A1189
Title: Learning subtle neurodegeneration for early Alzheimer's disease detection
Authors: Zheyu Wang - Johns Hopkins University (United States) [presenting]
Abstract: Early Alzheimer's Disease (AD) is characterized by subtle and spatially distributed neurodegenerative changes that are difficult to detect with conventional MRI summaries. While MRI is non-invasive and widely available, standard volumetric and regional measures have limited sensitivity for identifying individuals in the earliest disease stages. While incorporating known risk factors such as age and ApoE4 status improves diagnostic performance, the added value of MRI in early detection remains limited. This leads to a crucial question: is this limitation inherent to MRI's temporal placement within the biomarker cascade (e.g., typically changing after amyloid pathology but before major cognitive decline), or because its full diagnostic potential is currently underutilized? Two novel approaches are explored to enhance MRI's capability for early AD detection: 1) high-dimensional distance measures that capture complex, multiregional patterns of subtle neurodegeneration, and 2) an attention-based model that learns how to integrate information across imaging representations to emphasize biologically meaningful signals. The results highlight the potential of MRI to capture early disease-related changes when coupled with modern approaches.
