Identification of epileptic discharge based on statistical analysis and fractal analysis

Abstract

Epilepsy is a relatively common brain disorder characterized by transient but recurrent abnormal discharge of neurons due to the dysfunction of the central nervous system. Brainwave EEG signals are customary used in clinical diagnosis and screening of epileptic seizure patients. EEG abnormalities include abnormal background waves, very short epileptic discharges (lasting only about several tens of milliseconds), and seizure signals (lasting a few seconds). There are 7 classes of short epileptic discharges, identification of which is often considered an effective screening of epileptic seizure patients. In this study, we consider classification of these short epileptic discharges. For this purpose, we analyzed 422 multi-channel EEG segments, each 4 $s$ long. Among these segments, 322 are short epileptic discharges, 100 are from healthy controls. We have first extracted features from these EEG segments using statistical analysis and Adaptive Fractal Analysis (AFA), then used Random Forest Classifier to identify and classify all 7 epileptic discharges. We have achieved very high recognition and classification accuracy with this synthesized approach.