📝 Task Description 📝

The core task of the RARE 2025 challenge is binary classification: determining whether an endoscopic image of a Barrett’s Esophagus (BE) patient contains early neoplasia or not. The goal is to build AI algorithms that can identify subtle but critical signs of early-stage cancer while maintaining a low false positive rate — a key requirement in real-world clinical use.

🧠 Open Development Phase 🧠

During the Open Development Phase, participants will have access to the training data to develop their models. They can submit their models to the validation set, and a live leaderboard will display performance based on these submissions.

The private validation set is primarily intended to help participants verify that their models are functioning correctly on the platform. While it provides performance feedback, it is not advised to solely rely on the validation set for final model selection.

Furthermore, the validation set is small compared to the eventual test set, so leaderboard results during this phase may not accurately reflect final performance. Participants are encouraged to focus on building models that generalize well to unseen data.

🧪 Closed Testing Phase 🧪

In the Closed Testing Phase, participants will submit their final models or predictions for evaluation on the full test set. The official results will be announced during the EndoVis workshop day at MICCAI 2025 in Daejeon, South Korea.

Following the workshop, the final leaderboard will be published, reflecting performance on the test set and determining the official ranking of all participants.

📝 Participation Rules 📝

To ensure fairness and transparency, all participants in the RARE 2025 challenge must adhere to the following rules:

• All participants must register as part of a team, even if the team has only one member. Each participant may only join a single team.
• Anonymous participation is not allowed. Verified profiles must accurately display each participant’s real name, affiliation, and country.
• Members of sponsoring or organizing institutions (Eindhoven University of Technology, Amsterdam Medical Centers) are welcome to participate but are not eligible for awards or final rankings.
• Use of external data or pre-trained models is allowed if they are publicly licensed and disclosed in the algorithm description and/or publication.
• To participate in the Closed Testing Phase, teams must submit a short arXiv paper (2–3 pages) and a public/private GitHub link to their code under a permissive license.
• All participants must publish their code under the MIT license in order to be considered for the final leaderboard.
• The organizers reserve the right to disqualify any team for unethical behavior or non-compliance with challenge rules.

A full set of rules will be available on the deployment platform. All participants are expected to review and follow them.