Hi, I'm Asbjørn Munk
I am a Ph.D. candidate at the University of Copenhagen and the Pioneer Centre for AI, supervised by Prof. Mads Nielsen. I'm funded by the 2023 PhD Fellowship from the Danish Data Science Academy. My research focuses on foundation models and self-supervised learning for medical imaging with a focus on brain MRI (and occasionally fetal ultrasound). Previously I was a visiting researcher at Juan Eugenio Iglesias' lab at Harvard & MGH.
asmu@di.ku.dk · twitter · linkedin · google scholar
News
- May 2026: FOMO26 is now out! Super proud to lead the second Foundation Model Challenge for Brain MRI at MICCAI 2026. Check it out at fomo26.github.io!
- April 2026: The FOMO25 challenge paper is now out on arXiv!
- February 2026: We have released the FOMO300K! A large-scale pretraining dataset for Brain MRI. Details available in this preprint and full dataset available at Hugging Face 🤗.
- July 2025: I'm in Boston! 🇺🇸 For the next six months I will be at Prof. Juan Eugenio Iglesias' lab at the Martinos Center, Harvard Medical School & Massachusetts General Hospital. If you are in the area, feel free to reach out!
- June 2025: Our paper on UltraDINO, a foundation model for fetal ultrasound, got accepted at MICCAI 2025! Paper is available on arXiv. Hope to see you in South Korea! 🇰🇷
- June 2025: In conjunction with the FOMO25 challenge we have released FOMO60k, a large-scale pretraining dataset for Brain MRI. It's available via Hugging Face 🤗, details available in this preprint.
- March 2025: Proud to be leading the first Foundation Model Challenge at MICCAI for Brain MRI! Read more and sign up at fomo25.github.io!
- January 2025: Our work on CLIP applied to Brain MRI to appear at ISBI 2025! This is my first publication as a last author, supervising two talented Master's students, Jakob and Valdemar. Hope to see you in Houston, TX! 🇺🇸
- August 2024: Our work on AMAES to appear at ADSMI @ MICCAI 2024 in Morocco! 🇲🇦
- May 2024: Excited to visit Prof. Juan Eugenio Iglesias at the Martinos Center, Harvard / MGH.
- January 2024: My work on the MDD-UNet was presented at the NLDL 2024 conference in Tromsø, Norway. 🇳🇴
- June 2023: I got awarded one of the prestigious DDSA PhD Fellowships!
Selected Publications
Full list of publications can be found on my google scholar. * denotes equal contribution.
-
Towards Brain MRI Foundation Models for the Clinic: Findings from the FOMO25 Challenge
Asbjørn Munk*, Stefano Cerri*, Vardan Nersesjan, Christian Hedeager Krag, Jakob Ambsdorf, Pablo Rocamora García, Julia Machnio, Peirong Liu, Mostafa Mehdipour Ghazi, Akshay Pai, Espen Jimenez Solem, Sebastian Nørgaard Llambias, Mikael Boesen, Michael Eriksen Benros, Juan Eugenio Iglesias, Mads Nielsen, et al.
PreprintResults and findings from the first Foundation Model Challenge at MICCAI, evaluating foundation models for brain MRI. Key findings:
arXiv · Website- Self-supervised pretraining improves generalization on clinical data under domain shift, with the strongest models trained out-of-domain surpassing supervised baselines trained in-domain.
- No single pretraining objective benefits all tasks: MAE favors segmentation, hybrid reconstruction-contrastive objectives favor classification.
- Strong performance was achieved by small pretrained models; improvements from scaling model size and training duration did not yield reliable benefits.
-
A large-scale heterogeneous 3D magnetic resonance brain imaging dataset for self-supervised learning
Asbjørn Munk*, Stefano Cerri*, Sebastian Nørgaard Llambias, Jakob Ambsdorf, Julia Machnio, Vardan Nersesjan, Christian Hedeager Krag, Peirong Liu, Pablo Rocamora García, Mostafa Mehdipour Ghazi, Mikael Boesen, Michael Eriksen Benros, Juan Eugenio Iglesias, Mads Nielsen
PreprintIntroduces FOMO300K, a large-scale dataset for pretraining foundation models on brain MRI from more than 900 sources.
arXiv -
General Methods Make Great Domain-specific Foundation Models: A Case-study on Fetal Ultrasound
Asbjørn Munk*, Jakob Ambsdorf*, Sebastian Llambias, Anders Nymark Christensen, Kamil Mikolaj, Randall Balestriero, Martin Tolsgaard, Aasa Feragen, Mads Nielsen
MICCAI 2025Introduces UltraDINO, a foundation model for fetal ultrasound trained on 2M images based on iBOT and DINOv2, achieving state-of-the-art performance on fetal ultrasound classification and segmentation tasks.
arXiv · GitHub -
AMAES: Augmented Masked Autoencoder Pretraining on Public Brain MRI Data for 3D-Native Segmentation.
Asbjørn Munk*, Jakob Ambsdorf*, Sebastian Llambias, Mads Nielsen
ADSMI @ MICCAI 2024Efficient pretraining for 3D segmentation models using MAE and augmentation reversal on a large brain MRI dataset. Introduces 🧠BRAINS-45K, at the time, the largest pretraining dataset used to train a foundation model for brain MRI.
Website · arXiv · GitHub
Projects
-
FOMO[25|26]: Foundation Model Challenges for Brain MRI at MICCAI
Leading the Foundation Model Challenge at MICCAI for Brain MRI. We hosted the first edition in 2025 (fomo25.github.io) and the second in 2026 (fomo26.github.io).
-
Asparagus: A configurable framework for large-scale medical imaging
I am contributing to Asparagus, an easily configurable and extendable framework built for pretraining, training, finetuning, and evaluating classification, regression, and segmentation models on medical imaging data.
GitHub -
Yucca: A modular deep learning framework for 3D medical data
I am contributing to Yucca, a framework that aims to make it effortless for researchers to reach SOTA performance (i.e. nnUNet performance) with minimal code complexity. Focus is on 3D data, in particular CT and MRI.
arXiv · GitHub