A Nonparametric Bayesian Spatial Point Process Approach to Neuroimaging Meta Analysis

Abstract

As the discipline of functional neuroimaging grows there is an increasing interest in meta analysis of brain imaging studies. A typical neuroimaging meta analysis collects peak activation coordinates (foci) from several studies and identifies areas of consistent activation. In our illustrative example, our colleagues collected 219 emotion studies consisting of five distinct emotion types: sad, happy, anger, fear, disgust. One psychological theory states that all emotions utilize the same functional brain regions, to varying degrees. Thus, the expected number of foci in these regions, across emotion types, should be correlated. To date, all imaging meta analysis methods have been developed to analyze a single population of studies. Furthermore, most imaging meta analysis methods do not provide an interpretable fitted model. To overcome these limitations, we propose a nonparametric Bayesian spatial point process model that generalizes the Poisson/gamma random field model (Wolpert and Ickstadt, 1998) in a hierarchical fashion. Our model simultaneously fits multi-type point pattern data, accounts for the (positive) correlation across the various types of point patterns and results in an easily interpretable posterior intensity function. Furthermore, our model can be used to accurately predict the emotion type from a new study---something of great interest to our collaborators.