It’s not brain surgery — it’s AI-assisted brain surgery.
Artificial intelligence is revolutionizing health care, and surgeons operating on brain tumors need the speed and accuracy that AI can deliver.
When a patient is on the operating table with their skull cut open and their brain exposed, surgeons have to know exactly what type of tumor they’re operating on — and quickly.
The surgeon also need to avoid cutting into brain tissue while remove the tumor (or glioma, the most common type of brain tumor).
“This process is not error proof,” Dr. Kun-Hsing Yu, a professor at Harvard Medical School, told the Guardian.
“When operating on brain cancer patients, doctors send a piece of sample [tissue] to the pathology,” he explained. “A pathologist can help tell them whether they are cutting the correct tissue, or what kind of specific cancer the patient has.”
But that process can take 15 minutes or longer, and mistakes can and do occur.
“People are under stress, and the quality of the slide is sometimes not great, so occasionally we will have misdiagnosis arising from this fast process,” Yu said.
That’s where the rapid processing speeds — and the learning ability — of AI can assist surgeons while the patient is still on the operating table.
There are three main types of gliomas, and each has a different molecular makeup, a different pattern of growth and speed and, therefore, differing forms of treatment.
Yu and his team have developed an AI tool named CHARM, for Cryosection Histopathology Assessment and Review Machine.
CHARM was developed using more than 2,300 brain tumor samples from 1,524 people with glioma. These samples “taught” CHARM what to look for when analyzing tumor samples.
When CHARM was tested on a new set of brain samples, the tool was 93% accurate at spotting tumors with specific molecular mutations and classifying them into the three major types of gliomas, helping to guide treatment.
Perhaps best of all, CHARM is fast — in some cases, it can provide an accurate analysis of tumor cells in less than one second, according to a recent study published in Med.
During surgery, getting an accurate molecular ID of a tumor can tell surgeons what type of drug treatment will work best for that type of tumor. Then, surgeons can implant a drug-coated wafer directly into the brain during the operation.
Medical researchers have already created AI models for other types of cancer, including colon, lung and breast cancers, but gliomas are tricky because of their molecular complexity and the variation in tumor cells’ shape and appearance.
And just like a good doctor, an AI tool like CHARM — if it gets approval from the FDA, which may be several years off — will need updating to learn new information as oncology research evolves.
“Just like human clinicians who must engage in ongoing education and training,” Yu said in a news release, “AI tools must keep up with the latest knowledge to remain at peak performance.”