Feature Analysis for Discrimination of Motor Unit Action Potentials

摘要: In electrophysiological signal processing for intramuscular electromyography data (nEMG), single motor unit activity is of great interest. The changes action potential (MUAP) morphology, (MU) activation, and recruitment provide the most informative part to study nature causality in neuromuscular disorders. practice, a nEMG recording, more than one activities (in surrounding area needle electrode) are usually collected. Such fact makes MUAP discrimination that separates crucial task. Most neurology laboratories worldwide still recruit specialists who spend hours manually or semi-automatically sort MUAPs. From machine learning perspective, this task analogous clustering-based classification problem which number classes other class information unfortunately missing. paper, we present feature analysis strategy help better utilize unsupervised (i.e., totally automated) methods discrimination. To end, extract large pool features from each MUAP. Then select top ranked candidates using clusterability scores as selection criteria. We found spectrograms wavelet decomposition top-ranking feature, highly correlated reference was separable existing features. Using correlation-based clustering technique, demonstrate sorting performance with set. Compared produced by human experts, our method obtained comparable result (e.g., equivalent found, identical morphology pair corresponding MU class, similar histograms MUs). Taking manual labels references, got much higher sensitivity accuracy compared method. reference.