Permutation-based true discovery proportions for functional magnetic resonance imaging cluster analysis

Area of Science:

• Neuroscience
• Statistics
• Medical Imaging

Background:

• Simultaneous hypothesis testing is crucial in neuroimaging to manage the high dimensionality of data.
• Functional Magnetic Resonance Imaging (fMRI) cluster analysis faces challenges with spatial specificity and multiple testing.
• Existing methods may lack power or robust confidence statements for data-driven cluster selection.

Purpose of the Study:

• To develop a permutation-based method for simultaneous hypothesis testing.
• To provide simultaneously valid lower bounds on true discoveries in selected hypothesis subsets.
• To offer a confidence statement on the percentage of truly activated voxels in fMRI clusters, addressing the spatial specificity paradox.

Main Methods:

• A novel permutation-based approach for testing large hypothesis collections.
• Calculation of lower bounds for true discoveries with high confidence validity.
• Development of a user-friendly tool for estimating true discovery percentages in fMRI clusters.
• Adaptation to the spatial correlation structure inherent in fMRI data.

Main Results:

• The proposed method yields simultaneously valid lower bounds for true discoveries.
• It provides a confidence statement on the percentage of truly activated voxels within selected fMRI clusters.
• The method effectively controls the family-wise error rate for multiple testing.
• It demonstrates increased power compared to parametric approaches by adapting to spatial correlations.

Conclusions:

• The permutation-based method offers a robust solution for simultaneous hypothesis testing in complex datasets like fMRI.
• It enhances the reliability of findings in fMRI cluster analysis by providing confidence in true discovery rates.
• The user-friendly tool facilitates accurate interpretation of brain activation patterns while controlling for statistical errors.