A Bayesian High-Dimensional Poisson Graphical Model for Identifying Functional Coactivation Patterns

• Activity Number: 604
• Type: Contributed
• Date/Time: Wednesday, August 12, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Caprichia Jeffers* and Jian Kang
• Companies: Emory University Rollins School of Public Health and Emory University
• Keywords: Poisson graphical model ; graphical model for count data ; identifying co-activation pattern ; functional neuroimaging meta-analysis

Abstract

Meta-analysis of functional neuroimaging data has become increasingly important recently. Much attention has been paid to detect consistent activation regions or locations across independently performed studies, while very limited works have been focusing on co-activation pattern identifications. To fill this gap, a Poisson graphical model (Xue et al, 2014) was proposed and the penalized likelihood approaches along with EM algorithms have been developed to make model inference. However, this method is not applicable for high-dimensional problems due to challenging computational difficulties. To mitigate this problem, we propose a Bayesian Poisson graphical model for which we introduce a new prior model for the intensity parameters in a multivariate Poisson distribution. We develop efficient posterior computational algorithms that are scalable for a high dimensional graphical model. We illustrate our methods via extensive simulation studies and a meta-analysis of functional neuroimaging data for emotion studies.