Simultaneous and proportional control of wrist and hand degrees of freedom with kinematic prediction models from high-density EMG.

To improve intuitive control and reduce training time for active upper limb prostheses, we developed a myocontrol system for 3 degrees of freedom (DoFs) of the hand and wrist. In an offline study, we systematically investigated movement sets used to train this system, to identify the optimal compromise between training time and performance. High-density surface electromyography (HDsEMG) and optical marker motion capture were recorded concurrently from the lower arms of 8 subjects performing a series of wrist and hand movements activating DoFs individually, sequentially, and simultaneously. The root mean square (RMS) feature extracted from the EMG signal and kinematics obtained from motion capture were used to train regression and classification models to predict the kinematics of wrist movements and opening and closing of the hand, respectively. Results showed successful predictions of kinematics when training with the complete training set (r2 = 0.78 for wrist regression and recall = 0.85 for hand closing/opening classification). In further analysis, the training set was substantially reduced by removing the simultaneous movements. This led to a statistically significant, but relatively small reduction of the effectiveness of the wrist controller (r2 = 0.70, p<0.05), without changes for the hand controller (closing recall = 0.83). Reducing the training time and complexity needed to control a prosthesis with simultaneous wrist control as well as detection of intention to close the hand can lead to improved uptake of upper limb prosthetics.