A0752
Title: Multimodal topological analysis of neuroimaging in Alzheimer's disease via persistent homology
Authors: Yuexuan Wu - University of South Carolina (United States) [presenting]
Abstract: Alzheimer's disease (AD) involves complex pathological processes, including the accumulation of beta-amyloid and tau proteins and progressive structural brain degeneration. Early and accurate prediction of disease onset and progression remains a critical challenge. A unified persistent homology framework is proposed to analyze multimodal neuroimaging data for both mechanistic understanding and predictive modeling. A bi-filtered approach is first introduced for positron emission tomography (PET), enabling threshold-free characterization of the spatial co-localization and interaction between beta-amyloid and tau. By embedding PET-derived brain networks into a dendrogram-based geometric space, multiscale topological patterns that reveal biologically meaningful features are captured, and thus, a stronger association with clinical AD. Additionally, topological features of brain morphology are extracted from structural MRI via cubical persistent homology, and these features are used to reveal patterns of neurodegeneration, classify disease stages, and forecast progression. The integration of topological features across multiple imaging modalities improves disease prediction performance over existing methods. Together, the utility of the persistent homology-based approach is demonstrated in neuroimaging for both biomarker discovery and robust prediction, offering valuable insights for the early detection of AD.
