Participation Rules

By participating in the ACROBAT challenge and using the data provided with the challenge, participants agree to adhere to the following participation rules.

Algorithm

Execution of the algorithm during registration of test images needs to be fully automated, without user input beyond the stain antibody of the WSI. Algorithms should aim to register all tissue regions in the WSIs that exist in both the moving and the target image that are to be registered. The target image for each case is the H&E WSI, with the IHC WSIs as the moving images.

Training data policy

There are no restrictions in training data, the use of additional external data is not prohibited, even if that data is not publicly accessible. This is to enable the use of registration software whose configuration was optimized previously using potentially non-public data.

Policy for members of the organizers' institutions

Members of the organizers' departments may not participate in the challenge. Members of the organizers' institutions are not excluded.

Publication policy

We aim to publish a thorough analysis of the challenge results both on arxiv and in a relevant journal. The publication will include statistical analysis of results and their sensitivity to particular samples, annotators and IHC stains in the test data. The 10 top-performing teams will be invited to contribute to this publication. At the challenge organizers’ discretion, teams with conceptually novel/interesting algorithms that did not reach a high performance may be invited to contribute as well. All team members that were registered in a submitting team at the end of the challenge will be invited to co-author the publication. Participating teams may publish their own results separately after the challenge organizers have published their findings from the challenge in a relevant journal.

Algorithm description

All participating teams who want feedback on their test set submission and who would like to be listed in the final challenge leaderboard need to submit a description of their algorithm. The details of this submission will follow at a later stage.

Algorithm code publication

We do not require teams to publish the code for their methods, but would like to strongly encourage teams to do so. There will be a set of prizes available to all participants, and a second set of prizes only available to those which publish an easily executable self-contained code example for their method. Teams can receive prizes from both sets of prizes.

Docker container privacy and intellectual property

All docker containers that are submitted for the ACROBAT 2023 challenge will be handled with high discretion and only executed on local, secured computing infrastructure. Docker containers will not be published or shared by the challenge organisers and will be deleted immediately upon the submitting participants' request. If a non-disclosure agreement is required in order to share docker containers with us, please contact us as soon as possible, as all NDAs need to be reviewed by Karolinska Institute's legal department, which is a process whose timeline we cannot influence. It is possible that this may make participations that are contingent on NDAs unfeasible.