A New imaging Biomarker of Gray-White Matter Covariance in Alzheimer's Disease Using a Variational Autoencoder-based Framework

Motivation: This study is for a better understanding of gray matter brain areas and white matter fiber tracts interactions on multi-center datasets in Alzheimer's Disease (AD). Goal(s): The primary goal is to discover a new biomarker of gray-white matter covariance patterns associated with cognitive decline in AD and MCI with neural networks. Approach: We employed a Variational Autoencoder-based framework combining diffusion and structural MRI metrics, training on Normal Control data and testing on MCI and AD datasets to reconstruct new metrics. Results: Our analysis revealed significant correlations between gray-white matter metrics and cognitive scores, highlighting critical fiber bundles and brain areas affected in AD. Impact: This study enhances our understanding of the interdependence between the structure of gray matter brain areas and white matter fiber tracts in Alzheimer's Disease, and a new cognitive-related biomarker was proposed.