Support vector machine-based classification of first episode drug-naïve schizophrenia patients and healthy controls using structural MRI

• Published in: Schizophrenia research Vol. 214; pp. 11 - 17
• Main Authors: Xiao, Yuan, Yan, Zhihan, Zhao, Youjin, Tao, Bo, Sun, Huaiqiang, Li, Fei, Yao, Li, Zhang, Wenjing, Chandan, Shah, Liu, Jieke, Gong, Qiyong, Sweeney, John A., Lui, Su
• Format: Journal Article
• Language: English
• Published: Netherlands Elsevier B.V 01.12.2019
• Subjects: Classification; Cortical thickness; Psychiatric/Mental Health; Schizophrenia; Support vector machine (SVM); Surface area; Support vector machine (SVM); Cortical thickness; ROC; Classification; Schizophrenia; SVM; Surface area; AUC; area under curve; support vector machine; receiver operating characteristic
• DOI: 10.1016/j.schres.2017.11.037
• ISSN: 0920-9964; 1573-2509

Although regional brain deficits have been demonstrated in schizophrenia patients by structural MRI studies, one important question that remains largely unanswered is whether the complex and subtle deficits revealed by MRI could be used as objective biomarkers to discriminate patients from healthy controls individually. To address this question, a total of 326 right-handed participants were recruited, including 163 drug-naïve first-episode schizophrenia (FES) patients and 163 demographically matched healthy controls. High-resolution anatomic data were acquired from all subjects and processed via Freesurfer software to obtain cortical thickness and surface area measurements. Subsequently, the Support Vector Machine (SVM) was used to explore the potential utility for cortical thickness and surface area measurements in the differentiation of individual patients and healthy controls. The accuracy of correct classification of patients and controls was 85.0% (specificity 87.0%, sensitivity 83.0%) for surface area and 81.8% (specificity 85.0%, sensitivity 76.9%) for cortical thickness (p<0.001 after permutation testing). Regions contributing to classification accuracy mainly included the gray matter in default mode, central executive, salience, and visual networks. Current findings, in a sample of never-treated FES patients, suggest that the patterns of illness-related gray matter changes has potential as a biomarker for identifying structural brain alterations in individuals with schizophrenia. Future prospective studies are needed to evaluate the utility of imaging biomarkers for research and potentially for clinical purpose.