KiTS will be held again in 2021!

Our proposal to hold a second edition of KiTS in conjunction with MICCAI 2021 in Strasbourg was accepted! The challenge will feature significantly more data, several annotations per case, and a number of additional annotated regions.

Annotated data will be released as it becomes available on the official challenge repository.

4 Likes

Congratulation! That is fantastic news!

1 Like

When the Kits21 challenge comes out, will the classification phase of Kit19 be closed? Or do you have precision when it will be closed?

1 Like

Great question! The test set from KiTS19 will be reannotated and included in the training set for KiTS21, but the KiTS19 leaderboard on grand-challenge.org will be left open with a disclaimer indicating that labels have been released.

But, Is there a date for the launch of the KITS21 challenge :question:

The timeline is as follows

Date Item
Mar 1 - Jul 1 Annotation, Release, and Refinement of Training Data
Aug 9 Deadline for Intention to Submit & Required Paper
Aug 16 - 30 Submissions Accepted
Sep 1 Results Announced
Sep 27 or Oct 1 Satellite Event at MICCAI 2021

So we’re currently in the process of collecting annotations for the training set, which will be available in its “final” form by July 1. You can track our progress on the challenge homepage.

At the moment, we’re ironing out the last details of the annotation pipeline. Once that is finished, we’ll make a formal announcement and include some examples of what the dataset will look like. My best guess for when this will happen is about a week from now.

1 Like

Just to make sure: does this mean, that the training set will be released in portions not all at once? I used data downloading script and only first 22 cases had full segmentation masks. The rest only have scan images.

Yep! We’re releasing annotations as soon as they’re performed.

I have a question about the segmentation. In case_00000, it seems that there are some slices that were not segmented. Top left and right in the image below are two adjacent, the right one is “closer to the middle of an entire scan” (don’t know how to express this in a better way). The artery and vein (if I’m not mistaken) are correctly segmented in the slice on the right (it’s segmentation masks are shown below it), while no segmentation is present in the adjacent slice, even though it clearly contains a vein and an artery. Is it a mistake? If not, how this will be scored in the predictions?

Thanks for bringing this up. In order to focus our annotation efforts near to the kidney, we bounded our artery and vein annotations above and below at consistent anatomical landmarks:

  • On the superior side: the IVC and aorta are bounded above by the axial slice at which the celiac artery bifurcates from the aorta – shown above the aorta in your top right image
  • On the inferior side: the IVC, aorta, and ureters are bounded below by the axial slice at which the aorta bifurcates into the common iliac arteries.

The kidney, tumor, and cyst regions have no such boundaries, which is why the kidney mask is still present in your top left slice.

Since this intent is consistent in our training data, we expect the nets to be able to identify these landmarks fairly easily and make predictions accordingly. That said, if this issue causes trouble for our internal baselines, we might revisit this point with either a modified metric or by cropping the imaging. We’re also open to suggestions :slight_smile: