In the Netherlands, Researchers Train AI to Diagnose Brain Tumors Before Surgery

Scientists in the Netherlands have taught artificial intelligence to determine the size of a patient's tumour on the operating table. This cutting-edge tool will advise surgeons on how much damaged tissue needs to be removed, as reported by researchers.

This study was published in The New York Times.

The method involves a computer scanning segments of the tumour's DNA and identifying specific chemical modifications. With this data, the type and even subtype of the brain tumor can be diagnosed.

"It's crucial to know the subtype of the tumour during surgery. What we've uniquely enabled now is the ability to perform this very intricate, reliable, detailed diagnosis during the operation," said Dr. Jeroen de Ridder, an associate professor at the Center for Molecular Medicine at UMC Utrecht and lead researcher.

This AI tool has been named "Sturgeon" and was initially tested on frozen tumour samples after prior surgeries. Sturgeon accurately diagnosed 45 out of 50 cases within 40 minutes of commencing genetic sequencing. In five additional cases, it refrained from making a diagnosis as the information was unclear.

Following this, the tool was used in 25 paediatric neurosurgeries, successfully diagnosing 18 patients. Researchers compared its findings with the standard method of examining tumour samples under a microscope.

In many hospitals, determining the tumour type can take weeks. In contrast, Sturgeon takes less than 90 minutes to establish a diagnosis. This is quick enough for a surgeon to make decisions before commencing the operation, according to the researchers.

Dr. de Ridder stated that the model is powerful enough to make diagnoses with rare genetic data, similar to how someone recognizes an image based on just one percent of its pixels from an unknown part of the picture.

"It can figure out by itself what it's looking at and provide a reliable classification," Dr. de Ridder explained.

However, diagnosing some tumours can be challenging. For instance, samples taken during surgery are approximately the size of a corn kernel. If these samples contain a small amount of healthy brain tissue, artificial intelligence might struggle to select a sufficient quantity of tumour-specific markers.

Moreover, there can be variations in tumour cells within a single patient, meaning the sequenced segment might not be representative of the entire tumour.

This new method is part of a broader movement to introduce molecular precision into tumour diagnosis, potentially enabling scientists to develop targeted treatments that are less damaging to the nervous system.