Brain Structure-Function Fusing Representation Learning Using Adversarial Decomposed-VAE for Analyzing MCI.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Integrating structural and functional brain connectivity is crucial for understanding brain science and diagnosing cognitive impairments.
• Current methods face challenges in effectively fusing multimodal brain network data for comprehensive analysis.

Purpose of the Study:

• To propose a novel brain structure-function fusing-representation learning (BSFL) model for analyzing mild cognitive impairment (MCI).
• To effectively learn fused representations from diffusion tensor imaging (DTI) and resting-state functional magnetic resonance imaging (fMRI) data.

Main Methods:

• Developed a decomposition-fusion framework to separate and adaptively fuse uniform and unique feature spaces from DTI and fMRI.
• Incorporated a knowledge-aware transformer module to capture local and global brain connectivity patterns.
• Utilized a uniform-unique contrastive loss to improve feature decomposition and complementarity.

Main Results:

• The proposed BSFL model demonstrated superior performance in predicting and analyzing MCI compared to existing methods.
• The model effectively learned fused representations, highlighting its potential for reconstructing unified brain networks.

Conclusions:

• The BSFL model offers a promising approach for integrating multimodal neuroimaging data in MCI analysis.
• This model can serve as a valuable tool for predicting abnormal brain connections during MCI progression.