Enhancing Major Depressive Disorder Diagnosis With Dynamic-Static Fusion Graph Neural Networks

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Major Depressive Disorder (MDD) is a complex mental health condition with poorly understood mechanisms.
• Functional magnetic resonance imaging (fMRI) shows altered brain connectivity in MDD patients.
• Current fMRI models for MDD diagnosis have limitations, including poor interpretability and inability to capture dynamic network changes.

Purpose of the Study:

• To develop a novel graph neural network framework, DSFGNN, for improved diagnosis of Major Depressive Disorder (MDD).
• To address limitations of existing fMRI-based models by incorporating dynamic network analysis and enhancing interpretability.

Main Methods:

• DSFGNN utilizes a graph isomorphism encoder to model both static and dynamic brain networks.
• A spatiotemporal attention mechanism fuses temporal and spatial information for enhanced interpretability.
• Causal disentangling and orthogonal regularization modules are incorporated to improve model expressiveness.

Main Results:

• DSFGNN achieved superior diagnostic performance on the Rest-meta-MDD dataset compared to existing methods.
• Ablation studies and interpretability analyses validated the model's effectiveness.
• The framework shows potential for identifying novel biomarkers for MDD.

Conclusions:

• DSFGNN offers a significant advancement in AI-driven diagnostic tools for Major Depressive Disorder.
• The model's ability to analyze dynamic brain connectivity and provide interpretable results is a key innovation.
• DSFGNN holds promise for improving diagnostic accuracy and facilitating biomarker discovery in MDD research.