Machine learning-based integration of surface morphometric features identifies diagnostic and symptom-related...

Schizophrenia (SCZ) is associated with widespread cortical abnormalities, however, neuroanatomical markers with robust diagnostic and clinical relevance remain limited. In this study, we integrated multiple surface-based cortical features-cortical thickness, gyrification, sulcal depth, cortical complexity, and surface ratio-from a large multi-site cohort (329 SCZ, 669 healthy controls) and implemented a rigorous cross-validated model with repeated random split framework to evaluate their relevance to SCZ. Models incorporating multiple feature metrics achieved robust and reproducible classification accuracy of up to 77%. Feature importance analyses revealed convergent neuroanatomical patterns across models, with cortical thickness and gyrification emerging as the most discriminative metrics. These alterations were predominantly localized to prefrontal, insular, and temporal regions. Notably, the identified cortical thickness in lateral prefrontal regions was significantly associated with symptom severity, as measured by Positive and Negative Syndrome Scale (PANSS) scores. Together, these findings highlight the utility of integrated surface-based cortical measures as neurobiologically grounded markers of both disease presence and clinical symptom burden in schizophrenia.