical engineering, computer engineering, and computer
science, are all represented in the neuromorphic computing community.
Neuromorphic technologies make use of substances with attributes com-
parable to those of biological neural systems; therefore, neuroscientists
explore, produce, and characterize new materials for application in these
devices [11]. Researchers in the field of neuroscience employ neuro-
morphic systems to imitate and analyze biological neural systems, and
they disseminate information about new findings from their investigations
that may have computational applications [12–14]. Device-level analog,
digital, mixed analog/digital, and non-traditional circuits are all tools of
the trade for electrical and computer engineers as they develop and

Figure 10.1 Various related fields with neuromorphic research.

190 Computational Techniques in Neuroscience

implement novel devices, systems, and modes of communication. Inspired

by biological systems and machine learning, computer scientists and en-
gineers aim to create novel network models that can be trained and/or
learn on their own. Aside from creating neuromorphic computing sys-
tems, they also create the enabling software for those systems to be used
in practice.
Both von Neumann and Turing discussed brain-inspired systems in the
1950s, so the idea of employing specialized hardware to construct neurally
inspired machines is as old as both computer science and computer en-
gineering [15]. For quite some time, it has been a goal of the computing
community to simulate the neuronal circuitry of the human brain. There
have been several significant advances in ANNs, artificial intelligence (AI),
and machine learning as a result of this quest. While ANNs and neuro-
science are not the primary emphasis of this study, the creation of non-von
Neumann hardware to model ANNs and biological brain systems is. Based
on incentives mentioned in the literature, we analyze many reasons why
neuromorphic systems have been developed over the years [16]. The ex-
ponential growth in the number of published studies related to neuro-
morphic computing over the past decade is shown in Figure 10.2. The
evolution of 10 of the most influential reasons given in the literature for
using a neuromorphic approach is depicted. Each of these top 10 inspira-
tions has been mentioned in the literature at least 15 times, making them
strong contenders for inclusion in the list.
Much of the pioneering accounts in neuromorphic computing was
motivated by the creation of hardware that could conduct parallel pro-
cesses, taking cues from the reported complexity in neurons and glia but
running on a single chip. Although parallel architectures existed, neuro-
morphic systems differentiated themselves by emphasizing a large number
of relatively simple processing components (often neurons) with a high
density of connections between them (typically synapses) [17–20]. The
intrinsic parallelism of neuromorphic systems was the most common
justification for bespoke practical systems in early efforts on neuro-
morphic computing.
The ability to do computations quickly was another motivation for early
neuromorphic and neural network hardware implementations. Particularly,

Figure 10.2 Graphical representation of research in neuromorphic and neural network

hardware.
 Analysis of neural network & neuromorphic computing with hardware 191

early system developers stressed that it was possible to achieve much faster
neural network computation with custom chips than was possible with
traditional von Neumann architectures, in part by extracting their natural
computation, as described above, but also by adding custom hardware to
finalize neural-style mathematical calculations [21–25]. This early emphasis
on speed suggested that neuromorphic devices could be used as boosters for
machine learning or neural network-style activities in the future.
One of the primary drivers of early neuromorphic systems was the need
for real-time performance. In applications like real-time control, real-
time digital image reconstruction, and autonomous robot control, the
devices’ natural computation and computational speed allowed neural
network simulations to be completed more quickly than in implementa-
tions on von Neumann architectures [26–28]. The demand for quicker
computing in these instances was driven more by the performance
requirements of the underlying applications than by research into the
topologies of neural networks. This is why we have separated it as a
driving force behind the evolution of machine learning from the pursuit
of speed and parallelism.
Due to various inherent single points of failure, both in the parallelized
representation and in the potential adaptation or self-healing capabilities
observed in ANN interpretations in software, developers have begun to
consider neural networks as a natural template for hardware design [29].
In the past and now, these traits mattered for making new hardware
implementations, as imperfections in both the created and used devices are
possible due to component and technique variance.
The possibility for exceptionally low power performance is the most
common reason given in the current research and discussions on neuro-
morphic devices in the cited papers [30]. The human brain is our primary
source of inspiration, but it only uses approximately 20 watts of electricity
and is capable of incredibly complicated computations and jobs. From its
inception, the goal of developing neuromorphic devices with comparably
low power consumption has been a driving force for neuromorphic com-
puting. This goal has just recently emerged as a key motive [31–33].
During this century of neuromorphic research, the development of por-
table devices with the computational power of neural networks but with
a minimal resource requirement (in terms of device size) has emerged as a
driving force [34]. As the prevalence of integrated systems and micro-
processors has increased, so has the need for designs that consume little
space, have different contexts, and use minimal energy.
Recently, reduced prevalence has been the driving force for the creation
of neuromorphic devices. As can be seen in Figure 10.2, this is the primary
driving force for neuromorphic computing. Major drivers for the advance-
ment of neuromorphic systems continue to be their inherent parallelism,
real-time performance, speed in both operation and training, and tiny
device footprint [35]. During this time, a few other motivations gained
192 Computational Techniques in Neuroscience

traction; one example is the proliferation of methods that employ neural

network-style architectures (i.e., architectures made up of neuron and
synapse-like components) due to their fault-tolerance characteristics or
reliability in the face of hardware errors. In light of the usage of innovative
materials for building neuromorphic systems, this has been an increasingly
popular rationale in recent years.
The research of neuroscience has also been a driving force behind the
development of neuromorphic systems in the last decade. Numerous
neuroscience-driven efforts, such as the European Union’s Human Brain
Project [36], have relied on the development of custom domain adapta-
tion because it is impractical to simulate genuine neural behavior on a
conventional supercomputer due to its size, speed, and energy require-
ments. Therefore, in order to efficiently run significant neuroscience
simulations, it is necessary to develop one’s own neuromorphic imple-
mentations. Similarly, scalability has emerged as a major driver in the
development of neuromorphic systems [37]. Most extensive neuro-
morphic projects discuss cascading their devices to achieve many neurons
and synapses. However, most of the other specified justifications are
related to the study of neuromorphic systems as a potential complemen-
tary architecture in the beyond Moore’s law computing landscape, which
is a common motivation that is not explicitly given in Figure 10.2. Many
neuroscientists do not believe neuromorphic structures will substitute von
Neumann architectures, but “building a better computer” is a driving
force behind the creation of neuromorphic devices. This motivation is
fairly all-encompassing, covering problems with conventional computers
such as the impending end of Moore’s law and Dennard scaling. The von
Neumann bottleneck, which occurs in von Neumann architectures due to
the separation of memory and processing and the performance gap
between processing and memory technologies in present-day systems, is
another driving force for the creation of neuromorphic computing. By
colocating memory and computing, neuromorphic systems reduce the
effects of the von Neumann bottleneck [38].
One of the driving forces behind the development of neuromorphic sys-
tems in recent years is the promise of online learning, or the capacity
to quickly and effectively adjust to new circumstances within a task as
they arise [39]. Though online learning mechanisms are currently poorly
defined, many neuromorphic systems feature online learning components
that could be used to accomplish learning tasks in an unsupervised, low-
power fashion. Systems that can process and analyze this data in an
unsupervised, online manner will be crucial in future computing platforms
because of the exponential expansion of information collection in recent
years. In addition, as our knowledge of biological brains expands, we
should be able to design more effective mechanisms for online learning,
with neuromorphic computing serving as an ideal platform for doing so.
 Analysis of neural network & neuromorphic computing with hardware 193

10.2 MODELS OF RESEARCH

Which model of neural networks to select is a crucial concern in neuro-

morphic computing. The components of the network, their functions, and
their relationships with one another are all spelled out in the neural net-
work model. Inspired by biological brain networks, basic types of neural
networks include neurons and synapses. One must establish models for
each component (e.g., neurons and synapses) while constructing the neural
system model, with each component model dictating the behavior of its
respective constituents [40].
How does one decide which model to use? It’s possible that the selected
model is inspired by a certain field of use. If, for the purposes of a neu-
roscience investigation, a more rapid simulation of biological brains is
required than can be achieved using conventional von Neumann struc-
tures, then a model that is biologically credible and/or realistic is required
for the neuromorphic device. A neuromorphic system that uses neural
networks with convolution may be the best option if the application
calls for very accurate picture recognition. Modifying the model itself is
an option.
A specific tool or material has limitations and/or properties. For instance,
models of spiking neural networks are best suited to the properties of
systems based on memristors, such as spike-timing sensitive plasticity-like
mechanismsn. In many other situations, it is unclear which model to use or
how complex it should be.
Neuromorphic and neural network equipment technologies have been
used to implement many different kinds of models [41–45]. The models
can be primarily computationally generated or primarily biologically
inspired. ANN models, rather than biological brains, serve as inspiration
for the latter. In this section, we’ll go over the various neuron models,
synapse models, and network models that have been implemented in
neuromorphic systems, and we’ll highlight some of the most important
articles for each.

10.2.1 Models of neurons

There are three main parts that make up a neuron in the body: the cell
body, the axon, and the dendrites. Neurons typically (but not always)
use their axon to send their signals away from the cell. Typically, den-
drites are the site of intake to neurons, and they are responsible for
relaying that knowledge to the cell body, though this is not always the
case. Neurons can take in data from many other neurons via chemical or
electrical signals. A presynaptic is the connection between the synapses
and the axon of two neurons that provides for the communication of
data or signals between them. Upon receiving a signal from another
194 Computational Techniques in Neuroscience

Figure 10.3 Hardware implementation in neuron models.

neuron via a synapse, a neuron will often begin to charge up as a result

of a shift in the power supply throughout the neuron’s cell membrane.
When the voltage inside a neuron reaches a certain threshold, the neuron
“fires,” or produces a muscle contraction that travels along the axon
and modulates the charge on neighboring neurons via synapses. Different
neuron models in neuromorphic systems may use different techniques to
accomplish the same ideas of charge transport and firing to influence
neighboring neurons. Similarly, axons and synapses are not often im-
plemented in artificial simulations that are not scientifically realistic (i.e.,
models inspired by neuroscience instead of copying neuroscience).
The various neuron model implementations are shown in Figure 10.3.
There are five main types of neuron models provided:
To be biologically credible, a model must explicitly simulate the varieties
of behavior observed in real-world neural networks.

