Attention-based Multimodal Spatiotemporal Enhanced Interaction Network For Major Depressive Disorder Detection.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Deep learning shows promise for detecting major depressive disorder (MDD).
• Existing models struggle with multimodal brain network interactions and spatiotemporal dependencies.
• Limitations hinder accurate MDD detection using neuroimaging data.

Purpose of the Study:

• To propose the Attention-based Multimodal Spatiotemporal Enhanced Interaction Network (AM-SEIN) for improved MDD detection.
• To address limitations in exploiting interactive information across multimodal brain networks.
• To develop adaptive mechanisms for capturing spatiotemporal dependencies among brain regions.

Main Methods:

• Integrated 3D structural magnetic resonance imaging (sMRI) and functional magnetic resonance imaging (fMRI) data.
• Designed Cross-Modal Interaction Network (CMIN) for enhanced mutual information aggregation.
• Developed attention-based adaptive spatiotemporal feature extraction modules (fASF and sRLCD).

Main Results:

• The AM-SEIN model achieved state-of-the-art performance on the Rest-meta-MDD(RMM) and Rest-meta-MDD-V2(RMM-V2) datasets.
• Effectively encoded inter-regional interactions crucial for MDD detection.
• Demonstrated enhanced mutual information aggregation and interactive guidance between sMRI and fMRI modalities.

Conclusions:

• The proposed AM-SEIN effectively addresses limitations in current deep learning models for MDD detection.
• The integration of multimodal data and adaptive spatiotemporal feature extraction improves diagnostic accuracy.
• AM-SEIN represents a significant advancement in computational approaches for diagnosing major depressive disorder.