A0577
Title: Bayesian quantile scalar on image quantile regression via a nonparametric method
Authors: Chuchu Wang - The Chinese University of Hong Kong (Hong Kong) [presenting]
Abstract: The motivation comes from research on Alzheimer's disease and the use of medical imaging data. A quantile scalar-on-image regression model enables a comprehensive study of the relationship between cognitive decline and various clinical covariates and imaging factors. A Bayesian nonparametric model is used here to cope with the complex spatially distributed imaging data. It is assumed that there is a latent Gaussian process to capture the sparse structure of the regression coefficients, and the soft-thresholded operator is used to shrink the coefficient function. The kernel basis functions approximate the latent Gaussian process, which promises an efficient MCMC computation algorithm and a consistent estimation result. The model is represented in a hierarchical form, and a fully Bayesian approach is constructed. A hybrid algorithm combining the Gibbs sampler and Metropolis-Hastings algorithm is adopted to conduct posterior sampling and make the inference. The method's performance is compared with the functional principal component analysis (FPCA) method in simulations, and finally, the proposed method is applied to a study of Alzheimer's disease.