Data-analytical stability of cluster-wise and peak-wise inference in fMRI data analysis

Area of Science:

• Neuroimaging
• Neuroscience
• Brain Imaging

Background:

• Methodological variability in functional Magnetic Resonance Imaging (fMRI) studies hinders reproducible research.
• Existing evaluations often focus on validity and reliability, overlooking data-analytical stability.

Purpose of the Study:

• To introduce and evaluate data-analytical stability as a key metric for assessing fMRI methods.
• To compare the data-analytical stability of topological feature-based inference (random field theory vs. permutation testing) and smoothing (Gaussian vs. adaptive) approaches.

Main Methods:

• Assessed data-analytical stability, a proxy for reproducibility, in real fMRI data.
• Compared random field theory and permutation-based testing for inference.
• Compared Gaussian and adaptive smoothing methods.
• Evaluated validity and reliability through simulations.

Main Results:

• Methodological choices demonstrably impact the validity, reliability, and stability of fMRI results.
• Permutation-based inference methods yielded the most valid results.
• Adaptive smoothing, while reliable, can reduce stability with larger kernel widths, particularly with cluster-based permutation inference.

Conclusions:

• Data-analytical stability is a crucial, easily integrated criterion for fMRI method evaluation protocols.
• Incorporating stability measures enhances the distinguishability and performance assessment of different fMRI analysis techniques.