BrainDINO: A Brain MRI Foundation Model for Generalizable Clinical Representation Learning

2026-04-29

AI Summary

This work proposes BrainDINO, a self-supervised learning framework that learns general-purpose representations from 6.6 million unlabeled axial brain MRI slices without requiring volumetric pretraining or full-network fine-tuning. Addressing the limited generalizability of existing task-specific methods that rely heavily on annotated data, BrainDINO leverages a self-distillation architecture to train a frozen encoder, which is then paired with lightweight task-specific heads for downstream applications. The approach demonstrates superior performance over both natural-image and MRI-specialized baselines across diverse tasks—including tumor segmentation, brain age estimation, and disease classification—exhibiting strong sensitivity to anatomical and pathological features. Notably, BrainDINO excels in label-scarce scenarios, highlighting its potential as a robust foundation model for neuroimaging analysis.

Abstract

Brain MRI underpins a wide range of neuroscientific and clinical applications, yet most learning-based methods remain task-specific and require substantial labeled data. Here we show that a single self-supervised representation can generalize across heterogeneous brain MRI endpoints. We trained BrainDINO, a self-distilled foundation model, on approximately 6.6 million unlabeled axial slices from 20 datasets encompassing broad variation in population, disease, and acquisition setting. Using a frozen encoder with lightweight task heads, BrainDINO supported transfer across tumor segmentation, neurodegenerative and neurodevelopmental conditions classification, brain age estimation, post-stroke temporal prediction, molecular status prediction, MRI sequence classification, and survival modeling. Across tasks and supervision regimes, BrainDINO consistently equaled or exceeded natural-image and MRI-specific self-supervised baselines, with particularly strong advantages under label scarcity. Representation analyses further showed anatomically organized and pathology-sensitive feature structure in the absence of task-specific supervision. Our findings indicate that large-scale slice-wise self-supervised learning can yield a unified brain MRI representation that supports diverse neuroimaging tasks without volumetric pretraining or full-network fine-tuning, establishing a scalable foundation for robust and data-efficient brain imaging analysis.

Problem

• brain MRI
• generalizable representation
• self-supervised learning
• label scarcity
• clinical tasks

Innovation

• self-supervised learning
• foundation model
• brain MRI
• representation learning
• transfer learning