Nature Medicine · 10 July 2026 · From issue 08
Health system learning enables generalist neuroimaging models
A brain scan AI trained on real hospital data beat top general AI systems at reading scans.
Akhil Kondepudi, Akshay Rao et al.
A computer model learned to read brain scans by studying millions of real hospital images. It now beats today's top general AI systems at this task. Its makers call this method "health system learning." General AI systems trained on public internet data do poorly with brain scans. This happens because privacy rules keep most medical scans off the public internet. The new model, called NeuroVFM, was trained on over 5 million brain scans from real patient care. It can spot disease patterns and write radiology reports. When paired with a separate tool that writes text, it produced reports that were more accurate. Experts also preferred these reports over ones from leading general AI systems. The model also made fewer false or missed findings.
Integrity screen: passed (3 checks) Checked 23 August 2026. Read the source
The plain-language reading of this paper is in the full issue, for subscribers.
Read the full issue