Diffusion-relaxation MRI as virtual histology: separable microstructural signatures of AD pathology in ex vivo human brain

Cognitive decline in Alzheimer's disease (AD) reflects progressive disruption of cellular and microstructural organization, yet the biological specificity of conventional MRI signals remains poorly understood. Multidimensional diffusion–relaxation MRI (MD-MRI) resolves sub-voxel tissue heterogeneity and may offer a framework for linking imaging signals to underlying neuropathology. We tested the hypothesis that neuronal, glial, and white matter pathologies in AD occupy separable regions of diffusion–relaxation space and generate spatially organized signatures associated with cognitive impairment. We integrated ex vivo MD-MRI with co-registered histology from 12 human donors spanning a range of Braak stages and pathological severity. Nested cross-validated elastic net models predicted voxelwise Aβ, pTau, microglia, and myelin burden from the multidimensional diffusion–relaxation density distribution. Regional associations were assessed across hippocampal subfields and white matter, and MRI-predicted pathology was related to ante-mortem Mini-Mental State Examination scores. Distinct diffusion–relaxation components were preferentially associated with different pathological markers, indicating separable microstructural signatures. MRI-derived predictions corresponded significantly with histological measures of myelin (ρ = 0.77), pTau (ρ = 0.62), and microglia (ρ = 0.61), with weaker correspondence for Aβ (ρ = 0.45). Regionally, predicted pathology recapitulated known patterns of selective vulnerability, with elevated pTau and microglial signal in hippocampal subfields and dominant myelin-associated signal in white matter (p < 0.0001). Higher predicted hippocampal pTau was strongly associated with worse cognitive performance (ρ = −0.88, p = 0.0014), with a moderate association in white matter (ρ = −0.66, p = 0.036). These findings demonstrate that AD-related pathological processes manifest as distinct, spatially organized diffusion–relaxation signatures, providing mechanistic insight into the microstructural basis of MRI contrasts. As clinically feasible MD-MRI protocols continue to emerge, translation of these signatures to in vivo imaging may enable more biologically informed assessment of neurodegeneration.