Diagnosis of Epileptic Seizures using Deep Learning and Optimization Technique

The research on Electroencephalogram (EEG) signals are gaining more attention for analysis of many disorders as the signal include crucial information on the electrical processes of the brain. One of the commonly employed disorders is analysis of epilepsy. Major identification of epileptic seizures involves three states one is interictal state, second one is preictal state and other is ictal state. Many studies have attempted to anticipate this preattack condition of a seizure, but an accurate prediction with high specificity and sensitivity still a difficulty. The current work suggests a system for seizure prediction based on deep learning and optimization techniques. The proposed methodology involves four stages. Initial preprocessing involves the scalp EEG signals, extraction of features using convolution neural network (CNN), optimization of features using modified particle swarm optimization (MPSO) algorithm and finally classified with the support vector machine (SVM) by which a new algorithm named MPSO-SVM is developed. The results obtained using optimization technique i.e. CNN-MPSO-SVM achieves promising results with accuracy of 95%, whereas CNN-SVM achieves accuracy of 91.9%. The F1 score achieved using the proposed model is 94.75%.