In a development published in Nature Medicine in July 2026, researchers trained a three-dimensional visual foundation model directly on 5.24 million routine clinical computed tomography (CT) and magnetic resonance imaging (MRI) image series. The architecture achieves state-of-the-art diagnosis, enabling preliminary report generation and triage in real health systems.
In Plain English: The Clinical Takeaway
- What happened: Researchers trained a three-dimensional visual foundation model on millions of routine clinical CT and MRI image series instead of relying on foundation models trained on public Internet and medical data.
- Why it matters: This approach allows the model to learn a shared representation of neuroanatomy and disease, resulting in state-of-the-art diagnosis.
- Patient impact: By enabling preliminary report generation and triage in real health systems, these tools aim to assist in clinical workflows.
Learning from Routine Clinical Data
Some foundation models are trained on public Internet and medical data. To bridge this gap, the research team leveraged a repository of 5.24 million routine clinical CT and MRI series. By ingesting this imaging data, the system learned a shared representation of neuroanatomy and disease.
According to the study published in Nature Medicine (doi:10.1038/s41591-026-04567-4), training a three-dimensional visual foundation model directly on clinical volumes allows it to learn a shared representation of neuroanatomy and disease.
Diagnostic Performance and Real-World Triage
When evaluated, the newly trained foundation model demonstrated state-of-the-art diagnosis. Beyond merely classifying images, the system enabled preliminary report generation and triage in real health systems.
| Metric / Feature | Internet and Medical Data Foundation Models | Routine Clinical Data Foundation Models |
|---|---|---|
| Training Dataset Source | Public Internet and medical data | 5.24 million routine clinical CT and MRI series |
| Dimensionality | Not specified | Three-dimensional visual foundation model |
| Clinical Integration | Not specified | Enables preliminary report generation and triage |
Regulatory Landscapes and Translation to Care
Contraindications & When to Consult a Doctor
Future Trajectory in Clinical Informatics
References
- Nature Medicine. (2026). Training a three-dimensional visual foundation model on routine clinical imaging. Published online 31 July 2026. doi:10.1038/s41591-026-04567-4.
Disclaimer: This article is for informational purposes only and does not constitute medical advice, diagnosis, or treatment. Always consult a qualified healthcare provider for any health-related concerns or clinical interpretations.
