1. fMRIPrep-Infants: Bringing standardized, easy-to-use, transparent preprocessing to the infant MRI space

Mathias Goncalves Presenter
Stanford University
Stanford, CA 
United States
Monday, Jun 24: 3:15 PM - 4:30 PM
Room: Grand Ballroom 101-102 
Functional magnetic resonance imaging (fMRI) preprocessing encompasses multiple steps aimed at cleansing and standardizing data prior to statistical analysis. Typically, researchers are faced with a few choices: create a customized workflow tailormade for their scanner acquisition protocol, or reuse an existing tool, many of which place the burden of dependency installation compatibility on the user. Furthermore, large, distributed data collection initiatives, such as the ABCD and HBCD studies, require reproducible and standardized data processing to ensure results are comparable across sites.

From this need came fMRIPrep, a preprocessing pipeline designed to provide an easily accessible, state-of-the-art interface that is robust across protocols and requires minimal user input, while providing easily interpretable and comprehensive output reporting. fMRIPrep established a model for developing standardized neuroimaging workflows that build on well-tested, existing tools. However, it was built for and its performance assessed on human adult MRIs, and does not account for the stark structural differences nor variance in tissue contrast across ages due to ongoing myelination. Because of this, we developed fMRIPrep-Infants, an adaptation of fMRIPrep across the entire infant spectrum, from newborns to two-year-olds.

Just as fMRIPrep revolutionized preprocessing for adult brains, fMRIPrep-Infants replicates its robust methodology, shares the 'glass-box philosophy' of scientific transparency, and minimizes misoperation with an easy-to-use interface. By adhering to the Brain Imaging Data Structure (BIDS), all available scans and necessary metadata can be easily collected, and used to create an adaptive workflow. Although fMRIPrep-Infants leverages much of the same underlying methodology as fMRIPrep, a few important deviations are taken to enhance performance.

Brain Extraction
fMRIPrep-Infants uses an atlas-based registration method for anatomical brain extraction, inspired by antsBrainExtraction. However, the infant’s T2-weighted (T2w) image is preferred for normalization to the target atlas over the T1-weighted (T1w) image, due to the increase in tissue contrast, most notably in the 0-8 month range. The default template used is the UNC 0-1-2 Infant atlas, which provides three atlases of distinct time points in infant development. By utilizing the participant’s age available via BIDS, a best match time point is selected as the target template. To ensure a consistent output space, the T2w image is coregistered to T1w space, as well as the brain mask.

Surface Reconstruction
Signal intensities in tissue exhibit variations between neonatal and adult brains. Consequently, finding the necessary contrast between gray and white matter, crucial for identifying cortical surfaces, varies depending on age. fMRIPrep-Infants provides three alternative methods to reconstruct cortical surfaces, each performant in a stage of development. With its utilization of the T2w over the T1w, Melbourne Children’s Regional Infant Brain Surface (M-CRIB-S) voxel-based parcellation excels in the early months (0-8). As the infant brain matures and myelination completes, T1w approaches become effective - Infant FreeSurfer can be used around 9-24 months. Upwards of 24 months, we have found FreeSurfer’s recon-all to perform reasonably well. These ranges are only recommendations; users can override and select the method they most prefer if desired.

Subcortical Structures Alignment
The compact overall anatomy of the infant brain generally results in a decreased signal-to-noise ratio (SNR) and an increased partial voluming, relative to an adult brain. These effects become evident during segmentation of subcortical brain areas, where a large amount of brain structures closely border each other in low resolution. Even if the segmentation is expert-validated, transforming these into a shared template space may result in misalignment. To remedy this, fMRIPrep-Infants incorporates a structure-by-structure alignment process, to protect from potential structure-specific distortions.

Offering maximal flexibility, fMRIPrep-Infants provides a containerized, easy-to-use interface to generate results that are minimally tied to a specific analysis, allowing seamless integration with other downstream tools or workflows. As part of the NiPreps community initiative, this open-source project is crafted with thorough documentation and actively encouraging contributions from the community. Its methods are continuously under evaluation to incorporate the latest advances in the field, and hopes to spark a community adoption similar to that of fMRIPrep’s.

Abbreviations: ABCD - Adolescent Brain Cognitive Development; HBCD - HEALthy Brain and Child Development