By now, it is widely accepted that artificial intelligence (AI) will reshape contemporary medicine. The question is simply when this hypothetical will become an everyday reality. For gastroenterologists involved in the management of inflammatory bowel disease (IBD), the waiting period may be ending.
AI “is the next step in clinical care,” Jacob Kurowski, MD, medical director of pediatric inflammatory bowel diseases at Cleveland Clinic Children’s in Cleveland, Ohio, said in an interview.
“In terms of technological breakthroughs, this is like going from some of the more rigid endoscopies to high-definition and white-light endoscopy or the upgrade from paper charts to the electronic medical record (EMR), but instead of making your life more difficult, it will actually make it a lot easier,” said Dr. Kurowski, who has researched and lectured on AI applications in IBD.
Simply put, “AI is when algorithms use data to simulate human intelligence,” said Seth A. Gross, MD, clinical chief in the Division of Gastroenterology and Hepatology at NYU Langone Health and a professor at NYU Grossman School of Medicine, New York City, who has studied the use of AI for polyp detection.
IBD is ideally served by AI because to diagnose and manage the disease, gastroenterologists must gather, analyze, and weave together a particularly heterogeneous mix of information — from blood tests and imaging to patient-reported symptoms and family history — often stored in different places or formats. And to ensure patient participation in their care plans, gastroenterologists also need to help them understand this complex disease.
Because of their potential to aid gastroenterologists with these tasks, three core AI technologies — some of which already have commercial applications — are likely to become foundational in clinical practice in the coming years: Image analysis and processing, natural language processing (NLP), and generative AI, according to experts familiar with AI research in IBD.
Image Analysis and Processing
One of AI’s most promising applications for IBD care is in medical image and video processing and analysis. Emerging AI tools convert the essential elements of medical images into mathematical features, which they then use to train and refine themselves. The ultimate goal is to provide fast, accurate, and granular results without inter- and intraobserver variation and human potential for bias.
Today’s techniques don’t quantify IBD very well because they’re qualitative and subjective, Ryan Stidham, MD, associate professor of gastroenterology and computational medicine and bioinformatics at the University of Michigan, Ann Arbor, and a leading researcher in AI applications in IBD, said in an interview.
“Even standardized scoring systems used by the US Food and Drug Administration and the European Medicines Agency to assess disease severity and measure therapeutic response are still pretty crude systems — not because of the gastroenterologists interpreting them, who are smart — but because it’s a very difficult task to quantify these features on imaging,” he said.
Another appeal of the technology in IBD care is that it has capabilities, including complex pattern recognition, beyond those of physicians.
“What we can’t do is things such as tediously measure every single ulcer, count how many different disease features are seen throughout the entire colon, where they are and how they’re spatially correlated, or what are their color patterns,” Dr. Stidham said. “We don’t have the time, feasibility, or, frankly, the energy and cognitive attention span to be able to do that for one patient, let alone every patient.”
AI-based disease activity assessments have yielded promising results across multiple imaging systems. The technology has advanced rapidly in the last decade and is beginning to demonstrate the ability to replicate near perfectly the endoscopic interpretation of human experts.
In separate studies, AI models had high levels of agreement with experienced reviewers on Mayo endoscopic scores and ulcerative colitis endoscopic index of severity scores, and they reduced the review time of pan-enteric capsule endoscopy among patients with suspected Crohn’s disease from a range of 26-39 minutes to 3.2 minutes per patient.
A report from the PiCaSSO study showed that an AI-guided system could distinguish remission/inflammation using histologic assessments of ulcerative colitis biopsies with an accuracy rate close to that of human reviewers.
In Crohn’s disease, research indicates that cross-sectional enterography imaging could potentially be made more precise with AI, providing hope that radiologists will be freed from this time-consuming task.
“As of today, several commercial companies are producing tools that can take an endoscopic image or a full-motion video and more or less give you a standardized score that would be akin to what an expert would give you on review of a colonoscopy,” Dr. Stidham said.
This is not to say there isn’t room for improvement.
“There’s probably still a bit of work to do when looking for the difference between inflammation and adenoma,” said Dr. Kurowski. “But it’s coming sooner rather than later.”