AI joins the scalpel in the brain surgeon’s toolbox

A new system analyzes multiple data types, including tumors, during surgery to help physicians make accurate life-and-death decisions.

In the news

For surgeons operating to excise brain tumors, there comes a moment in the already-fraught process in which they face an agonizing choice: cut away healthy tissue in an effort to remove the entire tumor, or leave as much healthy tissue as possible and risk leaving part of the growth?

A team in the Netherlands says it has developed an artificial intelligence system that delivers key information to surgeons much more quickly than before, improving their chances of reaching the best decision. And not to be outdone by the brain surgeons, a team of California cardiologists say they can use AI to detect AFib—an irregular heart rhythm—before an attack occurs.

The critical factor is AI’s ability to process and analyze vast amounts of data quickly. In deciding how aggressively to operate, surgeons require DNA-encoded information on the tumor itself. Typically, this would involve examining samples under a microscope. More thorough analysis is possible through genetic sequencing, but it’s not universally available and takes weeks to perform.

The newly developed AI tool, however, can begin determining the tumor’s type and subtype early in the surgery—and reach a decision in time to assist the surgeon during the lengthy procedure.

The story is similar in the AFib detection. Every electrocardiogram, or EKG, creates about 20,000 data points. Generative AI’s data-crunching abilities make it an excellent tool for analyzing this data, comparing it with known cases of AFib, and determining risk.

According to the Dutch development team, the AI system accurately diagnosed 45 of 50 cases within 40 minutes during early testing on frozen tumor samples, and then 18 of 25 within 90 minutes during “live” tests of actual in-process surgeries.

Not everyone in the medical community is ready to embrace the AI tool. Some say it requires specialized knowledge that will be hard to reproduce. Also, brain tumors are easier to analyze than most types of tumors using this particular method.

The Cognizant take

Perhaps the most exciting thing about this AI system is that it analyzes actual tissue samples, says Niloy Chakrabarty, Senior Director, Healthcare Consulting at Cognizant. Traditional machine learning models rooted in computer vision “are hardly new,” he points out, and indeed are used extensively in the healthcare field.

The Dutch tool, by contrast, is an example of multimodal AI; it can use multiple data types to “create more accurate determinations and really draw conclusions,” Chakrabarty notes, to make more precise predictions—indeed, potentially life-or-death predictions—in the brain surgery use case.

The convergence of computer vision-based models with multimodal capability has considerable potential, he adds. “Now you can overlay generative AI on computer vision output,” Chakrabarty says. "You can work with other clinical data, you can create a proper radiological note—this is very promising.”

Nevertheless, he cautions, “You’ve got to take risk into consideration when implementing it. AI will never replace the physician. Rather, it’s a tool to assist the physician.”