Analysis of Myoelectric Signals to Prosthesis Applications

Bruna Souza Morais

University of São Paulo, São Carlos, Brasil

Dr Samuel Lourenço Nogueira

Federal University of São Carlos, São Carlos, Brasil

Dr Thiago Luiz de Russo

Federal University of São Carlos, São Carlos, Brasil

Dr Gabriela Lopes dos Santos

Federal University of São Carlos, São Carlos, Brasil
Faculty Alfredo Nasser, Goiânia, Brasil

Luisa Fernanda García Salazar

Federal University of São Carlos, São Carlos, Brasil /
Universidad del Rosario, Bogotá, Colômbia

Dr Arlindo Neto Montagnoli

Federal University of São Carlos, São Carlos, Brasil

[email protected]

[email protected]

[email protected]

[email protected]

[email protected]

[email protected]

Abstract. This work presents a myoelectric system measurement incorporated in an arm band
device, particularly designed for application in upper-limb myoelectric prosthesis with pattern
recognition-based sEMG control. For this particular use, specifications such as low cost, high
sampling frequency, size and weight, may be observed. Moreover, it is introduced an acquisition
protocol to evaluation of the myoelectric signal in daily activities.
 Keywords. Electromyography, Upper-limb Myoelectric Prosthesis, Myoelectric Signal,
Wearable sensors

1. Introduction
The perfection of the human body has been motivating engineers and researchers to reproduce it in
technologies to enhance human capabilities. One of the most complex pieces of natural engineering are
the hands. Characteristics such as opposable thumb and the unique arrangement of muscles give them a
great movement versatility going from a power grip to precision manipulation. But their functionality
goes beyond just handling objects, they are also involved in communication and tactile sensing. [1]
Therefore, the amputation of a hand by congenital, traumatic or disease causes, has Psychological,
Sociological and Ergonomic impacts. However, the active upper-limb prosthesis, considered the best
current palliative solution, has high rates of rejection estimated ranging from 25% to 50% [2] especially
when compared with a low-limb prosthesis [3]. Indeed it has been more than 70 years since the first
concept of the myoelectric prosthesis was introduced by Reinhold Reiter [4], and still, there has not been
a good enough solution for a reasonable price found.
The challenges developing an upper limb prosthesis reflect the complexity of the human hand.
Further the low acceptance of the currently available devices motivates additional researches. In fact
developing effective and intuitive myoelectric control could not just help individuals with amputation,
but may further expand possibilities of humans interacting with the environment and control other
systems.
The myoelectric signal, the classic control signal used on an active prosthesis, is the evoked response
by muscle contraction generated by the electrical action potential firing through the muscle fibers. It can
be measured in a non-invasive way by the surface electromyography (sEMG) sensors. [5]
The myoelectric prosthesis models, such as i-Limb from Touch Bionics, Bebionic and Michelangelo
from Ottobock, has proven the feasibility of mechanically dexterous prostheses, even in a commercial
level despite the still high price. Therefore the system-user interface still remains as one of the main
issues causing insufficient functionality, rather than the hardware component itself. [6]
These commercial devices use the Direct Control approach, which uses two opposing muscle groups
to control each direction of one DoF (degree of freedom) and a voltage threshold-based trigger to the
motion. If the prosthesis has multiple DoFs, then the subject through a co-contraction sequentially switch
the different setting modes. A rather more sophisticated and intuitive solution is the Pattern Recognition
Control, which extracts features from the myoelectric signal aiming to identify patterns associated with
the movements, eliminating the need of switching modes. [7] Some of the several parameters common
used to quantify the myoelectric signal are Root-Mean-Square (RMS), Mean Absolute Value, Zero
Crossings, Slope Sign Changes, Waveform Length, Histogram (HIST), marginal Discrete Wavelet
Transform (mDWT). [8]
This work presents a measurement system of the myoelectric signal to be future applied on a pattern
recognition-based sEMG control upper limb prosthesis. Also an acquisition protocol is presented to
evaluate the myoelectric signal from different muscles in daily activity.
The following Section 2 covers the Methodological Approach of this work presenting the acquisition
setup and data acquisition protocol. Section 3 details and discuss the Results. Finally, in Section 4, the
conclusions and future works are presented.

2. Methodology

2.1. Acquisition Setup

In this subsection will be described the main hardware parts that composes the measurement system
and the software used on the data acquisition process.

Surface Electromyography System: The myoelectric measurement system consist of an elastic and
adjustable arm band with three active (indicator) electrodes plus one forth reference electrode as
shown in Figure 1 and Figure 2 - Detail D. These are non-invasive dry active differential electrodes
formed by a sensor of silver coated with silver-chloride, embed in a pre-processing circuit board. Each
electrode was placed in a 3D printed box specially designed to accommodate it and attachable to the
arm band.
The pre-processing circuit was divided in three stages: (I) an instrumentation amplifier Burr-
Brown INA118P; a bandpass filter with (II) a passive high-pass filter and (III) a first-order inverting
low-pass filter. The primary functions of this circuit is signal acquisition, filtering the frequencies
between 48.228 ~ 482.28Hz and amplify (G = 2199.5) the signal adjusting within the operating range
of the microcontroller’s A/D converter.

