Muscle activity mapping by single peak localization from HDsEMG

Abstract

Human-machine interfaces using electromyography (EMG) offer promising applications in control of prosthetic limbs, rehabilitation assessment, and assistive technologies. These applications rely on advanced algorithms that decode the activation patterns of muscles contractions. This paper presents a new approach to assess and decode muscle activity by localizing the origin of individual temporal peaks in high-density surface EMG recordings from the dorsal forearm during low force finger extensions. Localization was performed using a surface Gaussian fit applied in the spatial domain to the varying amplitudes across the channels of the electrode grids. Localized EMG peaks were used to estimate different muscle volumes for each finger, showing high consistency across 10 subjects. The results suggest that muscle regions generating each action are highly distinct and indicate potential structural differences of muscle fibres between digits. The estimated volumes were further used to classify individual EMG peaks into each corresponding action. The percentage of correctly classified peaks for each action across 10 participants were 79 ± 18, 84 ± 9, 76 ± 13, and 79 ± 9 percent for index, middle, ring, and little finger extension, respectively. The presented volume analysis provides a new approach to assessing the spatial activation patterns in compact muscle anatomies; and the single peak classification approach opens up possibilities for near-instantaneous identification of muscle activations.