From 100,000+ images to winning the first brain MRI foundation model challenges: Sharing lessons and models
🤖 AI Summary
This work addresses the lack of efficient and lightweight foundation models for 3D brain MRI analysis by proposing a U-Net-based CNN architecture that integrates anatomical priors and domain-specific neuroimaging knowledge. Leveraging self-supervised learning, tailored preprocessing, and optimized training strategies, the model achieves competitive performance while accelerating training by 10–100× and reducing model size to one-tenth that of comparable Transformer-based approaches, substantially lowering computational costs. The method secured first place in both the MICCAI 2025 SSL3D and FOMO25 Brain MRI Foundation Model Challenges and has been publicly released.
Application Category
📝 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.
Problem
Research questions and friction points this paper is trying to address.
Foundation Models
brain MRI
medical image analysis
radiological tasks
3D imaging
Innovation
Methods, ideas, or system contributions that make the work stand out.
Foundation Model
U-Net CNN
Anatomical Priors
3D Brain MRI
Efficient Training
🔎 Similar Papers
No similar papers found.
