Estimating sample size in functional MRI (fMRI) neuroimaging studies: statistical power analyses.

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Statistical Modeling

Background:

• Statistical power is crucial for functional MRI (fMRI) study design.
• Estimating power requires understanding percent signal change and variability (intra- and inter-subject).

Purpose of the Study:

• To estimate percent signal change and variability from fMRI data.
• To use these estimates in simulations to generate power curves.
• To determine optimal subject numbers and time points for fMRI cognitive function studies.

Main Methods:

• Estimated intra-subject variability from resting-state fMRI data.
• Estimated inter-subject variability and percent signal change from verbal working memory fMRI data.
• Conducted simulation experiments to generate power curves based on estimated parameters.

Main Results:

• Simulations demonstrated the impact of signal change, variability, and time points on statistical power.
• An empirical test showed good agreement between simulation-based power predictions and observed power.
• Approximately 12 subjects are needed for 80% power at the single voxel level with a liberal threshold (p=0.05).

Conclusions:

• Power is influenced by signal change, intra- and inter-subject variability, and the number of time points.
• More realistic thresholds (corrected for multiple comparisons) necessitate doubling the subject count for equivalent power.
• Findings provide guidance for optimizing fMRI study design for cognitive function research.