EEG based automated identification of schizophrenia from FBSE-EWT technique

Abstract

Monitoring of the brain’s activity usually done by analyzing electroencephalogram (EEG) signals, EEG signals are helpful to predict abnormal behavior of the brain. The purpose of this thesis is to develop an efficient method for diagnosis of schizophrenia. Already there are many existing techniques for the classification of EEG signals. In previous studies, EEG signal has been analysed for many applications and for diagnosis of many diseases using empirical wavelet transform (EWT) and Fourier-Bessel series expansion (FBSE) techniques. The proposed method uses FBSE-EWT as a combined tool to analyse and detect the important characteristic of the EEG signals. In our method, we have used this FBSE-EWT technique to decompose EEG signal into sub-bands. Here FBSE uses Bessel functions as bases, Bessel function are non-stationary signals which are suitable for analyzing non-stationary signals like EEG. Now after getting sub-bands we have applied Hilbert spectral analysis method for getting time-frequency representation (TFR) of these sub-bands. Here, these TFRs have been fed as input to the convolution neural network (CNN) classifier and this classifier has been obtained a maximum accuracy of 100% in classifying normal and schizophrenia disease classes of EEG signal.