Benchmarking common preprocessing strategies in early childhood functional connectivity and intersubject correlation
Kirk Graff1, Ryann Tansey1, Amanda Ip1
1Child and Adolescent Imaging Research Program, University of Calgary, Calgary, AB, Canada; Alberta Children's Hospital Research Institute, University of Calgary, Calgary, AB, Canada; Hotchkiss Brain Institute, University of Calgary, Calgary, AB, Canada; Department of Neuroscience, University of Calgary, Calgary, AB, Canada.
Developmental Cognitive Neuroscience
|February 23, 2022
Summary
This study benchmarks functional magnetic resonance imaging (fMRI) preprocessing for children, finding that censoring and global signal regression (GSR) improve data quality and individual differences. Optimal pipelines balance noise reduction with preserving unique brain connectivity patterns.
Area of Science:
- Neuroimaging
- Developmental Neuroscience
- Data Science
Background:
- Functional magnetic resonance imaging (fMRI) preprocessing is challenging for young children due to high head motion.
- Existing benchmarking studies often use adult data, which may not generalize to pediatric populations.
- Mitigation strategies like censoring and global signal regression (GSR) are debated.
Purpose of the Study:
- To benchmark fMRI preprocessing pipelines in a pediatric cohort (ages 4-8) with significant head motion.
- To compare the efficacy of different preprocessing steps, including GSR, censoring, and ICA-AROMA.
- To evaluate pipelines based on noise reduction, individual differences (connectome fingerprinting), and stimulus-evoked responses (intersubject correlations).
Main Methods:
- Systematic investigation of preprocessing pipelines combining global signal regression (GSR), volume censoring, and ICA-AROMA.
- Utilized a longitudinal, passive viewing fMRI dataset from children aged 4-8 years.
- Compared pipelines using metrics for noise removal, connectome fingerprinting, and intersubject correlations (ISC).
Main Results:
- The most effective pipeline incorporated censoring, GSR, bandpass filtering, and head motion parameter (HMP) regression for noise removal and information recovery.
- ICA-AROMA showed comparable performance to HMP regression and did not eliminate the need for censoring.
- GSR minimally affected connectome fingerprinting but enhanced ISC, while strict censoring reduced motion artifacts but decreased identifiability.
Conclusions:
- A combined approach of censoring, GSR, and HMP regression is recommended for preprocessing pediatric fMRI data with high motion.
- Careful consideration of censoring strictness is necessary to balance motion artifact reduction with preserving individual brain connectivity.
- Findings provide crucial guidance for analyzing pediatric fMRI data, improving reliability and interpretability.
