HBCD Processing Pipelines🔗

The HBCD processing pipelines are a collection of modular tools used to process data from the HBCD study. These pipelines are not exclusive to HBCD, but are general-purpose tools for analyzing a variety of data modalities, including magnetic resonance imaging (MRI), electroencephalography (EEG), magnetic resonance spectroscopy (MRS), and biosensor data. The pipelines are designed to be modular and flexible, allowing for customization to meet the specific needs of the HBCD study.

The following processing pipelines were used for the HBCD study. Also see a more detailed breakdown by modality with links to derivative folder documentation for HBCD here.

Modality	Processing Pipeline	Derivatives	Pipeline Description
MRI	MRIQC	mriqc/	Automated extraction of image quality metrics from structural and functional MRI data
	qMRI Postproc	symri/ qmri_postproc/	Minimal post-processing for SyMRI synthetic images derived from QALAS acquisition
	BIBSNet	bibsnet/	Deep learning model for brain segmentation
	infant-fMRIPrep	nibabies/ freesurfer/ mcribs/	Structural and functional MRI preprocessing pipeline
	XCP-D	xcpd/	Functional MRI post-processing and noise regression pipeline
	QSIPrep and QSIRecon	qsiprep/ qsirecon/ qsirecon-*/	Diffusion-weighted MRI (dMRI) data processing pipelines
MRS	OSPREY-BIDS	osprey/	MRS data processing pipeline
EEG	HBCD-MADE	made/	Maryland Analysis of Developmental EEG (MADE) pipeline adapted for HBCD
Biosensors	HBCD-Motion	hbcd_motion/	Leg movement sensor data processing

Pipeline Standards & Design🔗

All pipelines used for HBCD data processing must adhere to HBCD Processing & Analytic Software Standards, which require, among other things:

• NMIND peer review and DOI publication for reproducibility.
• Compliance with the Brain Imaging Data Structure (BIDS) standard.
• Implementation as BIDS Apps (Gorgolewski et al.,2017), ensuring containerized, standardized processing.

Why BIDS & BIDS Apps?🔗

BIDS is a community-driven standard for organizing neuroimaging and behavioral data, making datasets structured, shareable, and reproducible. BIDS Apps are containerized applications that run on any system supporting Docker or Apptainer (Singularity).

Benefits of Containerization:
Ensures all software dependencies are included.
Guarantees consistent processing environments across systems.
Simplifies reproducibility and collaboration.

Running an HBCD Processing Pipeline🔗

To process HBCD study data using one of these pipelines, follow the installation and usage instructions on the specific pipeline's documentation page (links below).

Choosing a Container System:
Singularity/Apptainer → Recommended for university-affiliated HPC clusters, where users lack administrative privileges.
Docker → Preferred for personal computers due to its ease of installation (may require extra setup for HPC use).

All processing containers are available on Docker Hub.

File Selection for Processing🔗

With the exception of TB1 MRI and electrocardiogram (ECG) data, raw BIDS files are included in the release only if they were used in at least one processing pipeline, ensuring alignment with derived pipeline outputs. Since HBCD employs multiple pipelines — each with its own requirements — the released data represent the union of all files that meet at least one pipeline’s criteria.

For some data categories, files are selected for processing based on pipeline-specific criteria detailed under Quality Control Selection Information in the Tool Names section of the HBCD Processing website. For imaging data that underwent raw data quality control, only files that pass are included in the data release and utilized for data processing. All quality control information is stored in the sub-{ID}_ses-{V0X}_scans.tsv file located in each BIDS session folder. This file is queried prior to processing to determine which files to include/exclude (e.g. based on thresholds for QU_motion, acq_motion, brain_SNR, etc.).

There are some exceptions: for instance, MRS localizers are not excluded from processing based on QC alone. Data curation is instead performed during OSPREY-BIDS processing, which prioritizes localizer timing over QC (see details here).

When additional QC criteria apply, filtering typically occurs in two stages: first, using both manual and automated QC fields, and second, using only automated fields. For example, only the highest-quality T1w and T2w are selected for structural MRI processing when multiple scans passing QC are present. In this first release, all high-resolution T1w and T2w scans — and most QALAS acquisitions — were selected using QU_Motion, a manual assessment of motion artifacts.