A quality-rated, analysis-ready release of 26,174 dMRI sessions from the ABCD Study
Poster No:
1810
Submission Type:
Abstract Submission
Authors:
Steven Meisler1, Matthew Cieslak1, Hamsanandini Radhakrishnan1, Taylor Salo1, Eric Feczko2, Kimberly Weldon2, Timothy Hendrickson2, rae McCollum2, Begim Fayzullobekova2, Tanya Pandhi2, Lucile Moore2, Bárbara Avelar-Pereira3, Joëlle Bagautdinova1, Sendy Caffarra4, Kelly Chang5, Philip Cook1, Teresa Gomez5, Mareike Grotheer6,7, McKenzie Hagen5, Zeeshan Huque8, Iliana Karipidis9,10, Arielle Keller11,12, John Kruper5, Audrey Luo1, Kahini Mehta13, Jamie Mitchell14, Adam Pines14, Ethan Roy14, Hannah Stone15, Valerie Sydnor16, Maya Yablonski14, Jason Yeatman14, Ariel Rokem5, Damien Fair2, Theodore Satterthwaite1
Institutions:
1University of Pennsylvania, Philadelphia, PA, 2University of Minnesota, Minneapolis, MN, 3Karolinska Institutet, Stockholm, Sweden, 4University of Modena and Reggio Emilia, Modena, Italy, 5University of Washington, Seattle, WA, 6Department of Psychology, Philipps-Universität Marburg, Marburg, Germany, 7Center for Mind, Brain and Behavior – CMBB, Philipps-Universität Marburg, Justus-Liebig-Universität Giessen and Technische Universität Darmstadt, Marburg, Germany, 8Temple University, Philadelphia, PA, 9Neuroscience Center Zürich, University of Zürich and ETH Zürich, Zürich, Switzerland, 10Department of Child and Adolescent Psychiatry and Psychotherapy, University Hospital of Psychiatry Zürich, University of Zürich, Zürich, Switzerland, 11Department of Psychological Sciences, University of Connecticut, Storrs, CT, 12Institute for the Brain and Cognitive Sciences, University of Connecticut, Storrs, CT, 13Columbia University, New York, NY, 14Stanford University, Stanford, CA, 15University of California, Santa Barbara, Santa Barbara, CA, 16University of Pittsburgh, Pittsburgh, PA
First Author:
Co-Author(s):
Mareike Grotheer, PhD
Department of Psychology, Philipps-Universität Marburg|Center for Mind, Brain and Behavior – CMBB, Philipps-Universität Marburg, Justus-Liebig-Universität Giessen and Technische Universität Darmstadt
Marburg, Germany|Marburg, Germany
Department of Psychology, Philipps-Universität Marburg|Center for Mind, Brain and Behavior – CMBB, Philipps-Universität Marburg, Justus-Liebig-Universität Giessen and Technische Universität Darmstadt
Marburg, Germany|Marburg, Germany
Iliana Karipidis, PhD
Neuroscience Center Zürich, University of Zürich and ETH Zürich|Department of Child and Adolescent Psychiatry and Psychotherapy, University Hospital of Psychiatry Zürich, University of Zürich
Zürich, Switzerland|Zürich, Switzerland
Neuroscience Center Zürich, University of Zürich and ETH Zürich|Department of Child and Adolescent Psychiatry and Psychotherapy, University Hospital of Psychiatry Zürich, University of Zürich
Zürich, Switzerland|Zürich, Switzerland
Arielle Keller, PhD
Department of Psychological Sciences, University of Connecticut|Institute for the Brain and Cognitive Sciences, University of Connecticut
Storrs, CT|Storrs, CT
Department of Psychological Sciences, University of Connecticut|Institute for the Brain and Cognitive Sciences, University of Connecticut
Storrs, CT|Storrs, CT
Introduction:
The Adolescent Brain and Cognitive Development (ABCD) Study (Casey, 2018) includes diffusion-weighted imaging (DWI), an essential approach for studying white matter development. The extensive computational resources required to produce derived imaging phenotypes for analysis pose a significant barrier for many research groups. Furthermore, data quality remains one of the most important considerations in studies of brain development (Koirala, 2022), and manually rating a dataset of ABCD's size is impractical. To address these challenges, we created a resource of 26,174 pre- and post-processed DWI images using state-of-the-art pipelines, and trained a classifier to automate image quality ratings.
Methods:
T1w, DWI, and field map images from 26,174 sessions were curated into BIDS format. Images originated from three different MRI vendors (Siemens: n=16,351, GE: n=6,911, Philips: n=2,912) at multiple timepoints (baseline: n=9100, 2 year follow-up: n=7351, 4 year follow-up: n=6053, 6 year follow-up: n=3670). Data were processed with QSIPrep and QSIRecon (Cieslak, 2021), generating an array of DWI image quality metrics and analyzable derivative images.
Image quality metrics included motion parameters, shell-wise contrast-to-noise ratios (CNR), average neighboring DWI coefficient (NDC), average DWI contrast, and the number of bad slices. NDC is defined as the voxel-wise correlation between a volume and its closest neighbor in q-space. DWI contrast is defined as the ratio of correlations between a volume and its nearest q-space neighbor (i.e., NDC) and between the same volume and that furthest away in q-space. We explored the heterogeneity of CNRs, NDCs, and DWI contrasts across MRI vendors.
