A Bayesian General Linear Modeling Approach to Cortical Surface fMRI Data Analysis.

Area of Science:

• Neuroimaging
• Statistical modeling
• Computational neuroscience

Background:

• Cortical surface fMRI (cs-fMRI) offers advantages over volumetric fMRI, including better visualization and alignment.
• Current cs-fMRI analyses often use the classical general linear model (GLM), a massive univariate approach, lacking spatial Bayesian modeling.
• Existing spatial Bayesian models are not optimized for cs-fMRI data.

Purpose of the Study:

• To propose a novel spatial Bayesian GLM specifically designed for cs-fMRI data.
• To introduce advanced computational techniques and a multi-subject modeling framework for cs-fMRI analysis.
• To enhance the identification of brain activation regions in cs-fMRI studies.

Main Methods:

• Developed a spatial Bayesian GLM utilizing sophisticated spatial processes to model latent activation fields.
• Employed integrated nested Laplacian approximations (INLA) for efficient Bayesian computation.
• Utilized an excursions set method based on joint posterior distributions for activation identification and introduced the first multi-subject spatial Bayesian model for cs-fMRI.

Main Results:

• The proposed spatial Bayesian GLM for cs-fMRI is computationally advantageous.
• The model demonstrated effectiveness in simulation studies.
• Validated through two task fMRI studies from the Human Connectome Project, showing robust performance.

Conclusions:

• The developed spatial Bayesian GLM provides a powerful and efficient new tool for cs-fMRI data analysis.
• This work addresses a significant gap by offering the first multi-subject spatial Bayesian modeling approach for cs-fMRI.
• The findings pave the way for more sophisticated analyses of brain activity using cs-fMRI.