Development of an Intelligent Seizure Prediction System

Epilepsy is a neurological disorder characterized by the occurrence of unexpected epileptic seizures. This unpredictable character of epilepsy tends to hinder the daily activity of the patients or even worse, leads to injuries in everyday duties. This paper aims to discover patterns in long term electroencephalogram (EEG) signals in order to predict seizures up to 30 minutes before the occurrence. We proposed an adaptive system for seizure prediction by extracting features based on spectral energy from surface EEG using a Multilayer Perceptron neural network classifier. Another objective of this work is to analyze the distribution of the significant spectral information relative to our task, the prediction of epilepsy seizures. Therefore, the results obtained using different features extraction methods based on linear and nonlinear bands were compared. To achieve the intended purpose, the system was trained and tested on 19 patients from chb-MIT database. The best results indicate an average of 95.51% accuracy, with a sensitivity of 95.76% and a specificity of 95.25%.