Nov. 11, 2022 – Artificial intelligence has great potential in medicine, helping doctors find skin cancer, spotting potential problems on a chest X-ray, and assisting with many other procedures. Another prime example is colorectal cancer screening as part of a colonoscopy.
A colonoscopy — recommended for Americans at average risk of cancer aged 45 and older — will not be much different for patients with the addition of artificial intelligence, or AI. But behind the scenes, AI could make detecting precancerous polyps, or cancerous lesions, more likely.
“AI-assisted colonoscopy effectively increases the physician’s ability to find even the most subtle precancerous polyps,” said Tyler M. Berzin, MD, gastroenterologist at the Center for Advanced Endoscopy at Beth Israel Deaconess Medical Center in Boston.
The technology is designed to flag anything the computer “sees” as suspicious, but does not replace the training and expertise of a gastroenterologist. Even with AI, the doctors stay at the patient’s side and carry out the procedure.
The doctor maintains full control, says Prateek Sharma, MD, a gastroenterologist and professor of medicine at the University of Kansas School of Medicine in Kansas City, KS. “AI helps them and warns them about colon polyps — the precancerous lesions in the colon — so the doctor can remove them.”
The controversy continues
Size, height and number play a role in polyps. Doctors generally remove or biopsy lesions that are 10 millimeters and larger.
But there is less consensus on the best approach to smaller polyps.
“The clinical relevance of detecting and removing small (5 to 9 mm) or tiny (less than 5 mm) adenomas is a subject of ongoing debates, Berzin and co-authors wrote in a leading gastroenterology journal in May 2020.
For example, one of the potential downsides to using AI polyp tools is “the risk of removing larger numbers of small or hyperplastic polyps, increasing cost and risk without benefit to the patient,” says Berzin.
“Trained gastroenterologists are experts at identifying and removing precancerous colon polyps,” says Berzin. “But a gastroenterologist working with an AI tool for polyp detection has a great advantage because AI computer vision tools can analyze at the same time every pixel of the endoscopy monitor and can be distracted or tired for even a millisecond.”
The benefit to patients is “another pair of eyes looking for polyps and helping the doctor,” says Sharma, who is also chair of the artificial intelligence task force at the American Society for Gastrointestinal Endoscopy.
How it works
AI is based on computer instructions, called algorithms, that learn the difference between worrisome and benign colonoscopy images and videos. AI is getting better at objects in a process called machine learning. When an AI system discovers a potentially problematic area, the technology draws attention to it by framing it in a box on the screen. Some systems also emit an audible alarm.
“We see greater interest in using these algorithms as they will standardize polyp detection by endoscopists, thereby reducing the number of colon cancers missed,” says Dr.
“These products are slowly gaining in importance. When planning colonoscopy, patients should ask if their endoscopist has access to advanced diagnostic tools,” she says.
The technology is not accurate 100% of the time – false alarms can occur if the system flags a bubble in the colon as potentially dangerous, for example. That’s just one of the reasons doctors still have the final say on whether a polyp is suspicious or not.
AI or no AI: “Coloscopy has long been our most effective tool for preventing colon cancer because it detects precancerous polyps earlier than any other screening method,” says Berzin, who is also an associate professor of medicine at Harvard Medical School.
AI can be expensive
AI and machine learning are already playing a role in “smart” technologies (smartphones, smartwatches, and smart speakers), self-driving cars, and speech-recognition software. But the use of AI in medicine is comparatively new. And like many new technologies, it is also expensive. “The AI equipment has to be bought and is expensive,” says Sharma.
“The cost of the algorithms may currently be prohibitive for some centers in the current healthcare landscape,” Parasa agrees. “The cost will likely come down as more algorithms enter the GI market, as is the case with other software solutions.”
Colon cancer is common
Aside from some types of skin cancer, colon cancer is the fourth most common cancer in Americans. It is also the fourth leading cause of cancer-related death in the United States CDC reports. According to this, more than 150,000 Americans will be diagnosed with colon cancer and more than 50,000 will die in 2022 Figures from the National Cancer Institute.
More research is needed to examine how humans and this technology interact, Berzin says. “The most interesting research in this area will not be comparing ‘doctor versus AI’ but will focus on understanding the nuances of ‘doctor plus AI’.”
There are at least three FDA-approved AI algorithms for polyp detection in the US, and more are being developed, Parasa says.
“In addition, other applications currently available in the European market may also be available in the US market in the near future, including polyp characterization.”
“As the field matures, we’ll likely see more AI augmentation tools that will help us detect and diagnose GI disorders in real time,” she adds. “A suite of algorithms like this will definitely improve patient care and outcomes.”
Even if AI in medicine is still a work in progress, Berzin expects that the combination of doctor and AI technology “will lead to the highest possible protection against colon cancer in the long term”.