• Biologically inspired: Attempting to mimic the behavior of biological

neural systems in a way that is not necessarily faithful to biology.
• Neuron+Other: Neuron models have additional biologically inspired
components, such as axons, dendrites, or glial cells, that are typically
absent from other neuromorphic neuron models.
• Integrate-and-fire models are a subset of the more complex biologi-
cally inspired spiking neuron models.
• Models of neurons based on the McCulloch-Pitts neuron, which is
widely utilized in the field of ANNs.
 Analysis of neural network & neuromorphic computing with hardware 195

Equation for the output of Neuron x for the given model as:

Y
Mx = f i =0
wi , jni

Mx is result of the equation, f is the function that is used for activation, Y is

the count of input for neuron x, wi,j is the value from I to j, and ni is the
result of neuron x.
Numerous hardware implementations of biologically credible and bio-
logically motivated neuron models have been constructed. Cell membrane
dynamics, which regulates factors like rate leakage throughout the neuron’s
biological membranes; ion stream dynamics, which regulates how ions
move from and to a neuron, switching the neuron’s charge level; axonal
models, which may incorporate delay ingredients; and dendritic concepts,
which regulate how many pre- and post-synapse neurons alter the current
neuron. Izhikevich [24] provides a useful summary of various spiking
neuron models.

1. Hodgkin-Huxley Model: Credible in terms of biology, the Hodgkin-

Huxley model is widely accepted as the most credible biologically
feasible model of a neuron. The Hodgkin-Huxley model, initially
proposed in 1952, is a sophisticated neuron model that accounts for
ion transport within and out of the neuron using four-dimensional
nonlinear differential equations [46]. Due to their biological plausi-
bility, Hodgkin-Huxley models have found widespread application
in neuromorphic implementations that aim to faithfully simulate
biological brain systems. The Morris Lecar model simplifies the model
to a two-dimensional nonlinear equation, which is nonetheless phy-
siologically reasonable. In neuroscience and neuromorphic systems,
this model is widely used.
2. Biologically Inspired: The Fitzhugh-Nagumo and Hindmarsh-Rose
simulations are two examples of reduced variations of the Hodgkin-
Huxley system that have been executed in hardware. In an effort to
represent behavior rather than aiming to imitate physical activity in
biological systems, these models tend to be simpler computationally
and simpler in terms of the number of features, while becoming more
scientifically inspired than specific biological systems. Simplifying
computation has the potential to yield more efficient and space-saving
solutions in neuromorphic computing devices [47,48]. From the point
of view of algorithms and learning methods, models with fewer
parameters can be simpler to set and/or train.
3. Biologically Inspired Mechanisms (Neurons + Others): There are many
other types of biologically inspired models out there. They typically
include a lot more detail about living organisms than concepts found
196 Computational Techniques in Neuroscience

in the machine learning and artificial intelligence literature, like mem-

brane dynamics, modeling ion-channel dynamics, incorporating axons
and/or dendrite models, and glial cell or astrocyte interactions.
Sometimes, brand-new models are made with the hardware in
mind from the start. To facilitate implementation in low-power VLSI,
on field-programmable gate array (FPGA), or using static comple-
mentary metal-oxide-semiconductor (CMOS), for example, equations
based on the Fitzhugh-Nagumo, Morris Lecar, Hodgkin-Huxley, or
other models have been modified or abstracted. The Hodgkin-Huxley
model has also been revised by other researchers to incorporate more
recent hardware advancements like the metal-oxide-semiconductor
field-effect transistor (MOSFET) transistor and the single-electron
transistor.
4. Integrate-and-Fire Neurons: The integrate-and-fire family of spiking
neuron models ranges in complexity from the simplest (the basic
integrate-and-fire) to that of the Izhikevich model and other compli-
cated biologically inspired models. While less realistic from a biological
perspective, integrate-and-fire models nonetheless provide sufficiently
complicated behavior for application in spiking neural systems. Even
in its simplest form, the integrate-and-fire model preserves the
neuron’s resting membrane potential. The probability on a neuron
decays over duration in the faulty combine approach, which is an ex-
tension of the basic implementation. It’s widely utilized in neuro-
morphic systems and considered a top model. The generic regressive
combine method, along with the quadratic integrate-and-fire model
utilized in some neuromorphic systems represents the next degree of
complexity. The reactive exponential integrate-and-fire model adds
another layer of complexity on top of the models already mentioned
(e.g., the Izhikevich model). Similarly, neuromorphic systems have
taken advantage of these.
There are currently digital adaptations of spiking neuron models,
which complement the formerly analog-style spiking neuron models.
Instead of using regressive or linear system of equations to describe
the dynamics, a cellular automaton is typically used in digital spiking
neuron models. For neuromorphic formulations, solutions of resonate-
and-fire, and adaptations of rotate-and-fire digital spiking neurons, a
hybrid analog/digital implementation was developed [49]. In order to
display nonlinear response characteristics, a modified send information
spiking model has been developed. Pulse-coupled networks have also
made use of digital spiking neurons. The first hardware implementation
of a neuron for a random neural network has been made.
Full network models will be referred to as “spiking neural networks”
in the following sections. We make no claims about which specific
neuron model is being used in the execution of these spiking networks.
In addition, the neuron model is customizable in some hardware
 Analysis of neural network & neuromorphic computing with hardware 197

implementations like SpiNNaker, allowing for many neuron models to

be realized within a single neuromorphic implementation.
5. When considering developing a reliable ANN, the standard McCulloch-
Pitts neuron architecture is available in a wide range of forms. The
processing element is a popular hardware version of the McCulloch-
Pitts models. It uses a simple thresholding function as the activation
function. Efforts have also been made to develop a system of different
activation patterns for neurons based on the McCulloch-Pitts model.
Although certain activation functions are computationally demanding,
others have seen more success in the neural network domain. This
computational burden can translate to configurability, and as a result,
there are various activation functions and implementations that try
to strike a balance between these two competing goals, as well as the
model’s overall accuracy and computing utility. Other hardware-based
activation functions have included the ramp-saturation feature, linear,
piecewise linear, step function, multi-threshold, radial basis function,
the tangent sigmoid function, and periodic activation functions. The
basic sigmoid function and the hyperbolic tangent function are the
most certain circuits because they incorporate not only the sigmoid
and hyperbolic tangent activation functions but also their derivatives
because the gradient is used in the pull training process. Some imple-
mentations have centered on the generation of neurons with program-
mable activation functions or on the generation of building blocks for
the generation of neurons.

Hardware implementations of neuron models have also been made for more
classic forms of ANNs. There are various neuron models, such as those found
in binary neural networks, fuzzy neural networks, and Hopfield neural net-
works. Generally speaking, many distinct neuron models have been realized
in hardware, and one of the choices a user might make is a compromise
between complexity and biological inspiration. There is a qualitative com-
parison between various neuron models with respect to these two criteria
in Figure 10.4.

10.2.2 Synapse models

A significant amount of attention has been paid to generating synapse
implementations separate from neuron models for neuromorphic systems,
just as there has been a lot of attention paid to generating neuron models.
Once again, we may break the synapse models into two groups: biologically
inspired synapse implementations, which include synapses for spike-based
systems, and synapse implementations for classic ANNs, including feed-
forward neural networks. It’s important to remember that in neuromorphic
systems, synapses are often the most numerous component, necessitating
the greatest space on a given chip. For several practical systems and notably
198 Computational Techniques in Neuroscience

Figure 10.4 Graphical representation of neurons model comparison.

for the creation and usage of novel materials for neuromorphic, the focus is
often on simplifying the synapse implementation. Therefore, unless an
attempt is made to describe biological activity explicitly, synapse models
tend to be quite straightforward. The strength or weight value of a neuron
can alter over time thanks to a plasticity mechanism, which is often
included in more complicated synapse models. In biological brains, mech-
anisms of plasticity have already been linked to knowledge.

