Categories
Resources

Diffusion MRI Data for Best Practices in Image Preprocessing

A project by the Diffusion Study Group of the International Society for Magnetic Resonance in Medicine (ISMRM)

Image preprocessing is essential to analysis of diffusion MRI data. However, there is currently no standard for best practices in designing a preprocessing pipeline. The lack of consensus inhibits reproducibility of diffusion MRI studies, which in turn impedes objective evaluation of new methods.

Help us understand how preprocessing tools affect the reproducibility of diffusion MRI analysis and develop best-practice recommendations for the diffusion MRI research community.

How to Contribute

  1. Download the raw diffusion MRI data available in this resource. 
  2. Process each data set with your typical preprocessing pipeline and tools.
  3. Upload the preprocessed data to a secured server through a dedicated link.

The deadline to contribute is October 23, 2021.

For more details, see the readme section below.

About the Data

We are sharing multi-shell and multi-site diffusion MRI data. The resource includes raw, unprocessed magnitude MRI data sets of multiple subjects. For each subject, data from multiple scans with distinct settings are included in order to study artifacts of varying shape and size while keeping the encoded diffusion information constant. For technical specifications of the data, see the attached readme.

Who Should Contribute

If you’re working with diffusion MRI data and using or developing image preprocessing tools, we want to see your contribution.

Your contributions will allow the Diffusion Study Group to evaluate how preprocessing pipelines and tools affect the reproducibility of diffusion MR analyses. The study group will assess inter-pipeline variability compared to inter- and intra-subject variability and evaluate the effects of individual image preprocessing steps. The results will be publicly shared and will form the basis for the development of best practices in image preprocessing of diffusion MRI data.

About the Best Practices Project

This project is part of an effort headed by the Diffusion Study Group of the ISMRM aimed at formulating informed, consensus-based best practices in diffusion MRI. The study group’s mission is to facilitate the technological development, evaluation, and clinical application of diffusion MRI. This component, also known as “diffusion-weighted image preprocessing” is led by Jelle Veraart, PhD, and Maxime Descoteaux, PhD.

Get the Data

The data available on this page are provided free of charge and come without any warranty. CAI²R and NYU Grossman School of Medicine do not take any liability for problems or damage of any kind resulting from the use of the files provided. Use of the data is solely at the user’s own risk. The data provided are not medical products and must not be used for making diagnostic decisions.

The data are provided solely for the purposes of the project “Best Practices for Image Preprocessing of Diffusion MRI Data” and subject to terms and conditions of this data sharing agreement:

  1. The data will be used for the purposes of the project “Best Practices for Image Preprocessing of Diffusion MRI Data” only.
  2. No transfers, distribution, release, or disclosure of the provided data shall be made except that necessary for participation in the “Best Practices for Image Preprocessing of Diffusion MRI Data” project.
Please spell out your affiliation (e.g. "New York University" rather than "NYU").

Readme

Detailed information about what you’re downloading, what to do with the data, what to upload, where, and when.

What You’re Downloading

This resource includes raw, unprocessed magnitude MRI data sets of multiple subjects. For each subject, data from 31 sessions with distinct settings are included. Each of the sessions, referred to as “scans”, is saved in a subfolder with a unique name (e.g. Y0423).

Note that for the purposes of this project, it is not relevant which scans are from the same subject, scanner, or settings.

For each scan, we provide the following data:

  • T1-weighted MPRAGE image [defaced]
    ./*/anat/t1w_anat.nii.gz
  • Multi-shell diffusion-weighted images
    ./*/dwi/dwi.nii.gz
  • The diffusion-weighted gradient scheme used in the acquisition in FSL bvecs/bvals format files
    ./*/dwi/dwi.bvec/bval
  • Non-diffusion-weighted images, acquired with reversed phase encoding
    ./*/dwi/b0rpe.nii.gz
  • B0 field map
    either
    ./*/fmap/B0_fieldmap.nii.gz
    or both
    ./*/fmap/gre_magnitude.nii.gz and ./*/fmap/gre_phasediff.nii.gz

For each file, a corresponding JSON file includes all relevant scan information.

What to Do (and Not Do) with the Data

For each scan, preprocess the diffusion-weighted images with your typical image preprocessing pipeline. Process each of the 31 scans independently, only using the information provided in the scan-specific folder. Do not customize your setup to suit the data—we are studying how users’ typical pipelines affect further analysis.

This project is not a challenge. Individual contributions will not be ranked or linked to your name or affiliation.

What to Upload, Where, and When

What to Upload

We define preprocessing as the series of operations that precede model fitting, and we ask that you upload the following data for each scan:

  • Preprocessed diffusion-weighted images
    ./*/preproc/dwi.nii.gz
  • The diffusion-weighted gradient scheme of the preprocessed data in bvecs/bvals format files †
    ./*/preproc/dwi.bvec/bval
  • Identified signal outliers that must be rejected from further analysis; provide only if outlier rejection without signal replacement is part of your pipeline
    ./*/preproc/outliers.nii.gz
  • A text summary of the steps and methods used in your preprocessing pipeline
    ./methods.txt (a template will be provided in the upload folder).

† If voxelwise operations are applied to bvec/bval, then please provide dwi.gradx, dwi.grady, and dwi.gradz instead.

Please use the suggested file names to help us streamline data analysis.

Where to Upload

Within 24 hours of requesting a download of the data on this page, you will receive an email from jelle.veraart@nyulangone.org with a personal link to a secured upload server. Use the link to upload your data after preprocessing.

If you don’t receive the email with an upload link within 24 hours of making a download request on this page, please check your spam folder. If you find no email with an upload link, contact jelle.veraart@nyulangone.org.

When to Upload

The deadline for contributing your preprocessed data is October 23, 2021.

Contact

Questions about this resource may be directed to Jelle Veraart, PhD.

Acknowledgments

Data were acquired at Sherbrooke Medical Imaging Center (CIMS), Université de Sherbrooke, Canada; Vanderbilt University Medical Center, Nashville, USA; Richard M. Lucas Center for Imaging, Stanford University, USA; and Max Planck Institute for Human Cognitive and Brain Sciences, Germany. 

Erpeng Dai, Maxime Descoteaux, Luke J. Edwards, Bennett Landman, Jennifer McNab, Kurt Schilling, and Nikolaus Weiskopf contributed to data collection.

Daan Christiaens, Maxime Descoteaux, Vladimir Golkov, Siawoosh Mohammadi, and Jelle Veraart contributed to the design of the acquisition protocol.

Data acquisition has been made possible with support from Canada Institutional Research Chair in NeuroInformatics; National Science Foundation CAREER award number 1452485; the National Institutes of Health award number R01EB017230; the European Research Council under the European Union’s Seventh Framework Programme (FP7/2007-2013; ERC #616905); the German Federal Ministry of Education and Research (BMBF) (01EW1711A & B) in the framework of ERA-NET NEURON; and the Max Planck Society.