Congratulations To All Who Participated!
Of course, everyone hopes that they win, but regardless of performance, every team who made a submission to this challenge has helped to move our collective understanding of this problem forward, and you all deserve recognition for that.
First Glance Takeaways
There will be lots of time to do a deep exploration of the leaderboard, but at first glance, I notice a few things:
1. The top performance is lower than in KiTS21
Based on our internal baselines we expected this, but it is an interesting phenomenon. KiTS23 featured a more difficult problem with the addition of the nephrogenic contrast phase, and this certainly could have played a role. It’s also possible that there was simply a tougher “draw” in test set cases this year. We’ve known for a while that cases vary significantly in their difficulty, and a greater fraction of difficult cases could bring down overall performance substantially. Rest assured that we will explore this further.
2. There are 25 participating teams
This is actually identical to KiTS21, but once again a large decrease from the 100 submissions to KiTS19. I think that this could be because in 2019 there were no widely-accepted baselines for kidney tumor segmentation. Now that there are, it is perhaps less exciting and more daunting to approach a problem that has become so well studied. I’d be happy to hear others’ perspectives on this.
3. Submission quality was very high
I really feel that with every KiTS challenge, the quality of the papers becomes more and more impressive. I believe that these papers are very important because they enable such unique apples-to-apples meta analyses that are otherwise really difficult in our field. The value of a leaderboard truly resides in these manuscripts.