Multiple testing correction over contrasts for brain imaging.

Area of Science:

• Neuroimaging
• Statistical analysis
• Brain mapping

Background:

• Multiple testing is a significant challenge in neuroimaging due to high dimensionality (voxels/vertices).
• Testing multiple hypotheses (contrasts) within a general linear model (GLM) necessitates correction for multiplicity.
• Failure to correct for multiple comparisons inflates the rate of false positives, compromising study validity.

Purpose of the Study:

• To address the multiple testing problem in brain imaging.
• To evaluate and compare various statistical correction methods for neuroimaging data.
• To identify the most suitable method for handling multiplicity in common brain imaging scenarios.

Main Methods:

• Discussion and comparison of different statistical correction methods.
• Application of methods to address multiplicity in general linear models for neuroimaging.
• Evaluation of permutation testing as a correction strategy.

Main Results:

• A classical, widely used correction method was found to be invalid for neuroimaging data.
• Permutation testing demonstrated exactness and flexibility across various common imaging situations.
• The study highlights the limitations of existing methods and advocates for permutation-based approaches.

Conclusions:

• Correction for multiple testing is essential in neuroimaging to maintain statistical rigor.
• Permutation testing is recommended as the optimal method for addressing multiplicity in brain imaging.
• The findings provide guidance for researchers to improve the reliability of neuroimaging statistical analyses.