Improving the recognition of grips and movements of the hand using myoelectric signals. Issue 2 (July 2016)

• Record Type: Journal Article
• Authors: Shuman, Gene; Durić, Zoran; Barbará, Daniel; Lin, Jessica; Gerber, Lynn
• Is Part Of: BMC medical informatics and decision making. Volume 16:Issue 2(2016)
• Page Start: 65
• Page End: 83
• Publication Date: 2016-07
• Subjects: Electromyograms -- Machine learning -- ADLs -- Prehensile patterns -- Classification -- SAX -- Dynamic time warping; Medical informatics -- Periodicals; Clinical medicine -- Decision making -- Periodicals; 610.285
• DOI: 10.1186/s12911-016-0308-1
• Languages: English
• ISSNs: 1472-6947

Abstract

Abstract Background People want to live independently, but too often disabilities or advanced age robs them of the ability to do the necessary activities of daily living (ADLs). Finding relationships between electromyograms measured in the arm and movements of the hand and wrist needed to perform ADLs can help address performance deficits and be exploited in designing myoelectrical control systems for prosthetics and computer interfaces. Methods This paper reports on several machine learning techniques employed to discover the electromyogram patterns present when performing 24 typical fine motor functional activities of the hand and the rest position used to accomplish ADLs. Accelerometer data is collected from the hand as an aid in identifying the start and end of movements and to help in labeling the signal data. Techniques employed include classification of 100 ms individual signal instances, using a symbolic representation to approximate signal streams, and the use of nearest neighbor in two specific situations: creation of an affinity matrix to model learning instances and classify based on multiple adjacent signal values, and using Dynamic Time Warping (DTW) as a distance measure to classify entire activity segments. Results Results show the patterns can be learned to an accuracy of 76.64 % for a 25 class problem when classifying 100 ms instances, 83.63 % with the affinity matrix approach with symbolic representation, and 85.22 % with Dynamic Time Warping. Classification errors are, with a few exceptions, concentrated within particular grip action groups. Conclusion The findings reported here support the view that grips and movements of the hand can be distinguished by combining electrical and mechanical properties of the task to an accuracy of 85.22 % for a 25 class problem. Converting the signals to a symbolic representation and classifying based on larger portions of the signal stream improve classification accuracy. This is both clinically useful and opens the way for an approach to help simulate hand functional activities. With improvements it may also prove useful in real time control applications.