Experts have weighed in on the future of artificial intelligence in rheumatology at ACR Convergence 2024.
Artificial intelligence has a long way to go before it could reliably replace doctors and traditional diagnostic pathways, two international experts have told ACR Convergence 2024.
Speaking to the college’s podcast series, ACR on Air, at the conference, Professor Bella Mehta and Professor Amanda Nelson discussed how the power of AI could be harnessed in the rheumatology landscape.
Professor Nelson is professor of medicine and adjunct associate professor of epidemiology at the University of North Carolina Gillings School of Global Public Health, and Professor Mehta is a rheumatologist at the Hospital for Special Surgery in New York. She is also an assistant professor of medicine at the Weill Cornell Medical College’s Department of Rheumatology.
Professor Mehta said there was still more work to be done to ensure AI could be used safely in a clinical setting.
“I don’t think that we are there,” she said.
“There are these articles that come out [saying] that AI will replace 80% of the doctors in 2035. I’m sure we’ve seen some of those. [But] I don’t think those are real. I don’t think anybody can replace doctors and diagnosis, at least at this stage.
“I mean, there’s a lot of error rates [with AI], and that’s not something that healthcare is ready for, at least right now.”
Professor Mehta said she believed there was potential to outsource areas like radiology to AI.
“I guess imaging is probably the one of the first ones that AI is showing more promise than others, especially because it’s a computer vision. Those sort of data sets are curated, and we are seeing good sort of results there in terms of diagnosis of diseases,” she said.
Professor Nelson agreed, and said she thought there was a lot of potential for AI to be applied to research and imaging and there was future promise that it could be useful in the clinic.
“But I think the work to understand their limits, their biases, the appropriate training and testing sets, their sort of characteristics around, you know, how accurate they’re going to be and which populations those things all need to be worked out before they’re rolled out broadly in a in a more open clinical setting,” she said.
“One of the things we struggle with in rheumatology is trying to sort of phenotype patients. And we get a patient with seronegative RA and trying to determine how are they’re going to behave in the future, what medicines they’re going to respond to.
“Bella, I think you’ve written a little bit about how artificial intelligence may help us with, you know, phenotyping and trying to classify patients, but I guess it’s something that’s still not there yet.
“I mean, I think it’s a futuristic concept of, like, exactly how the concept of a digital twin, which is basically a virtual representation of individuals physical health. So basically, if it’s me, all of my data, multimodal data, which could include medical records, you know, wearables such as Apple watches or whatever that I wear, all sorts of imaging and even generic data. So basically, you create sort of a twin of yourself on the computer, and then you simulate a person’s biological systems in this digital environment.”
There was plenty of potential for AI in the clinical setting, as well as research and patient engagement and education, the speakers said.
The use of AI in documentation is valuable but challenging when it comes to the translation to notes.
“It’s all about prompting, too. So again, the concept of an alien, it’s a smart alien, but you have to exactly tell it what to do, and you have to be very explicit,” said Professor Mehta.
“So, if you say, ‘in my assessment, I think that this patient has lupus and is flaring’, it will come to you [through in the notes] – we also need to change our behaviours for that to be picked up, and you’ll be surprised, it starts picking up that pretty quickly.”
Hear the full podcast here.
