AI, long hyped for radiology, is still early in application

With some of the hype dying down, healthcare organizations are looking for more commonsense ways to use the technology and gain support for it.

Hopes have been high to integrate artificial intelligence into healthcare, with early sights set on radiology.

Relating that to the Gartner Hype Cycle, the expectations are following the traditional and expected rise and fall, said Paul Chang, MD, a professor at the University of Chicago and a well-known futurist in radiology and informatics. Right now, those expectations are down off the giddy tops supported by the heady predictions of many.

Reality is setting in, but that’s good, Chang contended during an Ask the Expert session at the recent annual meeting of the Radiological Society of North America (RSNA) in Chicago.

Paul Chang, MD

“There are hopes to improve quality and reduce variability” by applying AI in various healthcare domains, especially radiology, he said. “AI is no different from (picture archiving and communications systems), in that there’s nothing new under the sun, that when it comes to disruptive technology and overhyping. But we’ll eventually get there; we’ll get to the point where we’ll appropriately consume any technology. We do tend to overhype new technologies, but in the end, we underestimate the effects of them in the long run.”

Even though AI has been bandied about as a technology for several years, widespread adoption of this advanced technology in healthcare – and even in an “obvious” place like radiology – is still early. Case in point – all the panelists on this Ask an Expert session, entitled “The Business of Artificial Intelligence in Radiology,” responding to a simple query by Chang, admitted that they considered their organizations early in adopting AI.

“We are right in the trough of disillusionment,” Chang said, referring back to the Gartner hype cycle. There’s been little consolidation among vendors of AI technology, he noted. “There’s lot of investment and interest, but relatively modest revenue realization by the vendors. We’re going to get there, but it will take years.”

Research organizations are pointing to a growth plan for revenue derived from AI, said Mona Flores, MD, head of medical AI for NVIDIA Corp. She cited Signify Research, which estimates $2 billion will be spent annually on medical AI by 2028, and she buttressed her positive view by noting that a third of radiology practices surveyed in 2019 said they planned to adopt AI over the next two years.

There are questions surrounding these implementations, though, as to whether the technology can offload low-level work. “It needs champions and practices need to determine how does it get integrated into workflow,” Flores added.

Not all organizations are willing to venture out with AI, noted Hari Trivedi, MD, assistant professor of radiology and biomedical informatics at Emory University and the Winship Cancer Institute. “There’s a dichotomy between academic medical centers that are more willing to take risks,” he said. “For large academic medical centers, there can be a thirst for AI, but the mechanism for actually doing it can be a multi-year process.”

At Trivedi’s Emory, “we’re in the early stages” of AI adoption; “In about 12 months, we hope to be using it; we’re not using it now.”

Even in radiology, where AI has been forecast to assist clinicians by scanning high volumes of images to identify those that might have potential concerns warranting further study by clinicians, there are concerns about how the technology is being trained and the algorithms that are initially built and refined. Fears about these “black box” issues suggest that organizations need to have a growth path for implementing AI that builds trust in its capabilities.

For example, Nina Kottler, MD, associate chief medical officer for California-based Radiology Partners, said her organization is putting AI in place to support radiologists as they dictate reports, while another effort helps in identifying factors in reports that lead to insights. “Our big ROI was natural language processing,” she said. “We want to use AI for things that we are not good at. We have to start small. