Deep association analysis framework with multi-modal attention fusion for brain imaging genetics

Area of Science:

• Neuroscience
• Genetics
• Medical Imaging

Background:

• Brain imaging genetics integrates genetic variation and brain imaging data for insights into neurological mechanisms.
• Advancements in medical imaging fuel interest in multi-modal imaging and genetic data correlation.
• Existing methods often fail to capture inter-modal interactions and complex associations between genetics and neuroimaging.

Purpose of the Study:

• To propose a deep association analysis framework with multi-modal attention fusion (DAAMAF) for early Alzheimer's disease (AD) diagnosis.
• To overcome limitations of simple feature concatenation and traditional correlation analysis in exploring intrinsic associations.

Main Methods:

• Extracting multi-modal feature representations from imaging genetics data for nonlinear mapping.
• Designing a cross-modal attention network to learn interactions between multi-modal imaging features.
• Mapping genetic information onto imaging representations via a generative network to capture intrinsic associations.

Main Results:

• DAAMAF demonstrated superior performance in early Alzheimer's disease diagnosis on the AD Neuroimaging Initiative dataset.
• The framework successfully identified disease-related biomarkers associated with AD.
• The approach effectively utilizes complementary information from multi-modal data.

Conclusions:

• DAAMAF offers a promising approach for early AD diagnosis by effectively integrating multi-modal imaging and genetic data.
• The framework contributes to a deeper understanding of Alzheimer's disease pathogenesis.
• The developed methods and findings have significant implications for neurological research and clinical applications.