This page was translated automatically from Turkish. Read the original

This page is an AI-assisted Turkish summary of the study presented at the 26th World Congress of Psychiatry. The study belongs to the presenting authors; please refer to the original source when citing it.

EV0922

Resting-state functional connectivity in the differential diagnosis of schizophrenia and mood disorders: a machine learning study

RESTING-STATE FUNCTIONAL CONNECTIVITY FOR THE DIFFERENTIAL DIAGNOSIS OF SCHIZOPHRENIA AND MOOD DISORDERS: A MACHINE LEARNING STUDY

Support vector machine models based on resting-state functional connectivity can distinguish schizophrenia from major depression and bipolar disorder with approximately 70% balanced accuracy.

The aim of this study was to test whether resting-state functional connectivity could distinguish schizophrenia from mood disorders within a machine learning framework. The sample consisted of 50 patients with schizophrenia (SCZ), 148 with major depressive disorder (MDD), and 125 with bipolar disorder (BD). Resting-state fMRI data were preprocessed using HALFpipe, and features were extracted using dual regression (DR) across 14 resting-state networks and seed-based connectivity (SBC) from the bilateral amygdala, striatum, anterior cingulate cortex, and hippocampus. Separate support vector machine (SVM) classifiers were built in PRoNTo for the SCZ versus MDD and SCZ versus BD comparisons, and balanced accuracy (BA) and AUC were calculated using five-fold nested cross-validation. For distinguishing SCZ from MDD, the best-performing model was right amygdala seed-based connectivity (BA 70%, AUC 0.75; p<0.001). In this comparison, 5 significant models were obtained with DR and 5 with SBC; other strong-performing models were DR component 11 (BA 70%, AUC 0.74), DR component 10 (BA 69%, AUC 0.73), and right striatum SBC (BA 69%, AUC 0.72). For distinguishing SCZ from BD, the best-performing model was DR component 13, which included the bilateral cuneus (BA 69%, AUC 0.69; p<0.001). Here, there were 4 significant DR models and 1 significant SBC model, and classification was driven more specifically by dual regression. Other models in this comparison were DR component 12 (BA 65%, AUC 0.70), DR component 2 (BA 65%, AUC 0.65), and left amygdala SBC (BA 61%, AUC 0.59). In conclusion, complementary connectivity signatures distinguish schizophrenia from mood disorders, with the seed-based approach standing out for the SCZ versus MDD distinction and dual regression for the SCZ versus BD distinction. These methods are presented as potential tools for providing objective support to clinical assessment.

All WCP 2026 summaries