Reducing type II error in fMRI analysis: cluster-extent threshold simulation results and an evaluation of current...

Area of Science:

• Neuroimaging
• Statistical analysis
• Brain imaging analysis

Background:

• Multiple comparison correction in fMRI often uses cluster-extent thresholds.
• Sample size (N) is frequently omitted from these thresholding models.
• This omission may lead to inaccurate statistical inferences in fMRI.

Purpose of the Study:

• To investigate the effect of sample size (N) on cluster-extent thresholds in fMRI.
• To determine if modeling N reduces cluster thresholds.
• To test the role of between-subject variability in observed cluster size reductions.

Main Methods:

• Extensive simulations varied N (10, 20, 30), corrected p-values, voxel p-values, FWHM, and voxel resolution.
• Bayesian analysis was used to assess the contribution of between-subject variability.
• Simulation findings were validated using a real fMRI dataset.

Main Results:

• Modeling sample size (N) resulted in approximately 18% smaller clusters.
• Larger sample sizes led to significantly increased cluster thresholds.
• Bayesian analysis strongly supported the role of between-subject variability.

Conclusions:

• Sample size (N) is a critical parameter that should be incorporated into fMRI analysis methods.
• Including N improves threshold accuracy and helps reduce type II errors.
• The study questions assumptions about non-task fMRI activity reflecting null data.