AI System Identifies Brain Tumors in Minutes Instead of Weeks

Researchers in Heidelberg, Germany, have developed an artificial intelligence system capable of classifying brain tumors with remarkable accuracy using standard microscopic tissue samples. The technology, known as Hetairos, can identify more than 100 molecular subtypes of central nervous system tumors within minutes, potentially transforming brain cancer diagnostics worldwide. The findings were published in the journal Nature Cancer.

Brain and spinal cord tumors are highly diverse, and accurate diagnosis increasingly depends not only on microscopic examination but also on molecular analysis. DNA methylation profiling is currently considered the gold standard for classifying many brain tumors, yet the process is expensive, requires specialized laboratories, and often takes up to two weeks to deliver results.

To address these limitations, scientists from the German Cancer Research Center (DKFZ), Heidelberg University, and Heidelberg University Hospital developed Hetairos, an AI model trained to predict a tumor’s molecular subtype directly from routinely prepared histological slides.

The system was trained and validated using more than 11,000 digitized tissue samples from 9,606 patients collected across 11 medical centers on four continents. Hetairos can distinguish 102 molecular tumor subtypes, covering nearly the entire current WHO classification of central nervous system tumors.

In cases where the AI expressed high confidence in its predictions—roughly 50% to 70% of all samples—it achieved an accuracy rate of approximately 87% to 88%. Even when confidence was lower, the system significantly narrowed the range of possible diagnoses, helping specialists select the most appropriate follow-up tests.

The researchers also compared Hetairos with experienced neuropathologists. In an assessment involving 210 cases, the AI achieved a diagnostic accuracy of 68%, while five specialists averaged 30%. When the three most likely diagnoses were considered, Hetairos reached 84% accuracy compared with about 50% for the human experts.

A prospective clinical study demonstrated the system’s speed advantages. While conventional molecular diagnostics required an average of 12 days, Hetairos generated results in just 12 minutes after digitalization of tissue slides. Including sample preparation, results could often be available within one to two days.

According to the developers, Hetairos is designed to support rather than replace molecular testing. The technology could be particularly valuable in regions with limited access to advanced diagnostic laboratories and may also reduce costs by using standard tissue sections already prepared for routine pathology.

Follow NEWS.am Medicine on Facebook and Twitter