Learning Explainable Imaging-Genetics Associations Related to a Neurological Disorder

Area of Science:

• Neuroscience
• Genetics
• Artificial Intelligence

Background:

• Traditional imaging-genetics methods lack interpretability.
• Understanding brain structure and genetic interplay in neurological disorders is crucial.

Purpose of the Study:

• Introduce NeuroPathX, an explainable deep learning framework.
• Capture interactions between brain MRI data and genetic pathways.
• Enhance interpretability and robustness of findings.

Main Methods:

• Utilize an early fusion strategy with cross-attention mechanisms.
• Incorporate sparsity and pathway similarity loss functions for attention matrix.
• Validate on autism spectrum disorder and Alzheimer's disease datasets.

Main Results:

• NeuroPathX outperforms existing baseline approaches.
• Identifies biologically plausible associations with neurological disorders.
• Demonstrates enhanced interpretability and robustness.

Conclusions:

• NeuroPathX offers a powerful tool for advancing the understanding of complex brain disorders.
• Highlights the potential of explainable AI in neuroimaging genetics.
• Facilitates discovery of novel genotype-phenotype relationships.