The MICCAI 2019 training data is now final


The training data for the 2019 MICCAI challenge has been finalized and posted! Thanks everyone for your help and your patience while we ironed out the various issues. You are still welcome to submit issues about the data if you run into them, but they will not be addressed until after the July 29 deadline.

The GitHub Repository now has two branches, master and interpolated. master has the data at the spacing at which it was captured and interpolated has the data at the median pixel width (0.78162497mm) and slice thickness (3.0mm). This will also be the case for the testing data, and you will be allowed to submit your model’s predictions at either the original or the interpolated spacing. Instructions for downloading one or both branches can be found in the readme.

Happy training!


Which testing set scores will be prefered?
I asume, models will perform differently on original and interpolated sets


That’s an interesting question! We’re leaving it up to individual teams to decide which dataset to make predictions on. Our evaluation script will infer which dataset you used from the size of your prediction arrays.

The interpolated data may be easier to work with since the spacing is standardized, but one might want to make use of the highly-granular information from the “thin-cut” studies in original spacing.