From 100,000+ images to winning the first brain MRI foundation model challenges: Sharing lessons and models
This work addresses the problem of efficient and effective medical image analysis for radiology, though it appears incremental as it builds on existing U-Net architectures with domain-specific enhancements.
The authors tackled the challenge of developing foundation models for 3D brain MRI analysis by winning first place in both tracks of the SSL3D and FOMO25 competitions, with models that trained 1-2 orders of magnitude faster and were 10 times smaller than transformer-based competitors.
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.