Decomposition of HDsEMG Signals Recorded from a Forearm Extensor Muscle Based on Blind Source Separation

In this paper a decomposition of isometric high-density surface electromyography (HDsEMG) signals was performed using the blind source separaion (BSS) algorithm. The algorithm is a combination of the modified convolution kernel compensation algorithm and the K-means clustering (kmCKC) algorithm. The obtained decomposition results represent discharge times of the reconstructed motor units. Signals from a publicly available database of high-density surface electromyograms comprising 65 isometric hand gestures were used as input. The movements of the fingers representing the isometric activity of the forearm extensor muscles were selected for decomposition. Five time intervals of muscle activity for a single movement were taken. Decomposition was started separately for each interval. On average, 12.40 ± 1.14 motor units were successfully reconstructed, with an average pulse-to-noise ratio (PNR) of 18.9272 ± 2.0383 dB and coefficient of variation of interspike interval (CoVISI) of 0.4298 ± 0.0383. The average firing rate was 12.3723 ± 3.7259 pps. The expected number of reconstructed motor units for the extensor muscle and the theoretical average firing rate are in accordance with the obtained results. In addition to using the metrics themselves, visual inspection of the motor units also provides additional validation. The mentioned motor units appear visually credible, and the PNR and CoVISI metrics confirm this. The overall conclusion is that the decomposition was successfully performed and that the algorithm proved to be reliable and robust enough.