NEURAL NETWORK CLASSIFIER FOR THE DETECTION OF EPILEPSY/SEIZURE BASED ON BISPECTRUM FEATURES

Electroencephalogram (EEG) is widely used for the clinical investigation of epilepsy. Epilepsy is a brain disorder normally characterized by repeated seizures. This paper presents a method to detect epilepsy/seizure based on bispectrum features. For signals that are non-Gaussian and which are generated by nonlinear mechanisms, higher order spectra are useful in quantifying the nonlinearity. Bispectrum which is a second order spectrum is used to compute the features. The proposed method is tested on three datasets for detecting ictal / inter ictal EEG from normal EEGs and also classifying ictal and inter ictal EEGs and is giving a classification accuracy of 70% to 100% for various classes.