Classification and the Feature Domain Extraction of EEG of Epilepsy using 1D-CNN and Grad-CAM

This is the authors' abstract. We don't add key points for this paper.

This study aimed to develop a model to classify the raw EEG data of epilepsy patients and healthy subjects using 1D-CNN and to extract the features in the EEG for determination using Grad-CAM. The dataset was constructed from EEG recordings of subjects with epilepsy and healthy subjects. We divided the EEG data into training and test datasets. The training dataset was divided into four folds for cross-validation. We searched for the optimal hyperparameters to develop the model. Then, the model was learned and evaluated using test dataset. To extract features for classification, we applied Grad-CAM to the trained model. The results showed that the model distinguished between epilepsy and healthy in EEG with high accuracy. The mean Grad-CAM value in epilepsy patients was higher than that in the healthy subjects. The Grad-CAM indicated that the 1D-CNN model captured the clinical features of epileptiform discharges in the EEG for classification.