Time-Frequency Classification of the Transient Mechanomyogram

Abstract

Muscle vibrations, known as the mechanomyogram (MMG), have been used in a limited manner to control upper-limb prostheses. MMG is dependent on the recruitment and firing rates of muscle motor units during contraction. It was thus hypothesized that MMG signals would reflect the distinctive patterns of muscle activation during the grasp of differently shaped objects. This study investigates this hypothesis, specifically focusing on forearm muscles. MMG patterns originating from grasps were classified. The classification performance was evaluated for a number of combinations of feature-sets and sensor sites. With seven able-bodied participants, MMG signal transients from three sites could be differentiated among three classes of forearm muscle activity with 78.6 ± 13% accuracy using wavelet packet-derived features and a linear classifier. These results suggest that, with additional research, MMG may indeed become a usable control signal for multifunction prostheses.