Integrating structural and functional brain features to classify major depressive disorder: a multi modal approach

Major Depressive Disorder (MDD) is a common mental disorder that markedly impairs psychosocial functioning and quality of life. Multi-modal fusion methods are essential for integrating diverse neuroimaging modalities to better understand complex psychiatric disorders such as MDD. This study aims to investigate neural differences between individuals with MDD and healthy controls (HCs), explore the relationship between identified neuroimaging components and the Beck Depression Inventory (BDI-II) score, and develop a predictive model capable of generalizing to new cases for classifying MDD patients and HCs using a supervised machine learning approach, Random Forest (RF). To achieve this, we applied an unsupervised data fusion machine learning approach (Parallel Independent Component Analysis) to the gray matter (GM), white matter (WM), and regional homogeneity (ReHo) images of 197 MDD patients and 172 HCs to identify structural and functional brain network alterations. Results showed GM reductions in the frontal lobe, particularly in regions implicated in emotional regulation, such as the anterior cingulate cortex and the middle frontal cortex. WM increases were observed in the cerebellum and regions related to the default mode network (DMN). Moreover, enhanced functional activity was found in the dorsomedial prefrontal areas of the DMN. These networks showed significant correlations with BDI scores. Importantly, the RF classifier, which identified the same regions as key classification features, achieved 75.68 % accuracy in distinguishing MDD patients from HCs. These findings underscore the value of combining multimodal, data-driven approaches in revealing the neural basis of MDD and advancing the development of precision diagnostic tools in psychiatric.