Bayesian inference for group-level cortical surface image-on-scalar regression with Gaussian process priors.

Area of Science:

• Neuroimaging
• Statistical Modeling
• Brain Research

Background:

• Current neuroimaging analyses often use marginal general linear models with spatial smoothing, which can lead to poorly calibrated inference.
• Standard spatial modeling approaches are computationally intensive for large neuroimage datasets.

Purpose of the Study:

• To develop a computationally tractable Bayesian spatial regression model for group-level neuroimaging analyses.
• To improve the accuracy and adaptability of spatial inference in neuroimage data.

Main Methods:

• A Bayesian spatial regression model utilizing Gaussian process priors for regularization of spatially varying coefficients.
• Incorporation of a non-stationary error process model for data-adaptive smoothing.
• Application of a Vecchia-type approximation for computational tractability while preserving spatial rank.

Main Results:

• The proposed model provides more data-adaptive smoothing than standard methods.
• The Vecchia approximation enables efficient computation for large-scale neuroimaging data.
• Performance comparisons demonstrate advantages over standard vertex-wise analyses.

Conclusions:

• The Bayesian spatial regression model offers a powerful and computationally feasible alternative for group-level neuroimaging studies.
• This approach enhances the reliability and precision of statistical inference in neuroimage analysis.
• The methods are illustrated using functional magnetic resonance imaging data from the Adolescent Brain Cognitive Development Study.