Year: 2026
Type: preprint
Abstract: Clinical deployment of automated brain MRI analysis faces a fundamental challenge: clinical data is heterogeneous and noisy, and high-quality labels are prohibitively costly to obtain. Self-supervised learning (SSL) can address this by leveraging the vast amounts of unlabeled data produced in clinical workflows to train robust \textit{foundation models} that adapt out-of-domain with minimal supervision. However, the development of foundation models for brain MRI has been limited by small pretraining datasets and in-domain benchmarking focused on high-quality, research-grade data. To address this gap, we organized the FOMO25 challenge as a satellite event at MICCAI 2025. FOMO25 provided participants with a large pretraining dataset, FOMO60K, and evaluated models on data sourced directly from clinical workflows in few-shot and out-of-domain settings. Tasks covered infarct classification, meningioma segmentation, and brain age regression, and considered both models trained on FOMO60K (method track) and any data (open track). Nineteen foundation models from sixteen teams were evaluated using a standardized containerized pipeline. Results show that (a) self-supervised pretraining improves generalization on clinical data under domain shift, with the strongest models trained \textit{out-of-domain} surpassing supervised baselines trained \textit{in-domain}. (b) No single pretraining objective benefits all tasks: MAE favors segmentation, hybrid reconstruction-contrastive objectives favor classification, and (c) strong performance was achieved by small pretrained models, and improvements from scaling model size and training duration did not yield reliable benefits.
Source: arXiv (Cornell University)
Authors: Munk, Asbjørn, Cerri, Stefano, Nersesjan, Vardan, Krag, Christian Hedeager, Ambsdorf, Jakob, García, Pablo Rocamora, Machnio, Julia, Liu, Peirong, Ahn, Suhyun, Akbari, Nasrin, Khalil, Yasmina Al, Amador, Kimberly, Amirrajab, Sina, Arbel, Tal, Cuadra, Meritxell Bach, Baid, Ujjwal, Baheti, Bhakti, Banus, Jaume, Barbierik, Kamil, Brune, Christoph, Bu, Yansong, Callard, Baptiste, Chen, Yuhan, Crijnen, Cornelius, Dancette, Corentin, Drotar, Peter, Dutande, Prasad, Forkert, Nils D., Garg, Saurabh, Gazda, Jakub, Gazda, Matej, Gérin, Benoît, Ghosh, Partha, Gong, Weikang, Gordaliza, Pedro M., Hashemi, Sam, Heimann, Tobias, Jia, Fucang, Jiang, Jiexin, Kaczmarek, Emily, Kang, Chris, Kang, Seung Kwan, Khazaei, Mohammad, Khlaut, Julien, Koutsouvelis, Petros, Lee, Jae Sung, Li, Yuchong, Lyu, Mengye, Ma, Mingchen, Madabhushi, Anant, Maier-Hein, Klaus H., Manceron, Pierre, Mora, Andrés Martínez, Mazher, Moona, Meister, Felix, Molchanova, Nataliia, Niederer, Steven A., Nürnberg, Leonard, Park, Jinah, Qayyum, Abdul, Richiardi, Jonas, Saporta, Antoine, Setlak, Branislav, Shen, Ning, Szeto, Justin, Ulrich, Constantin, Vaish, Puru, Vigneshwaran, Vibujithan, Volmer, Leroy, Wang, Zihao, Wei, Siqi, Winder, Anthony, Wolterink, Jelmer M., Wynen, Maxence, Yang, Chang, Yie, Si Young, Ghazi, Mostafa Mehdipour, Pai, Akshay, Solem, Espen Jimenez, Llambias, Sebastian Nørgaard, Boesen, Mikael, Benros, Michael Eriksen, Iglesias, Juan Eugenio, Nielsen, Mads
Institutions: University of Copenhagen, Pioneer (United States), Johns Hopkins University, Korea Advanced Institute of Science and Technology, Artificial Intelligence in Medicine (Canada), Centre Hospitalier Universitaire Vaudois, University of Lausanne, University of Twente, Siemens (Slovakia), Technical University of Košice, German Cancer Research Center, Fudan University, Centre d'Imagerie BioMedicale, Maastro Clinic, Eindhoven University of Technology, Chinese Academy of Sciences, Shenzhen Institutes of Advanced Technology, University College London, Siemens (Germany), Siemens Healthineers (Germany), Maastricht University, Imperial College London, University of British Columbia, Mila - Quebec Artificial Intelligence Institute, McGill University, University of Calgary, Hotchkiss Brain Institute
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7154321941