Abstract
Schizophrenia(SZ) classification and treatment response prediction hold substantial clinical application value. However, only a limited number of researchers have exploited the multi-feature information derived from resting-state functional magnetic resonance imaging (rs-fMRI) to achieve short-term drug-treatment SZ classification and treatment response prediction. We developed a multi-feature fusion recursive feature elimination random forest (RFE-RF) approach for SZ classification and treatment response prediction. Initially, we computed multiple features, such as regional homogeneity, fractional amplitude of low-frequency fluctuations, and functional connectivity. Subsequently, the RFE-RF method was employed to conduct SZ classification. Moreover, we utilized the rate of score reduction (RR) of the Positive and Negative Symptom Scale (PANSS) to forecast the treatment response of individual patients. Finally, we identified the neuroimaging biomarkers for SZ classification and drug-treatment response prediction. This method achieved the classification results (accuracy = 91.7%, sensitivity = 90.9%, and specificity = 92.6%), and the abnormalities in the visual and default mode networks emerged as potential neuroimaging biomarkers for differentiating SZ from healthy controls (HC). Additionally, we predicted the drug-treatment response of SZ patients in terms of their total PANSS scores, as well as negative and positive symptom scores after eight weeks of treatment. Specifically, the abnormalities in the visual network, sensorimotor network, and right superior frontal gyrus are crucial biomarkers for the short-term drug-treatment response of negative symptoms in SZ patients. Meanwhile, the abnormalities in the visual and default mode networks serve as important biomarkers of the short-term drug-treatment response of positive symptoms. There findings offer novel insights into the neural mechanisms underlying SZ following eight weeks of short-term drug treatment. With further clinical validation in the future, this research may provide potential biomarkers and intervention targets for personalized treatment of SZ.