10.2.2.1 Network model

Network models explain the relationships between neurons and synapses.
The preceding paragraphs should have given you a sense of the breadth of
the research into neural network models for computational models. Again,
they span from statistically controlled, non-spiking neural networks to
those that aim to emulate biological behavior as precisely as possible. When
deciding on a network model, there are many aspects to take into account.
As has been established in prior sections, one of the elements is undoubtedly
the biological inspiration and complexity of neuron and synapse models.
Network topology is another important aspect to think about. Network
topologies that can be employed in different kinds of networks are illus-
trated in Figure 10.5. These topologies range from those influenced by
biology to those based on electrical impulses. It’s possible that the con-
nectivity will be limited according on the devices selected, which will in turn
limit the feasible topologies. Section will go into greater depth on a third
consideration: whether or not the selected network model can make use of
and benefit from preexisting training or learning methods. Finally, the
 Analysis of neural network & neuromorphic computing with hardware 199

Figure 10.5 Network arrangement used for neuromorphic system.

network model’s potential for broad use across a variety of applications is a

factor that should not be discounted.
In hardware, conventional spiking neural networks come in a wide range
of forms. The several types of integrate-and-fire neurons, as well as more
physiologically plausible or biologically inspired models, are used in these
implementations. The synapse architecture of a spiking biological system
often includes spike-timing-dependent plasticity (STDP). Spiking models
are often used in neuromorphic implementations due to their event-drive
nature and superior energy efficiency compared to other systems. Thus,
spiking neuromorphic systems have been used to implement various other
types of neural network models, such as spiking feed-forward networks,
spiking recurrent networks, spiking deep neural networks, spiking deep
belief networks, spiking Hebbian systems, spiking Hopfield networks or
associative memories. Another kind of neural network model was employed
in these implementations, and it makes use of a spiking neural network
architecture in neuromorphic devices. These techniques are often trained on
a classic neural network architecture (e.g., feed-forward network) before
the resulting network solution is modified to work with a spiking neuro-
morphic implementation. As a result, the spiking neural network’s full
capabilities may not be exploited.
Figure 10.6 is a high-level depiction of the modeling approaches used in
computational models. In Figure 10.7, we can see the development of
several popular models used in neuromorphic implementations. According
to the data, spiking and feed-forward implementations account for the vast
majority of real-world examples, with spiking’s popularity growing over
the past decade. Convolutional neural networks in deep learning have
become increasingly popular and fruitful over the past five years, while
generalized feed-forward models have begun to decline.
Deep neural networks are one sort of artificial brain network that has
frequently looked to sensory information for influence. Pulse-coupled
human brains and cellular neural networks are two more models used in
neuromorphic systems that were motivated by the visual system. While
input signal networks were all the rage in the early 2000s, cellular neural
networks were widely used in early neuromorphic implementations and are
seeing a renaissance.
200 Computational Techniques in Neuroscience

Figure 10.6 Basic neuromorphic application with network prototypical.

Figure 10.7 Neuromorphic execution over time with different models.

 Analysis of neural network & neuromorphic computing with hardware 201

Cellular automata; fuzzy neural networks, which integrate fuzzy approach

and ANNs; and the patriarchal temporal recollection schematic diagram
introduced by Hawkins are all examples of less widely accepted neural net-
work and neural network-neighboring models introduced in domain
adaptation.
It is evident that there is a rich landscape of significant biological and
artificial models for neural networks from which to choose when designing
neuromorphic implementations. The goal of the neuromorphic system will
significantly influence the selection of a suitable model. Models in neuro-
science research projects tend to choose the side of being biologically
plausible or at minimum biologically inspired. ANN-like technologies with
shown strengths in these areas may be most relevant for algorithms that
have been ported to equipment for a specific application, such as computer
vision on a distant sensor or autonomous robots. For event-driven spiking
neural network systems, it is common for the model to be selected or
modified so that it fits within specific hardware features (e.g., picking
models that employ STDP for memristors). The majority of neural network
models have evidently been applied to hardware at a certain point in time.

10.3 ALGORITHM AND LEARNING

Procedures are at the heart of many of the outstanding concerns with

neuromorphic systems. Selecting an appropriate algorithm is contingent on
the neuron, synapse, and network models selected, while certain algorithms
are tailored to particular network configurations, neuron simulations, or
other aspects of modelling choices. In addition, a second concern is whether
or not a system’s retraining or learning should occur on-chip, or if systems
should be learned off-chip and then transmitted to the neuromorphic
architecture. Third, we must decide if online, unsupervised algorithms are
required (in which case they must be implemented on-chip) or if off-line,
supervised approaches are adequate, or if a hybrid approach is best. As a
result of their potential for online learning, neuromorphic systems have
gained popularity in the post-law Moore’s era as a supplementary archi-
tecture. However, even the best-funded neuromorphic systems have had
trouble developing algorithms for configuring their hardware. Algorithms
that are developed specifically for hardware implementation, including
those that run entirely on the chip or with the chip present, are the primary
emphasis of this section, as shown in Figure 10.8.

10.3.1 Supervised learning concept

Back-propagation represents the most popular method for coding compu-
tational models. As a supervised learning technique, back-propagation isn’t
often considered an online approach. Feed-forward neural nets, recurrence
202 Computational Techniques in Neuroscience

Figure 10.8 Step-by-step procedure of on-chip learning.

nets (often back-propagation through time), spiking nets (where often feed-
forward nets are converted to spiking systems), and convolutional nets can
all be trained with back-propagation and its many variants. Since there are
numerous highly optimized software implementations available, using
back-propagation off-line on a conventional host system is the simplest
option. These methods mostly employ elementary back-propagation, which
has been thoroughly explored elsewhere in the neural network literature.

10.4 CONCLUSION

The purpose of this paper is to provide a review of the literature on neu-

romorphic computing. Although the reasons for creating neuromorphic
computers have evolved over time, the requirement for a non-von Neumann
architecture that is low-power, massively parallel, able to perform in real
time, and potentially capable of training or learning online has not. We
went over several models of neurons, synapses, and networks that have
been implemented in neuromorphic and neural network hardware before,
stressing the wide range of options available when designing a neuro-
morphic system.
 Analysis of neural network & neuromorphic computing with hardware 203

Each of these models has its own advantages and disadvantages, making
it unlikely that they will ever be merged into a single, unified theory. It’s safe
to assume that anything from simple feed-forward neural networks to
complex simulations of biological brain networks will continue to coexist in
the neuromorphic computing ecosystem.
The many learning and training algorithms developed for and used by
neuromorphic systems were discussed. Moving forward, we need to focus
on developing dedicated training and learning algorithms for neuromorphic
systems, as opposed to just adopting those built for other architectures.
Our research suggests that this subfield of neuromorphic computing holds
some of the greatest promise for future advancements. We talked about
the big picture of neuromorphic system hardware and the cutting-edge
device-level components and materials that are powering their develop-
ment. As time goes on, there is also plenty of room for improvement here.
We briefly covered some of the ancillary systems for neuromorphic com-
puters, like ancillary software, of which there is relatively little and which
would substantially assist the community. Finally, we went over some of
the uses of neuromorphic computing systems.
With this work, we aimed to provide a comprehensive overview of
neuromorphic computing research across multiple domains. Therefore, we
have added all citations in this revision. The results of this study, we believe,
will serve as motivation for others to create similarly novel approaches
to help fill in the blanks with their own research and to think about how
neuromorphic computers might work for their own purposes.

REFERENCES

[1] C. Mead, “Neuromorphic electronic systems,” Proceedings of the IEEE,

vol. 78, no. 10, pp. 1629–1636, Oct 1990.
[2] D. Monroe, “Neuromorphic computing gets ready for the (really) big time,”
Communications of the ACM, vol. 57, no. 6, pp. 13–15, 2014.
[3] J. Von Neumann and R. Kurzweil, The computer and the brain. Yale
University Press, 2012.
[4] A. M. Turing, “Computing machinery and intelligence,” Mind, vol. 59,
no. 236, pp. 433–460, 1950.
[5] A. F. Murray and A. V. Smith, “Asynchronous vlsi neural networks using
pulse-stream arithmetic,” Solid-State Circuits, IEEE Journal of, vol. 23,
no. 3, pp. 688–697, 1988.
[6] S. Lakkadi, A. Mishra, and M. Bhardwaj, “Security in ad hoc networks,”
American Journal of Networks and Communications, vol. 4, no. 3–1,
pp. 27–34, 2015.
[7] I. Jain and Dr. M. Bhardwaj, “A Survey Analysis of COVID-19 Pandemic
Using Machine Learning (July 14, 2022),” Proceedings of the Advancement in
Electronics & Communication Engineering, 2022, Available at SSRN: https://
ssrn.com/abstract=4159523 or http://dx.doi.org/10.2139/ssrn.4159523
204 Computational Techniques in Neuroscience

[8] F. Blayo and P. Hurat, “A vlsi systolic array dedicated to hopfield

neural network,” in VLSI for Artificial Intelligence. Springer, 1989,
pp. 255–264.
[9] F. Salam, “A model of neural circuits for programmable vlsi im- plementa-
tion,” in Circuits and Systems, 1989. IEEE International Symposium on.
IEEE, 1989, pp. 849–851.
[10] S. Bibyk, M. Ismail, T. Borgstrom, K. Adkins, R. Kaul, N. Khachab, and
S. Dupuie, “Current-mode neural network building blocks for analog
mos vlsi,” in Circuits and Systems, 1990. IEEE International Symposium on.
IEEE, 1990, pp. 3283–3285.
[11] A. Sharma, A. Tyagi, and M. Bhardwaj, “Analysis of techniques and at-
tacking pattern in cyber security approach: A survey,” International Journal
of Health Sciences, vol. 6, no. S2, pp. 13779–13798, 2022. 10.53730/
ijhs.v6nS2.8625
[12] A. Tyagi, A. Sharma, and M. Bhardwaj, “Future of bioinformatics in
India: A survey,” International Journal of Health Sciences, vol. 6, no. S2,
pp. 13767–13778, 2022. 10.53730/ijhs.v6nS2.8624.
[13] F. Distante, M. Sami, and G. S. Gajani, “A general configurable architec-
ture for wsi implementation for neural nets,” in Wafer Scale Integration,
1990. Proceedings.,[2nd] International Conference on. IEEE, 1990,
pp. 116–123.
[14] J. B. Burr, “Digital neural network implementations,” Neural networks,
Concepts, Applications, and Implementations, vol. 3, pp. 237–285, 1991.
[15] M. Chiang, T. Lu, and J. Kuo, “Analogue adaptive neural network circuit,” IEE
Proceedings G (Circuits, Devices and Systems), vol. 138, no. 6, pp. 717–723,
1991.
[16] K. Madani, P. Garda, E. Belhaire, and F. Devos, “Two analog counters
for neural network implementation,” Solid-State Circuits, IEEE Journal of,
vol. 26, no. 7, pp. 966–974, 1991.
[17] A. F. Murray, D. Del Corso, and L. Tarassenko, “Pulse-stream vlsi neural
networks mixing analog and digital techniques,” Neural Networks, IEEE
Transactions on, vol. 2, no. 2, pp. 193–204, 1991.
[18] P. Hasler and L. Akers, “Vlsi neural systems and circuits,” in Computers and
Communications, 1990. Conference Proceedings, Ninth Annual International
Phoenix Conference on. IEEE, 1990, pp. 31–37.
[19] P. Chauhan, and M. Bhardwaj, “Analysis the Performance of Interconnection
Network Topology C2 Torus Based on Two Dimensional Torus,” International
Journal of Emerging Research in Management & Technology, vol. 6, no. 6,
pp. 169–173, 2017.
[20] N. S. Pourush, and M. Bhardwaj, “Enhanced Privacy-Preserving Multi-
Keyword Ranked Search over Encrypted Cloud Data,” American Journal of
Networks and Communications, vol. 4, no. 3, pp. 25–31, 2015.
[21] J. Wu, S. A. Haider, M. Bhardwaj, A. Sharma, and P. Singhal, “Blockchain-
Based Data Audit Mechanism for Integrity over Big Data Environments,”
Security and Communication Networks, vol. 2022, 2022.
[22] J.-C. Lee and B. J. Sheu, “Parallel digital image restoration using adaptive
vlsi neural chips,” in Computer Design: VLSI in Computers and Processors,
1990. ICCD’90. Proceedings, 1990 IEEE International Conference on.
IEEE, 1990, pp. 126–129.
 Analysis of neural network & neuromorphic computing with hardware 205

