Providing Evidence for the Null Hypothesis in Functional Magnetic Resonance Imaging Using Group-Level Bayesian...

Area of Science:

• Neuroimaging
• Statistical Inference
• Brain Activity Analysis

Background:

• Classical null hypothesis significance testing (NHST) is limited to rejecting point-null hypotheses, hindering the interpretation of non-significant findings and introducing bias.
• Bayesian parameter inference (BPI) offers a framework for assessing 'null effects' by considering the probability of effects within a region of practical equivalence.

Purpose of the Study:

• To discuss and demonstrate the advantages of Bayesian parameter inference (BPI) for assessing null effects in functional magnetic resonance imaging (fMRI) data group analysis.
• To compare BPI with classical null hypothesis significance testing (NHST) using both empirical and simulated fMRI data.
• To provide practical guidance and tools for implementing BPI in fMRI studies.

Main Methods:

• Bayesian parameter inference (BPI) was employed, focusing on the posterior probability of effects within a defined region of practical equivalence.
• fMRI data group analysis was performed using both BPI and classical NHST for comparative analysis.
• Simulated data with varying effect sizes, noise levels, distributions, and sample sizes were used to evaluate BPI performance.

Main Results:

• BPI allows for the identification of 'not activated' voxels and provides a single decision rule for inference, unlike NHST.
• BPI facilitates adaptive experimental design by enabling evaluation of data sufficiency as sample size increases, potentially allowing for early stopping.
• Comparative analyses on empirical and simulated data demonstrated the utility and performance of BPI in fMRI group analysis.

Conclusions:

• Bayesian parameter inference (BPI) provides a more comprehensive approach to null effect assessment in fMRI compared to classical NHST.
• BPI enhances the interpretation of neuroimaging results by allowing for evidence of absence of effect, not just absence of evidence.
• A toolbox for Statistical Parametric Mapping 12 was developed to facilitate the adoption of BPI for fMRI practitioners.