Signal Decompostion and Diagnostic Classification of the Electromyogram Using a Novel Neural Network Technique.

Abstract Interpretation of physiologic signals to assist medical diagnosis requires human expertise. Success in automating this process has been limited. We present a three-step method for automated interpretation of the EMG. Signal decomposition and classification steps, which have not been automated using traditional computer methods, utilize neural networks. To deal with poorly described signals, a novel decomposition method, pseudoun-supervised learning, has been developed. The resulting method is considerably more robust than prior methods.