Figure 1 - The myoelectric measurement arm band.

Acquisition Hardware: The Power Supply Circuit has an unique Li-Po battery (Figure 2 - Detail A) of
2 cells, 7.4V, provides power to the system. Moreover the power supply board (Figure 2 - Detail B)
regulates the voltage in -5V and +5V to properly power the electrodes.
The Microcontroller used was the STM32F103, developed with the ARM Cortex-M3 32-bit RISC
core operating at a 72 MHz frequency built in the STM32F103c8t6 board (Figure 2 - Detail C),
programed at Mbed OS environment. The pre-processed signal once received by the microcontroller
A/D converter, with 12-bit resolution and 0-3.3V operating voltage range, was sampled at 1600Hz
frequency forming data chunks containing 512 samples from each of the three sensors.
Moreover to provide a supplementary data, an Inertial Sensor Xsens MTw Awinda, which is a
wireless motion tracker (Figure 2 - Detail E), was placed in a separated compartment at the bottom of
the water bottle providing inertial data of the movement performed by the subject. This data, which
includes acceleration, angular velocity and magnetic field (earth), will be used afterwards to validation
of the myoelectric data.

Software: A Python 2.7.1 application receives through Serial communication data containing the
measurements from the three channels. Moreover, a feature of this application allows the researcher
manually insert markers corresponding to the executed movement by the voluntary. Therefore during
the analysis process, the different tasks are easily identified on the myoelectric signal.
The incoming data is transferred from Python to Matlab R2015a using the communication protocol
Lab Streaming Layer. This synchronization of the streaming data enables real-time displaying, live
analysis and recording on Matlab. This feature of simultaneously tracking all channels during the trial
is important to ensure that all electrodes are correctly measuring.
(E)

(C)

(B) (D)
(A)

Figure 2 – Subject wearing the measurement system during trial.

2.2. Data Acquisition

The objective of the proposed trial is to collect data to identify common characteristics from the
myoelectric signal related to a daily task. This research project will be conducted in accordance with the
Guidelines and Regulatory Standards for Research Involving Humans (Resolution 196/1996 of the
National Health Council and Declaration of Helsinki), and was submitted to the Human Research Ethics
Committee from the School of Physical Education and Sport of the University of São Paulo (EEFE-
USP).
Each subject signed a consent term agreeing to participate of the research and acknowledging the
risks involved. While the physiotherapist prepared the subject to the trial, a series of questions were
asked regarding name, birth date, age, gender, weight, height, job, fitness, arm related diseases or
previously arm injuries. Moreover an Edinburgh Handedness Inventory was filled in order to assess
hand dominance in everyday activities.
The physiotherapist palpating the arm of the subject while he repeatedly contracts the muscle,
identified the Extensor Digitorum (Channel 1 - Blue), Brachioradialis (Channel 2 - Red) and Flexor
Digitorum (Channel 3 – Yellow) muscles in both arms. This muscle group where selected considering
a transradial amputation. Then the position of higher myoelectric activity for each muscle were shaved,
cleaned and marked.
The acquisition is made for both arms, but a computational algorithm randomly select the starting
side. Then the measurement system is attached to the subject as shown in Figure 2. After inviting the
subject to sit at the experimental table observing his posture, a bottle with water is positioned at a
distance 80% of his arm length far away from his the xiphoid process.
The trial for each arm is divided in three sections. Each section, the subject are asked to perform
three different movements 10 times following the same random order, previously presented before the
beginning of the exercise.
The three movements repeated in a random order each section are:
 M1 - Grab the bottle;
 M2 - Grab and rotate (pronation) the bottle;
 M3 - Pick the bottle up until the shoulder’s height and rotate (pronation) it.
Considering human reaction times and attention spans, the subjected was reminded of each
movement before executing it and should perform it just when signalized. The research after inserting
the corresponding movement marker on Python, would wait around 5 seconds before signalling to the
volunteer execute the task to guarantee identification of the beginning of contraction.
The data for each section were recorded separately
3. Results and Discussion
Data from intact subjects were collected as described in Section 2.2 and Figure 3 shows the sample of a
myoelectric signal measured during 10 repetitions of the randomly generated sequence M1, M2 and M3,
executed with the right hand. It was a male volunteer, left-handed, with 30 years old, height of 1.72m
and weighing 80Kg. Figure 4 shows the RMS value calculated for each channel.

