Dr. Sravanthi Parasa stresses the importance of clinical training on AI
It is extremely important for gastroenterologists to understand the development of artificial intelligence (AI) algorithms and whether the questions they are designed to answer are relevant to their patient population, said Sravanthi Parasa, MD, gastro -enterologist at Swedish Gastroenterology in Seattle, Washington, at Digestive Illness Week 2022.
In what ways is AI being used to help process medical data?
There are different ways to use AI. When we think of AI, it’s not pictures, dictation or notes that we do, it’s a bunch of different things. A simple example, for at least in gastroenterology, where it has already made progress is in computer vision, where we already have 2 FDA-approved algorithms for detecting polyps. It is a vision-based algorithm, but there are several other documentation software based on natural language processing, which can help us read your pathology reports. Often we know that it takes a lot of resources to report some of our quality metrics, almost like 1 FTE [full-time equivalent] a year, so that’s another area that’s almost mature.
The third is to recruit patients into clinical trials using computer vision as a source to screen patients who might be eligible for IBD trials. The other side is big data, your electronic health records, how to make better risk prediction models. The list goes on and on, so there are so many different ways to use AI in medicine at this point.
How does AI help clinicians and what challenges do clinicians face when using it?
The whole point of AI is that it’s going to be more augmented intelligence. At this point, it’s just going to help us make better decisions. Among people who are already at the 95th percentile, you probably won’t see much difference because they’re already great, but it helps standardize the quality of care we can provide.
When we start using AI, for example an off-the-shelf device, we need to understand how the AI will interact with humans. Sometimes we get distracted because of the bounding boxes or something, so it’s almost like we’re transitioning back into a different environment. It’s as if you were used to driving a 1994 Ford to a Tesla in 2020 or 2022: it’s a different system but it’s not really difficult to adapt to it. I think that’s the step forward we need to take. But, more importantly, it is very important to understand how these algorithms are created and whether the question the AI is trying to answer is relevant to your patient population. That’s where we want to know how to interpret the results and stuff, that’s where a little bit of clinical training in the basics of AI becomes important.