[23] L. Tarassenko, M. Brownlow, G. Marshall, J. Tombs, and A. Murray, “Real-

time autonomous robot navigation using vlsi neural networks,” in Advances
in neural information processing systems, 1991, pp. 422–428.
[24] L. Akers, M. Walker, D. Ferry, and R. Grondin, “A limited- interconnect,
highly layered synthetic neural architecture,” in VLSI for artificial intelli-
gence. Springer, 1989, pp. 218–226.
[25] P. H. Leong and M. A. Jabri, “A vlsi neural network for morphology clas-
sification,” in Neural Networks, 1992. IJCNN., International Joint
Conference on, vol. 2. IEEE, 1992, pp. 678–683.
[26] M. Bhardwaja and A. Ahlawat, “Evaluation of Maximum Lifetime Power
Efficient Routing in Ad hoc Network Using Magnetic Resonance Concept,”
Recent Patents on Engineering, vol. 13, no. 3, pp. 256–260, 2019.
[27] M. Bhardwaj and A. Ahalawat, “Improvement of Lifespan of Ad hoc
Network with Congestion Control and Magnetic Resonance Concept,” in
International Conference on Innovative Computing and Communications.
Springer, Singapore, 2019, pp. 123–133.
[28] M. Bhardwaj and A. Ahlawat, “Optimization of Network Lifetime with
Extreme Lifetime Control Proficient Steering Algorithm and Remote Power
Transfer,” DEStech Transactions on Computer Science and Engineering,
2017.
[29] G. Cairns and L. Tarassenko, “Learning with analogue vlsp mlps,” in
Microelectronics for Neural Networks and Fuzzy Systems, 1994., Proceedings
of the Fourth International Conference on. IEEE, 1994, pp. 67–76.
[30] H. Markram, “The human brain project,” Scientific American, vol. 306,
no. 6, pp. 50–55, 2012.
[31] J. Schemmel, D. Bruderle, A. Grubl, M. Hock, K. Meier, and S. Millner,
“A wafer-scale neuromorphic hardware system for large- scale neural
modeling,” in Circuits and Systems (ISCAS), Proceedings of 2010 IEEE
International Symposium on. IEEE, 2010, pp. 1947–1950.
[32] J. Backus, “Can programming be liberated from the von neumann style?: a
functional style and its algebra of programs,” Communications of the ACM,
vol. 21, no. 8, pp. 613–641, 1978.
[33] W. S. McCulloch and W. Pitts, “A logical calculus of the ideas immanent
in nervous activity,” The Bulletin of Mathematical Biophysics, vol. 5, no. 4,
pp. 115–133, 1943.
[34] M. Bhardwaj and A. Ahlawat, “Enhance Lifespan of WSN Using Power
Proficient Data Gathering Algorithm and WPT,” DEStech Transactions on
Computer Science and Engineering, 2017.
[35] M. Sharma, S. Rohilla, and M. Bhardwaj, “Efficient Routing with
Reduced Routing Overhead and Retransmission of Manet,” American
Journal of Networks and Communications. Special Issue: Ad Hoc
Networks, vol. 4, no. 3-1, pp. 22–26, May 2015. doi: 10.11648/j.ajnc.s.
2015040301.15
[36] M. Bhardwaj, “7 Research on IoT Governance, Security, and Privacy Issues
of Internet of Things,” Privacy Vulnerabilities and Data Security Challenges
in the IoT, p. 115, 2020.
[37] E. M. Izhikevich, “Which model to use for cortical spiking neurons?”
IEEE Transactions on Neural Networks, vol. 15, no. 5, pp. 1063–1070,
2004.
206 Computational Techniques in Neuroscience

[38] A. L. Hodgkin and A. F. Huxley, “A quantitative description of membrane

current and its application to conduction and excitation in nerve,” The
Journal of Physiology, vol. 117, no. 4, p. 500, 1952.
[39] A. Basu, C. Petre, and P. Hasler, “Bifurcations in a silicon neuron,” in
Circuits and Systems, 2008. ISCAS 2008. IEEE International Symposium
on. IEEE, 2008, pp. 428–431.
[40] F. Castanos and A. Franci, “The transition between tonic spiking and
bursting in a six-transistor neuromorphic device,” in Electrical Engineering,
Computing Science and Automatic Control (CCE), 2015 12th International
Conference on. IEEE, 2015, pp. 1–6.
[41] S. P. DeWeerth, M. S. Reid, E. A. Brown, and R. J. Butera Jr, “A com-
parative analysis of multi-conductance neuronal models in silico,”
Biological Cybernetics, vol. 96, no. 2, pp. 181–194, 2007.
[42] D. Dupeyron, S. Le Masson, Y. Deval, G. Le Masson, and J.-P. Dom, “A
bicmos implementation of the hodgkin-huxley formalism,” in Microelectronics
for Neural Networks, 1996., Proceedings of Fifth International Conference on.
IEEE, 1996, pp. 311–316.
[43] A. Kumar, S. Rohilla, and M. Bhardwaj, “Analysis of Cloud Computing
Load Balancing Algorithms,” International Journal of Computer Sciences
and Engineering, vol. 7, pp. 359–362, 2019.
[44] M. Bhardwaj, A. Ahlawat, and N. Bansal, “Maximization of Lifetime
of Wireless Sensor Network with Sensitive Power Dynamic Protocol,”
International Journal of Engineering & Technology, vol. 7, no. 3.12,
pp. 380–383, 2018.
[45] M. Bhardwaj and A. Ahlawat, “Wireless Power Transmission with Short and
Long Range Using Inductive Coil,” Wireless Engineering and Technology,
vol. 9, pp. 1–9, 2018. doi: 10.4236/wet.2018.91001.
[46] F. Grassia, T. Le´vi, S. Sa¨ıghi, and T. Kohno, “Bifurcation analysis in
a silicon neuron,” Artificial Life and Robotics, vol. 17, no. 1, pp. 53–58,
2012.
[47] M. Grattarola, M. Bove, S. Martinoia, and G. Massobrio, “Silicon neuron
simulation with spice: Tool for neurobiology and neural networks,” Medical
and Biological Engineering and Computing, vol. 33, no. 4, pp. 533–536,
1995.
[48] A. M. Hegab, N. M. Salem, A. G. Radwan, and L. Chua, “Neuron model with
simplified memristive ionic channels,” International Journal of Bifurcation
and Chaos, vol. 25, no. 06, p. 1530017, 2015.
[49] K. M. Hynna and K. Boahen, “Thermodynamically equivalent silicon models
of voltage-dependent ion channels,” Neural Computation, vol. 19, no. 2,
pp. 327–350, 2007.
Chapter 11

Analysis of technology research and

ADHD with the neurodivergent
reader
A survey
Manish Bhardwaj1, Jyoti Sharma2, Analp Pathak2,
Vinay Kumar Sharma3, and Mayank Tyagi2
1
Department of Computer Science and Information Technology,
KIET Group of Institutions, Delhi-NCR, Ghaziabad, India
2
Department of Information Technology, KIET Group of Institutions,
Delhi-NCR, Ghaziabad, India
3
School of Computing Science and Engineering, Galgotias University,
Greater Noida, India

11.1 INTRODUCTION

The word “neurodiversity” describes the wide variation in people’s neu-

rological make-ups, with a focus on the many different ways people think.
This movement is driven by self-advocates and opposes the deficit or
impairment labeling of neurotypes [1].
This was originally brought to human–computer interaction (HCI) by
Dalton, who, in discussing the possibilities of the notion of neurodiversity
for technology design and research, urged HCI researchers to work with
neurodiverse populations and to recognize and promote the talents of
individuals with such differences [2–4]. Neurodiversity, as used here, is
the lived experience of cognitive and/or expressive differences from the
neurotypical (including but not limited to medical designations such as
autism, dyslexia, and attention deficit hyperactivity disorder [ADHD]).
Similarly, Mankoff et al. advocated for research on assistive technology
to be guided by the lens of disability studies and to actively connect with
disability communities. So, there has been a rise in research on neurodi-
vergent populations, and as a result, neurodiverse academics have con-
tributed to a better understanding of the language used in the fields of
computer science and HCI [5]. First recommendations for the design of
technology related to ADHD already exist, but primarily from a medically
educated perspective.
Recent work by Cibrian et al. provides an overview of tools that can help
children with ADHD learn to self-regulate [6,7]. What has been lacking up
to this point is a critical investigation that focuses on how technological

