A parametric approach to voxel-based meta-analysis

Abstract

A key aim of meta-analysis in functional neuroimaging is to quantify the consistency of regional cerebral responses across studies. We derive here a parametric approach to voxel-based meta-analysis, based on spatial statistics. In this method, the value of each voxel in the meta-analysis summary map reflects the proportion of studies reporting an activation within a specified local neighborhood. We threshold this map by testing whether voxel scores could have been expected had the activation peaks been generated at random locations. Our aim is to detect ‘signal’ regions, with scores that are unlikely to arise under such null hypothesis. The method is applicable to both fixed-effects (in which each study is considered as deriving from the same generating process) and random-effects (in which the process varies across studies) meta-analysis. Simulations show strict false positive control, and this approach leads to increased power and substantial gains in computational time relative to existing simulation-based alternatives. We illustrate the technique by performing a random-effects meta-analysis of word production. Parametric voxel-based meta-analysis provides a powerful and practical tool for neuroimaging meta-analysis.