Epileptic Seizure Detection & Prediction using Deep Learning based EEG Signal Analytic Mobile Smart Device Technology

A convolutional neural network (CNN) based extended reality (XR) deep learning method is proposed to detect and predict seizures from electroencephalography (EEG) signals. The frequency domain feature is extracted from the EEG signals averaging 18 channels from FP1-F7 to CZ-PZ. The proposed model can detects seizures with a training accuracy of 94.111%. An additional post-processing filtering algorithm is proposed to reduce the false seizure alarm rate. The proposed algorithm is effective in reducing the false seizure alarms achieving a higher seizure prediction accuracy. The accuracy performance increases from 79.911% to 82.133%, precision from 11.277% to 12.159%, and F1-score from 0.201 to 0.214, even when the recall performance is degraded from 91.878% to 88.325%.