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

• Published in: Human brain mapping Vol. 46; no. 7; pp. e70080 - n/a
• Main Authors: Goffi, Federica, Bianchi, Anna Maria, Schiena, Giandomenico, Brambilla, Paolo, Maggioni, Eleonora
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 01.05.2025
• Subjects: Adult; benchmarking; Brain - diagnostic imaging; Brain - physiology; Connectome - methods; Connectome - standards; denoising pipeline; Female; functional magnetic resonance imaging; Humans; Image Processing, Computer-Assisted - methods; Image Processing, Computer-Assisted - standards; Magnetic Resonance Imaging - methods; Magnetic Resonance Imaging - standards; Male; Nerve Net - diagnostic imaging; Nerve Net - physiology; preprocessing; resting state; Young Adult; resting state; benchmarking; preprocessing; functional magnetic resonance imaging; denoising pipeline
• ISSN: 1065-9471; 1097-0193; 1097-0193
• DOI: 10.1002/hbm.70080

ABSTRACT Despite the increasing use of resting‐state functional magnetic resonance imaging (rs‐fMRI) data for studying the spontaneous functional interactions within the brain, the achievement of robust results is often hampered by insufficient data quality and by poor knowledge of the most effective denoising methods. The present study aims to define an appropriate denoising strategy for rs‐fMRI data by proposing a robust framework for the quantitative and comprehensive comparison of the performance of multiple pipelines made available by the newly proposed HALFpipe software. This will ultimately contribute to standardizing rs‐fMRI preprocessing and denoising steps. Fifty‐three participants took part in the study by undergoing a rs‐fMRI session. Synthetic rs‐fMRI data from one subject were also generated. Nine different denoising pipelines were applied in parallel to the minimally preprocessed fMRI data. The comparison was conducted by computing previously proposed and novel metrics that quantify the degree of artifact removal, signal enhancement, and resting‐state network identifiability. A summary performance index, accounting for both noise removal and information preservation, was proposed. The results confirm the performance heterogeneity of different denoising pipelines across the different quality metrics. In both real and synthetic data, the summary performance index favored the denoising strategy including the regression of mean signals from white matter and cerebrospinal fluid brain areas and global signal. This pipeline resulted in the best compromise between artifact removal and preservation of the information on resting‐state networks. Our study provided useful methodological tools and key information on the effectiveness of multiple denoising strategies for rs‐fMRI data. Besides providing a robust comparison approach that could be adapted to other fMRI studies, a suitable denoising pipeline for rs‐fMRI data was identified, which could be used to improve the reproducibility of rs‐fMRI findings. The present study aims at defining an appropriate denoising strategy for resting‐state fMRI data by quantitatively comparing the performance of multiple pipelines in terms of both artifact removal and signal preservation, including resting‐state network identifiability. This will ultimately contribute to standardizing fMRI denoising steps, reducing analytic flexibility, and improving reproducibility.