Varying Coefficient Models for Neuroimaging Data

• Activity Number: 512
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Linglong Kong*+ and Hongtu Zhu
• Companies: The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
• Keywords: Varying coefficient models ; Neuroimaging ; weighted least square ; quantile regression ; multiscale adaptive estimation ; Functional principal analysis

Abstract

Motivated by recent work studying massive imaging data in the neuroimaging literature, we propose three varying coefficient models: multivariate varying coefficient models (MVCM) for modeling the relation between multiple functional responses and covariates, spatial varying coefficient models (SVCM) for spatially modeling the relation between high-dimensional imaging measures on a 2D or 3D region covariates, and multiscale adaptive composite quantile regression models (MACQRM) to robustly model those associations between image measures and covariates. All the three models allow the coefficients change to characterize the varying associations between responses and a set of covariates. Weighted least squares and multiscale adaptive estimation methods are used to estimate the varying coefficient functions. We use functional principal component analysis to delineate the structure of the variability of responses. The asymptotic properties of the estimation and inference procedures are established. We conduct Monte Carlo simulation to examine the finite-sample performance of the proposed procedures. We then illustrate real data analysis on various neuroimaging data.