Miniature EMG Sensors for Prosthetic Applications

Abstract

Poly-articulated, myoelectric hand prostheses reproduce complex multi-degree of freedom movements. Typically, a pattern recognition algorithm translates the recorded electromyographic (EMG) activity into joint movements. Control algorithms may benefit by adding more EMG sensors, however, their mechanical integration within the socket strongly affects the physical robustness of the prosthetic system. Their typical size and rectangular shape are indeed nonoptimal for the limited amount of space within the socket. To solve this issue, here we present and test custom-made sensors for decoding multi-joint hand movements from EMG recordings of arm muscles. The sensors have circular shape and smaller size with respect to their standard counterpart, thus allowing a higher number of channels for multi-DOF control strategies. In order to evaluate their performance for multi-joint decoding, we tested a Non-Linear Logistic Regression classifier on both healthy and amputated subjects. We optimized the classifier in terms of F1Score, depending on the number of EMG sensors, and in terms of Embedding Optimization Factor, depending on the polynomial complexity degree. We then compared performance with that of standard rectangular EMG sensors and found no significant difference. Our custom-made sensors achieved higher F1Score for all the patients. This result, coupled with more effective integration with the socket, suggests effective prosthetic applications for our sensors.