Attention-Diffusion-Bilinear Neural Network for Brain Network Analysis.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Medical Imaging Analysis

Background:

• Brain network analysis is crucial for diagnosing neurological disorders.
• Integrating functional connectivity (FC) and structural connectivity (SC) offers complementary information for improved diagnostic accuracy.
• Traditional methods often analyze FC or SC in isolation and focus on pairwise node interactions.

Purpose of the Study:

• To propose a novel Attention-Diffusion-Bilinear Neural Network (ADB-NN) framework for integrative brain network analysis.
• To develop a method that couples FC and SC to capture wider node interactions for disease diagnosis.
• To generate joint FC-SC representations for enhanced patient identification.

Main Methods:

• Defined a brain network graph where nodes represent brain regions and edges represent SC derived from Diffusion Tensor Imaging (DTI).
• Utilized functional magnetic resonance imaging (fMRI) for functional activity (FC) features.
• Developed two Attention-Diffusion-Bilinear (ADB) modules trained end-to-end, incorporating attention for interaction strength, diffusion for node representation, and bilinear pooling for feature extraction.

Main Results:

• The proposed ADB-NN framework effectively integrates FC and SC for brain network analysis.
• Experiments on an epilepsy dataset demonstrated the method's effectiveness and advantages over traditional approaches.
• The framework successfully generated joint FC-SC connectivity-based features for disease prediction.

Conclusions:

• The ADB-NN framework provides a powerful tool for integrative brain network analysis.
• This approach enhances the identification of brain disorders by leveraging complementary connectivity information.
• The study highlights the potential of deep learning models for advancing neurological disease diagnosis.