Figure 3 – Example of data collected for motion sequence M1, M2 and M3

Figure 4 – RMS value

 The results was compared with a dataset from the Ninapro Project Database of a male intact
subject, right-handed, with 29 years old, height of 1.83m and weighing 75Kg, executing a movement
similar to M1 (grabbing a bottle) with the right hand. The myoelectric signals from the extensor and
flexor muscles were recorded with Delsys Trigno Wireless electrodes and sampled at a frequency rate
of 2 kHz. Also different from the presented protocol, he repeated the movements 6 consecutive times
in a non-randomized order. [8]
The signal envelope was defined for both datasets calculating the Moving Average, which is an
amplitude behaviour estimator indicating the muscle activation interval. The Figures 5 and 6 shows
the signals recorded with both measurement systems with the envelope line. Comparing them it was
observed a similar shapes for each muscle. The activation interval of the extensor muscle has a more
linear flat shape. While the activation interval of the flexor muscle has two distinguished peaks in the
beginning and at the end of the task. These patterns can be observed for both systems. However, it is
clearer for the measurements with the Delsys Trigno Wireless electrodes. This is explained by the
differences of protocol: while the work presented in [8] used a more controlled setting, the proposed
method aimed to replicate a natural movement reproducing a daily task.

Figure 5 – Myoelectric signal recorded of the subject grabbing a bottle with the developed electrodes
and the signal envelope (red line)
 Figure 6 – Myoelectric signal recorded of the subject grabbing a bottle with the Delsys Trigno
Wireless electrodes and the signal envelope (red line)

4. Conclusion
In this project, the proposal of a low cost electrographic signal acquisition system for use in upper limb
orthoses was presented. The system proved reliable and with good repeatability of signal reading. In the
future stages, the data collected during this research will be further analysed identifying more patterns
for each movement considering the same subject and also comparing the variation between different
subjects. Then a classifier associated with a control system will be implemented to power an upper-limb
prosthesis with two degrees of freedom: flexion/extension and pronation/supnation.

References
[1] National Academies of Sciences, Engineering, and Medicine; Health and Medicine Division;
Board on Health Care Services; Committee on the Use of Selected Assistive Products and
Technologies in Eliminating or Reducing the Effects of Impairments; Flaubert JL, Spicer CM,
Jette AM, editors. The Promise of Assistive Technology to Enhance Activity and Work
Participation. Washington (DC): National Academies Press (US); 2017 May 9. 4, Upper-
Extremity Prostheses. Available from: https://www.ncbi.nlm.nih.gov/books/NBK453290/
[2] Atzori, M; Gijsberts, A; Heynen, S; Hager, AG M.; Deriaz, O; Smagt, vdP; Castellini, C.; Caputo,
B, and Müller, H. Building the Ninapro database: A resource for the biorobotics community.
2012 4th IEEE RAS & EMBS International Conference on Biomedical Robotics and
Biomechatronics (BioRob). 2012, pp. 1258-1265. doi: 10.1109/BioRob.2012.6290287
Atzori, M; Gijsberts, A; Kuzborskij, I; Elsig, S; Hager, AG M.; Deriaz, O; Castellini, C; Müller,
H; and Caputo, B. Characterization of a Benchmark Database for Myoelectric Movement
Classification. IEEE Transactions on Neural Systems and Rehabilitation Engineering. 2015,
23(1): 73-83. doi: 10.1109/TNSRE.2014.2328495.
[3] Raichle KA, Hanley MA, Molton I, Kadel NJ, Campbell K, Phelps E, Ehde D, Smith DG.
Prosthesis use in persons with lower- and upper-limb amputation. J Rehabil Res Dev.
 2008;45(7):961-72. PubMed PMID: 19165686; PubMed Central PMCID: PMC2743731.
More references
[4] Zuo KJ, Olson JL. The evolution of functional hand replacement: From iron prostheses to hand
transplantation. Plast Surg (Oakv). 2014 Spring;22(1):44-51. PubMed PMID: 25152647;
PubMed Central PMCID: PMC4128433.
[5] Brunelli, D; Tadesse, A M; Vodermayer, B; Nowak, M; and Castellini, C. Low-cost wearable
multichannel surface EMG acquisition for prosthetic hand control. 2015 6th International
Workshop on Advances in Sensors and Interfaces (IWASI), Gallipoli. 2015, pp. 94-99.
[6] Kuzborskij, I; Gijsberts, A; and Caputo, B. On the challenge of classifying 52 hand movements
from surface electromyography. 2012 Annual International Conference of the IEEE
Engineering in Medicine and Biology Society, San Diego, CA, 2012, pp. 4931-4937. doi:
10.1109/EMBC.2012.6347099
[7] Kuiken T A, Miller L A, Turner K, Hargrove L J. A Comparison of Pattern Recognition Control
and Direct Control of a Multiple Degree-of-Freedom Transradial Prosthesis. IEEE J Transl
Eng Health Med. 2016 Nov 22;4:2100508. doi: 10.1109/JTEHM.2016.2616123. PubMed
PMID: 28560117; PubMed Central PMCID: PMC5396910.
[8] Atzori, M; Gijsberts, A; Castellini, C; Caputo, B; Hager, AG M; Elsig, S; Giatsidis, G; Bassetto,
F; and Müller, H. Electromyography data for non-invasive naturally-controlled robotic hand
prostheses. Scientific Data. 2014, 1 - Article number: 140053. DOI: 10.1038/sdata.2014.53