Spatial Bayesian latent factor regression modeling of coordinate-based meta-analysis data.

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Statistical Modeling

Background:

• Functional MRI (fMRI) has a vast literature, necessitating meta-analytic synthesis.
• Coordinate-Based Meta-Analysis (CBMA) is crucial due to limited sharing of fMRI image data, relying instead on peak activation coordinates (foci).
• Current neuroimaging meta-analysis aims to identify consistent activation areas and enable predictive modeling for new studies (reverse inference).

Purpose of the Study:

• To propose a novel Bayesian point process hierarchical model for CBMA.
• To simultaneously identify consistent activation areas and build predictive models for cognitive processes.
• To expand the capabilities of existing neuroimaging meta-analysis methods.

Main Methods:

• Modeling study foci as a doubly stochastic Poisson process.
• Utilizing a high-dimensional basis set for the study-specific log intensity function.
• Employing latent factor modeling for sparse representation of intensities and incorporating meta-regression for study-level covariates.

Main Results:

• The proposed Bayesian model effectively synthesizes fMRI data using peak activation coordinates.
• The methodology allows for both identification of consistent activation and predictive modeling.
• The framework successfully incorporates study-level covariates, enhancing meta-analytic capabilities.

Conclusions:

• The developed Bayesian point process hierarchical model offers a powerful new tool for coordinate-based meta-analysis in neuroimaging.
• This approach advances the ability to synthesize fMRI findings and perform reverse inference.
• The model's flexibility with meta-regression broadens its applicability in cognitive neuroscience research.