Meet the winner of the 2nd edition of the QuanTIM Webinar PhD Award !
This work advances Alzheimer's disease (AD) diagnostics by investigating signs of vascular damage like white matter hyperintensities (WMHs) as potential biomarkers. Initially, it focused on optimizing the automatic segmentation pipelines for WMHs by validating a harmonization strategy for reliable segmentation across multiple datasets and scanners. Then, it explored Quantitative Susceptibility Mapping's contribution to WMH detection, finding significant improvements when incorporated into the imaging protocol. Finally, it used deep learning (DL) models for AD classification across multiple populations, employing explainability techniques to compare WMHs' relevance against healthy tissues and established AD biomarkers.
Remarkably, WMHs demonstrated higher relevance than both normal-appearing white matter and medial temporal loberegions. This finding remained consistent across DL learning architectures and explainability approaches. Our results strongly support including WMHs in AD diagnostic criteria, potentially enhancing early detection and treatment efficacy.
Keywords. White Matter Hyperintensities, Alzheimer's Disease Diagnosis, Imaging Biomarkers, Artificial Intelligence
