Welcome to FOMO26
The Foundation Model Challenge for Brain MRI at MICCAI 2026
Same core philosophy as FOMO25, now expanded with 306,303 brain MRI scans for pretraining and seven downstream tasks evaluating generalization in realistic clinical settings.
Towards Clinical-Grade Foundation Models:
The FOMO26 Challenge Setup
Track 1: Methods
Pretrain on FOMO300K.
Participants pretrain using the official FOMO26 large-scale dataset to compare methods under a shared data regime.
Comprehensive code-release.
Track 2: Open
Pretrain on Any Data.
Participants may use additional public/private data and custom pipelines to maximize generalization.
Open-data and proprietary pipelines allowed.
Evaluation
Few-shot and out-of-domain clinical benchmarks.
As in FOMO25, all methods are evaluated on clinically relevant downstream tasks; FOMO26 expands the suite with additional tasks and broader scanner/domain coverage.
No restrictions on fine-tuning strategy unless explicitly stated in task rules.
Timeline
April 2026
Challenge registration opens. Access to pretraining dataset and finetuning data. Public GitHub repo with example pretraining and finetuning code (MAE + U-Net) released.
15 May 2026
Sanity-check pipeline on Synapse opens — technical validation to confirm containers are correctly configured.
15 June 2026
Validation leaderboard opens. Final submission pipeline opens.
21 August 2026
Challenge submission deadline.
MICCAI 2026 — 4 or 8 October
Results announced. Top teams present their methods.
Tasks
Seven downstream tasks spanning classification, segmentation, regression, and representation quality. Tasks 1–3 continue from FOMO25; Tasks 4–7 are new.
Infarct Classification
Few-shot binary classification of acute ischemic infarcts from clinical MRI acquired in Denmark and India.
Meningioma Segmentation
Few-shot segmentation of meningioma tumors from clinical brain MRI.
Brain Age Estimation
Regression of chronological brain age from T1w and T2w MRI scans across a broad age range.
Trigeminal Neuralgia Segmentation
Multiclass segmentation of the trigeminal nerve and surrounding vessels, relevant to surgical planning.
Polymicrogyria Classification
Few-shot binary classification of polymicrogyria, a rare cortical malformation, from T1w MRI.
Linear Probing
Embedding quality assessment via linear separability — no fine-tuning allowed. Measures what the model learned during pretraining.
Bias and Fairness
Group fairness evaluation using the same linear probing setup as Task 6. Assesses whether embeddings encode or suppress demographic and acquisition biases.
Team
Organizing Committee
University of Copenhagen & Pioneer Centre For AI
University of Copenhagen & Pioneer Centre For AI
University of Copenhagen & Pioneer Centre For AI
University of Copenhagen & Pioneer Centre For AI
University of Copenhagen & Pioneer Centre For AI
University of Copenhagen & Pioneer Centre For AI
University of Copenhagen & Pioneer Centre For AI
University of Copenhagen & Pioneer Centre For AI
University of Copenhagen
University of Copenhagen
University of Copenhagen
University of Copenhagen
University of Copenhagen
University of Copenhagen & Copenhagen University Hospital, Rigshospitalet
Copenhagen University Hospital, Rigshospitalet
Radiological AI Testcenter, Denmark
Radiological AI Testcenter & Copenhagen University Hospital
Copenhagen University Hospital
Copenhagen University Hospital, Rigshospitalet
Mental Health Centre Copenhagen & University of Copenhagen
MGH / Harvard Medical School & MIT
Johns Hopkins University
Johns Hopkins University
University of Copenhagen & Pioneer Centre For AI
News
Announcing FOMO26
April 3, 2026
We are excited to announce FOMO26, the Foundation Model Challenge for Brain MRI at MICCAI 2026. Full documentation and registration details will be published soon.
Join the Challenge
Interested in participating, sponsoring, or collaborating? Reach out and we will notify you when full registration opens.
Contact information will be published when registration opens.