Fuzzy clustering and feature selection analysis toward improved identification of MUAP in needle EMG signal

Decomposing the EMG signal and determining fiber´s motor units is a challenging work in this area. The high amounts of resident noise and artifacts distort the signal. But, the decomposition gets more complicated, when we have motor unit action potential MUAP, interference during one channel recording. Considering this distortion, recognition and separation of motor units from each other, and determining the degree of their membership to each fiber gives valuable information. In this paper. First, motor units are determined and separated according to a set of filters similar to Gabor filters, then by extracting some different time-frequency and morphology features, the feature space will be determined. In the next step, the number of clusters which are the number of fibers will be determined. The clustering method used for this purpose is FCM clustering method. One of the innovations of the proposed method in this study is using an algorithm which improves the accuracy of the decomposition. This algorithm employs the membership information of each motor unit in fuzzy clustering along with the feature selection using mutual information of each motor unit. The results indicate 7.3% improvement while decreasing computational costs.