Title: A Bayesian approach to the joint analysis of multi-type image-based and coordinate-based neuroimaging meta-analysis data
Authors: Silvia Montagna - University of Turin (Italy) [presenting]
Thomas Nichols - University of Oxford (United Kingdom)
Timothy Johnson - University of Michigan (United States)
Abstract: As the popularity of functional MRI (fMRI) has grown exponentially over the years, so does the need to aggregate and summarise different fMRI studies via meta-analysis. Neuroimaging meta-analysis is used to 1) identify areas of consistent activation; and 2) build a predictive model of task type or cognitive process for new studies (reverse inference). Currently, two types of meta-analyses are possible. Namely, coordinate-based meta-analyses (CBMA); when data from different studies are available only as peak activation coordinates (foci) in a three dimensional coordinate system. And, image-based meta-analyses (IBMA); when the statistical parametric maps resulting from a group-level analysis (voxel-level data) are shared. We propose a Bayesian hierarchical model for neuroimaging meta-analysis which allows for the joint modelling of CBMA and IBMA data, whilst simultaneously addressing the two aims above. Specifically, we build a spatial process for voxel-level (IBMA) data and a spatial point process model for point-pattern foci-based (CBMA) data, then combine the two component models via a latent factor framework that allows for the borrowing of information across the different studies. We apply our methodology to a neuroimaging meta-analysis dataset of pain and emotions studies.