Classification of neuromuscular disorders using features extracted in the wavelet domain of sEMG signals: a case study

The present study introduces a method for detecting possible neuropathy or myopathy cases of a subject based on surface electromyograms signals; the same method has been developed as a classification tool for movements of the upper arm. This research is proposed for its capability to classify subjects from a clinical dataset in healthy, myopathic and neuropathic cases. The extraction of features with simple morphology but estimated on the signals wavelet domain increases the classification rate of the system drastically. Therefore, a set of features based mainly on energies of the EMG signals along with the Hudgins’ measurements, all estimated on the wavelet domain create a feature space consisted of highly discriminant subspaces for the three classes healthy, neuropathies or patients with myopathies. For the classification task the k-NN algorithm used and the validation performed with k-folds method; the predictions for the performance on unknown data was close to the actual validation results. Overall accuracy of the system for all three classes is 98.36 ± 0.79%, and it is safe to state that based on the different tests performed, it is a robust approach for the classification of subjects.