Researchers of a recent study aimed to improve the classification of schizophrenia and its subtypes by applying machine learning (ML) techniques to magnetic resonance imaging (MRI) data. Using a public dataset from University of California, Los Angeles, researchers analyzed structural and resting-state functional MRI data from 50 patients with schizophrenia and 50 matched healthy controls. By extracting cortical and subcortical volumetric features and graph-based measures (e.g., degree, betweenness centrality), the team trained ML models to distinguish between groups. The k-nearest neighbor algorithm achieved 79% classification accuracy using a reduced set of 12 neuroimaging features selected via the minimum redundancy maximum relevance method. When tested on an external dataset, the model maintained a respectable 72% accuracy.

Additionally, the study explored the classification of schizophrenia subtypes using a linear support vector machine, achieving a 64% accuracy with 62 selected features. Notably, neuroimaging markers were linked with cognitive performance—specifically, the degree of the postcentral gyrus showed a significant correlation with reaction time on a verbal task. These results highlight the potential of combining MRI with ML to improve diagnostic precision in schizophrenia, identify neural subtypes, and better understand associated cognitive impairments.

Reference: Tavakoli H, Rostami R, Shalbaf R, Nazem-Zadeh MR. Diagnosis of Schizophrenia and Its Subtypes Using MRI and Machine Learning. Brain Behav. 2025;15(1):e70219. doi: 10.1002/brb3.70219.