At the Super Bowl of gastroenterology—the 50th annual Digestive Diseases Week in San Diego—some of the 14,000 attendees crowded the display booth of Virgo Surgical Video Solutions.
What held them captivated was not just another video feed of a simulated colonoscopy as Virgo software engineer Ryon Sabouni fed an endoscope—a flexible tube with a camera and a light on the end—through a model colon. It was what happened whenever the scope spotted a simulated polyp. A blue rectangle appeared, highlighting the polyp and keeping it framed whenever it was in view.
To Dr. Mankanwal Sachdev, a gastroenterologist at the Arizona Centers for Digestive Health, part of Covenant Surgical, the demo looked like the future.
“AI is coming to everything we do,” Dr. Sachdev said. He, for one, welcomes the technology. “There’s up to a 3% miss rate, even in the best hands,” he says of spotting polyps without help from AI. In other words, three out of every hundred patients have precancerous growths that go undetected through colonoscopies, the best detection method. That significantly increases their risk of becoming one of the more than 144,000 people developing colorectal cancer in the U.S. this year, and ultimately one of the more than 50,000 people who die from it.
“You need all the help you can get. If you’re going to put a patient through this semi-invasive procedure and take a day off from work, you have to be as effective as you can,” Dr. Sachdev said. Rather than seeing AI as a potential replacement for the work of doctors, he believes it could become another tool to help them do their jobs better.
“There’s up to a 3% miss rate, even in the best hands… If you’re going to put a patient through this semi-invasive procedure, you have to be as effective as you can,”
Dr. Mankanwal Sachdev
Gastroenterologist at the Arizona Centers for Digestive Health
Early detection to save lives
Colorectal cancer is the second most lethal cancer, after breast cancer, according to the Centers for Disease Control. That’s why gastroenterologists perform some 15 million colonoscopies in the United States alone to spot the abnormalities known as polyps before they can turn into cancer.
Nevertheless, some polyps inevitably go undetected. “Some polyps are so subtle,” says Dr. Sachdev. “It’s almost like trying to find a scorpion in your backyard or a spider.” Virgo aims to increase the odds with its surgical video system. “We are super excited about the potential for running computer vision systems on the edge,” says Virgo CEO and co-founder Matt Schwartz.
Virgo’s video capture device is about the size and shape of a small home WiFi router. It plugs into a standard endoscopy system for use in a wide range of procedures, including colonoscopies. One of its key benefits is its ability to remain tucked away, with few added cables. Schwartz appreciates the small form factor and low power consumption of the Coral Dev Board while still enabling real-time AI inference. He says the company will likely move to Coral’s even smaller System-on-Module units for production.
The company’s team of just five employees put together the demo in only five weeks, with the help of Coral’s platform for local AI. Schwartz says the company doesn’t yet have a timeline for human trials of the technology but that the company’s mission is to bring this and other AI-powered functionality to endoscopies over the next few years. “We’re also developing detection and classification models beyond just polyp detection for things like Barrett’s esophagus, instrument detection, and technique analysis,” Schwartz explained.
“It’s a really fun system to use,” says Schwartz of the Coral AI platform. “We were blown away by how we were able to spin up a demo so rapidly based on the training data that we had. The entire project was a blast.”
To be able to do something like that in about two weeks is definitely very encouraging…
CEO & Founder, Virgo
Real-time detection at the edge
Until the demo at the conference, the Virgo system relied on AI in the cloud for added functionality, initially to trim lengthy videos of colonoscopies for download after the procedure. With the Digestive Diseases Week demo, however, the Virgo team brought AI down from the cloud to the edge. “We used an object detection model that was specifically trained on a subset of our video data,” said Schwartz.
This lets the system analyze visual information in real-time rather than waiting for processing in the cloud. Since the system generates high resolution, 1080p video at up to 60 frames per second and files up to 2GB per procedure, processing in the cloud for real-time polyp detection isn’t possible, Schwartz says. But processing at the edge allows the system to detect polyps in about 17 milliseconds, fast enough to seem instantaneous for practitioners.
The result is a kind of AI copilot for practitioners as they perform procedures, potentially reducing the need for follow-ups to catch missed polyps,depending on how effective it turns out to be in actual practice beyond the demonstration stage.
AI processing at the edge is essential for real-time computer vision processing because the relatively rapid movements of an endoscope don’t allow time for uploading video data to the cloud for processing. Processing at the edge also alleviates privacy concerns that might come with uploading and processing sensitive medical data remotely. Using Coral’s AI-at-the-edge technology enables Virgo to bring all the processing into the room with the patient, where it can stay if required by hospitals and patients.
“To be able to do something like that in only a few weeks is definitely very encouraging to us for what we’re going to be able to do when we start using our larger data set and working towards real-time human use applications,” says Schwartz.
Even without AI at the edge in the system’s current production models, Virgo’s video recording capabilities — which have now captured over 60,000 procedure videos — have proven invaluable to Sachdev and his colleagues for educating each other and sharing video with patients. But, he says, additional AI-driven functionality, like that enabled by Coral’s platform, will take his practice to the next level. “I think it would make me a better endoscopist.” Not to mention improve the lives of those up-to-3% of patients whose precancerous growths go undetected during colonoscopies today.
If you’re interested in how Coral can help you build your next product please contact us.