Classification of Healthy People and Schizophrenics Using Time- Frequency Domain Features Extracted from Electroencephalogram Signals

Abstract

Schizophrenia (SZ) is a chronic and complex mental disorder associated with neurobiological deficits. The complexity and heterogeneity of schizophrenia symptoms pose challenges for objective diagnosis, which is currently based on behavioral and clinical manifestations. Furthermore, other psychiatric disorders such as bipolar disorder or major depressive disorder are often misdiagnosed as schizophrenia. Consequently, manual screening through psychiatrist-patient interviews is not entirely reliable. This study aims to develop an automated SZ diagnosis scheme using electroencephalogram (EEG) signals as a complementary tool to assist psychiatrists. A novel method is proposed, utilizing features from time, frequency, and time-frequency domains to classify EEG signals from schizophrenia patients and healthy individuals. Time-domain features, frequency-domain features, as well as nonlinear and statistical features were extracted, and 10 feature combinations were selected based on importance using a hybrid mutual information and Sequential Forward Feature Selection approach. Classification was performed using K-nearest neighbors (KNN), weighted KNN, linear and nonlinear support vector machines (SVM) with radial basis function kernels, decision trees, linear discriminant analysis, and the naive Bayes method. Remarkably, three classifiers achieved 100% accuracy.