Analysis of Electromyogram (EMG) for Detection of Neuromuscular Disorders

In this paper, we conducted time domain analysis on EMG signals for the development of neuromuscular disorder classification algorithm based on Support Vector Machine (SVM). The type of EMG data considered are from a healthy individual, from one individual with neuropathy and one individual with myopathy condition. The extracted feature from the EMG data comprises of statistical features namely mean, mean absolute deviation, standard deviation, interquartile range, energy, coefficient of variation and Auto-Regressive (AR) model. The statistical features are then ranked using ReliefF algorithm in order to identify the most prominent features and to be used for selection of the best features. In classification method, 2-stage cascaded SVM is used in which the 1st-stage SVM involved classification of healthy versus unhealthy (neuropathy + myopathy), whereas the 2nd-stage SVM involve classification of data coming from the 1st-stage SVM. This means, the 2nd-stage SVM will only involve classification of neuropathy versus myopathy EMG data. The proposed EMG classification method gave good performance with average accuracy of 95.7%, average sensitivity of 100% and average specificity of 91% generated over the two SVM stages.