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

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Psychiatry

Background:

• Accurate diagnosis of Major Depressive Disorder (MDD) is crucial but challenged by subjective traditional methods.
• T1-weighted MRI offers stable, interpretable data but automatic classification of MDD remains difficult due to disease heterogeneity and complex brain structures.

Purpose of the Study:

• To develop a hierarchical multi-scale feature fusion network for accurate, multi-center classification of MDD using T1-weighted MRI.
• To improve upon existing methods for automated MDD diagnosis by integrating advanced feature extraction and fusion techniques.

Main Methods:

• Proposed a novel network combining 3D re-parameterized Vision Transformer (3D RepViT) for local and global feature extraction of gray and white matter.
• Implemented a 3D hierarchical multi-scale feature fusion (3D HMSFF) module to integrate multi-scale structural information across four stages.
• Utilized the large-scale, multi-center REST-meta-MDD dataset with 2,226 subjects for validation.

Main Results:

• The proposed model achieved an overall accuracy of 74.89% on the REST-meta-MDD dataset.
• Demonstrated high performance with a sensitivity of 0.7850, specificity of 0.7077, and an Area Under the Curve (AUC) of 0.8525.
• Outperformed existing methods in MDD classification accuracy.

Conclusions:

• The developed hierarchical multi-scale feature fusion network provides an efficient and generalizable solution for automated MDD classification using T1-weighted MRI.
• The model shows significant potential for clinical applicability and aiding in the assisted diagnosis of MDD.