DOI: 10.1201/9781003398066-11 207

208 Computational Techniques in Neuroscience

Figure 11.1 Basic conditions under neurodiversity.

research occurs in the context of ADHD, especially from a perspective

expressly molded by persons with ADHD themselves.
The first-person perspective of people with disabilities (whether re-
searchers or not) is crucial to understanding how these technologies work,
and while modern HCI research in the realm of technologies and dis-
abilities is increasingly informed by theories and practices from disability
studies, this is not always the case [8,9]. Although autoethnographic
studies and research focused on agency and self-determination have
recently entered the field, these viewpoints have historically been less
prevalent. Figure 11.1 shows the conditions that are present under one
name “neurodiversity”.
We contribute to this expanding body of work by conducting a system-
atic literature review of HCI and computer science studies that focus on
ADHD [10]. This is consistent with earlier studies of technology and neu-
rodivergence or disabilities more generally, such as a review by Spiel et al.
of technologies for autistic children, a survey by Mack et al. of accessibility
research, studies of wearables for autism intervention, and studies of
research pertaining to neurodivergence and play [11].
Moreover, our work is a reaction to and an expansion on prior recom-
mendations for learning prospective design techniques relating to technolo-
gies for people with ADHD [12]. Since neurodivergent people are often left
out of studies focusing on them, we think it’s important to expressly adopt
this subjective stance [13]. This level of participation, however, affects how
we read, analyze, and understand the corpus content. So, our emotional
 Analysis of technology research and ADHD 209

Figure 11.2 Research fields related to human–computer interaction.

reactions to how our coworkers characterize us, our communities, and our
loved ones play a significant role in our writing [14].
We now provide context for our study by outlining our knowledge of
ADHD, describing our previous work in the context of HCI and neurodi-
versity, and introducing our theoretical support from Critical Disability
Studies and, more specifically, Crip Technoscience. Figure 11.2 shows the
various research fields that are related to the HCI field.
After that, we explain our methodology in further detail. Our analysis
and results show how existing research exposes ADHD as a problem space
for technology design due to solutionist and paternalistic perspectives
of the intended audience [15]. We draw conclusions for the technological
research community, propose hypotheses about possible solutions, con-
sider the consequences of engaging with works on a deeply personal level,
and sketch the contours of possible future developments based on these
results [16–18].
210 Computational Techniques in Neuroscience

11.2 DOMAIN PREVIOUS STUDY

First, we explain ADHD briefly while addressing the ongoing research in

the field of HCI and neurodivergence [19], before our discussion of relevant
hypotheses building on Radical Theory and practice and specifically the
concept of “Crip Technoscience”.

11.2.1 ADHD
Clinically, hyperactivity, impulsivity, and inattention are the hallmarks of
ADHD. Predominant inattentiveness [20–22], hyperactive-impulsiveness,
and a mixed profile are frequently used diagnostic criteria [ibid]. For a
long time, ADHD, or hyperkinetic behavior syndrome, was thought of
as a condition that only affected kids [23].
Because of its early definition as a childhood condition, adults who
sought a diagnosis were overlooked, and its failure to account for the
“inattentive” (daydreamer) type led to a long-held misconception that
ADHD mostly affects boys [24].
Research suggests that the prevalence of ADHD in the general population
diminishes progressively throughout age groups, lending credence to the
idea that one might “grow out of” the disorder [25].
This misconception persists despite growing evidence that the under-
lying neurological distinctions that define ADHD persist over the course
of a person’s lifetime [26–29]. ADHD may be less noticeable in adults
because of differences in their environments (work, family life, etc.;
whereas in most countries, children all go to school), and because they
have learned to adapt to their surroundings through time, using a variety
of contextual masking tactics.
Misdiagnosis of depression or oppositional defiant disorder might occur
because of the presentation of traits being connected to conventional
standards along gender and race.
In recent years, those who have been diagnosed with ADHD have con-
tributed to our growing body of knowledge about what it’s like to live
with and manage ADHD in the real world.
Research that takes both a critical and appreciative stance and the par-
ticipation of people with ADHD in study both add to the credibility of these
narratives [30–33]. In this context, lobbying focuses mostly on dispelling
harmful myths and rebutting internalized societal stigma.

11.2.2 Neurodivergence and HCI

Proponents in HCI often claim to do accessibility research, although the
precise nature of this work is rarely defined or agreed upon. Mack et al. found
that only around 5% of self-identified accessibility research publications
 Analysis of technology research and ADHD 211

published between 1994 and 2019 focused on neurodivergent communities

(which includes autistic populations).
In a similar vein, a recent analysis by Spiel and Gerling on HCI games
research and neurodivergent communities found that a disproportionate
amount of attention was paid to autistic populations compared to those
with other conditions [34].
As a result, autism spectrum disorder (ASD) and children with ASD
are common topics of discussion in the existing reviews in this field.
One review, driven primarily by a medical and deficiency-oriented
approach [35], provides implications for technological research in the
setting of ADHD, and the other, focused on children with ADHD, does
the same.
Here, we offer a quick recap of the current literature devoted to
designing technology for people with neurodiversities. This study includes
instances in which neurodivergent people were involved in the design of
data visualizations due to nonstandard processing of visual information
or the creation of shared spaces in which neurotypical and neurodivergent
people can meet [36], for example, in the context of children’s play or
adults in the context of the workplace. Figure 11.3 shows the graphical
representation of neurodiversity between low cognative and high cogna-
tive ability.
Self-determined design experiments are rarely seen in published work,
but one such example is Damiani’s creative examination of the particulars
of neurodivergent embodiments.
We see these as crucial examples of what research and design may look
like when driven by neurodiverse people, and this bodes well for future

Figure 11.3 Graphical representation of neurodiversity between low cognative and high
cognative ability.
212 Computational Techniques in Neuroscience

studies on ADHD. Following the proverb in disability activism that claims

“Nothing about us without us,” Parsons et al. show the knowledge- and
human-rights-based implications of including autistic people in research
about them [37,38], which is a response to the overwhelming amount of
research being driven from the outside in the context of autism.
They provide a real example to show how one author’s declaration of an
ADHD diagnosis led to them being disqualified from consideration for
an otherwise suitable job posting [ibid].
For both neurotypical and neurodiverse academics, the inclusion of
neurodiverse communities in their studies has important consequences for
the negotiation of a shared impaired (or nondisabled) identity.
However, little attention has been paid to the potential forms that self-
determined and participatory research can take in the context of ADHD
[39], and a few of researchers who have ADHD have chosen not to declare
their condition publicly.
As a result, our study contributes to the literature on neurodiversity and
technology development by offering a dedicated analysis to the ADHD
setting carried out by writers who do so expressly as impacted parties.

11.2.3 Studies of disabilities with a critical lens

and crip technology
Disability Studies was first introduced to HCI scholars by Mankoff et al. in
2010. To be more specific, the autonomous premise of Disability Studies
provides a critical counterpoint to the common understanding of assistive
technologies for people with disabilities, which sees them as devices whose
primary purpose is to maintain a corporeal standard, rather than to aid people
with disabilities in developing and employing their own unique strategies for
navigating environments that were not designed with them in mind [40].
To describe this bias in technological innovation and research, Shew
created the term “technoableism”. Feminist and Queer theories, as well as
those from other critical fields, have all contributed to the development
of modern Disability Studies’ investigation into different ways of under-
standing disability.
Because of the efforts of disability activists, the term “crip” has been
reclaimed as a positive identity marker (similar to “queer”) to describe
people with these types of identities.
Workshops, lively debates, and analytical lenses are only some examples
of the growing popularity of incorporating ideas from disability studies into
the field of HCI.
As such, the concept of diverse models of disability is important to our
work, positionality, and analysis. Many other perspectives exist on dis-
ability, but the most common divide is between the medical (or deficiency-
oriented) model, which emphasizes the individual, and the social (or access-
oriented) model, which emphasizes structural impediments.
 Analysis of technology research and ADHD 213

Living with an ADHD neurotype, however, is characterized by a dis-

crepancy between internal expectations and the way in which the individual
processes information from the outside world.
Disabling experiences become ingrained in our life alongside the wide-
spread stigma of declaring ADHD and requesting adjustments. So, we have
chosen to use the term “disability” in order to show our support for those
who share these experiences, both as a person looking for crip kinship and
as a person describing aspects of our own identity.
Because not doing so or working from a medical viewpoint on disabilities
is itself a political act – any presumption of a default status is fundamentally
political – we interact with this identity as disabled on purpose, despite the
fact that doing so is a political act.
Hamraie and Fritsch provide a variety of perspectives on the nature
of these technologies within the framework of viewing disability, and by
extension, the space in which people with disabilities live, as political.
Their Crip Technoscience framework prioritizes people with disabilities
by pledging to increase accessibility, encourage collaboration, and pro-
mote disability justice. While HCI has been exposed to the concept of
neurodiversity, no prior analysis of Crip Technoscience has focused on its
relevance to neurodivergent populations.

