Virtual histology of Alzheimer’s disease pathology in the ex vivo human brain using diffusion-T2 relaxation correlation multidimensional MRI

Motivation: Due to the complexity of brain tissue, single-contrast MRI approaches often lack the sensitivity and specificity to identify a specific pathology. Goal(s): Our goal was to develop a method to predict specific Alzheimer's disease (AD) pathologies or their cellular features from diffusion-T2 relaxation correlation density distributions. Approach: We adopted a multimodal approach by integrating and co-registering histological and MRI datasets from 12 human donors with mixed AD diagnoses. We used elastic net-regularized linear regression to predict histological stains from the diffusion-T2 density distribution. Results: We could predict voxelwise amyloid beta, phosphorylated tau, microglia, and myelin content (cross-validated R2=0.32,0.55,0.49,0.60 and Pearson correlation 0.57,0.74,0.70,0.78, respectively). Impact: Ex vivo diffusion-T2 relaxation correlation MRI can be used to detect multiple Alzheimer's disease pathologies. This approach could lead the way for advanced in vivo characterization of pathologies related to aging and dementia.