IntroEvidence from neuroimaging has implicated abnormal cerebral cortical patterns in schizophrenia. Application of machine learning techniques is required for identifying structural signature reflecting neurobiological substrates of schizophrenia at the individual level. We aimed to detect and develop a method for potential marker to identify schizophrenia via the features of cerebral cortex using high-resolution magnetic resonance imaging (MRI).MethodIn this study, cortical features were measured, including volumetric (cortical thickness, surface area, and gray matter volume) and geometric (mean curvature, metric distortion, and sulcal depth) features. Patients with first-episode schizophrenia (n = 52) and healthy controls (n = 66) were included from the Department of Psychiatry at Xijing Hospital. Multivariate computation was used to examine the abnormalities of cortical features in schizophrenia. Features were selected by least absolute shrinkage and selection operator (LASSO) method. The diagnostic capacity of multi-dimensional neuroanatomical patterns-based classification was evaluated based on diagnostic tests.ResultsMean curvature (left insula and left inferior frontal gyrus), cortical thickness (left fusiform gyrus), and metric distortion (left cuneus and right superior temporal gyrus) revealed both group differences and diagnostic capacity. Area under receiver operating characteristic curve was 0.88, and the sensitivity, specificity, and accuracy of were 94%, 82%, and 88%, respectively. Confirming these findings, similar results were observed in the independent validation. There was a positive association between index score derived from the multi-dimensional patterns and the severity of symptoms (r = 0.40, P < .01) for patients.DiscussionOur findings demonstrate a view of cortical differences with capacity to discriminate between patients with schizophrenia and healthy population. Structural neuroimaging-based measurements hold great promise of paving the road for their clinical utility in schizophrenia.
Full Text via ClinicalKey