Deep Learning-Based Regional White Matter Hyperintensity Mapping as a Robust Biomarker for Alzheimer's Disease

TL;DR

The paper tackles the limitation of global WMH burden by introducing a deep learning pipeline that segments WMHs and maps them to 34 white matter subregions. Trained on the MICCAI WMH Challenge data and validated on ADNI, the approach yields regional WMH loads that generalize across datasets and outperform global WMH in AD-status prediction, achieving up to $0.97$ AUC when combined with brain volumetrics. Key contributions include a robust 3D U-Net segmentation framework, atlas-based regionalization, and demonstration that region-specific WMH patterns—especially in anterior white matter tracts—are associated with diagnostic status. This regional WMH quantification, combined with atrophy markers, enhances early diagnosis and stratification potential in Alzheimer's disease and related dementias, and supports scalable analyses in large cohorts. The work provides a practical pathway to incorporate spatial WMH biomarkers into clinical research and predictive modeling pipelines.

Abstract

White matter hyperintensities (WMH) are key imaging markers in cognitive aging, Alzheimer's disease (AD), and related dementias. Although automated methods for WMH segmentation have advanced, most provide only global lesion load and overlook their spatial distribution across distinct white matter regions. We propose a deep learning framework for robust WMH segmentation and localization, evaluated across public datasets and an independent Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort. Our results show that the predicted lesion loads are in line with the reference WMH estimates, confirming the robustness to variations in lesion load, acquisition, and demographics. Beyond accurate segmentation, we quantify WMH load within anatomically defined regions and combine these measures with brain structure volumes to assess diagnostic value. Regional WMH volumes consistently outperform global lesion burden for disease classification, and integration with brain atrophy metrics further improves performance, reaching area under the curve (AUC) values up to 0.97. Several spatially distinct regions, particularly within anterior white matter tracts, are reproducibly associated with diagnostic status, indicating localized vulnerability in AD. These results highlight the added value of regional WMH quantification. Incorporating localized lesion metrics alongside atrophy markers may enhance early diagnosis and stratification in neurodegenerative disorders.

Figures (4)

• Figure 1: Overview of the proposed pipeline for AD status prediction. (A) Regional WMH lesion and white matter region maps are obtained using 3D U-Net models [machnio2025deepmachnio2025towards] trained on the WMH Challenge dataset. (B) The trained models are applied to ADNI scans to extract local and global WMH volumes, which are subsequently used for downstream AD status prediction.
• Figure 2: (A–B) Distribution of WMH load across datasets: (A) WMH Challenge training and test sets, (B) Our predicted WMH load compared to UCD reference estimates in ADNI. (C) Bland–Altman plot of our predicted versus reference WMH volumes in ADNI, with each point representing a subject colored by diagnostic group.
• Figure 3: Predicted white matter lesions and regional white matter segmentations for three representative ADNI subjects with high WMH loads.
• Figure 4: ROC curves for AD prediction tasks using different MRI feature sets. Regional WMH consistently outperforms global WMH, and combining local WMH with brain volumetrics further boosts performance.