Adjusting for Variable Brain Coverage in Voxel-Based fMRI Meta-Analysis

Abstract

Meta-analyses of fMRI studies are vital to establish consistent findings across the literature. However, fMRI data are susceptible to signal dropout (i.e. incomplete brain coverage), which varies across studies and brain regions. In other words, for some brain regions, only a variable subset of the studies included in an fMRI meta-analysis have data present. These missing data can mean activations in fMRI meta-analysis are underestimated (type II errors). Here we present SPM (MATLAB) code to run a novel method of adjusting random-effects models for meta-analytic averaging of a group of studies and mixed-effects models for comparison between two groups of studies. In two separate datasets, meta-analytic effect sizes and z-scores were larger in the adjusted, compared to the unadjusted analysis. Relevantly, these changes were in regions such as the ventromedial prefrontal cortex where coverage was lowest. Limitations of the method, including issues of how to threshold the adjusted maps are discussed. Code and demonstration data for the adjusted method are available at https://doi.org/10.25377/sussex.c.4223411.