AI system Hetairos cuts brain tumor diagnosis time to minutes

Researchers have unveiled an artificial‑intelligence tool called Hetairos that can predict the molecular subtype of brain and spinal‑cord tumors in about twelve minutes, replacing the conventional DNA‑methylation analysis that typically takes around twelve days. The system was developed by the German Cancer Research Center (DKFZ) and the University of Heidelberg and was trained on more than 11,000 digitized histology slides from 9,600 patients across four continents.

Hetairos can differentiate 102 molecular subtypes defined by the WHO classification and provides a confidence score for each prediction. In a head‑to‑head trial involving 210 complex cases, the AI achieved a diagnostic accuracy of 68%, far above the 30% average of five senior neuropathologists, and reached 84% accuracy when the top three predictions were considered. For the 50‑70% of cases flagged as high‑certainty, accuracy rose to 87‑88%. The tool also highlights the specific tissue regions that drive its decision, allowing pathologists to verify results. By drastically shortening diagnosis time and reducing reliance on specialized laboratories, Hetairos could expand access to advanced tumor classification, especially in low‑resource settings.