Factors Affecting the Performance of FastICA Algorithm for Decomposition of High-Density Surface Electromyogram

The motor unit (MU) physiology and the neural control of human movement can be studied via surface electromyogram (sEMG) decomposition techniques, which estimate the discharge times of MUs during a given contraction. The recent decomposition methods are based on the blind source separation (BSS) methods, and some approaches such as independent component analysis (ICA) can be applied to solve the problem. This study aims to investigate the influence of several parameters of the fast independent component analysis (FastICA) algorithm in the decomposition process of high-density sEMG. The extension factor, the number of iterations, and the method of initialization of the separation vector were explored for three evaluation metrics, namely, the execution time of decomposition, the absolute number of unique decomposed MUs, and the ratio of the number of unique MUs to the total number of extracted MUs. The number of iterations increased the execution time of decomposition and the absolute number of unique decomposed MUs. Besides, the method of initialization of the separation vector had little influence on the execution time of decomposition, an optimal value can be achieved for the absolute number of unique decomposed MUs. Moreover, the repeated convergence of the FastICA algorithm to the same source was mainly affected by the number of iterations. The study has reinforced the importance of evaluating the parameter combinations to achieve a better FastICA decomposition performance based on the chosen evaluation metric.