MICCAI 2026 Challenge

Benchmarking Medical MLLM Reasoning under Domain Shift

A benchmark challenge evaluating whether specialized medical reasoning remains reliable when evidence quality drops and clinical data shift.

Challenge status Final submission Closed on Jul 22, 2026 Final leaderboard Expected around Aug 16, 2026 Paper track OpenReview deadline Aug 18, 2026, 12:00 AM SGT Contact medreason26@googlegroups.com

Final submission closed; results are in progress

Current status

Final evaluation is underway

The final challenge submission deadline closed on July 22, 2026. The organizing team is completing the final evaluation, and the final leaderboard is expected to be released around August 21, no later than August 23, 2026.

Final submission Closed July 22, 2026
Final leaderboard ~ Aug 21 Expected release
Previous results

Pre-evaluation leaderboard

Review the published pre-evaluation results on Synapse while final evaluation is being completed.

View pre-evaluation leaderboard →

MedReason individual method paper track

Following many requests from participating teams, MedReason has opened an optional archival paper track for all registered teams—not only the Top 3. Accepted papers will be included in the MICCAI 2026 Springer LNCS proceedings.

We sincerely apologize for the short notice, and we warmly encourage interested teams to submit an individual paper describing their team’s work.

EligibilityAll registered teams
Submission limitOne paper · Up to three authors
ReviewBlind review
Initial submission deadlineAug 18, 2026 · 12:00 AM SGT
Review comments and decisionAug 21, 2026Two or three reviewers per paper; no rebuttal stage
Camera-ready paper dueAug 24, 2026

Why MedReason?

MedReason evaluates whether medical MLLM systems can perform clinically grounded visual question answering when image quality and acquisition conditions shift.

The challenge builds on Med-CMR and focuses on one medical VQA task, while extending evaluation to more difficult out-of-distribution settings such as low-dose X-ray and 7T brain MRI.

The central question is not only whether a model can answer correctly, but whether its reasoning remains reliable, visually grounded, and robust under domain shift.

At-a-glance challenge task

Participants build one medical multimodal system for two question formats: multiple-choice and open-ended questions grounded in medical imaging evidence.

What the system receives

One medical image, one question, and optional context depending on the case.

What the system must return

For multiple-choice questions, the system returns one option. For open-ended questions, the system returns both a reasoning_trace and a final answer.

How performance is evaluated

Official ranking is based on MCQ Accuracy, Open-ended GT, and Open-ended VA, with organizer-side evaluation on hidden public and private evaluation datasets that contribute equally to the final ranking.

Challenge timeline

The challenge follows a simple progression: registration and data releasepre-evaluationfinal submission and workshop.

1

Launch and registration

Website and registration open on Apr 22, 2026.

2

Train release

Public materials and train split are released on Apr 22, 2026.

3

Validation release

Participant-facing validation resources open on May 15, 2026.

4

Pre-evaluation submission

Pre-evaluation submissions are accepted until Jul 15, 2026. New teams may submit before this date, with one organizer-side pre-evaluation opportunity per team.

5

Final submission

The final Docker submission deadline closed on Jul 22, 2026.

6

Final evaluation and leaderboard

Final evaluation is in progress. The final leaderboard is expected to be released around Aug 16, 2026.

7

MICCAI workshop

The MedReason workshop activities will take place during MICCAI 2026.