The dark side of the mood: structural and functional fronto-insular and cerebellar alterations classify major depression

Despite major depressive disorder (MDD) being the leading cause of disability worldwide, the exact characterization of its neural bases and the development of reliable biomarkers are still at an early stage. A possible solution lies in multimodal analysis approaches, which integrate cross-modal data to investigate the relationship between structural and functional network disruptions, potentially improving the accuracy of machine learning (ML) models for individual-level predictions. In this study, we employed a data fusion unsupervised ML method called transposed Independent Vector Analysis (tIVA) to investigate joint functional and structural brain networks to classify MDD. To this aim, the amplitude of low-frequency fluctuations (ALFF) and gray matter density (GMD) of 461 participants (MDD = 226, HC = 235) were taken into consideration. The analysis revealed a multimodal link between reduced functional activity in the cerebellum and structural deficits in subcortical regions (primarily including the anterior cingulate cortex (ACC) and insula) implicated in emotional regulation, highlighting how these structural and functional changes can mutually influence and reinforce each other. Moreover, enhanced functional activity was found in dorsomedial prefrontal areas of the default mode network (DMN), with concurrent reduced activity in dorsolateral prefrontal regions of the executive control network (CEN). Importantly, a Random Forest (RF) classifier, which identified the same areas as important classification features, achieved a 69.89% accuracy in distinguishing MDD patients from HC. These findings underscore the value of combining multimodal data-driven approaches to investigate the neural basis of MDD, possibly enhancing diagnostic precision and advancing precision psychiatry.