06/07/2024
news.ddw.org
Artificial intelligence (AI) has great potential to improve patient care, but providers must remain cognizant of its limitations, according to Nayantara Coelho-Prabhu, MBBS, associate professor of medicine at the Mayo Clinic Rochester, who is speaking on the topic during Digestive Disease Week® (DDW) 2024.
“Physicians need to be aware of both the opportunities and limitations of AI — not only when we build these algorithms, but also when we use them,” says Dr. Coelho-Prabhu. “We need a lot of information before we apply AI. We must understand the model, what it was built upon, and what disease states it should be applied to.”
Dr. Coelho-Prabhu outlined the potential applications of AI in gastroenterology, some with FDA approval and others still in the investigational stage.
- Computer-aided detection (CADe) and diagnosis (CADx): Some systems such as GI Genius, Skout and Magentiq are FDA-approved for detecting polyps during colonoscopy. “These types of tools can really improve our endoscopy. Size estimation is hard,” she says, “but it’s important to get sizing and histology right because it determines your surveillance guidelines.”
- Quality metrics: Coelho-Prabhu sees quality metrics as one of the biggest opportunities for AI. ASGE has defined quality metrics for endoscopists, such as withdrawal time and adenoma detection rate. In addition, many more parameters such as speed, percent of the surface inspected, and percentage of missed blind spots can be calculated automatically, she says. She showed examples of real-time colonoscopy quality monitoring where endoscopists can get feedback during a procedure.
- Inflammatory bowel disease (IBD) scoring: IBD activity-scoring is often subjective, but it determines how the patient is managed. “There’s a lot of work being done to standardize IBD scoring to minimize variability, and we believe AI algorithms will be robust at achieving this goal,” she says.
- Bile duct strictures: AI algorithms are being developed to help determine whether bile duct strictures are benign or malignant.
- Automated note writing: The use of AI to create notes during endoscopic procedures is coming soon. “We will have AI programs that can not only detect what you’re seeing during a procedure but also create your note for you so that you don’t have to record it afterward,” she says. “It will note that it found the polyp and record its size and location and which tool is being used to remove it.”
- Personalized endoscopy: “AI may let you overlay your last procedure onto the current one so that you have a personalized road map for that patient and can tailor interventions based on their history,” says Dr. Coelho-Prabhu.
She also cautioned providers about the limitations of AI, based on the type of data used to build the algorithm:
- Lack of specificity and standardization in electronic health data. “A lot of data types lack standardized terminology, which makes it hard to extract the correct information,” she says. “If the algorithms aren’t built with the right information, they can miss details and important findings.”
- Risk of bias. “If AI algorithms are built using only hospital or doctor visits, they end up targeting sicker populations and those with access to medical care,” she says. “They can end up under-sampling a large portion of the population.”
- Use of billing codes. “We know that in the hospital, people put in high-compensation codes as their highest diagnosis. AI algorithms built on billing code data may not reflect what is most important for that patient. To know what the patient is dealing with just from a billing code is very difficult.”
- Wearable technology. “Many patients who use wearable technology may be healthier or more health conscious and almost certainly more wealthy than average,” says Dr. Coelho-Prabhu.
Dr. Coelho-Prabhu’s oral presentation, “ASGE consensus on AI in endoscopy?” on Tuesday, May 21, at 8:05 a.m. EDT is part of the session “Artificial Intelligence: Endoscopy Reimagined.”
