CDMRI'23 – QuantConn Challenge

QuantConn: Quantitative Connectivity through Harmonized Preprocessing of Diffusion MRI

Challenge Organisers

Nancy Newlin	Vanderbilt University, Nashville, TN
Neda Jahanshad	Keck School of Medicine of USC, Los Angeles, CA
Kurt Schilling	Vanderbilt University, Nashville, TN
Daniel Moyer	Vanderbilt University, Nashville, TN
Eleftherios Garyfallidis	Indiana University Bloomington, Bloomington, IN
Bennett Landman	Vanderbilt University, Nashville, TN
Serge Koudoro	Indiana University Bloomington, Bloomington, IN
Bramsh Chandio	Keck School of Medicine of USC, Los Angeles, CA

Contact: nancy.r.newlin@vanderbilt.edu

Update from the Team (15 AUG 2023)

For those interested, we have a distortion corrected version of the data available under "Testing_preprocessed" and "Training Preprocessed"! This data has been distortion corrected using PreQual (https://github.com/MASILab/PreQual).

The Task

We have provided DW images from two sites with very different acquisition protocols. Your team is tasked with making these two sites as similar as possible, or “harmonizing” them. There is no limit to what methods you can use! For example, we envision explicit image harmonization methods, denoising approaches, super resolution, and anything within the preprocessing pipeline that retain biological differences while mitigating differences due to different acquisition protocols. In summary:

Participants can do any preprocessing and/or harmonization to the data that they think might minimize differences between scanners.
Harmonization can be from A to B (or vice versa), or to any desired space.
Data from both sites can be submitted at any resolution, reconstructed with any associated b-table, and in any desired space.
Evaluation will be performed on the submitted datasets only (test dataset, N=25), and in the space dataset is submitted in.

About the data

CLICK HERE to access the data.

Thanks to our colleagues at QIMR Berghofer Medical Research Institute, we have curated a newly released dataset comprising 100 subjects with paired data from two distinct diffusion acquisition protocols.

The data is organized as follows. There are 25 subjects in the "Testing" folder. This is the subset of data to harmonize and submit. We provided 77 additional subjects to be used for training, if needed (in "Training" folder). All subjects have three sub-folders: Diffusion data from site A ("A"), Diffusion data from site B ("B"), and a T1-weighted image in "anat" folder. In addition, we have provided age and sex demographic data ("QuantConn_DemographicData_Testing_and_Training.xlsx"). Inside of the /anat/ folder, we provide a Freesurfer based atlas (Desikan-Killany) in T1-space, called "atlas_freesurfer.nii.gz"; the lookup table is also provided ("fs_default.txt").

Scanning was performed at the QIMR Berghofer Medical Research Institute on a 4 tesla Siemens Bruker Medspec scanner. T1-weighted images were acquired with an inversion recovery rapid gradient-echo sequence (inversion/repetition/echo times, 700/1500/3.35 ms; flip angle, 8°; slice thickness, 0.9 mm; 256 × 256 acquisition matrix).

Site A DW images were acquired using single-shot echo-planar imaging with a twice-refocused spin echo sequence to reduce eddy current-induced distortions. A 3-min, 30-volume acquisition was designed to optimize signal-to-noise ratio for diffusion tensor estimation (Jones 1999). Imaging parameters were repetition/echo times of 6090/91.7 ms, field of view of 23 cm, and 128 × 128 acquisition matrix. Each 3D volume consisted of 21 axial slices 5 mm thick with a 0.5-mm gap and 1.8 × 1.8 mm2 in-plane resolution. Thirty images were acquired per subject: three with no diffusion sensitization (i.e., T2-weighted b0 images) and 27 DW images (b = 1146 s/mm2) with gradient directions uniformly distributed on the hemisphere.

Site B DW images were acquired using single-shot echo planar imaging (EPI) with a twice-refocused spin echo sequence to reduce eddy-current induced distortions. Acquisition parameters were optimized to improve the signal-to-noise ratio for estimating diffusion tensors (Jones 1999). Imaging parameters were: 23 cm FOV, TR/TE 6090/91.7 ms, with a 128 × 128 acquisition matrix. Each 3D volume consisted of 55 2-mm thick axial slices with no gap and a 1.79 × 1.79 mm2 in-plane resolution. 105 images were acquired per subject: 11 with no diffusion sensitization (i.e., T2-weighted b0 images) and 94 DWI (b = 1159 s/mm2) with gradient directions distributed on the hemisphere. HARDI scan time was 14.2 minutes.

Dataset contributions:
Margaret J. Wright, Lachlan Strike
Lachlan T. Strike, Gabriella A.M Blokland, Narelle K. Hansell, Nicholas G. Martin, Arthur W. Toga, Paul M. Thompson, Greig I. Zubicaray, Katie L. McMahon and Margaret J. Wright. (2023). Queensland Twin IMaging (QTIM). OpenNeuro. Dataset: doi

How to register

Please fill out THIS FORM to register.

How to submit

24-48 hours after registering with the form above, you will receive an email from our team with a link to a box folder specific to your team. Upload your DW images, bvecs, and bvals to this folder. You only need to process the 25 subjects in the "Testing" folder. Once done, send an email to nancy.r.newlin@vanderbilt.edu with your team's report! Please title the email with "MICCAI 2023 Challenge Submission – [YOUR TEAM NAME]".

We provided two example submissions in the correct format and associated report ("TesSubmission_1" and "TestSubmission_2"). Please keep the same directory organization as the data provided. Note: TestSubmission_2 is ~50GB and may not download in one go. We suggest downloading a single subject, if needed.