A Spatial Bayesian Approach for Interpretable Modeling of Neuroimaging Meta-Analysis Data

• Activity Number: 56
• Type: Invited
• Date/Time: Sunday, July 29, 2012 : 4:00 PM to 5:50 PM
• Sponsor: IMS
• Author(s): Thomas E. Nichols*+
• Companies: University of Warwick
• Keywords: fMRI ; meta-analysis ; point process

Abstract

Meta-analysis for functional brain imaging is growing in importance as the number of available studies grows. Unfortunately, the only data widely available for meta-analysis are the [x,y,z] coordinates of peak activation foci. Most existing methods convert these coordinate data into images and then apply traditional imaging modelling and inference proceedures. However, these methods are mass-univariate, make inference on individual voxels (or clusters) and require fixed tuning parameters. Such methods do not provide an interpretable fitted model, and cannot produce spatial confidence intervals on location of activation. Using a Bayesian hierarchical point process approach, modelling brain activation as a mixture of latent activation centers, and these activation centers are in turn modelled as off-spring from latent population centers. When our fully Bayesian model fit to multiple groups it is trival to make predictions of the group label for new data and, through an importance sampling trick, leave-one-out cross validation accuracy for the sample studied.

Joint work Jian Kang, Emory Biostatistics, and Tim Johnson, University of Michigan Biostatistics.