CBMR: Meta Regression and Inference for Coordinate Based Meta Analysis Data Across Multiple Groups
Poster No:
1557
Submission Type:
Abstract Submission
Authors:
Yifan Yu1, Lauren Hill-Bowen2, Michael Riedel3, Katherine Bottenhorn4, Angela Laird5, Thomas Nichols6
Institutions:
1Oxford Big Data Institute, University of Oxford, Old road campus, Oxford, Oxfordshire, 2Department of Psychiatry and Behavioral Sciences, Vanderbilt University Medical Center, Nashville, TN, 3Department of Physics, Florida International University, Miami, FL, 4Keck School of Medicine of USC, Los Angeles, CA, 5Florida International University, Miami, FL, 6University of Oxford, Oxford, Oxfordshire
First Author:
Co-Author(s):
Lauren Hill-Bowen
Department of Psychiatry and Behavioral Sciences, Vanderbilt University Medical Center
Nashville, TN
Department of Psychiatry and Behavioral Sciences, Vanderbilt University Medical Center
Nashville, TN
Introduction:
Meta-analysis synthesizes studies to evaluate consistency and heterogeneity, improving the precision of inference about brain activations. Existing coordinate-based meta-analysis (CBMA) methods often lack interpretability, fail to accommodate study-level covariates, cannot model spatial dependence in the data and are limited to compare imbalanced groups. While Bayesian model-based CBMA offers greater accuracy, interpretability and explicit spatial modelling, they are more computationally expensive. We have developed a coordinate-based meta-regression and inference (CBMR) framework, a Python-based tool that jointly models all voxels while controlling for study-level covariates (e.g. sample size, publication year). Here, we describe new developments for flexible group-wise comparison.
Methods:
Our CBMR is based on generalised linear model (GLM) with log link function, where the log intensity is modelled spatially using 3D B-spline bases, along with (standardised) study-level covariates. We developed model factorisation to replace the full dataset with sufficient statistics, ensuring dimensionality does not exceed the number of studies or foci in each group. We consider two models for random variation: the Poisson model; the Negative Binomial (NB) that allows for excess variation, and we further simplify their log-likelihood functions using either additive property of Poisson or moment matching approach. Regression coefficients are optimised via the L-BFGS algorithm. CBMR allows both voxelwise tests of spatial homogeneity and tests for equality of group-wise intensities: for datasets with at least 200 foci per group, we conduct Wald test at the voxel level; for datasets with insufficient foci, we apply a parametric bootstrap method instead.
Results:
Extensive simulation experiments show CBMR inference using parametric bootstrap method effectively controls false positives for both voxel-wise tests of spatial homogeneity and group comparisons. In contrast, p-values from Wald tests exhibit inflation of false positives (not shown). We applied CBMR to a Cue Reactivity dataset consisting of 275 studies and 3197 foci, categorised into three groups. Using a spatial model with spline parameterization (10mm knot spacing, 2624 basis elements), we compared activation regions identified with meta-regression with a Poisson model and inference based on either the Wald test or parametric bootstrap against ALE. Figure 1 presents activation maps for tests of the null hypothesis of spatial homogeneity (uncorrected Z-score, thresholded 5% FDR) for the drug group and natural group. Activation areas are identified in the left cerebral cortex, frontal orbital cortex, insular cortex, and left and right accumbens. We observed ALE provided the smoothest activation regions, while parametric bootstrap method detected more voxels with significant p-values, and Wald test identified the most stringent and localised activation regions. Figure 2 presents group comparison maps testing the equality of activation between the drug group and natural group. The strongest differences occurred in cingulate gyrus, superior frontal gyrus, precuneus cortex, left thalamus and occipital pole. Model comparison via Likelihood ratio test indicates NB model is preferred over Poisson (p<10^-8). Study-level effects showed publication year was non-significant (p=0.6681), while sample size had a significant effect (p<10^-8).
Conclusions:
We have presented a multi-group CBMR, implemented as a module in the open-source Python package NiMARE. It has two stages: meta-regression employs a spatial model based on spline parameterisation with a roughness penalty; while meta-inference fits a GLM at the voxel level using either a Poisson or NB distribution and accommodates study-level covariates, through the Wald test or parametric bootstrap method. We demonstrated it detects spatial clustering of activation regions, similar to existing CBMA methods, and allows for group comparisons and inference on study-level covariates.
Modeling and Analysis Methods:
Activation (eg. BOLD task-fMRI)
Methods Development 1
Multivariate Approaches 2
Keywords:
Cognition
Meta- Analysis
Modeling
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
Provide references using APA citation style.
Bickel, P. J. (1981). Some asymptotic theory for the bootstrap. The annals of statistics, 9(6), 1196-1217.
Eickhoff, S. B. (2012). Activation likelihood estimation meta-analysis revisited. Neuroimage, 59(3), 2349-2361.
Geoffroy, P. (2001). A Poisson-Gamma model for two-stage cluster sampling data. Journal of Statistical Computation and Simulation, 68(2), 161-172.
Hall, P. (2013). The bootstrap and Edgeworth expansion. Springer Science & Business Media.
Hill-Bowen, L. D. (2021). The cue-reactivity paradigm: An ensemble of networks driving attention and cognition when viewing drug and natural reward-related stimuli. Neuroscience & Biobehavioral Reviews, 130, 201-213.
Lawless, J. F. (1987). Negative binomial and mixed Poisson regression. The Canadian Journal of Statistics/La Revue Canadienne de Statistique, 209-225.
Salo, T., Yarkoni, T. (2022). NiMARE: neuroimaging meta-analysis research environment. NeuroLibre, 1(1), 7.
Samartsidis, P. (2019). Bayesian log-Gaussian Cox process regression: applications to meta-analysis of neuroimaging working memory studies. Journal of the Royal Statistical Society Series C: Applied Statistics, 68(1), 217-234.
Yu, Y. (2024). Neuroimaging meta regression for coordinate based meta analysis data with a spatial model. Biostatistics, 25(4), 1210-1232.