Tensor Generalized Estimating Equations for Longitudinal Imaging Analysis

• Activity Number: 407
• Type: Topic Contributed
• Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Xiang Zhang* and Lexin Li and Hua Zhou and Dinggang Shen
• Companies: and UC Berkeley and North Carolina State University and The University of North Carolina at Chapel Hill
• Keywords: generalized estimating equations ; longitudinal data ; magnetic resonance imaging ; tensor regression

Abstract

In increasing number of neuroimaging studies, brain images, which are in the form of multidimensional arrays, or tensors, have been obtained on each subject at multiple time points. Of scientific interest is to analyze such massive and complex longitudinal images to diagnose neurodegenerative disorders, to predict onset of neuropsychiatric diseases, and to identify disease relevant brain regions or activity patterns. In this article, we treat those problems in a unifying regression framework with image predictors, and propose tensor generalized estimating equations (GEE) for longitudinal imaging analysis. The GEE approach effectively takes into account intra-subject correlation of responses, whereas a low rank tensor decomposition of the coefficient array enables effective estimation and prediction under limited sample size. We propose an efficient estimation algorithm, study the asymptotics and also investigate tensor GEE with regularization that is particularly useful for region selection. The efficacy of the proposed tensor GEE is demonstrated on both simulated data and a real data set from the Alzheimer's Disease Neuroimaging Initiative (ADNI).