HBCD MR Quality Control Procedures🔗

Raw MR Data QC🔗

Raw MRI QC combines automated and manual checks to evaluate unprocessed data and identify acquisition errors, image artifacts, or corrupted files before downstream processing. Automated QC is applied to all data. Due to the large data volume and time-intensive nature of manual inspection, manual visual review is only performed for series that fail automated QC. Although automated tools detect most quality issues, some artifacts may be missed if misclassified or not assessed as part of automated QC. Raw MR QC metrics are provided in the raw BIDS scans.tsv files in the release (see details).

Acquired imaging data are automatically uploaded to central servers, where they undergo automated protocol compliance and completeness checks. Data that fail are flagged for review and excluded from release until the issues are resolved.

Protocol compliance is performed by extracting imaging parameters from DICOM headers to confirm that key parameters (e.g., voxel size, TR, orientation) match the expected scanner protocol. Out-of-compliance series are flagged for review, with site followup as necessary.

Completeness checks verify that all expected series are present in each imaging session. Missing data may be caused by aborted scans, incomplete sessions, and/or incomplete data transfer. Valid sessions are expected to include: T1w & T2w, 2 resting state functional runs (each accompanied by fieldmaps acquired in AP/PA phase encoding directions), diffusion scans (acquired AP/PA), quantitative QALAS and B1 maps, and MRS scan and SVS localizer.

Automated QC🔗

Data that pass protocol adherence and completeness checks move to the next stage of automated QC. Automated QC metrics are calculated for modalities as follows:

Modality Automated QC Metrics
sMRI & qMRI
  • Estimate motion artifacts using a deep learning model
  • Compute signal-to-noise ratio (SNR)
fMRI
  • Estimate head motion with average framewise displacement and data (sec) at FD thresholds of 0.2/0.3/0.4 mm (Power et al., 2012)
  • Detect line artifacts and FOV cutoff
  • Compute spatial smoothness (FWHM) and temporal SNR (tSNR) after motion correction (Triantafyllou et al., 2005)
dMRI
  • Estimate head motion (framewise displacement, FD)
  • Refine motion estimates via registration to tensor-synthesized images (Hagler et al., 2009)
  • Identify dark slices (caused by abrupt head movements) using RMS difference between raw and tensor-fitted data
  • Calculate total slices and frames with motion artifacts
  • Detect line artifacts and field-of-view (FOV) cutoff
Field Maps Detect line artifacts and field-of-view (FOV) cutoff
All Compute SNR where applicable

Manual Review🔗

Data are flagged for manual review based on automated QC results using multivariate prediction and Bayesian classifiers, so only a subset undergoes both automated and manual review. When a series is flagged, trained technicians perform visual review and rate artifact severity on a 0–3 scale: none (0), mild (1), moderate (2), or severe (3). Series rated 3 (severe) are automatically assigned an overall QC score of 0 (Fail) and excluded from downstream processing. For all others, final selection is informed by manual ratings, reviewer notes, and automated QC metrics.

Modality Manual QC Procedures & Scoring
sMRI
  • Motion artifacts (ripples, blurring), scored 0–3
  • Document additional issues (e.g., intensity inhomogeneity, ghosting
qMRI
  • Same artifact scoring (0–3)
  • Inspect derived data (parametric maps, ROI analysis, quantitative checks for 3D-QALAS)
dMRI, fMRI, fmaps
  • Score susceptibility artifacts, FOV cutoff, and horizontal line artifacts (present in sagittal view)
  • Note susceptibility artifacts, including signal dropout (common in posterior occipital cortex of infant fMRI data acquired in PA phase encoding direction), signal bunching, and warping
MRS
  • Visual inspection and overall QC only of SVS localizer (used to define spectroscopy ROI)

BrainSwipes🔗

BrainSwipes is a gamified crowdsourcing platform used to perform manual QC of processed MRI data. Reviewers assess images from XCP-D visual reports, displaying a series of brain images in coronal, axial, and sagittal planes, and classify each report as Pass (1) or Fail (0).

BrainSwipes results are included in the Tabular Imaging domain:

  • Report-level QC metrics: mean QC score and number of reviewers for each visual report
  • Subject-level QC metrics: mean QC score and average number of reviewers across all reports for a participant
BrainSwipes example
QC Assessment What is Evaluated
Surface Delineation Accuracy of cortical surface placement and gray/white matter boundaries
Atlas Registration Alignment between the participant's anatomical image and the reference atlas
Functional Registration Alignment between functional and structural images and detection of major artifacts such as signal dropout

Where to find QC data in release🔗

MR quality control (QC) is performed at multiple stages of data processing. The release includes the following QC metrics, reports, and manual review results, displayed below in roughly chronological order (from raw data QC to group-level analyses provided in the release documentation):

QC Source Stage Description Release Data
Raw MR QC Raw DICOM Automated and manual QC metrics, including compliance/completeness checks scans.tsv files
MRIQC Raw BIDS Automated image quality metrics MRIQC derivatives
Pipeline QC Reports Processing Pipeline-generated visual reports and automated metrics Derivatives
BrainSwipes Post-processing Manual review results for structural and functional XCP-D QC reports Tabular Imaging
Release QC Summaries Group analysis Group-level analyses/QC summaries provided in documentation, e.g., fMRI and dMRI QC Summary Statistics Release documentation

Dean III, D. C., Tisdall, M. D., Wisnowski, J. L., Feczko, E., Gagoski, B., Alexander, A. L., ... & HBCD MRI Working Group. (2024). Quantifying brain development in the HEALthy Brain and Child Development (HBCD) Study: The magnetic resonance imaging and spectroscopy protocol. Developmental Cognitive Neuroscience, 70, 101452. https://doi.org/10.1016/j.dcn.2024.101452

Gard, A. M., Hyde, L. W., Heeringa, S. G., West, B. T., & Mitchell, C. (2023). Why weight? Analytic approaches for large-scale population neuroscience data. Developmental Cognitive Neuroscience, 59, 101196. https://doi.org/10.1016/j.dcn.2023.101196

Hagler, D. J., Jr, Ahmadi, M. E., Kuperman, J., Holland, D., McDonald, C. R., Halgren, E., & Dale, A. M. (2009). Automated white-matter tractography using a probabilistic diffusion tensor atlas: Application to temporal lobe epilepsy. Human Brain Mapping, 30(5), 1535–1547. https://doi.org/10.1002/hbm.20619

Power, J. D., Barnes, K. A., Snyder, A. Z., Schlaggar, B. L., & Petersen, S. E. (2012). Spurious but systematic correlations in functional connectivity MRI networks arise from subject motion. NeuroImage, 59(3), 2142–2154. https://doi.org/10.1016/j.neuroimage.2011.10.018

Triantafyllou, C., Hoge, R. D., Krueger, G., Wiggins, C. J., Potthast, A., Wiggins, G. C., & Wald, L. L. (2005). Comparison of physiological noise at 1.5 T, 3 T and 7 T and optimization of fMRI acquisition parameters. NeuroImage, 26(1), 243–250. https://doi.org/10.1016/j.neuroimage.2005.01.007