11.3 RESEARCH METHODS

This literature evaluation took around 2.5 years to complete, beginning in

early 2019.
We show how we got here by first describing the reading perspective we
used, which was also the driving force behind our study.
Furthermore, we detail the methods we used to build and analyze the
corpus as a whole. All of the writers’ professional interest in technology
research pertaining to ADHD inspired this undertaking. The first author
reached out to other neurodiverse people to talk about problems they had
noticed in self-descriptions.
We all got together at a conference at the beginning of 2019 to introduce
ourselves to one another and launch the project. We also passed out reading
glasses and established preliminary areas of interest at this time.
Our goal in demonstrating the relevance of this work to people with
ADHD is, in part, to “support the construction of ‘safer’ settings for
marginalized academics and students,” in this case, neurodiverse peers.
Following Kafer’s lead in adopting crip time, which “bends the clock to meet
crippled bodies and minds” instead of vice versa, we spread this effort out over
the course of 2.5 years to make explicit space for multiple temporalities.
Moreover, we must recognize that we had to factor in sorrow time into
this crip period, as we lost peers and friends or had to deal with the re-
percussions of a global pandemic, including individual disease, ourselves.
214 Computational Techniques in Neuroscience

11.3.1 Development of corpus

We built not one, but two corpora, the main one and an expansion on
top of it. Our combined corpus excludes workshop descriptions, panel
proposals, and journal articles, but includes full/long papers, short papers,
works-in-progress and posters, demonstrations, and journal articles re-
porting on completed research.
On February 19, 2019, we compiled the initial core corpus by searching
for the terms “ADHD” and “ADD” in the titles, abstracts, and keywords of
publications in the ACM Digital Library Guide to Computing Literature
and the hcibib indexing platform (although the latter stopped indexing
content in 2018).
For this reason, we have chosen to restrict our review to papers dealing
with computing and HCI. As a result, 56 papers were produced. After
reading each abstract, we culled the corpus down to 52 papers that
directly addressed our research questions. We refrained from making
any more cuts to the corpus, which enabled us to compare and contrast
longer and shorter publications and to report on current trends. The use
of merely these two references suggests that we have not made a concerted
effort to read through the technical articles published in other clinical
or medical journals. Yet, we zero in on the ways in which computing,
and especially HCI-oriented literature, constructs ADHD and related
technologies.
Afterwards, we surveyed more recent papers from the same source,
covering the time period from around the middle of 2018 to the end of
2020, to see if the patterns we had noticed in the original corpus were still
there, and/or if any new patterns emerged.
One workshop, one paper written in a language none of the authors
can read fluently, and full online versions of PhD theses were all dismissed
as irrelevant to our investigation. This process added 48 more publica-
tions to the ‘expanded corpus’ (after duplicates were removed). With the
growing body of literature on neurodiversity in HCI, we decided to do a
broad survey of these additional studies to test the validity of our earlier
conclusions.

11.4 RESULT AND DISCUSSION

Each of the four authors read the core corpus with a focus on either
participants (i.e., who was included or addressed in the research and how
did potential participants behave), disability (i.e., how did authors con-
ceptualize and explain ADHD), researchers (i.e., the larger research
framing and disciplinary origin of the work), or technology (i.e., the
design and development processes, including the artifacts and their
purpose).
 Analysis of technology research and ADHD 215

To a) facilitate chunky reading in a systematic way and b) permit possibly

divergent reading of the same publications to correctly account for varied
strengths in the research we surveyed, we defined separate reading lenses.
We then compared and contextualized our studies across different au-
thors by using a procedure of procedural coding along subcategories of
these focuses. The authors addressed several papers that had received
contrasting ratings after perusing each other’s remarks and codes. The goal
here was not to force consistency among the many evaluations, but rather
to make effective use of them in order to better comprehend the paper’s
merits and shortcomings.
In order to conduct our theoretical investigation, we first created a
conceptual map of the publications in the corpus. After that, we used
Critical Disability Studies theories to analyze the publications’ implications
from the neurodiverse perspective. We followed Boyatzis’ method of theme
analysis, which permits inductive and deductive coding by numerous coders
while taking into account the situatedness and subjective quality standards
of acceptable categorization.
We avoided a codebook method, which is more numerically focused, on
purpose so that we could include more voices in our study and bolster
our findings by working through our disagreements.
Two of the four original coders similarly analyzed the expanded corpus
to spot themes in recent research that compared to our initial findings.
With the exception of when a new pattern emerged, this study relied on
previously defined codes. Discordant findings are presented in separate
paragraphs adjacent to the relevant sections. Many of the papers in this
extended corpus are marked as part of the “extended corpus” because they
are excellent illustrations of patterns observed in the main corpus.
We make no assumptions about the neurotypes of the people who par-
ticipated in the research that is reflected in our database. Indeed, we have
a keen awareness of the potential adverse effects of disclosure. Instead,
we evaluate the research strategy and viewpoint taken from a discursive
perspective.
To paraphrase Haraway, we intend to “take the privilege of our limited
perspective[s]” seriously, rather than make generalizations based on our
reading.
Our findings highlight the prevalence of tools for intervention and
diagnosis in the field of ADHD. We also found that those with ADHD
are often left out of the loop when it comes to the development of tools
meant specifically for them. This results in a solitary framing of the target
population as a source of “issues to address” and technology that pre-
dominantly mirror neurotypical expectations rather than neurodivergent
wants and desires.
The implications of these findings for identifying the “user” of a given
technology in neurodiverse settings are discussed, along with what Crip
Technoscience-based alternatives to current technologies would entail.
216 Computational Techniques in Neuroscience

11.5 RESEARCH GAPS

In addition to the obvious implications for the field of ADHD technology

research as a whole, we also found a number of open areas where additional
investigation into the role(s) technologies play in this setting would be beneficial.
Our presentation of these alternatives is meant to be illustrative and
exploratory, and not exhaustive. It does reveal, however, that there are
fundamental gaps in the knowledge production about ADHD and tech-
nologies, particularly in regards to language, the inclusion of persons with
ADHD, and a closer examination of the discursive and material repercus-
sions that technology embodies.
Most of the studies we looked at were conducted on children, and even
those that weren’t frequently employed children or adolescents as subjects
(i.e., children or adolescents).
The lack of research involving adults, and especially older persons with
ADHD, stands out to us as a major hole. The function that technology plays in an
adult’s life is likely to change from that of a child’s since adults have less external
structure in their lives. Also, compared to adults, children are given less credit for
having the ability to make their own choices regarding their interactions with
technology and are more likely to become embedded in existing power systems.
As a result, purposeful interaction with adults has the potential to un-
dermine the paternalistic assumptions that now underpin much techno-
logical investigation into ADHD.
We also see promise in tools that enhance executive function during
freely chosen activities. Instead of forcing persons with ADHD to follow a
predetermined set of tasks, we propose looking for ways to help them figure
out how to accomplish the goals they have for themselves.
Finally, we think there’s value in studying how persons with ADHD are
adapting and using current technologies. One way this could manifest itself is in
the way these people use social media as a means to both find and share infor-
mation and form groups centered on the construction of personal meaning.
There is also a lack of research on why many life-organization methods
that are effective for neurotypical people are ineffective for people with
ADHD (e.g., weekly calendar planning).
Since we recognize that researchers in this field of study must first es-
tablish a foundation of trust, we call on our peers and colleagues alike to
help bring this potential to fruition.
Here, we’re hoping that our work might help lay the groundwork for a
fruitful discussion.

11.6 CONCLUSION

From the perspective of neurodiverse readers, we analyzed works con-

cerning technology in the context of ADHD. Our findings highlight the
 Analysis of technology research and ADHD 217

barriers that prevent people with ADHD from participating in the design
of aiding technologies.
We also found that until recently, researchers’ perspectives remained
static, with the exception of a shift toward a concentration on diagnostic
and interventionist approaches in technology development.
Consequences of the current research’s reliance on a deficit model of
ADHD features were then discussed. We next provide a number of sug-
gestions for developments in this area based on our speculative choices.
In this way, we advocate for researchers to abandon their adherence to
neurotypical standards and approach their work with us and other mar-
ginalized communities less from a paternalistic and more from a solidaric
and community-oriented stance.
There are constraints on this work, like there are on any. For one,
all of us writing this are white, we live and work in the Global North,
and, as we’ve already mentioned, we have the advantage of having
access to a diagnosis, which in turn governs our ability to receive social
accommodations.
As a matter of fact, we owe our academic success to the fortunate
circumstances of our own educational backgrounds. We can thus only
speak for ourselves in our mutual assessment and make no claims to
represent all people with ADHD. However, we provide a close reading
of existing works and outline the future potential of such appreciative
approaches in light of the growing number of researchers who openly
disclose their ADHD, also within HCI, and the emerging research into
self-determined options for technologies in this context.
We can’t exist without technology, but if it’s only used to sort individuals
into boxes and control their actions, we have a responsibility as technolo-
gists to find better uses for it.

