Longitudinal Scalar-on-Image Regression

• Activity Number: 227
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Lei Huang*+ and Jeff Goldsmith and Philip Reiss and Ciprian Crainiceanu
• Companies: The Johns Hopkins University and The Johns Hopkins University and New York University and The Johns Hopkins University
• Keywords: Bayesian Variable Selection ; Binary Markov Random Field ; Guassian Markov Random Field ; Markov Chain Monte Carlo ; Diffusion Tensor Imaging ; Longitudinal Functional Data

Abstract

Recently, regression of scalar responses on functional predictors has been extended to the case of high-dimensional images as predictors. This talk will present a new method that introduces a latent indicator coefficient with an Ising prior to achieve the sparsity of estimated coefficient image, and a Gaussian Markov random field prior to control the spatial smoothness of the non-zero coefficient regions. Estimation is fast and relies on a combination of empirical Bayes estimation and a highly efficient Gibbs sampler. The methodological development was motivated by our current studies of the association between intracranial white matter microstructure and cognitive disability in multiple sclerosis patients; full brain fractional anisotropy images are treated as predictors and cognitive measures as scalar responses. We extend our methodology to the longitudinal case to analyze cognitive outcomes and brain imaging data collected at multiple visits.