Improving power in functional magnetic resonance imaging by moving beyond cluster-level inference

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Brain Network Analysis

Background:

• Neuroimaging inference traditionally focuses on focal brain areas or circuits.
• Emerging research indicates widespread brain effects, suggesting focal findings may be incomplete.
• The impact of focal versus broad-scale inference on study outcomes requires empirical evaluation.

Purpose of the Study:

• To compare the sensitivity and specificity of different inference levels in neuroimaging.
• To evaluate how broad-scale versus focal analysis impacts statistical power.
• To assess the utility of false discovery rate (FDR) control versus familywise error rate (FWER) control.

Main Methods:

• Empirical benchmarking using task-based connectomes from the Human Connectome Project dataset.
• Resampling of data across approximately 1,000 subjects and 7 tasks.
• Comparison of 7 inferential procedures at varying resampling group sizes.

Main Results:

• Broad-scale (network and whole-brain) procedures achieved 80% statistical power, exceeding focal (edge and cluster) procedures by over 20%.
• False discovery rate (FDR) control significantly increased statistical power compared to familywise error rate (FWER) control.
• Loss in specificity for broad-scale and FDR procedures was modest relative to the substantial gains in power.

Conclusions:

• Broad-scale inference methods offer a powerful and practical approach to neuroimaging analysis.
• Shifting to broad-scale analysis and employing FDR control can effectively address statistical power limitations in the field.
• The introduced broad-scale methods are simple, fast, and provide a foundation for more advanced techniques in functional connectivity and related areas.