Heidelberg AI cuts brain‑tumor diagnosis from days to minutes

Researchers at the German Cancer Research Center (DKFZ) and the University of Heidelberg have created an artificial‑intelligence system called Hetairos that can classify brain and spinal‑cord tumours from a routine stained tissue slide. The algorithm distinguishes 102 molecular sub‑types in an average of twelve minutes, reducing the typical diagnostic interval from twelve days to a matter of minutes.

Hetairos was trained on more than 11,000 digitised slides from 9,606 patients collected at eleven cancer centres on four continents, using DNA‑methylation profiling as the reference standard. In internal validation the system matched the reference in 75 % of cases on its first prediction and reached 87 % accuracy when the top three predictions were considered. In a blinded test against five experienced neuropathologists, Hetairos gave the correct first‑choice diagnosis in 68 % of 210 cases, compared with about 30 % for the human experts. The method requires only basic microscopy, simple staining and standard computer hardware, making it potentially transformative for regions lacking specialised molecular‑diagnostic facilities. The results were published in Nature Cancer.