A New Brain Network Construction Paradigm for Brain Disorder via Diffusion-Based Graph Contrastive Learning.

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Brain network analysis is crucial for understanding brain function and disease mechanisms.
• Current tools for brain network construction suffer from user dependency, inconsistent results, and inefficiency.

Purpose of the Study:

• To introduce DGCL, a diffusion-based pipeline for automated and consistent end-to-end brain network construction.
• To enhance the accuracy and generalization of brain network analysis for disease research.

Main Methods:

• DGCL utilizes a brain region-aware module (BRAM) with diffusion processes for precise spatial localization, avoiding subjective parameter choices.
• Graph contrastive learning optimizes brain connections by removing irrelevant individual differences, improving network consistency.
• Jointly applied node-graph contrastive loss and classification loss refine the model for network reconstruction and analysis.

Main Results:

• DGCL demonstrated superior performance over traditional and deep learning methods in predicting disease progression stages on ADNI and ABIDE datasets.
• The model significantly improved the efficiency and generalization capabilities of brain network construction.
• DGCL effectively identified critical brain connections using generative paradigms.

Conclusions:

• DGCL offers a universal scheme for brain network construction, enhancing consistency and efficiency.
• The method has the potential to provide valuable disease interpretability support in neuroscience research.
• DGCL facilitates the identification of important brain connections for a deeper understanding of neurological conditions.