Classification of Neuromuscular Disorders Using Wavelets and Entropy Metrics on Surface Electromyograms

The aim of this study was an investigation whether the examined subjects can be classified as normal, myopathic or neuropathic, based on 4-channel surface electromyographic (SEMG) recordings of the biceps brachii muscle recorded at isometric voluntary contractions. Five different force levels were used: 10, 30, 50, 70 and 100 % of maximum voluntary contraction (MVC), with each recording lasting 5 seconds. The amplitude of SEMG recordings was processed using dyadic Haar wavelet and entropy metrics. In each scale, the entropy of transformed SEMG was calculated and a set of 80 features per subject was formed. From existing data, we created five dataset, where subject were labelled based on the type as: normal/abnormal, normal/myopathic, normal/neuropathic, myopathic/neuropathic and normal/myopathic/neuropathic. Five classification techniques from the WEKA machine learning package were used, i.e. decision trees j48, random trees, random decision forests, support vector machines and ensemble of support vector machine with polynomial kernel. Resulting classifiers were obtained by using 3-fold cross-validation with 50 iterations for every machine learning technique. Classification accuracy and deviation of 150 classifiers for each dataset and machine learning technique were computed. The results show that normal and neuropathic groups are the most distinguishable (79±8%), while myopathic and neuropathic groups are most hardly distinguished (50±15%). Other classifications scored normal/abnormal (64±11%), normal/myopathic (73± 7%), normal/neuropathic/myopathic (63±8 %). The results are quite promising, taking into account that only the SEMG amplitude was considered. The introduced method can be further upgraded with a decomposition technique which would extract individual motor unit action potentials (MUAPs) from SEMG signal, so that also their properties would be used in classification criteria. Key-Words: surface electromyography, SEMG, neuromuscular disorders, neuropathy, myopathy, isometric voluntary contraction, entropy, wavelet transform