Revisiting multi-subject random effects in fMRI: advocating prevalence estimation.

Area of Science:

• Neuroimaging
• Statistical analysis
• Cognitive neuroscience

Background:

• Random effect analysis in fMRI aims to generalize findings to the population.
• Classic random effect models assume normally distributed effects and a non-null mean, which may not reflect true population variability.
• Inter-subject variability due to functional and anatomical differences poses challenges for fMRI generalization.

Purpose of the Study:

• To propose a more realistic statistical model for fMRI data that accounts for between-subject spatial disagreement.
• To introduce a Gaussian-mixture-random-effect model to quantify activation prevalence at each brain location.
• To provide a formal definition and estimation procedure for activation prevalence.

Main Methods:

• Developed a Gaussian-mixture-random-effect model to handle spatial inconsistencies across subjects.
• Defined and established an estimation procedure for activation prevalence.
• Applied the model to fMRI data to generate prevalence maps.

Main Results:

• The proposed model quantifies activation prevalence, reflecting the proportion of subjects showing activation at a given location.
• Prevalence maps highlight brain regions with consistent activation across many individuals.
• The method provides more informative results compared to traditional active/inactive paradigms and p-value displays.

Conclusions:

• The Gaussian-mixture-random-effect model offers a more realistic approach to fMRI analysis by incorporating population variability.
• Prevalence estimation is a valuable metric, providing a more nuanced understanding of brain activation patterns.
• This method enhances the generalizability and interpretability of fMRI findings.