The models submitted and to be considered for this challenge must be fully automated. The submissions will be evaluated using the weighted Sorensen-Dice coefficient across cartilage and bone on a held-out test set that will be under-sampled by 8x in the equivalent pattern as training data and generates multi-class segmentation masks as output. Relevant scripts for computing the evaluation metric and a template for creating the submission file will be shared.
No additional data allowed for training.
Members associated with UCSF may participate but are not eligible for awards and not listed in leaderboard.
Participants with the top weighted dice scores for multi-class segmentation (cartilage, bone) will be announced publicly on a leaderboard with potential cash prizes to be determined for the top 3 submissions.
The performance of all submissions will be made public on the leaderboard unless a specific team withdraws from the competition. The top 3 scorers will be highlighted and will be awarded cash prizes.
Top 5 submissions will be invited to contribute to a full-length challenge summary research article.