Objective outcome prediction in depression through functional MRI brain network dynamics

Research purpose: Subjective clinical decision-making in major depressive disorder (MDD) may result in low treatment effectiveness. This study aims to identify objective predictors of MDD outcome using resting-state functional MRI scans, acquired from 25 MDD patients at baseline. Over a year, patients were assessed every 3 months, labeled as positive or negative outcome (change in depression severity). Group independent component analysis (GICA) identified (sub)networks at different orders, from which static and dynamic (wavelet) fMRI features were extracted. Binary classifiers performed MDD outcome prediction at each follow-up. Principal Results: The total coherence feature, reflecting network interactivity, yielded the highest performance (area under the curve (AUC) of 0.70). In the positive outcome group, total coherence between the default mode network and ventral salience network was increased at all follow-ups. Classification using this feature alone further demonstrated its discriminating capability (AUC of 0.76 ± 0.10 over all follow-ups). These results suggest that a higher switching capability between internal and external brain states predicts symptom improvement. Higher GICA orders, where major networks are divided into subnetworks, yielded optimal classification performance. Major Conclusions: Total coherence, a dynamic fMRI measure, achieved the highest classification performance. These findings contribute to the identification of prognostic biomarkers in MDD.