Diagnosis of Schizophrenia from Electroencephalogram using Machine Learning: A Brief Review

This study reviewed some studies that applied machine learning algorithms for schizophrenia diagnosis from EEG signals, which focus on segregating between schizophrenia subjects and healthy controls. The majority of the studies utilized a hand-crafted feature-extracted approach, with just a few adopting automatic feature extraction with deep learning techniques. Some of the studies with hand-crafted feature extraction also incorporated feature selection such as Fisher's technique, recursive feature selection, black hole, and the relief algorithm. A good number of the works achieved high classification accuracy above 90%, with a particular study based on hand-crafted feature extraction that utilized feature fusion as against feature selection, achieving excellent classification performance of 100%. With the encouraging results from the literature review, features extracted from EEG are now established biomarkers for the diagnosis of schizophrenia. Only one of the studies developed a regression model that is capable of estimating schizophrenia symptom severity, which will be useful in the management of the ailment because it provides a means of assessing treatment effectiveness, and more research needs to be focused on this direction.