Adaptive Epileptic Seizure Prediction Based on EEG Synchronization

Abstract

Epilepsy is type of neurological disorder characterized by recurrent seizures that may cause injury to self and others. The ability to predict seizure before its occurrence, so that counter measures are considered, would improve the quality of life of epileptic patients. This research work proposes an adaptive seizure prediction approach based on electroencephalography (EEG) signals analysis. We use cross-correlation to estimate synchronization between EEG channels. Abnormal synchronization between brain regions may reveal brain condition and functionality. Two EEG synchronization baselines, normal and pre-seizure, are used to continuously monitor sliding windows of EEG recording to predict the upcoming seizure. The two baselines are continuously updated using distance-based method based on the most recent prediction outcome. Up to 570 hours continuous EEG recording taken from CHB-MIT dataset is used for validating the proposed method. An overall of 84% sensitivity (46 out of 55 seizures are correctly predicted) and 63% specificity are achieved with one hour prediction horizon. The proposed method is suitable to be implemented in mobile or embedded device which has limited processing resources due to its simplicity.