Prediction of epileptic attacks from electroencephalogram signal using deep learning

Predicting epileptic attacks before they occur can be effective in preventing them through therapeutic intervention. Electroencephalogram (EEG) signals are used to diagnose epileptic seizures. However, the screening system cannot accurately detect epileptic seizure states. In the proposed method, in order to predict epileptic attacks before seizures based on electroencephalogram signals, a series of preprocessings, artificial neural networks and EEG deep learning have been used to prevent the occurrence of epileptic attacks. First, the CHB-MIT Scalp EEG dataset is used for experiments, and the EEG signals are divided into ۴ periods of ۱۰ minutes after preprocessing. Then, for pre-processing, a notch filter is used to reduce noise, the inherent components of EMD and wavelet transform are used to analyze EEG signals into different sub-bands and extract features. Then these features are given to the support vector machine (SVM) in order to classify the signal of the healthy and affected person, and the signal of the affected person is considered in two states, ictal and preictal, and finally, we use the convolutional neural network (CNN) with the AlexNet architecture in MATLAB program by comparing the features to predict epileptic attacks. The architectural sensitivity in predicting epileptic attacks is ۹۹%, the rate of wrong prediction of epileptic attacks is on average ۰.۰۹ per hour, and the duration of seizure prediction until seizure occurrence is ۳۰-۴۰ minutes. According to the results of the proposed method, the prediction of epileptic attacks has shown a better performance compared to the compared related works, because it is possible to predict and prevent epileptic attacks in a longer time before the occurrence with higher accuracy and new architecture.