Artificial intelligence (AI) has arrived in the colonoscopy procedure rooms of UCHealth University of Colorado Hospital (UCH) on the Anschutz Medical Campus.
It’s not taking over, but rather assisting gastroenterologists in spotting polyps that can become cancerous.
The AI comes inside Medtronic’s GI Genius intelligent endoscopy module. It’s an unassuming blue box that hooks into standard endoscopy tools in three of the UCH’s five colonoscopy procedure rooms, but it packs some serious diagnostic power.
“I don’t see how this isn’t standard of care at some relatively short interval,” said Dr. Blair Fennimore, a University of Colorado School of Medicine gastroenterologist.
The U.S. Food and Drug Administration (FDA) approved GI Genius in April 2021 based on a 2020 study that demonstrated that it can spot small, potentially precancerous lesions called adenomas – colorectal polyps – that gastroenterologists missed. A 2022 study showed even stronger results: GI Genius reduced the number of missed polyps by about a factor of two.
Those numbers underscore how hard it is even for well-trained eyes to spot tiny polyps. They are often scarcely differentiated in color or texture from the colon’s mucosal layers, and they’re often less than 0.5 millimeters (0.2 inches) across, or about the height of a computer USB connector.
AI-assisted colonoscopies
GI Genius system works by tapping into the endoscope’s feed before it hits the screen the physician observes during the colonoscopy. If GI Genius spots a suspected polyp, it puts a digital green box around it onscreen. The gastroenterologist then makes the call as to whether it’s actually a polyp and, if so, whether to remove it during the procedure.
The studies showed GI Genius to have little advantage over trained physicians regarding larger, more dangerous polyps. But small polyps can grow into problem ones: For each 1% increase in detecting these adenomas, the risk of being diagnosed with colon cancer in the next five years falls by 3%.
More than 150,000 colorectal cancer cases are diagnosed annually in the United States alone, making it the third-most common form of cancer. The five-year survival rate for localized colon cancer detected early is 91%. That drops to 72% if it’s spread to nearby structures or lymph nodes and plummets to 13% if it ends up in distant organs such as the liver or lungs. Finding more polyps, earlier, can make a real difference.
Fennimore says that 30% to 50% of screening colonoscopies spot an adenoma. He says it’s hard to know whether GI Genius has picked up adenomas that he would have missed, mainly because the green boxes appear so quickly on the screen. But, he says, “I’m convinced that there were small polyps that I would have missed – ones that blended in and looked exactly like the background mucosa.”
That there have also been instances where he noted a polyp that GI Genius missed, he says, and times when GI Genius puts a green box around something he decides isn’t a concern.
“It’s not the end-all, be-all,” Fennimore said. “But it’s a tool to decrease your rates of missed polyps.”
He expects that tool to continue to improve: Because GI Genius is AI-driven, it will improve with increasing volumes of real-world data to train on. He also anticipates that UCHealth will acquire more of the devices at UCH and elsewhere in the system.
Do gastroenterologists feel threatened by that prospect?
“As a GI doctor and somebody who’s done thousands of colonoscopies, it’s almost like, ‘I don’t want there to be a computer better than me. I feel like I do a good job of this,” Fennimore said. “But this is going to make me better. We’re in the business of preventing colon cancer, not of massaging egos. And so I think this is yet another tool to help us do as much as we can prevent colon cancer.”