Quantifying anomalies in patients with neuropathologies via diffusion magnetic resonance imaging (application to alzheimer diagnosis)

Details

Supervisors

Macq, Benoît ; Dricot, Laurence

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil biomédical, à finalité spécialisée

Abstract

This thesis investigates the white matter microstructural signature of Alzheimer's disease (AD) using a multi-model diffusion MRI framework anchored to local tau pathology. Three diffusion models with complementary biophysical assumptions, DTI, NODDI, and DIAMOND, were applied to the UCL2016-121 clinical cohort of 120 participants (28 cognitively normal, 65 mild cognitive impairment, 27 dementia; 112 with amyloid status). Microstructural metrics were extracted both regionally, using a dilated Desikan-Killiany atlas, and along seventeen reconstructed white matter bundles, then related to regional tau-PET (¹⁸F-MK-6240) SUVr and to MMSE cognitive scores. The work follows a region-tract-tau triangulation: for each cortical region it asks whether the region is microstructurally altered, whether its connecting bundles are altered, and whether local tau burden is associated with that alteration. Cortical alterations followed a Braak-like topography, with the largest effect sizes in lateral frontal (Braak IV-V) territories rather than classical medial-temporal structures, an asymmetry attributed to a ceiling effect in already-impaired participants and to partial-volume limitations of dMRI in thin curved cortex. The connecting bundles, principally the inferior longitudinal, uncinate, and inferior fronto-occipital fasciculi and the corpus callosum, were substantially altered, and a subset of these alterations was anchored to local tau, with the strongest tau-bundle coupling at the cortical endpoints. Methodologically, the three models showed distinct sensitivity profiles: DTI gave the broadest tau-coupled detection including subcortical and commissural single-fibre structures, NODDI sacrificed regional coverage for compartment-specific biological interpretability and provided the strongest single bundle-level tau coupling through f_intra, and DIAMOND resolved crossing-fibre territories and yielded the largest fascicle-resolved tau slopes. The biological (amyloid) classification produced systematically larger effects than the cognitive one, with a substantial preclinical signal already detectable in cognitively unimpaired amyloid-positive participants. The convergence of three models on the same tau-coupled bundle topography supports a multi-model approach for clinical AD pipelines, and the doubly validated regions and bundles are candidate substrates for individual-level biomarker development.