Hospitals can now train one shared AI together — without ever handing over a single patient scan

A hospital's scans are among the most private data in medicine, which is why medical AI usually learns from one hospital at a time. Researchers at the Karlsruhe Institute of Technology have shown a different route, called federated learning: each hospital trains the model on its own images, and only the model's improvements travel between institutions — the scans never leave home. Across studies in medical-imaging journals, the approach produced usable synthetic CT images for planning cancer radiotherapy, and more efficient training cut computing time and carbon emissions by 9% to 23% with no loss of image quality. It is groundwork, not a clinic-ready tool — and it points at collaboration without exposure.

SDG 3 · Good health and well-being — Mechanism: tech innovation