Spatio-Temporal Gaussian Process Models for Neuroimaging Data

• Activity Number: 375
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Jung Won Hyun* and Yimei Li and Chao Huang and Hongtu Zhu
• Companies: St. Jude Children's Research Hospital and St. Jude Children's Research Hospital and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
• Keywords: Neuroimaging ; Spatio-temporal modeling ; Prediction ; Functional principal component analysis

Abstract

Functional neuroimaging plays an important role in defining the neural basis of illness and risk factors for major psychiatric disorders. Recently, there is interest in using functional neuroimaging to provide clinicians with predictive information about individual brain activity. Functional neuroimaging data naturally exhibit complex spatial dependence structures in addition to temporal dependencies between repeated scanning sessions. The aim of this paper is to develop a spatio-temporal Gaussian process framework to achieve better prediction by borrowing strength from spatial and temporal correlations in neuroimaging data. Our proposed model uses a functional principal component model (FPCA) to capture the medium-to-large scale (or global) spatio-temporal dependence while using a parametric space-time covariance model to capture the small scale (or local) spatio-temporal dependence. We develop a three-stage estimation procedure, along with a predictive method to employ a kriging technique. We illustrate the proposed method using a simulated data set and application to positron emission tomography data from a longitudinal study of Alzheimer's disease.