Bayesian Spatial Tensor Regression for Neuroimages

• Activity Number: 514 - Recent Advances in Imaging Statistics: Bayesian Methods and Beyond
• Type: Contributed
• Date/Time: Wednesday, August 2, 2017 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Hossein Rekabdarkolaee* and Montserrat Fuentes
• Companies: and Virginia Commonwealth University
• Keywords: Bayesian inference ; Spatial statistics ; Tensor regression ; Neuroimage

Abstract

Data in the form of a multidimensional array, often refered to as tensor data, are used in neuroimaging and other big data applications. We proposed an innovative Bayesian approach to a neuroimaging regression problem with a tensor response against a vector predictor. We adopt tensor images across different groups of interest after adjusting for additional covariates and genetic information, which is of central interest in neuroimaging analysis. We propose a multivariate spatial version of a horseshoe prior to select informative genetic variables. This novel approach provides estimates for the parameters of interest by adopting a generalized sparsity principle while still being able to capture the spatial correlation among the voxels. We study the posterior consistency and develop an efficient Markov chain Monte Carlo algorithm for posterior computation. We implement this method to identify geneotypes linked to cocaine addiction.