Overview
- Researchers published a multicenter Nature Cancer study in June 2026 that prospectively evaluated Hetairos on 210 tumor samples without using the AI results to change patient care.
- The team trained and validated Hetairos on more than 11,000 digitized H&E sections from 9,606 patients across 11 centers on four continents to map routine histology to molecular subtypes.
- In a head-to-head test the AI outscored five experienced neuropathologists, with Hetairos reaching 68% top-1 accuracy versus a 30% average for the specialists and 84% top-3 accuracy versus about 50% for the experts.
- Hetairos returns predictions in roughly 12 minutes after slide digitization, reports a confidence score that is correct about 87–90% of the time for high-confidence calls, and highlights tissue regions that drove each prediction.
- Developers frame Hetairos as a diagnostic support and triage tool that could broaden access and cut time and cost where DNA methylation testing is scarce, but they warn performance lags for very rare subtypes and call for larger, more diverse datasets and further validation.