Neuromuscular Disease Detection by Neural Networks and Fuzzy Entropy on Time‐frequency Analysis of Electromyography Signals

Analysis of electromyography (EMG) signals is a necessary step in the diagnosis of neuromuscular diseases. Automatic classification systems can assist specialists and optimize the diagnostic process by applying time‐frequency analysis, fuzzy entropy, and neural networks to EMG signals in order to identify the presence of characteristics of a specific disorder, such as myopathy and amyotrophic lateral sclerosis. The performance of a decision support system depends on three important issues: the correct estimation of features from the EMG signal, the proper criteria for relevance analysis, and the learning process of the classification algorithm. In this paper, Discrete Wavelet Transform and Fuzzy Entropy are used to extract and select features from EMG signals, whereas Artificial Neural Networks are used to give the recognition result. The database used in this study is available for public use in EMGLAB, which is a website for sharing data, software, and information related to EMG decomposition. Results using the combination of these techniques show an accuracy around 98% for identifying EMG signals from three classes: healthy, patients with myopathy, or evidence of amyotrophic lateral sclerosis.