Analysis of first-episode and chronic schizophrenia using multi-modal magnetic resonance imaging

The brain structure and function differences among first-episode schizophrenia (FESZ) patients, chronic schizophrenia (CSZ) patients, and normal control (NC) subjects were investigated using structural and functional magnetic resonance imaging (MRI). Also, a support vector machine (SVM) combined with recursive feature elimination (RFE) was used for classification.

First, 44 FESZ patients, 44 CSZ patients, and 56 NC subjects were recruited, and structural MRI images were acquired. The regional gray matter volumes (GMVs) of 90 regions of interest (ROIs) were calculated, two-sample t-tests were conducted to analyze the GMV differences among the groups, and the partial correlations between the Positive and Negative Syndrome Scale (PANSS) scores and altered regional GMVs were calculated. Individual functional MRI images of the three groups were measured. The individual regional homogeneity (ReHo), amplitude of low-frequency fluctuations (ALFF), and degree of centrality (DC) values of the 90 ROIs were calculated and used to evaluate the differences among the groups. Then, the partial correlations between the PANSS scores and altered regional ReHo, ALFF, and DC were determined. An SVM combined with RFE was employed for classification using both structural and functional MRI input features. The sensitivity and specificity were measured to quantify the SVM performance.

The GMVs in the bilateral calcarine of FESZ and CSZ patients were significantly lower than that of NC subjects. Compared to the NC group, the GMV was significantly reduced in numerous additional brain regions of the CSZ group. In comparison to the NC group, the patient groups exhibited significant ReHo increases in several regions and ReHo reductions in the occipital lobe. ReHo in the insula and left postcentral gyrus of CSZ patients were significantly lower than that of the NC subjects. Compared with the NC group, both patient groups exhibited ALFF aberrances in numerous regions. A significant reduction of ReHo, ALFF, and DC in certain regions were also found in patient groups compared with that of NC group. Significant positive correlations were found between the PANSS scores and ReHo and ALFF of the temporal and frontal lobes, while these correlations were negative in the occipital lobe. The SVM with RFE achieved excellent classification performance. The best performance was obtained using the following inputs: the ReHo and ALFF for FESZ/NC classification; the DC, ReHo, and ALFF for FESZ/CSZ classification; and the ReHo and ALFF for CSZ/NC classification.

Our data indicate that compared with the FESZ patients, brain GMV aberrances was increased in the CSZ patients. The functional features including DC, ReHo, and ALFF, could facilitate FESZ diagnosis, which is more sensitive than structural features in classification. The SVM with RFE presents excellent classification performance and assists SZ diagnosis.