Epileptic seizure stage classification from EEG signal using ResNet18 Model and Data Augmentation

Abstract

Epilepsy is a widespread neurological disorder nowadays. It can occur to people of any age and any physiological condition. For the proper treatment of epilepsy, the accurate detection of epileptic seizures is crucial. There are many procedures available, and research is still going on. This study focuses on detecting and classifying epileptic seizures from EEG signals using the Convolutional Neural Network. For this purpose, data augmentation was implemented on EEG signals first. We divided the raw signal into four parts of 5 seconds and then rearranged it with four different combinations. We have applied Continuous Wavelength Transform in these newly formed signals to construct the scalogram images. These images were later classified using ResNet-18. The stages classified were Ictal vs. Interictal, Normal vs. Ictal, Normal vs. Interictal. We found the accuracy of 98.4%, 99.1%, and 98.2%, respectively. The accuracy and acceptance of the method can be developed further by applying different epileptic datasets and tuning other neural network parameters.