B1707
Title: Longitudinal mixed membership image-on-scalar model for learning the progression of Alzheimers disease
Authors: Zhihao Wu - The Chinese University of Hong Kong (Hong Kong) [presenting]
Abstract: While magnetic resonance imaging (MRI) data has been widely used for the diagnosis and/or prediction of Alzheimers disease (AD), very little attention has been paid to the individual heterogeneity in terms of longitudinal MRI data. We propose a novel modeling framework for describing the dynamic pattern of longitudinal imaging data and use the proposed model to learn the progression of AD. First, a basis expansion approach is adopted to approximate the longitudinal images. Then, we introduce a vector of probability weights characterizing mixed membership to capture the individual heterogeneity. Finally, the approximated longitudinal images are modeled using regression models under typical membership. A Bayesian approach coupled with MCMC methods is developed to conduct statistical inference. Simulation results demonstrate a good performance of estimation under a medium sample size. The approach is applied to the Alzheimers Disease Neuroimaging Initiative (ADNI) to discover the dynamic pattern of brain regions in AD progression.
