Multi-Metric Approach for the Comparison of Denoising Techniques for Resting-State fMRI

Area of Science:

• Neuroimaging
• Neuroscience
• Medical Imaging Analysis

Background:

• Resting-state functional magnetic resonance imaging (rs-fMRI) is crucial for studying brain functional interactions.
• Robust rs-fMRI results are often limited by data quality and suboptimal denoising methods.
• Standardization of rs-fMRI preprocessing and denoising is needed for reliable findings.

Purpose of the Study:

• To establish an effective denoising strategy for rs-fMRI data.
• To quantitatively compare multiple denoising pipelines using the HALFpipe software.
• To identify a denoising approach that enhances reproducibility in rs-fMRI studies.

Main Methods:

• rs-fMRI data from 53 participants and synthetic data were analyzed.
• Nine distinct denoising pipelines were applied to minimally preprocessed fMRI data.
• Performance was evaluated using artifact removal, signal enhancement, and network identifiability metrics, alongside a novel summary performance index.

Main Results:

• Significant heterogeneity was observed in the performance of different denoising pipelines.
• The optimal denoising strategy involved regressing mean signals from white matter, cerebrospinal fluid, and the global signal.
• This preferred pipeline demonstrated the best trade-off between noise reduction and preservation of resting-state network information.

Conclusions:

• The study provides valuable methodological insights into rs-fMRI denoising.
• A recommended denoising pipeline was identified, balancing artifact removal and information preservation.
• This identified pipeline can improve the consistency and reproducibility of rs-fMRI research findings.