In a major development in how tumors are excised, researchers at the University of Michigan have shown that it’s possible to accurately analyze brain tumor tissue within the operating room and assess its nature using artificial intelligence.
Tumor tissues typically look just like the healthy stuff around them. When a tumor is removed, parts that are near the edges (margins) are sent to the pathology lab for review. After staining and observations using a microscope, the pathologist can let the surgical team know whether it removed all of the tumor or left some behind. This takes a long time, so much so that typically a follow-up surgery is required if the margins are not completely excised.
The new technology comes in the form of the NIO Imaging System from Invenio, a company out of Santa Clara, California. It uses stimulated Raman histology, developed at the University of Michigan, to quickly image tissues at the microscopic scale without any staining, completely bypassing the pathology lab. The technology is so fast that surgeons can take follow up actions that may prevent the tumor from regrowing without having to schedule another costly procedure.
Since surgeons are not pathologists, and even pathologists make mistakes, the imaging system was augmented by artificial intelligence software that was taught to learn what various types of brain tumor look like. This involved feeding a convolutional neural network powering the software with more than 2.5 million tissue sample images from 415 patients. It is impressively accurate and provides a quick prediction of the kind of tissue it is looking at. Evaluated on 278 patients undergoing brain surgery, the new method had a diagnostic accuracy slightly better than conventional histology (94.6% vs 93.9% respectively).
“This is the first prospective trial evaluating the use of artificial intelligence in the operating room,” said Todd Hollon, M.D., lead author of the study appearing in Nature Medicine. “We have executed clinical translation of an AI-based workflow. It’s so quick that we can image many specimens from right by the patient’s bedside and better judge how successful we’ve been at removing the tumor. The surgeon and pathologist determine whether they can make the diagnosis using the SRH image, or whether there is a need to send additional tissue to the pathology lab, the way we used to in the past.”
Here’s a Michigan video about the new technology:
Study in Nature Medicine: Near real-time intraoperative brain tumor diagnosis using stimulated Raman histology and deep neural networks
Product info page: NIO Imaging System