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

TL;DR

This paper presents a novel approach for 3D brain MRI analysis using a U-Net CNN that outperforms transformer-based models in speed, size, and accuracy, demonstrated through winning the MICCAI 2025 challenges.

Contribution

The authors developed a lightweight, fast-training CNN model leveraging domain knowledge, achieving top performance in brain MRI challenges and sharing their models publicly.

Findings

Model trained 1-2 orders of magnitude faster.

Models are 10 times smaller than transformer approaches.

Achieved first place in MICCAI 2025 brain MRI challenges.

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.