REFERENCES

[1] Ashwag Al-Shathri, Areej Al-Wabil, and Yousef Al-Ohali. 2013. Eye-
Controlled Games for Behavioral Therapy of Attention Deficit Disorders.
In HCI International 2013 – Posters’ Extended Abstracts, Constantine
Stephanidis (Ed.). Springer Berlin Heidelberg, Berlin, Heidelberg, 574–578.
[2] Shahab U. Ansari. 2010. Validation of FS+LDDMM by Automatic
Segmentation of Caudate Nucleus in Brain MRI. In Proceedings of the
8th International Conference on Frontiers of Information Technology
(Islamabad, Pakistan) (FIT ’10). Association for Computing Machinery,
New York, NY, USA, Article 10, 6 pages. 10.1145/1943628.1943638
[3] J. Anuradha, Tisha, Varun Ramachandran, K. V. Arulalan, and B. K.
Tripathy. 2010. Diagnosis of ADHD Using SVM Algorithm. In Proceedings
of the Third Annual ACM Bangalore Conference (Bangalore, India)
(COMPUTE ’10). Association for Computing Machinery, New York, NY,
USA, Article 29, 4 pages. 10.1145/1754288.1754317.
218 Computational Techniques in Neuroscience

[4] A. Kumar, S. Rohilla, and M. Bhardwaj (2019). Analysis of Cloud

Computing Load Balancing Algorithms. International Journal of Computer
Sciences and Engineering, 7, 359–362.
[5] M. Bhardwaj, A. Ahlawat, and N. Bansal (2018). Maximization of
Lifetime of Wireless Sensor Network with Sensitive Power Dynamic
Protocol. International Journal of Engineering & Technology, 7(3.12),
380–383.
[6] M. Bhardwaj and A. Ahlawat (2018) Wireless Power Transmission with
Short and Long Range Using Inductive Coil. Wireless Engineering and
Technology, 9, 1–9. doi: 10.4236/wet.2018.91001.
[7] Othman Asiry, Haifeng Shen, and Paul Calder. 2015. Extending Attention
Span of ADHD Children through an Eye Tracker Directed Adaptive User
Interface. In Proceedings of the ASWEC 2015 24th Australasian Software
Engineering Conference (ASWEC’ 15 Vol. II). Association for Comput-
ing Machinery, New York, NY, USA, 149–152. 10.1145/2811681.
2824997
[8] Kathleen B Aspiranti and David M Hulac. 2021. Using Fidget Spinners to
Improve On-Task Classroom Behavior for Students With ADHD. Behavior
Analysis in Practice (2021), 1–12.
[9] Peter Balan, Eva Balan-Vnuk, Mike Metcalfe, and Noel Lindsay. 2015.
Concept mapping as a methodical and transparent data analysis process.
In Handbook of Qualitative Organizational Research. Routledge, 350–362.
[10] Kess L. Ballentine. 2019. Understanding Racial Differences in Diagnosing
ODD Versus ADHD Using Critical Race Theory. Families in Society 100,
3 (2019), 282–292. 10.1177/1044389419842765 arXiv: https://doi.org/
10.1177/1044389419842765
[11] M. Bhardwaj and A. Ahlawat (2017). Enhance Lifespan of WSN Using Power
Proficient Data Gathering Algorithm and WPT. DEStech Transactions on
Computer Science and Engineering.
[12] Megha Sharma, Shivani Rohilla, and Manish Bhardwaj, Efficient Routing
with Reduced Routing Overhead and Retransmission of Manet, American
Journal of Networks and Communications. Special Issue: Ad Hoc
Networks. Volume 4, Issue 3–1, May 2015, pp. 22–26. doi: 10.11648/
j.ajnc.s.2015040301.15
[13] M. Bhardwaj (2020). 7 Research on IoT Governance, Security, and Privacy
Issues of Internet of Things. Privacy Vulnerabilities and Data Security
Challenges in the IoT, 115.
[14] Elizabeth Barnes. 2016. The minority body: A theory of disability. Oxford
University Press.
[15] M. Á. Bautista, A. Hernández-Vela, S. Escalera, L. Igual, O. Pujol, J. Moya,
V. Violant, and M. T. Anguera. 2016. A Gesture Recognition System for
Detecting Behavioral Patterns of ADHD. IEEE Transactions on Cybernetics
46, 1 (2016), 136–147.
[16] Robert Beaton, Ryan Merkel, Jayanth Prathipati, Andrew Weckstein, and
Scott McCrickard. 2014. Tracking Mental Engagement: A Tool for Young
People with ADD and ADHD. In Proceedings of the 16th International
ACM SIGACCESS Conference on Computers & Accessibility (Rochester,
New York, USA) (ASSETS ’14). Association for Computing Machinery,
New York, NY, USA, 279–280. 10.1145/2661334.2661399
 Analysis of technology research and ADHD 219

[17] M. Bhardwaja and A. Ahlawat (2019). Evaluation of Maximum Lifetime

Power Efficient Routing in Ad hoc Network Using Magnetic Resonance
Concept. Recent Patents on Engineering, 13(3), 256–260.
[18] M. Bhardwaj and A. Ahalawat (2019). Improvement of Lifespan of Ad
hoc Network with Congestion Control and Magnetic Resonance Concept.
In International Conference on Innovative Computing and Communications
(pp. 123–133). Springer, Singapore.
[19] M. Bhardwaj and A. Ahlawat (2017). Optimization of Network Lifetime with
Extreme Lifetime Control Proficient Steering Algorithm and Remote Power
Transfer. DEStech Transactions on Computer Science and Engineering.
[20] P. Chauhan and M. Bhardwaj (2017). Analysis the Performance of
Interconnection Network Topology C2 Torus Based on Two Dimensional
Torus. International Journal of Emerging Research in Management &
Technology, 6(6), 169–173.
[21] Cynthia L. Bennett, Erin Brady, and Stacy M. Branham. 2018. Inter-
dependence as a Frame for Assistive Technology Research and Design. In
Proceedings of the 20th International ACM SIGACCESS Conference on
Computers and Accessibility (Galway, Ireland) (ASSETS ’18). Association for
Computing Machinery, New York, NY, USA, 161–173. 10.1145/3234695.
3236348
[22] Cynthia L. Bennett and Daniela K. Rosner. 2019. The Promise of
Empathy: Design, Disability, and Knowing the “Other”. Association for
Computing Machinery, New York, NY, USA, 1–13. 10.1145/3290605.
3300528
[23] Jared David Berezin. 2014. Disabled capital: A narrative of attention deficit
disorder in the classroom through the lens of Bourdieu’s capital. Disability
Studies Quarterly 34, 4 (2014).
[24] K Birth. 2017. Time Blind. Springer.
[25] N. S. Pourush and M. Bhardwaj (2015). Enhanced Privacy-Preserving Multi-
Keyword Ranked Search over Encrypted Cloud Data. American Journal
of Networks and Communications, 4(3), 25–31.
[26] J. Wu, S. A. Haider, M. Bhardwaj, A. Sharma, and P. Singhal (2022).
Blockchain-Based Data Audit Mechanism for Integrity over Big Data
Environments. Security and Communication Networks, 2022.
[27] A. Sharma, A. Tyagi, and M. Bhardwaj (2022). Analysis of techniques and
attacking pattern in cyber security approach: A survey. International Journal
of Health Sciences, 6(S2), 13779–13798. 10.53730/ijhs.v6nS2.8625
[28] Mark Blythe, Jamie Steane, Jenny Roe, and Caroline Oliver. 2015.
Solutionism, the Game: Design Fictions for Positive Aging. Association for
Computing Machinery, New York, NY, USA, 3849–3858. 10.1145/2702123.
2702491
[29] Natale Salvatore Bonfiglio, Roberta Renati, Davide Parodi, Eliano Pessa,
Dolores Rollo, and Maria Pietronilla Penna. 2020. Use of training with
BCI (Brain Computer Interface) in the management of impulsivity. In 2020
IEEE international symposium on medical measurements and applications
(MeMeA). IEEE, 1–5.
[30] LouAnne Boyd. 2019. Designing Sensory-Inclusive Virtual Play Spaces for
Children. In Proceedings of the 18th ACM International Conference on
Interaction Design and Children. 446–451.
220 Computational Techniques in Neuroscience

[31] LouAnne Boyd, Kendra Day, Ben Wasserman, Kaitlyn Abdo, Gillian Hayes,
and Erik Linstead. 2019. Paper Prototyping Comfortable VR Play for
Diverse Sensory Needs. In Extended Abstracts of the 2019 CHI Conference
on Human Factors in Computing Systems (CHI EA ’19). Association for
Computing Machinery, New York, NY, USA, 1–6. 10.1145/3290607.
3313080
[32] Emeline Brulé and Katta Spiel. 2019. Negotiating Gender and Disability
Identities in Participatory Design. In Proceedings of the 9th International
Conference on Communities & Technologies - Transforming Communities
(Vienna, Austria) (C&T ’19). Association for Computing Machinery,
New York, NY, USA, 218–227. 10.1145/3328320.3328369
[33] Fiona Campbell. 2009. Contours of ableism: The production of disability
and abledness. Springer.
[34] A. Tyagi, A. Sharma, and M. Bhardwaj (2022). Future of bioinformatics in
India: A survey. International Journal of Health Sciences, 6(S2), 13767–13778.
10.53730/ijhs.v6nS2.8624.
[35] S. Lakkadi, A. Mishra, and M. Bhardwaj (2015). Security in ad hoc net-
works. American Journal of Networks and Communications, 4(3-1), 27–34.
[36] Jain, Ishita and Dr. Manish Bhardwaj, A Survey Analysis of COVID-19
Pandemic Using Machine Learning (July 14, 2022). Proceedings of the
Advancement in Electronics & Communication Engineering 2022, Available
at SSRN: https://ssrn.com/abstract=4159523 or http://dx.doi.org/10.2139/
ssrn.4159523
[37] Will H Canu, Matthew L Newman, Tara L Morrow, and Daniel L. W Pope.
2008. Social Appraisal of Adult ADHD: Stigma and Influences of the
Beholder’s Big Five Personality Traits. Journal of Attention Disorders 11,
6 (2008), 700–710.
[38] Dario Cazzato, Silvia M Castro, Osvaldo Agamennoni, Gerardo Fernández,
and Holger Voos. 2019. A Non-Invasive Tool for Attention-Deficit Disorder
Analysis Based on Gaze Tracks. In Proceedings of the 2nd International
Conference on Applications of Intelligent Systems (Las Palmas de Gran
Canaria, Spain) (APPIS ’19). Association for Computing Machinery,
New York, NY, USA, Article 5, 6 pages. 10.1145/3309772.3309777
[39] Kuo-Chung Chu, Hsin-Jou Huang, and Yu-Shu Huang. 2016. Machine
learning approach for distinction of ADHD and OSA. In 2016 IEEE/ACM
international conference on advances in social networks analysis and mining
(ASONAM). IEEE, 1044–1049.
[40] Franceli L Cibrian, Kimberley D Lakes, Sabrina Schuck, Arya Tavakoulnia,
Kayla Guzman, and Gillian Hayes. 2019. Balancing caregivers and
children interaction to support the development of self-regulation skills
using a smartwatch application. In Adjunct Proceedings of the 2019 ACM
International Joint Conference on Pervasive and Ubiquitous Computing
and Proceedings of the 2019 ACM International Symposium on Wearable
Computers. 459–460.
Index

