What is the core task of MedReason?
MedReason is a medical visual question answering challenge. Participants answer both multiple-choice and open-ended questions from medical images and optional high-level context.
What does “unseen data” mean in this challenge?
It refers specifically to the out-of-distribution conditions introduced by the challenge: low-dose X-ray and ultra-high-field 7T brain MRI.
Do I need to train a medical MLLM from scratch?
No. You submit a Dockerized runnable system. Publicly available datasets and open-source pre-trained models are allowed.
How is MedReason related to Med-CMR?
MedReason builds on Med-CMR as its benchmark backbone and extends it into a challenge with explicit domain shift.
How are open-ended answers evaluated?
Open-ended answers are evaluated for both ground-truth correctness and visual grounding quality, not only for fluency or coherence.
Who may submit to the archival paper track?
All registered MedReason teams—not only the Top 3—may submit one paper based primarily on the method used in their final challenge submission. Each paper may have up to three authors, and the team determines its author list and author order in accordance with standard authorship requirements. See the full Paper Track guidelines.
Does a workshop paper affect participation in the joint journal paper?
No. Teams may participate in both opportunities. The individual workshop paper focuses on the team’s own method and results, while the joint journal paper focuses on challenge design, aggregate evaluation, cross-team comparison, and broader findings.
What should authors do before OpenReview opens?
Create and complete all author profiles as early as possible, preferably using institutional email addresses. New profiles without an institutional email may require moderation that can take up to two weeks, while profiles created with an institutional email are activated automatically.