False discovery rate control in magnetic resonance imaging studies via Markov random fields

Area of Science:

• Neuroimaging
• Statistical genetics
• Computational neuroscience

Background:

• Magnetic resonance imaging (MRI) studies population effects on brain morphometry, involving millions of statistical hypotheses.
• High-dimensional inference leads to numerous false positives, commonly managed by controlling the false discovery rate (FDR).
• Existing FDR methods ignore spatial relationships between MRI voxels, limiting their accuracy.

Purpose of the Study:

• To develop novel methods for controlling FDR that explicitly incorporate spatial dependencies between MRI voxels.
• To improve the statistical rigor of neuroimaging studies by accounting for contextual information.

Main Methods:

• Utilized Markov random fields to model spatial dependencies between statistical hypotheses in MRI data.
• Employed maximum pseudo-likelihood estimation and the pseudo-likelihood information criterion for relationship estimation.
• Simulated dependent hypothesis scenarios to evaluate method performance.

Main Results:

• The proposed methods demonstrate desirable statistical properties for FDR control.
• The novel approach outperformed non-contextual methods in simulations of spatially dependent hypotheses.
• Applied to the PATH study, the method identified age-related effects on brain morphometry at whole-brain and component levels.

Conclusions:

• Incorporating spatial dependencies via Markov random fields enhances FDR control in neuroimaging.
• The developed method offers a more accurate approach to analyzing large-scale neuroimaging data.
• Findings from the PATH study align with existing literature on aging and brain morphometry.