Experts in Heidelberg have developed an AI system that can classify brain tumors with unprecedented accuracy using standard microscopic tissue sections. The system uses digitized standard stains to identify more than 100 molecular subtypes of central nervous system tumors and provides results within minutes, potentially accelerating brain tumor diagnosis around the world.
Brain and spinal cord tumors are highly variable. In recent years, it has become clear that many of these tumors can only be reliably diagnosed when examining their molecular characteristics in addition to their microscopic appearance. Of particular importance here is the so-called DNA methylation analysis, which is currently considered the gold standard for accurately classifying many brain tumors.
However, such tests are complex and require specialized laboratories, expensive equipment, and sufficient tumor material. Also, it often takes about two weeks for results to appear. In many parts of the world, the necessary technology is not even available.
AI learns from over 11,000 tissue sections
A new AI system called “Hetairos” is expected to bring significant improvements. It was developed by a team led by Moritz Gerstung (German Cancer Research Center, DKFZ) and Felix Sahm (Heidelberg Medical University and Heidelberg University Hospital). The goal of this project was to predict which molecular subgroup a tumor belongs to based solely on routinely prepared and stained tissue sections.
Hetailos was trained and validated using more than 11,000 digitized tissue sections from 9,606 patients. Diagnosis was primarily determined using DNA methylation diagnostics. Data were obtained from 11 medical centers on four continents. In total, Hetairos identified 102 different molecular tumor subtypes, covering nearly the entire spectrum of the current WHO classification of central nervous system tumors.
AI not only evaluates a diagnosis, but also indicates how confident it is in that diagnosis. In approximately 50-70% of all cases, Hetairos made predictions with high certainty. In these cases, the accuracy was approximately 87-88%. Even when the AI was uncertain, it was usually able to significantly narrow down the number of possible diagnoses.
Instead of differentiating over 100 tumor subtypes, Hetailos often provides neuropathologists with only a few likely candidates. This greatly simplifies the selection of further diagnostic tests.
This study shows that artificial intelligence can derive molecular information directly from routine tissue sections, thereby fundamentally changing cancer diagnosis. ”
Dalui Jin, one of the study’s lead authors
Hetairos outperform experienced specialists
Of particular note is the direct comparison with human experts. Five experienced neuropathologists from different international centers were given 210 cases and asked to make a diagnosis based solely on histological sections. Hetairos’ accuracy rate was 68%, while specialists’ average accuracy rate was 30%. Considering the three most likely diagnoses for each case, the AI’s score was 84 percent and the expert’s score was about 50 percent.
“These results show that modern AI systems can now recognize very subtle morphological patterns that are difficult to distinguish even by experienced experts,” said Felix Sahm.
“At the moment, the diagnosis of very rare tumor types remains a major challenge for Hetairos. Experienced neuropathologists seem to be at least on par in this respect. However, we expect that larger and more diverse datasets will further improve the performance of the system,” added Moritz-Gerstung.
Diagnosis in 12 minutes instead of 12 days
In a prospective study, Hetailos was used in parallel with routine clinical practice. The system analyzed 210 tumor samples without the AI results influencing actual diagnosis or treatment decisions.
While a full molecular diagnosis took an average of about 12 days, Hetailos produced results in just 12 minutes on standard computer hardware after digitizing stained tissue sections. Results are often available within 24 hours to 2 days, including tissue section preparation and digitization.
Support for difficult or unclear cases
Hetairos may be particularly valuable when traditional molecular methods have reached their limits, when there is insufficient tumor material for genetic testing, or when molecular testing does not yield unambiguous results. Additionally, the system highlights areas within the tissue section that are particularly important for determination. This allows physicians to understand the basis for AI diagnoses and identify which areas are appropriate for further investigation.
“We developed Hetairos primarily as a tool to support diagnosis,” explains neuropathologist Felix Sahm. “It is intended specifically to complement and accelerate molecular analysis, rather than replace it. As this technology is based on standard tissue sections used around the world, it has the potential to make an important contribution, especially in countries and regions with limited resources.”
This method may also have economic benefits. DNA methylation analysis typically costs hundreds of euros, but Hetailos uses existing tissue sections for analysis.
Moritz Gerstung admits: “Hetairos demonstrates the huge potential of AI-assisted digital pathology to provide rapid and widely available diagnostic methods that were previously not possible without significant technological efforts.”
sauce:
German Cancer Research Center (German Cancer Research Center, DKFZ)
Reference magazines:
Jin D, others. (2026) Hetailos is a histology-based artificial intelligence model for predicting methylation subtypes of central nervous system tumors. natural cancer. DOI: 10.1038/s43018-026-01186-3. https://www.nature.com/articles/s43018-026-01186-3
