Voxel-Wise or Region-Wise Nuisance Regression for Functional Connectivity Analyses: Does It Matter?

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Functional Neuroimaging

Background:

• Resting-state functional connectivity (rs-FC) analysis requires removing nuisance signals from BOLD time series.
• Voxel-level denoising is standard but computationally intensive.
• Region-level denoising offers potential computational efficiency but its impact on rs-FC analysis is not well understood.

Purpose of the Study:

• To systematically compare the effects of denoising at voxel-level versus region-level on rs-FC.
• To investigate the influence of aggregation methods (Mean, EV) and parcellation granularity on denoising outcomes.
• To evaluate the impact of denoising resolution on individual specificity and prediction of age and cognitive scores.

Main Methods:

• Analysis of 370 unrelated subjects from the HCP-YA dataset.
• Comparison of voxel-level and region-level denoising strategies.
• Evaluation using Mean and first eigenvariate (EV) aggregation methods across 100, 400, and 1000 regions.
• Assessment of individual specificity (fingerprinting) and prediction of age and cognitive scores.

Main Results:

• Region-level denoising performed equally or better than voxel-level denoising across various metrics.
• Mean aggregation resulted in comparable individual specificity and prediction performance for both denoising resolutions.
• EV aggregation showed reduced individual specificity for voxel-level denoising compared to region-level denoising.
• Increased parcellation granularity generally enhanced individual specificity.

Conclusions:

• Region-level denoising is a viable and computationally efficient alternative to voxel-level denoising for rs-FC analysis.
• The choice of aggregation method significantly influences the comparison between voxel-level and region-level denoising.
• Region-level denoising with Mean aggregation is recommended for brain-behavior studies due to its efficiency and comparable performance.