Benchmarking fMRI Denoising Pipelines
Tianye Zhai1, Hong Gu1, Anika Holton1
1Neuroimaging Research Branch, Intramural Research Program, National Institute on Drug Abuse, National Institutes of Health, Baltimore, Maryland, USA.
Human Brain Mapping
|July 1, 2026
Summary
Denoising functional magnetic resonance imaging (fMRI) data requires careful pipeline selection. Combining physiological nuisance regressors, especially temporally shifted ones, significantly improves fMRI data quality for better neuronal activity analysis.
Area of Science:
- Neuroimaging
- Data Science
Background:
- Functional magnetic resonance imaging (fMRI) is crucial for neuroscience research but suffers from inherent noise.
- Existing denoising strategies (e.g., aCompCor, ICA-based methods, multi-echo approaches) are often evaluated on limited datasets and may not incorporate recent advancements.
Purpose of the Study:
- To benchmark a comprehensive range of fMRI denoising pipelines incorporating recent methodological advances.
- To evaluate pipeline performance across diverse fMRI data types (task/resting-state, single/multi-band, single/multi-echo).
Main Methods:
- Developed a framework to test various denoising pipelines, including order-independent regression, pre-whitening, and temporal shifting of nuisance regressors.
- Evaluated pipelines using metrics like temporal signal-to-noise ratio (tSNR), remaining degrees-of-freedom (DoF), motion correction effectiveness, and signal preservation.
- Tested pipelines on multiple fMRI datasets with varying combinations of confounds.
Main Results:
- Pipelines relying solely on Independent Component Analysis (ICA) were insufficient for effective denoising.
- Pipelines incorporating physiological nuisance regressors demonstrated strong performance.
- Further performance enhancements were achieved by accounting for temporally shifted physiological regressors.
Conclusions:
- Recommends specific denoising pipeline strategies based on empirical benchmarking.
- Highlights the importance of continuous evaluation and adaptation of denoising methods as new techniques emerge and are applied to novel contexts.