A Machine-Learning-Based Investigation of Schizophrenia Using Structural MRI

Schizophrenia is a serious mental health concerns that affects 1% of the population (Jones et al., 2005). This study aimed to create objective tools that can correctly classify people with schizophrenia according to their diagnosis, predominant symptoms, illness duration, and illness severity based on their structural brain imaging variables. 1087 brain images (700=healthy controls, 387=people with schizophrenia) included in the analysis. Support Vector Machines, random forests, logistic regression, and XGBoost were used for diagnostic classification and reached 71% of maximum accuracy. Sulcal width was found to be the most important brain imaging variable that differed between groups. Support vector machines and random forests were used to classify patients according to their predominant symptoms and these classifications reached a maximum accuracy of 66%. Support vector machines could correctly classify people with schizophrenia according to their illness duration with a 75% accuracy and according to their illness severity with 69%. The result of the study shows that using machine learning methods, it is possible to create objective tools for schizophrenia that can be later used in clinics. Keywords: Schizophrenia, Structural MRI, Machine Learning Classification