en.Wedoany.com Reported - A new artificial intelligence system, Hetairos, can predict the molecular subtypes of brain tumors using only routinely prepared and stained histological slides, reducing diagnosis time from approximately two weeks to tens of minutes. The system was developed by a team led by Moritz Gerstung from the German Cancer Research Center (DKFZ) and Felix Sahm from the Heidelberg Medical Faculty of Heidelberg University and Heidelberg University Hospital.
Brain and spinal cord tumors are highly diverse. In recent years, the medical community has established that many tumors must be examined for both their molecular characteristics and microscopic appearance to achieve a reliable diagnosis. DNA methylation analysis is considered the gold standard for accurately classifying many types of brain tumors. However, this testing process is complex, requiring specialized laboratories, expensive equipment, and sufficient tumor material, and it typically takes about two weeks to obtain results. In many parts of the world, the necessary technical conditions are even lacking.
The Hetairos system was trained and validated using over 11,000 digitized tissue sections from 9,606 patients, with diagnoses primarily determined through DNA methylation analysis. The data came from eleven medical centers across four continents. The system can distinguish a total of 102 different molecular tumor subtypes, covering nearly the entire scope of the current World Health Organization (WHO) classification of central nervous system tumors.
The AI not only makes a diagnosis but also indicates the confidence level of its judgment. In approximately 50% to 70% of all cases, Hetairos can make predictions with high certainty, and in these high-confidence cases, the accuracy rate is about 87% to 88%. Even when the AI's judgment is uncertain, it can typically significantly narrow down the possible diagnoses. Instead of having to distinguish from over a hundred tumor subtypes, Hetairos usually provides neuropathologists with just a few possible candidates, which can greatly simplify the selection of subsequent diagnostic tests.
In a direct comparison with human experts, five experienced neuropathologists from different international centers were given 210 cases and asked to make diagnoses based solely on tissue sections. Hetairos achieved an accuracy of 68%, while the experts averaged 30%. When considering the three most likely diagnoses for each case, the AI scored 84%, compared to about 50% for the experts. Darui Jin, one of the co-lead authors of the study, stated that the research demonstrates artificial intelligence's ability to extract molecular information directly from routine tissue sections.
In a prospective study, Hetairos was used in parallel with routine clinical practice to analyze 210 tumor samples, with the AI results not influencing actual diagnosis or treatment decisions. Complete molecular diagnosis took an average of about twelve days, whereas Hetairos generated results in just twelve minutes on standard computer hardware after the stained tissue sections were digitized. Including the preparation and digitization of tissue sections, results can typically be obtained within 24 hours to two days.
The system may be particularly valuable in cases where traditional molecular methods reach their limits, when there is insufficient tumor material for genetic testing, or when molecular tests do not yield clear results. It also highlights areas of the tissue section that are particularly important for decision-making, helping doctors understand the basis of the AI's diagnosis. The development team stated that the technology is intended as a tool to support diagnosis, not to replace molecular analysis, but to complement and accelerate it in a targeted manner. It could make a significant contribution, especially in countries or regions with limited resources, as it is based on globally standard tissue sections. From an economic perspective, a single DNA methylation analysis typically costs several hundred euros, while Hetairos analyzes existing tissue sections.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
