Home Rules Results Register Download Submit FAQ

Download data sets

We have a total set of 150 MRI volumes available for this challenge. All datasets are completely anonymized and originate from the surgical planning program of Biomet, Inc. To access any of the data, you need to register a team first.

Team Name
What to download Training data 1-40 (444 MB)
Training data 41-60 (248 MB)
Training data 61-100, including ROIs (498 MB)
Test data 1-25 (308 MB)
Test data 26-50 (320 MB)
C++ source code for evaluation (5 KB)

Data format

All data is stored in Meta format containing an ASCII readable header and a separate raw image data file. This format is ITK compatible. Full documentation is available here. An application that can read the data is MITK-3M3. If you want to write your own code to read the data, note that in the header file you can find the dimensions of each file. In the raw file the values for each voxel are stored consecutively with index running first over x, then y, then z. The pixel type is short for the image data and unsigned char for the segmentations of the training data. Segmentations are multi-label images with the following codes: 0=background, 1=femur bone, 2=femur cartilage, 3=tibia bone, 4=tibia cartilage.

The last training data set (images 61-100) includes corresponding ROI images; these specify regions of interest where cartilage segmentations will be evaluated. Segmentations of the femoral cartilage will be evaluated in regions where bit 1 is set (i.e. values 1 and 3). Segmentations of the tibia cartilage will be evaluated in regions where bit 2 is set (i.e. values 2 and 3). ROI images are required by the provided evaluation source code, i.e. you can use the last training data set to estimate the performance of your algorithm before submission.

The content of the SKI10 website is copyrighted Tobias Heimann and Bram van Ginneken.