Hierarchical Multi-Scale Feature Fusion Network for Multi-Center Major Depressive Disorder Classification with...

Accurate diagnosis of Major Depressive Disorder (MDD) is critical for effective clinical intervention. However, traditional diagnostic methods heavily rely on subjective assessments, which increases the risk of misdiagnosis. T1-weighted magnetic resonance imaging (MRI) has shown great promise in MDD research due to its stability and interpretability. Nevertheless, automatic classification remains challenging due to the heterogeneity of MDD and the complex structural characteristics of the brain. To address this issue, we propose a hierarchical multi-scale feature fusion network for multi-center MDD classification with T1-weighted MRI. This model separately extracts gray matter (GM) and white matter (WM) features with 3D re-parameterized Vision Transformer (3D RepViT), and fuses multi-scale structural information of GM and WM via the 3D hierarchical multi-scale feature fusion (3D HMSFF) module. The 3D RepViT combines CNN-based local feature extraction with Transformer-based global modeling, while re-parameterization improves computational efficiency. The 3D HMSFF module extracts and fuses hierarchical multi-scale features across four stages. Experimental results on the multi-center, large-scale REST-meta-MDD dataset, comprising 2,226 subjects, demonstrate that our method achieves an overall accuracy of 74.89%, a sensitivity of 0.7850, a specificity of 0.7077, and an AUC of 0.8525, outperforming existing methods. These results suggest that our method may serve as an efficient and generalizable solution for automatic MDD classification, with potential clinical applicability.Clinical Relevance-This study proposes an automated MDD classification model using T1-weighted MRI, validated on a multi-center, large-scale dataset (REST-META-MDD) with 2,226 subjects. The model achieves overall accuracy of 74.89% and AUC of 0.8525, demonstrating its potential for pathological interpretability research and clinical-assisted diagnosis of MDD.