BDHT: Generative AI Enables Causality Analysis for Mild Cognitive Impairment.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Effective connectivity estimation is vital for understanding brain function but is prone to errors from software parameter variations.
• Existing methods struggle to accurately model complex causal relationships between brain regions, impacting analyses like mild cognitive impairment (MCI).

Approach:

• Proposes a novel brain diffuser with hierarchical transformer (BDHT), the first generative model using diffusion models for multimodal brain network analysis.
• BDHT utilizes structural connectivity to guide the denoising process, enhancing reliability and accuracy of effective connectivity estimation.
• Incorporates a hierarchical denoising transformer and a GraphConformer module (multi-head attention + GCN) to learn multi-scale features and improve structure-function complementarity.

Key Points:

• The BDHT model demonstrates superior accuracy and robustness in effective connectivity estimation compared to existing methods.
• The hierarchical denoising transformer effectively learns multi-scale topological features for improved denoising.
• The GraphConformer module enhances the integration of structural and functional brain connectivity data.

Conclusions:

• The proposed BDHT model offers a powerful new tool for analyzing brain networks and estimating effective connectivity, particularly for MCI.
• This approach can identify altered directional brain connections, providing insights into the pathogenesis of MCI.
• The study highlights the potential of diffusion models in advancing neuroimaging analysis and understanding neurological disorders.