Schizophrenia Detection from Resting State Functional MR Images Using Machine Learning

Psychiatric illnesses like schizophrenia present a great challenge to radiologists trying to diagnose them using biomarkers from magnetic resonance (MR) images. Machine learning algorithms show great promise in the tasks of automatic detection and classification of various psychiatric illnesses, including schizophrenia, from MR brain scans. We implemented and compared three machine learning algorithms (support vector machines, random forest, neural network) for detecting schizophrenia from resting state functional MR images from the COBRE dataset comprising 72 subjects with schizophrenia and 74 healthy controls. We also make a comparison of how different brain atlases impact the performance metrics. Our results showed that it is possible to achieve up to 80% accuracy in schizophrenia detection with machine learning methods, even with a relatively small dataset.