MICROSTRUCTURAL VULNERABILITY IN A PRECLINICAL ALZHEIMER'S DISEASE COHORT: MODEL FITTING OF DIFFUSION MRI WITH CONVENTIONAL AND SIMULATION-BASED INFERENCE APPROACHES

Neuroinflammation is an early and dynamic feature of preclinical Alzheimer's disease (AD) that may influence amyloid-β (Aβ) pathology. Diffusion-weighted MRI (DWI) offers a sensitive means of probing subtle tissue alterations in vivo, particularly when combined with multi-compartment biophysical modelling. However, clinical datasets are often not acquired with protocols suited for advanced DWI models. Here, we present preliminary evidence of microstructural alterations in cognitively unimpaired preclinical AD participants and evaluate simulation-based inference (SBI) as a flexible framework for diffusion model estimation. DWI data from a subset of participants in the ALFA+ study was analysed. Participants were classified as Aβ-positive or Aβ-negative using cerebrospinal fluid Aβ_42/40 cut-offs. Diffusion tensor and free-water imaging metrics were quantified in predefined white matter tracts and AD-signature cortical grey matter regions. Fractional anisotropy (FA), mean diffusivity (MD), and free-water fraction (FW_F) were estimated using conventional weighted least-squares fitting and SBI. Agreement between approaches was assessed using Lin’s concordance correlation coefficient (CCC, voxel-wise agreement) and the structural similarity index (SSIM, spatial correspondence). Following quality control, 179 datasets were available for analysis, including 68 Aβ-positive participants. Aβ-positive participants exhibited lower MD in the uncinate fasciculus and inferior temporal cortex. A significant Aβ-by-C-reactive protein interaction was observed for uncinate FW_F. SBI-derived diffusion maps showed excellent correspondence with conventional fitting (FA: CCC = 0.94, SSIM = 0.94), supporting its validity. These findings identify the uncinate fasciculus as a potential locus of early microstructural change and highlight SBI as a promising framework for future estimation of diffusion indices derived from multi-compartment biophysical DWI models.