Mayo Clinic researchers are developing ways to employ artificial intelligence (AI) to increase the detection of potentially cancerous polyps during colonoscopy. The procedure, which remains the standard of care for screening for and preventing colorectal cancer may miss lesions that later lead to more than half of post-colonoscopy cancer cases according to some studies.
Colonoscopy’s success lies in its ability to detect and allow a gastroenterologist to eliminate pre-cancerous polyps to help prevent colorectal cancer from developing. However, polyps are often hardest to spot in patients most in need of screening—patients with inflammatory bowel diseases IBD) like Crohn’s disease or ulcerative colitis who are at the highest risk of developing colorectal cancer. Pre-cancerous lesions in this population of patients tend to present as flat, or only slightly raised, making them easy to miss during the procedure.
The team at Mayo Clinic, however, are developing and AI application trained to recognize these hard-to-see features to work alongside physicians in real time to alert them to their presence by drawing red boxes around the polyps that could get overlooked.
“We’re all familiar with facial recognition software,” said James East, M.D., a gastroenterologist at Mayo Clinic Healthcare in London. “Instead of training AI to recognize faces, we train it to recognize polyps.”
To help train the AI model, Mayo Clinic turned to its large databank of surveillance colonoscopies it has conducted in the past. According to Nayantara Coelho-Prabhu, M.B.B.S., a gastroenterologist at Mayo Clinic in Rochester, Minnesota, the organization uniquely qualified to develop this form of AI as it routine conducts between 800 and 900 colonoscopies each year. She noted that the Mayo databank contains what is referred to as “ground truth”—real-world observations and measurements that can be used to train and test the AI algorithms.
Coelho-Prabhu and team have selected a subset of 1,000 patients and are now meticulously annotating data from them by watching each colonoscopy video to mark lesions in every fram from every angle that it will use to train the algorithm to recognize IBD-specific polyps.
In addition, Coelho-Prabhu and her colleague gastroenterologist Cadman Leggett, M.D., are developing a new digital endoscopy platform to allow for the filming of all in-house procedures, correlate them with patient medical records, and them integrate AI back into standard procedures as applicable.
“Once we develop algorithms, we can run them in our procedure videos to test their performance,” Coelho-Prabhu said.
The development of an AI algorithm to aid in colorectal cancer, while the most advanced, is not the only project at Mayo Clinic looking to using its automated learning capabilities to aid in disease detection and treatment.
The academic medical center also has an AI-focused program that is seeking to use natural-language processing (NLP) to scan medical notes in patients’ electronic health records to spot risk factors that may make them susceptible to developing pancreatic cancer. The program, led by Shounak Majumder, M.D., hopes that the NLP screening to identify those most at risk can be employed as a tool to identify patients who should receive additional screening.
