From 100,000+ images to winning the first brain MRI foundation model challenges: Sharing lessons and models

Pedro M. Gordaliza; Jaume Banus; Benoît Gérin; Maxence Wynen; Nataliia Molchanova; Jonas Richiardi; Meritxell Bach Cuadra

From 100,000+ images to winning the first brain MRI foundation model challenges: Sharing lessons and models

Computer Vision and Pattern Recognition 2026-01-21 v1 Artificial Intelligence Machine Learning

Authors: Pedro M. Gordaliza , Jaume Banus , Benoît Gérin , Maxence Wynen , Nataliia Molchanova , Jonas Richiardi , Meritxell Bach Cuadra

View on arXiv ↗ PDF ↗

Abstract

Developing Foundation Models for medical image analysis is essential to overcome the unique challenges of radiological tasks. The first challenges of this kind for 3D brain MRI, SSL3D and FOMO25, were held at MICCAI 2025. Our solution ranked first in tracks of both contests. It relies on a U-Net CNN architecture combined with strategies leveraging anatomical priors and neuroimaging domain knowledge. Notably, our models trained 1-2 orders of magnitude faster and were 10 times smaller than competing transformer-based approaches. Models are available here: https://github.com/jbanusco/BrainFM4Challenges.

Keywords

foundation model brain MRI medical image analysis U-Net CNN anatomical priors self-supervised learning

Cite

@article{arxiv.2601.13166,
  title  = {From 100,000+ images to winning the first brain MRI foundation model challenges: Sharing lessons and models},
  author = {Pedro M. Gordaliza and Jaume Banus and Benoît Gérin and Maxence Wynen and Nataliia Molchanova and Jonas Richiardi and Meritxell Bach Cuadra},
  journal= {arXiv preprint arXiv:2601.13166},
  year   = {2026}
}

Comments

Work presented at the SSL3D Challenge (1st place, ResEnc-L track) and FOMO Challenge (1st place, Methods track) on Brain MRI Foundation Models at MICCAI 2025

From 100,000+ images to winning the first brain MRI foundation model challenges: Sharing lessons and models

Abstract

Keywords

Cite

Comments

Related papers