Accessibility, 208, 210, 213 Brain waves, 85, 86, 87, 89, 92, 97, 98,
Acoustic neuroma, 8 99, 102, 104, 109
Adaptive filtering, 148, 149, 150, 154,
156, 157, 166, 175, 176 Canonical neural computation, 54
ADHD, 8, 75, 86, 87, 88, 89, 96, 100 Carcinoma, 26
Aggregation, 21, 22, 25, 28, 34, 35 Cell membranes, 37
Algorithm, 25, 29, 30, 33, 41, 43, 47, Classical feedforward approach, 43
50, 51, 62, 73, 115, 123, 124, Classical set theory, 19
128, 131, 135, 136, 137, 138, Clinical examination, 20
140, 141, 142, 143, 162, 166, CLM, 53, 54, 56
167, 169, 171, 172, 173, 174, CNN, 50, 51, 53, 54, 56, 104, 105,
175, 178, 182, 184, 185, 186, 106, 118
187, 188, 194, 198, 200, 201, Code efficiency, 54
203, 204, 219 Cognition, 9, 10, 37, 57, 60, 75, 86, 87,
Alternatives, 21, 22, 27, 28, 33, 162 89, 96, 97, 101, 111
Anatomical, 42 Computational tool, 20
ANN, 8, 50, 51, 52, 53, 119, 120, Correlation learning mechanism, 53
123, 124, 125, 147, 151, 152, Craniopharyngioma, 26
154, 157, 171, 209, 211, CT brain images, 53
215, 219
Artifacts, 163, 165, 170, 173, 191, 193 Decision, 1, 7, 10, 19, 20, 21, 22, 23,
Artificial intelligence, 1, 2, 9, 11, 13, 14, 25, 28, 29, 33, 34, 35, 36, 48,
37, 62, 77, 78, 81, 104, 207, 67, 68, 73, 98, 107, 108, 111,
208, 214 131, 134
Astrocytoma, 8 Decision making, 1, 10, 20, 21, 22, 23,
Attributes, 22, 25, 27, 33, 207 25, 33, 34, 35, 107, 108, 111,
Autism spectrum disorder, 57, 211 131, 134
Autistic, 60, 63, 83, 84 Decision support tools, 19
Autoradiography, 26 Decision theory, 19
Decision tree, 21, 48
Backpropagation method, 41 Deep learning, 37, 49, 50, 64, 104,
Beta waves, 87 105, 106
Biased competition, 55 Deterministic, 19, 147, 159, 196
Blind source separation (BSS), 147, 150, Diagnosis, 7, 8, 19, 20, 21, 22, 23, 25,
179, 180, 182, 193, 196, 29, 33, 35, 91, 93, 94, 95, 102,
200, 202 128, 145, 146, 147, 149, 151
Brain tumor, 7, 19, 22, 25, 27, 29, 30, Diagnostic process, 20
33, 56 DIFWA, 25, 27, 28, 33

221
222 Index

Dynamic intuitionistic fuzzy weighting Intuitionistic fuzzy set, 19

averaging operator, 27 Intuitionistic fuzzy weighted average, 21
Intuitionistic fuzzy weighted
EEG data, 45, 103, 104, 105, 106, 114, geometric, 21
162, 164, 165, 166, 168, 169,
170, 171, 172, 173, 174, 176, Jacobian matrix, 42
177, 180, 182, 185, 190, 195,
196, 199, 204 Linear classifier, 55
EEG signals, 103, 108, 109, 110, 111, Logistic regression, 51, 52
114, 115, 117
Electrochemical signals, 85 Machine learning, 9, 47, 48, 50, 104,
Electroencephalography, 85, 87, 195 105, 128, 171
Electrophysiological, 60, 73 Mathematics, 19, 107
Embryonal, 26 Max-pooling, 37
Enhancing, 54, 171 MCDM, 20, 21, 22, 25, 28, 33, 36
EOG signals, 166 Medical decision making, 20, 22
Ependymoma, 26 Medical knowledge, 20
Evolutionary, 25 Medulloblastoma, 26
Meningioma, 26
Fast Discrete S Transform (FDST), 167 Metastases, 26
Forecasts, 37, 51 Multi-criteria decision making, 2
Functional neuroimaging, 43, 44, 57, Multidisciplinary, 37, 81
85, 90, 93, 94, 95, 99, 161
Fuzziness, 20 Neural networks, 39, 40, 50, 51, 52, 53,
Fuzzy, 7, 19, 20, 21, 22, 23, 24, 25, 27, 54, 57, 63, 65, 66, 67, 68, 69,
29, 33, 34, 35, 36, 120, 147, 145, 151, 171, 197, 206, 209,
151, 153, 154, 155, 156, 158, 211, 212, 214, 215, 216,
160, 191, 215, 219 217, 219
Fuzzy set, 19 Neurodegeneration, 119, 145
Fuzzy set theory, 19 Neurodivergent, 8, 60, 63, 74
Fuzzy-based models, 20 Neurodiversity, 7, 60, 61, 63, 65, 67,
69, 71, 73, 74, 75, 76, 77, 78
Generalization, 19, 64, 66 Neuroimaging, 7, 43, 45, 61, 64, 65, 85,
Gliomas, 26 87, 89, 90, 91, 92, 93, 95, 97,
98, 99, 100
Heterostructure, 19 Neurological, 61, 62, 63, 75, 85, 91, 94,
HHM, 22 95, 96, 101, 107, 108, 114,
Hidden layer, 52, 67, 69 141, 145, 157
Hybrid harmonic mean, 22 Neurology, 60, 61, 63, 99, 138
Hypotheses, 37, 61, 72 Neurons, 9, 10, 26, 37, 38, 41, 42, 45,
46, 47, 51, 54, 55, 56, 62, 67,
IFNIS, 28, 33 69, 70, 71, 72, 85, 86, 92, 93,
IFOWG, 21, 25 97, 98, 101, 104, 108, 109,
IFPIS, 28, 33 119, 123, 124, 145, 151, 152,
IFS, 19, 20, 21, 23 161, 162, 208, 210, 211, 212,
IFWA, 21 213, 214, 215, 216, 217, 220
IFWG, 21, 25 Neuroscience, 1, 3, 11, 20, 22, 24, 26,
Inception, 19, 209 28, 30, 32, 34, 36, 38, 40, 42,
Innovative, 2, 25, 68, 79, 80, 210 44, 46, 48, 50, 52, 54, 56, 57,
Interdisciplinary, 9, 11, 19, 50, 58, 59, 60, 61, 62, 64, 66, 68,
107, 108 70, 72, 74, 76, 78, 80, 81, 82,
Intuitionistic fuzzy ordered weighted 84, 85, 86, 88, 90, 92, 94, 96,
geometric, 21 98, 100, 102, 104, 108, 110,
 Index 223

111, 114, 116, 118, 120, 122, Prioritized weighted average (PWA), 22
124, 126, 128, 130, 132, 134,
136, 138, 140, 142, 144, 146, Radiological, 20
148, 150, 152, 154, 156, 158, Real-life, 25
160, 162, 164, 166, 168, 170, Real-life problems, 1, 9, 19
172, 174, 186, 188, 190, 191, Recurrent neural networks (RNN),
192, 193, 194, 196, 197, 198, 50, 61
200, 202, 204, 208, 210, 212, Reinforcement learning, 1, 10, 49
214, 216, 218 RNNs, 61, 67, 68, 69, 70, 72, 73
Neuroscientists, 9, 11, 37, 107, 108,
207, 210 Sensitivity, 54, 55, 124, 125, 126, 146,
Nodes, 37, 51, 67, 123, 152 154, 170
Non-deterministic, 19 Spiking neuron, 37, 212, 213, 214
Normalization, 54, 55, 56, 57 Statistics, 19, 37, 51, 159, 169,
195, 196
Oligoden- droglioma, 26 Symptom, 20, 145
Ordered weighted averaging, 22 Synthetic intelligence, 81
Ordered weighted geometric, 22
Ordered weighted harmonic mean, 22 Techniques, 1, 8, 10, 11, 20, 21, 25, 39,
OWA, 22, 33, 34, 35 59, 61, 63, 70, 72, 77, 85, 90,
OWG, 22 91, 97, 98, 103, 104, 105,
OWHM, 22 106, 107, 108, 115, 119, 120,
123, 128, 131, 132, 150, 151,
Parameters, 39, 42, 119, 123, 141, 146, 161, 162, 163, 165, 166, 167,
151, 152, 154, 158, 159, 168, 168, 169, 170, 171, 173, 174,
171, 174, 185, 213 175, 176, 177, 178, 185, 187,
Pathological, 20 189, 190, 191, 193, 194, 195,
Pattern recognition, 19, 41, 77, 128 197, 199, 201, 202, 203
Perception, 1, 10, 37, 62, 86, 106, Traditional approaches, 19
107, 108 Treatment, 20, 21, 22, 23, 25, 40, 85,
Phenomenological approach, 78 93, 94, 97, 106, 114, 120,
Phenomenology, 40, 78 127, 146
Pineoblastoma, 26
Primary brain tumors, 26 Weighted harmonic mean (WHM), 22
Prioritized, 22, 25, 35