The empirical replicability of task-based fMRI as a function of sample size

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Psychology

Background:

• Replicability is crucial for scientific validity, especially in neuroimaging.
• Functional magnetic resonance imaging (fMRI) is a key tool, but its results' replicability is sensitive to study design.
• The impact of sample size on task-based fMRI replicability requires further empirical investigation.

Purpose of the Study:

• To empirically assess the relationship between sample size and replicability in task-based fMRI.
• To investigate how different measures of replicability are affected by sample size.
• To explore the interplay between spatial resolution and replicability in fMRI studies.

Main Methods:

• Utilized a large dataset (1400 subjects) from the IMAGEN project with four distinct tasks.
• Created subsamples of increasing size from the independent dataset to evaluate replicability.
• Assessed replicability using a multi-dimensional framework including test-retest reliability, coherence, and stability.

Main Results:

• Demonstrated a positive correlation between sample size and fMRI result replicability.
• Identified a trade-off between spatial resolution and replicability; higher resolution demands larger sample sizes.
• Observed increased replicability when analyzing clusters of voxels compared to individual voxels.
• Found variability in activation cluster sizes across different experimental paradigms.

Conclusions:

• Increasing sample size is essential for enhancing the replicability of task-based fMRI findings.
• fMRI studies need larger sample sizes than typically employed for robust voxel-wise replicability.
• Analyzing results at the cluster level can yield higher replicability, but this varies by experimental design.