Walking gait event detection based on electromyography signals using artificial neural network

Nurhazimah Nazmi, Mohd Azizi Abdul Rahman, Shinichirou Yamamoto, Siti Anom Ahmad

Research output: Contribution to journalArticle

5 Citations (Scopus)

Abstract

In many gait applications, the focal events are the stance and swing phases. Although detecting gait events using electromyography signals will help the development of assistive devices such as exoskeleton, orthoses, and prostheses, stance and swing phases have yet to be observed using electromyography signals. The core of this study is to propose a classification system for both stance and swing phases based on electromyography signals. This is to be done by extracting the patterns of electromyography signals from time domain features and feeding them into an artificial neural network classifier. In addition, a different number of input features and two prominent training algorithm of artificial neural network have been employed in this study. Eight subjects that participated in this study were divided into two categories namely, learned (first seven subjects) and unlearned data (the remaining one subject). It was observed that Levenberg-Marquardt algorithm with five time domain features performed better than other features with an average percentage of classification accuracy of 87.4%. This system was further tested with electromyography signals of learned and unlearned data to identify the stance and swing phases in order to detect the timing of heel strike and toe off. The mean absolute different values between artificial neural network and footswitch data for learned data were 16 ± 18 ms and 21 ± 18 ms for heel strike and toe off, respectively. For this case, no significant differences (p < 0.05) were observed in mean absolute different for heel strike and toe off detections. Besides, the mean absolute different values of unlearned data were shown to be acceptable, 35 ± 25 ms for heel strike and 49 ± 15 ms for toe off. By the end of this experiment, basing the examination of gait events with electromyography signals using artificial neural network is possible.

Original languageEnglish
Pages (from-to)334-343
Number of pages10
JournalBiomedical Signal Processing and Control
Volume47
DOIs
Publication statusPublished - 2019 Jan 1

Fingerprint

Electromyography
Neural networks
Prosthetics
Classifiers

Keywords

  • Artificial neural network
  • EMG signals
  • Gait event
  • Time domain features

ASJC Scopus subject areas

  • Signal Processing
  • Health Informatics

Cite this

Walking gait event detection based on electromyography signals using artificial neural network. / Nazmi, Nurhazimah; Abdul Rahman, Mohd Azizi; Yamamoto, Shinichirou; Ahmad, Siti Anom.

In: Biomedical Signal Processing and Control, Vol. 47, 01.01.2019, p. 334-343.

Research output: Contribution to journalArticle

@article{3e35f9874c9e473294f03392bb63ba93,
title = "Walking gait event detection based on electromyography signals using artificial neural network",
abstract = "In many gait applications, the focal events are the stance and swing phases. Although detecting gait events using electromyography signals will help the development of assistive devices such as exoskeleton, orthoses, and prostheses, stance and swing phases have yet to be observed using electromyography signals. The core of this study is to propose a classification system for both stance and swing phases based on electromyography signals. This is to be done by extracting the patterns of electromyography signals from time domain features and feeding them into an artificial neural network classifier. In addition, a different number of input features and two prominent training algorithm of artificial neural network have been employed in this study. Eight subjects that participated in this study were divided into two categories namely, learned (first seven subjects) and unlearned data (the remaining one subject). It was observed that Levenberg-Marquardt algorithm with five time domain features performed better than other features with an average percentage of classification accuracy of 87.4{\%}. This system was further tested with electromyography signals of learned and unlearned data to identify the stance and swing phases in order to detect the timing of heel strike and toe off. The mean absolute different values between artificial neural network and footswitch data for learned data were 16 ± 18 ms and 21 ± 18 ms for heel strike and toe off, respectively. For this case, no significant differences (p < 0.05) were observed in mean absolute different for heel strike and toe off detections. Besides, the mean absolute different values of unlearned data were shown to be acceptable, 35 ± 25 ms for heel strike and 49 ± 15 ms for toe off. By the end of this experiment, basing the examination of gait events with electromyography signals using artificial neural network is possible.",
keywords = "Artificial neural network, EMG signals, Gait event, Time domain features",
author = "Nurhazimah Nazmi and {Abdul Rahman}, {Mohd Azizi} and Shinichirou Yamamoto and Ahmad, {Siti Anom}",
year = "2019",
month = "1",
day = "1",
doi = "10.1016/j.bspc.2018.08.030",
language = "English",
volume = "47",
pages = "334--343",
journal = "Biomedical Signal Processing and Control",
issn = "1746-8094",
publisher = "Elsevier BV",

