Epileptic Seizure Motion Classification based on sEMG and Artificial Neural Network

According to the World Health Organization (WHO), 70% of all epileptic seizures are convulsive, making the EMG signal one of the essential signals used for epilepsy treatment. Nevertheless, epileptic seizure classification remains a challenge. Suitable machine learning classifiers and features are the critical metrics for good performance. In this paper, a collection of widely used time-domain features of surface electromyography (sEMG) signals is illustrated and an investigation of their ability to distinguish epileptic seizures is reported for five subjects performing myoclonic and tonic seizures using eight sEMG sensors placed on different muscles: Biceps brachii, flexor carpi ulnaris, gastrocnemius, and quadriceps. Feature selection step is required to eliminate information redundancy and irrelevant before epileptic seizure classification. The results show that extracting Waveform Length, Integrated EMG, Variance, Shannon Entropy, Complexity, Skewness, and Root Mean Square values are more significant for myoclonic-tonic seizure classification based on sEMG signal analysis. Selected features have been subjected to the Artificial Neural Network classifier. An overall epileptic seizure accuracy of 96% could be achieved by the ANN algorithm based on seven selected features.