Machine learning model for schizophrenia diagnosis using electroencephalogram (EEG)

Abstract

Schizophrenia is a chronic mental condition that affects a significant number of the population, the disease shows psychotic symptoms which manifest as delusions, hallucinations and cognitive deficits. The signs and severity of the symptoms often progresses quite rapidly though individual patients may experience varying levels of progressions. Early identification of the onset and advanced stages of the mental illness is crucial for prompt and efficient therapy to stop or lessen the disease's progression. Currently the main means of diagnosis of different mental illnesses (including Schizophrenia) can be found in the Diagnostic and Statistical Manual of Mental Disorders (DSM-5). The patient is questioned about the disease's symptoms and prognosis as part of the conventional diagnostic procedure. Clinical assessments, including the Positive and Negative Syndrome Scale (PANSS) is often used to gauge how severe the symptoms are. Even seasoned psychiatrists may find it challenging to regularly diagnose patients accurately, largely because of the erroneous information given by patients. To increase the accuracy of the diagnosis, it is crucial to create an objective technique that measures the intensity of the symptoms using quantitative biomarkers. EEG signals can be a helpful supplementary diagnostic tool for people with schizophrenia to be used by psychiatrists. It has been observed that time frequency modification of EEG signals obtained from electrodes is enough for the detection of schizophrenia. This study proposes the use of machine learning methodology to build a Schizophrenia Detection model through ERP (Event Related Potential) data obtained from EEG (Electroencephalograph). The data was sourced from an experiment conducted to determine a lack of neural and motor suppression from internal stimulus in patients with Schizophrenia. The data was used to train various machine learning algorithms and KNN ( K Nearest Neighbor) provided the highest accuracy with 93%.

Keywords: Schizophrenia detection, Schizophrenia prediction, Electroencephalography, EEG, ERP, Event Related Potential, Machine Learning, mental disorder