Continuous Dictionary of Nodes Model and Bilinear-Diffusion Representation Learning for Brain Disease Analysis.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Functional magnetic resonance imaging (fMRI) is vital for brain disease diagnosis.
• Representation learning shows promise for brain disease analysis but struggles with higher-order network interactions.
• Existing methods often overlook indirect and extensive node interactions in brain networks.

Purpose of the Study:

• To develop a novel method for analyzing brain networks to improve brain disease diagnosis.
• To address limitations in traditional representation learning for capturing higher-order brain network interactions.
• To enhance the understanding of brain disease pathology through advanced network analysis.

Main Methods:

• Proposed the Continuous Dictionary of Nodes model and Bilinear-Diffusion (CDON-BD) network.
• Utilized the CDON model to learn original brain networks and extract latent features.
• Employed Bilinear Pooling to construct higher-order brain networks and a Diffusion Module to capture extensive node interactions.

Main Results:

• The CDON-BD network achieved competitive classification performance on two real-world datasets.
• Higher-order representations learned by CDON-BD identified brain regions relevant to specific diseases.
• Demonstrated superior capability in capturing indirect and extensive node interactions compared to state-of-the-art methods.

Conclusions:

• The CDON-BD network offers a powerful new tool for brain disease analysis using fMRI data.
• The method effectively constructs higher-order brain networks, revealing crucial disease-related brain regions.
• This approach contributes to a deeper understanding of the underlying pathology of brain diseases.