Prediction of Epilepsy Seizures and Design Warning System by Using Single Channel EMG Signals

Abstract

Epilepsy is a neurological disease that affects many people worldwide and makes daily life difficult for individuals. The disease's basic mechanism is sudden and uncontrolled discharges between brain cells known as neurons, which result in convulsions in the body and even loss of consciousness may occur. However, the main factor that poses the risk of death for epilepsy patients is falls and blows to the head. This study proposes a machine-learning-based prototype warning system to minimize the risk of sudden falls and death by predicting tonic-clonic seizures, and then warning the patient. EMG signals of epilepsy patients and healthy control subjects were recorded from the right and left deltoids and biceps. Time domain signal processing analysis methods were done to extract features. To select single channels t-test was performed on each channel of EMG and a right deltoid channel was selected due to the highest number of features that show a significant difference (p<0.05) between the pre-seizure and healthy segments. A 5-fold cross-validation method was used to split data to train and test subsets. Between SVM and KNN classifiers, 7-NN achieved the highest average accuracy with a 98.63 rate.