We have gathered for this challenge a total of 100 MS patients for which the following images will be provided:
- 3D FLAIR sequence at a first timepoint
- 3D FLAIR sequence at a second timepoint (1 to 3 years after the first one)
A total of 15 different MRI scanners (all part of the centers acquiring for the OFSEP cohort) are represented:
- Three GE scanners: Optima MR450w 1.5T, SIGNA HDx 3T, SIGNA HDxt 1.5T
- Six Philips scanners: Ingenia 1.5T, 2 Ingenia 3T, 1 Achieva dStream 3T, 1 Achieva 1.5T, 1 Achieva 3T
- Six Siemens scanners: 1 Aera 1.5T, 1 Skyra 3T, 1 Verio 3T, 1 Prisma 3T, 2 Avanto 1.5T
The repartition of the data into a training and testing sets has been done prior to the challenge into:
- 40 patients for training
- 60 patients for testing
Reference segmentation on these data was defined in two steps. First, 4 expert neuroradiologists were given the task to segment new lesions on the 100 cases. Then, a consensus was formed in two steps: a senior expert neuroradiologist looked and confirmed (or declined) disputed lesions among the experts; then a fusion using majority voting was performed. This consensus will be the reference for evaluation but all manual segmentations are provided to the challengers for the training set.
As for the first challenge in 2016, we have excluded all images from GE scanners from the training set to see the generalizability of the algorithms. The training set is of course here for you to test your algorithm and optimize it. This time however, we do not provide pre-processed data: the only pre-processing applied on the images is a rigid registration bringing the two timepoints of each patient to a middle point where the ground truth was computed. If you wish to use a default pre-processing for the data, we have built one for you in Anima scripts available via these links:
- Anima scripts: https://anima.irisa.fr (you will have to get the current master to get the script)
- Docker and singularity configuration files: https://github.com/Inria-Empenn/lesion-segmentation-challenge-miccai21/
You have to subscribe to the challenge to get data access via Shanoir-NG (next generation). To do so, follow this link to ask for an account on shanoir: select the MSSEG-2 challenge and fill in your details in the form to send your account request.
Once your registration has been accepted, you will have access to the challenge training set from the Shanoir interface, after validating that you accept its data usage agreement (DUA). We put the DUA pdf here for information as well, please read it carefully before validating.
Then you will be able to download the data as a single zip file (around 6 Gb). Warning: depending on your connection quality, the download may take quite some time to start (from 5 minutes to some hours) as the browser in fact downloads the file before asking you where to save it. You may thus have to wait but something is happening ! The zip file includes a README.md file describing the contents of the training data. Please contact us at firstname.lastname@example.org if you have any query.