Three SRH images from biopsies: lymphoma, piilocytic astrocytomas brain tumor, and front lobal lesion (from l to r). Photo: Daniel Orringer/NYU Langone Health
In a clinical trial, a team of doctors found a combination of deep-learning algorithms and laser-imaging technology was as effective as pathologists — but faster — at diagnosing brain tumors in near real time, according to a study published Monday in Nature Medicine.
Why it matters: There is a shortage of pathologists and an increase in demand, with roughly 15.2 million people diagnosed with cancer each year globally, most of whom will have surgery. In the U.S. alone, there are roughly 1.1 million biopsies annually.
- "Surgeons can analyze tissue on a microscopic scale and use this information to guide decision-making. This is important because it means we can perform safer, more complete cancer surgeries," co-author Daniel Orringer told Axios.
The big picture: Traditional brain tumor biopsy surgery is intensive in time, resource and labor, says Orringer, who is an associate professor of neurosurgery at NYU Langone Health.
What they did: The research team developed an AI model trained on more than 2.5 million images. They combined the algorithm with a laser-based optical imaging technique called stimulated Raman histology (SRH). They tested this combination on 278 brain tumor patients at three hospitals for the 10 most common types of brain cancer.
What they found: In 150 seconds, the algorithm predicted the brain tumor diagnosis with a 94.6% accuracy, compared with the 93.9% accuracy of the traditional diagnosis that generally takes 20–30 minutes.
Yes, but: The algorithm is trained to recognize the 10 most common brain tumors, which limits its value in diagnosing all types of brain tumor, Orringer adds.
- "To address this issue, we developed a separate algorithm to determine whether a given tumor matches one of the 10 categories ... If not, it's a good signal to the clinicians that the AI may not be reliable and that they might be dealing with a rare tumor."
What they're saying: Manish Aghi, professor of neurosurgery at UC San Francisco who was not involved in the study, says it improves upon brain tumor diagnostics and could eventually help with pathology shortages.
What's next: Orringer expects further research will show the benefits of using the technology for diagnosis in multiple types of head and neck and GI tumors.
The bottom line: "The AI algorithm is not FDA approved and is not being used currently in clinical practice. Our report suggests that if it were implemented it would have great value in guiding surgical decision-making," Orringer tells Axios.