TY - JOUR
T1 - Dual window pattern recognition classifier for improved partial-hand prosthesis control
AU - Earley, Eric J.
AU - Hargrove, Levi J.
AU - Kuiken, Todd A.
N1 - Funding Information:
The contents of this paper were developed under a grant from the National Institute on Disability, Independent Living, and Rehabilitation Research (NIDILRR grant number 90RE5014-02-00). NIDILRR is a Center within the Administration for Community Living (ACL), Department of Health and Human Services (HHS). The contents of this paper do not necessarily represent the policy of NIDILRR, ACL, HHS, and you should not assume endorsement by the Federal Government. We would also like to thank Ann Barlow, Ph. D., and Sheila Burt for editing the manuscript.
Publisher Copyright:
© 2016 Earley, Hargrove and Kuiken.
PY - 2016/2/23
Y1 - 2016/2/23
N2 - Although partial-hand amputees largely retain the ability to use their wrist, it is difficult to preserve wrist motion while using a myoelectric partial-hand prosthesis without severely impacting control performance. Electromyogram (EMG) pattern recognition is a well-studied control method; however, EMG from wrist motion can obscure myoelectric finger control signals. Thus, to accommodate wrist motion and to provide high classification accuracy and minimize system latency, we developed a training protocol and a classifier that switches between long and short EMG analysis window lengths. Seventeen non-amputee and two partial-hand amputee subjects participated in a study to determine the effects of including EMG from different arm and hand locations during static and/or dynamic wrist motion in the classifier training data. We evaluated several real-time classification techniques to determine which control scheme yielded the highest performance in virtual real-time tasks using a three-way ANOVA. We found significant interaction between analysis window length and the number of grasps available. Including static and dynamic wrist motion and intrinsic hand muscle EMG with extrinsic muscle EMG significantly reduced pattern recognition classification error by 35%. Classification delay or majority voting techniques significantly improved real-time task completion rates (17%), selection (23%), and completion (11%) times, and selection attempts (15%) for non-amputee subjects, and the dual window classifier significantly reduced the time (8%) and average number of attempts required to complete grasp selections (14%) made in various wrist positions. Amputee subjects demonstrated improved task timeout rates, and made fewer grasp selection attempts, with classification delay or majority voting techniques. Thus, the proposed techniques show promise for improving control of partial-hand prostheses and more effectively restoring function to individuals using these devices.
AB - Although partial-hand amputees largely retain the ability to use their wrist, it is difficult to preserve wrist motion while using a myoelectric partial-hand prosthesis without severely impacting control performance. Electromyogram (EMG) pattern recognition is a well-studied control method; however, EMG from wrist motion can obscure myoelectric finger control signals. Thus, to accommodate wrist motion and to provide high classification accuracy and minimize system latency, we developed a training protocol and a classifier that switches between long and short EMG analysis window lengths. Seventeen non-amputee and two partial-hand amputee subjects participated in a study to determine the effects of including EMG from different arm and hand locations during static and/or dynamic wrist motion in the classifier training data. We evaluated several real-time classification techniques to determine which control scheme yielded the highest performance in virtual real-time tasks using a three-way ANOVA. We found significant interaction between analysis window length and the number of grasps available. Including static and dynamic wrist motion and intrinsic hand muscle EMG with extrinsic muscle EMG significantly reduced pattern recognition classification error by 35%. Classification delay or majority voting techniques significantly improved real-time task completion rates (17%), selection (23%), and completion (11%) times, and selection attempts (15%) for non-amputee subjects, and the dual window classifier significantly reduced the time (8%) and average number of attempts required to complete grasp selections (14%) made in various wrist positions. Amputee subjects demonstrated improved task timeout rates, and made fewer grasp selection attempts, with classification delay or majority voting techniques. Thus, the proposed techniques show promise for improving control of partial-hand prostheses and more effectively restoring function to individuals using these devices.
KW - Electromyography (EMG)
KW - Intrinsic hand muscles
KW - Myoelectric control
KW - Partial-hand prosthesis
KW - Pattern recognition
UR - http://www.scopus.com/inward/record.url?scp=84964329867&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=84964329867&partnerID=8YFLogxK
U2 - 10.3389/fnins.2016.00058
DO - 10.3389/fnins.2016.00058
M3 - Article
C2 - 26941599
AN - SCOPUS:84964329867
SN - 1662-4548
VL - 10
JO - Frontiers in Neuroscience
JF - Frontiers in Neuroscience
IS - FEB
M1 - 58
ER -