Data

MS segmentation challenge using a data management and processing infrastructure

A journal article is available at Nature scientific reports on the challenge results, it should be cited whenever using the challenge data or results. It is available at the following DOI: https://dx.doi.org/10.1038/s41598-018-31911-7. Supplemental results data is avilable at zenodo: http://doi.org/10.5281/zenodo.1307653

Data and format

53 datasets from 4 different sites and 4 different MRI scanners will be used. The provided MR sequences will include:

  • 3D FLAIR
  • 3D T1-w
  • 3D T1-w GADO
  • 2D DP/T2

The provided data are split into two sets (15 training data and 38 testing data) as shown in the following table. All 53 datasets include raw data in Nifti format. As of now, the challenge is currently closed for submissions. We are actively looking for ways to revive submissions for new algorithms to be evaluated in the same framework. For this reason, you may download in the meantime only the training database (after account registration) and no access to the testing database will be given yet.

# exams Training Testing
GE Discovery 3T 0 8
Philips Ingenia 3T 5 10
Siemens Aera 1.5T 5 10
Siemens Verio 3T 5 10
Total 15 38
Since April 8 2016, we have made available the 15 training datasets, allowing the teams to train their algorithms and tune their parameters, as two packages (zip files):
  • one package with unprocessed datasets and manual lesion delineations from seven independent experts
  • one package with pre-processed datasets (denoised, registered, brain extracted, bias field corrected) using state-of-the-art pre-processing algorithms

Access to the complete dataset (training and testing) is open with access through shanoir. To get access, please register an account on shanoir if you do not already have one on that page, and select the MSSEG-2016 study. If you use this data, please cite the data paper at Neuroimage: https://doi.org/10.1016/j.neuroimage.2021.118589

Comments are closed.