DWI derivatives (Figure 1) included scalar maps from diffusion kurtosis imaging, neurite orientation dispersion and density imaging, mean apparent propagator MRI, and generalized q-space modeling. All derived images were generated in both subject and MNI152NLin2009cAsym space. Additionally, 77 tractography bundles were generated using DSI Studio's Autotrack function, along with bundle-wise morphometric and microstructural measures.
A team of 23 trained reviewers rated 1,771 Siemens baseline images on a -2 (definitely fail) to 2 (definitely pass) integer scale with the dmriprep-viewer interface (Richie-Halford 2021). Ratings were based on visual inspection of DWI time series, motion measurements, and colored fractional anisotropy maps. Each image was rated by 3 reviewers. The subset of rated images spanned the full spectrum of quality, as indexed by average NDC.
Similar to Richie-Halford (2022), we used this subset of rated images to train an XGBoost model (Chen, 2016) to predict the average manual image quality rating from the full set of automated QSIPrep quality metrics. A Bayesian search with cross-validation (3 folds, 100 iterations, repeated twice) identified optimal model parameters. The best-performing model was applied across the Siemens dataset. Feature importance was measured by SHapley Additive exPlanations (SHAP) scores.
Image quality metrics included motion parameters, shell-wise contrast-to-noise ratios (CNR), average neighboring DWI coefficient (NDC), average DWI contrast, and the number of bad slices. NDC is defined as the voxel-wise correlation between a volume and its closest neighbor in q-space. DWI contrast is defined as the ratio of correlations between a volume and its nearest q-space neighbor (i.e., NDC) and between the same volume and that furthest away in q-space. We explored the heterogeneity of CNRs, NDCs, and DWI contrasts across MRI vendors.
DWI derivatives (Figure 1) included scalar maps from diffusion kurtosis imaging, neurite orientation dispersion and density imaging, mean apparent propagator MRI, and generalized q-space modeling. All derived images were generated in both subject and MNI152NLin2009cAsym space. Additionally, 77 tractography bundles were generated using DSI Studio's Autotrack function, along with bundle-wise morphometric and microstructural measures.
A team of 23 trained reviewers rated 1,771 Siemens baseline images on a -2 (definitely fail) to 2 (definitely pass) integer scale with the dmriprep-viewer interface (Richie-Halford 2021). Ratings were based on visual inspection of DWI time series, motion measurements, and colored fractional anisotropy maps. Each image was rated by 3 reviewers. The subset of rated images spanned the full spectrum of quality, as indexed by average NDC.
Similar to Richie-Halford (2022), we used this subset of rated images to train an XGBoost model (Chen, 2016) to predict the average manual image quality rating from the full set of automated QSIPrep quality metrics. A Bayesian search with cross-validation (3 folds, 100 iterations, repeated twice) identified optimal model parameters. The best-performing model was applied across the Siemens dataset. Feature importance was measured by SHapley Additive exPlanations (SHAP) scores.
Results:
Siemens data consistently showed higher CNR and NDC than GE and Philips data (Figure 2A). Preprocessing increased image quality (Figure 2B). The image quality classifier achieved 98% accuracy in predicting passing (average rating > 0) and failing (average rating < 0) Siemens images, with an AUC of 0.98. SHAP analysis revealed the most predictive features were the number of bad slices, the contrast-to-noise ratio of the innermost shell, peak framewise displacement, and max relative translation during acquisition.
Conclusions:
This curated dataset of 26,174 processed DWI images with analysis-ready derivatives and quality ratings represents a valuable resource for studying white matter development. Researchers should be aware of and account for systematic image quality differences across vendors. We are expanding quality ratings to Philips and GE data. The dataset will be available in 2025 for researchers with an ABCD data use certification.
Lifespan Development:
Early life, Adolescence, Aging 2
Modeling and Analysis Methods:
Diffusion MRI Modeling and Analysis
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
White Matter Anatomy, Fiber Pathways and Connectivity
Neuroinformatics and Data Sharing:
Databasing and Data Sharing 1
Informatics Other
Keywords:
Development
Informatics
NORMAL HUMAN
Open Data
PEDIATRIC
White Matter
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Cieslak, M. (2021). QSIPrep: an integrative platform for preprocessing and reconstructing diffusion MRI data. Nature methods, 18(7), 775-778.
Chen, T. (2016). XGBoost: A scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 785–794. Association for Computing Machinery. https://doi.org/10.1145/2939672.2939785
Koirala, N. (2022). Widespread effects of dMRI data quality on diffusion measures in children. Human brain mapping, 43(4), 1326-1341.
Richie-Halford, A. (2021). nipreps/dmriprep-viewer: v0.1.0: First release of dmriprep-viewer (v0.1.0). Zenodo. https://doi.org/10.5281/zenodo.5076263
Richie-Halford, A. (2022). An analysis-ready and quality controlled resource for pediatric brain white-matter research. Scientific data, 9(1), 616.