Independent component analysis based algorithms for high-density electromyogram decomposition: Experimental evaluation of upper extremity muscles

Motor unit firing activities can provide critical information regarding neural control of skeletal muscles. Extracting motor unit activities reliably from surface electromyogram (EMG) is still a challenge in signal processing. We quantified the performance of three different independent component analysis (ICA)-based decomposition algorithms (Infomax, FastICA and RobustlCA) on high-density EMG signals, obtained from arm muscles (biceps brachii and extensor digitorum communis) at different contraction levels. The source separation outcomes were evaluated based on the degree of agreement in the discharge timings between different algorithms, and based on the number of common motor units identified concurrently by two algorithms. Two metrics, the separation index (silhouette distance or SIL) and the rate of agreement, were used to evaluate the decomposition accuracy. Our results revealed a high rate of agreement (80%-90%) between different algorithms, which was consistent across different contraction levels. The RobustlCA tended to show a higher RoA with the other two algorithms (especially with Infomax), whereas FastICA and Infomax tended to yield a greater number of common MUs. Overall, through an experimental evaluation of the three algorithms, the outcomes provide information regarding the utility of these algorithms and the motor unit filter criteria involving EMG signals of upper extremity muscles.