Patient-Specific Seizure Prediction Using the Convolutional Neural Networks

Abstract

An epileptic seizure is a disease of the central nervous system caused by abnormal activity generated by neurons in the brain. Seizure reduces the quality of life of epileptic patients due to unconsciousness. In this paper, an efficient seizure prediction system is proposed to improve the quality of life. The raw EEG signal is converted into the EEG signal image. Then, a convolutional neural network is used for training the prediction system. The performance of the proposed system is evaluated using the CHB-MIT dataset. The classification accuracy of interictal and preictal states is achieved up to 94.33% using 10-fold cross-validation. Due to the presence of noise in the EEG signal, a pool based technique is used to make the decision on the majority of a 1 min EEG signal that increase the accuracy of the prediction of upcoming seizures.