Revolutionizing Brain Tumor Diagnosis: AI System "Hetairos" Matches Subtypes from Standard Tissue Slides
Researchers from the German Cancer Research Center (DKFZ) and Heidelberg University have developed a powerful new artificial intelligence system named "Hetairos." This tool can classify brain tumors into their precise molecular subtypes using only standard microscopic tissue sections.
System Overview
Hetairos was trained on a massive dataset of over 11,000 digitized tissue sections from 9,606 patients across eleven medical centers on four continents. The system is remarkably comprehensive, distinguishing 102 different molecular tumor subtypes—covering nearly the entire current World Health Organization (WHO) classification for central nervous system tumors.
A key advantage is that Hetairos requires only digitized versions of routinely prepared and stained tissue sections, eliminating the need for specialized and expensive molecular testing equipment.
Performance Metrics
When the AI is highly confident in its diagnosis—which occurs in 50–70% of cases—its accuracy is a robust 87–88% . When uncertain, the system narrows its results to a few likely candidates, which can guide neuropathologists in their further investigation.
In a direct head-to-head test against five experienced neuropathologists on 210 cases, the results were striking:
- Hetairos achieved 68% accuracy, while the specialists' average was just 30% .
- When considering the top three most likely diagnoses, Hetairos scored 84% , compared to the specialists' average of 50% .
Speed is another major advantage. In a prospective study, Hetairos analyzed 210 tumor samples in just 12 minutes on standard computer hardware. In contrast, complete molecular diagnostics took approximately 12 days. Including preparation and digitization, results could be available within 24 hours to two days.
Limitations
The system is not without its challenges. It struggles with very rare tumor types, where experienced neuropathologists perform at least as well. Researchers are confident that performance will improve as the system is trained on larger and more diverse datasets.
Purpose and Intended Use
Hetairos is explicitly designed as a diagnostic support tool, not a replacement for molecular analysis. It can highlight the specific tissue areas that were most important for its decision, helping pathologists understand its reasoning and guiding further investigation.
This tool is particularly valuable when molecular methods are limited due to insufficient tumor material, unclear results, or a lack of resources.
Economic and Accessibility Aspects
Current DNA methylation analysis can cost several hundred euros per test. Hetairos, by contrast, works with existing, routinely prepared tissue sections.
- The technology uses standard hematoxylin and eosin (H&E) stained sections, which are globally available and inexpensive.
- It requires only a digital scanner to prepare the slides for analysis.
This approach could dramatically broaden access to state-of-the-art brain tumor diagnostics, particularly in regions with limited financial resources or specialized laboratory infrastructure.