Dear participants,

Today we have pushed some important updates to the challenge repository:

  • We have identified a number of faulty annotations in the training dataset. Where possible we have corrected the masks and otherwise removed them. You can pull from the challenge repository to get the updates, or look at the commit to find out what was changed. The majority of errors were from the bone lesion dataset and LiTS, and were due to annotation artifacts or partial annotations. If you have found other cases with faulty annotations, please feel free to contact us or start an issue on GitHub and we will look into correcting them.

  • We have also fixed an error in the implementation of the long- and short-axial error calculation. The metric is now symmetrical, doesn't favor under or over-prediction, scaled from 0-1, and when there is no prediction from the model the maximum possible error is used. We have re-evaluated the submissions to the development leaderboard, and subsequent submissions will use the new implementation.

Kind regards,

The ULS23 organizers