}

TY - JOUR

T1 - Walking gait event detection based on electromyography signals using artificial neural network

AU - Nazmi, Nurhazimah

AU - Abdul Rahman, Mohd Azizi

AU - Yamamoto, Shinichirou

AU - Ahmad, Siti Anom

PY - 2019/1/1

Y1 - 2019/1/1

N2 - In many gait applications, the focal events are the stance and swing phases. Although detecting gait events using electromyography signals will help the development of assistive devices such as exoskeleton, orthoses, and prostheses, stance and swing phases have yet to be observed using electromyography signals. The core of this study is to propose a classification system for both stance and swing phases based on electromyography signals. This is to be done by extracting the patterns of electromyography signals from time domain features and feeding them into an artificial neural network classifier. In addition, a different number of input features and two prominent training algorithm of artificial neural network have been employed in this study. Eight subjects that participated in this study were divided into two categories namely, learned (first seven subjects) and unlearned data (the remaining one subject). It was observed that Levenberg-Marquardt algorithm with five time domain features performed better than other features with an average percentage of classification accuracy of 87.4%. This system was further tested with electromyography signals of learned and unlearned data to identify the stance and swing phases in order to detect the timing of heel strike and toe off. The mean absolute different values between artificial neural network and footswitch data for learned data were 16 ± 18 ms and 21 ± 18 ms for heel strike and toe off, respectively. For this case, no significant differences (p < 0.05) were observed in mean absolute different for heel strike and toe off detections. Besides, the mean absolute different values of unlearned data were shown to be acceptable, 35 ± 25 ms for heel strike and 49 ± 15 ms for toe off. By the end of this experiment, basing the examination of gait events with electromyography signals using artificial neural network is possible.

AB - In many gait applications, the focal events are the stance and swing phases. Although detecting gait events using electromyography signals will help the development of assistive devices such as exoskeleton, orthoses, and prostheses, stance and swing phases have yet to be observed using electromyography signals. The core of this study is to propose a classification system for both stance and swing phases based on electromyography signals. This is to be done by extracting the patterns of electromyography signals from time domain features and feeding them into an artificial neural network classifier. In addition, a different number of input features and two prominent training algorithm of artificial neural network have been employed in this study. Eight subjects that participated in this study were divided into two categories namely, learned (first seven subjects) and unlearned data (the remaining one subject). It was observed that Levenberg-Marquardt algorithm with five time domain features performed better than other features with an average percentage of classification accuracy of 87.4%. This system was further tested with electromyography signals of learned and unlearned data to identify the stance and swing phases in order to detect the timing of heel strike and toe off. The mean absolute different values between artificial neural network and footswitch data for learned data were 16 ± 18 ms and 21 ± 18 ms for heel strike and toe off, respectively. For this case, no significant differences (p < 0.05) were observed in mean absolute different for heel strike and toe off detections. Besides, the mean absolute different values of unlearned data were shown to be acceptable, 35 ± 25 ms for heel strike and 49 ± 15 ms for toe off. By the end of this experiment, basing the examination of gait events with electromyography signals using artificial neural network is possible.

KW - Artificial neural network

KW - EMG signals

KW - Gait event

KW - Time domain features

UR - http://www.scopus.com/inward/record.url?scp=85053192544&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85053192544&partnerID=8YFLogxK

U2 - 10.1016/j.bspc.2018.08.030

DO - 10.1016/j.bspc.2018.08.030

M3 - Article

AN - SCOPUS:85053192544

VL - 47

SP - 334

EP - 343

JO - Biomedical Signal Processing and Control

JF - Biomedical Signal Processing and Control

SN - 1746-8094

ER -