Colon cancer is one of the most deadly cancers in the world today and its incidence rate has been steadily increasing among younger people. Colonoscopy is currently considered the best way to detect and prevent colorectal cancer. However, it is hard to spot precancerous polyps in patients who are most at risk, especially those with IBDs like Crohn's Disease and Ulcerative Colitis.
The pre-cancerous lesions they develop can be very flat or only slightly raised, making them difficult to catch during colonoscopy. Recent studies suggest that over half of post-colonoscopy cases of colon cancer appear due to missed lesions in previous examinations. To address this problem, scientists are now exploring the use of AI to locate these hard-to-see polyps.
Researchers at Mayo Clinic (Rochester, MN, USA) are studying the use of AI to enhance polyp detection rate in colonoscopy. AI is being used by gastroenterologists for a variety of gastrointestinal conditions with the aim of identifying signs earlier and making them easier to treat. In the case of colon cancer, the AI system scans the real-time video feed from the colonoscopy and highlights potential polyps with small red boxes, helping doctors to spot them more quickly. Adding AI to traditional colonoscopies can help physicians better detect polyps that may have been missed otherwise.
Mayo Clinic carries out around 800 to 900 surveillance colonoscopies on IBD patients annually which has provided it with a rich databank for developing AI systems to improve the process. This data serves as "ground truth" or real-world examples that are used to train AI algorithms. The team will annotate images from a sample of 1000 patients, watching all colonoscopy videos and marking lesions in frames from different angles. The annotated images will then be fed to a computer to create AI machine learning algorithms that can teach the machine how to detect polyps associated with IBD. The researchers are also creating a new digital endoscopy platform that will film all in-house procedures, correlate them with medical records, and then integrate AI back into the procedures as applicable.
"We’re all familiar with facial recognition software. Instead of training the AI to recognize faces, we train it to recognize polyps," said James East, M.D., a gastroenterologist at Mayo Clinic Healthcare in London.