EEG Signals Feature Extraction and Artificial Neural Networks Classification for The Diagnosis of Schizophrenia

Abstract

This paper presents the design of artificial neural networks (ANN) for the classification of Electroencephalograph (EEG) signals collected from 49 Schizophrenia patients and 32 healthy controls. The EEG signals are captured based on event-related potential (ERP) corresponding to button pushing and audio tone playback. Five temporal features extracted from the EEG signals, and two demographic features are used for ANN training and testing. The best classification accuracy of above 98.5% is achieved. Additional time-frequency features are extracted after applying wavelet transform to the ERP EEG signals for ANN classification. The research outcomes show that there is great potential in developing effective and subjective diagnosis tool for Schizophrenia based on EEG signals. Two software environments RStudio and MATLAB are used for the design of ANN classifiers. The latter offers more flexibility and design options such as training functions. The training performances are comparably measured.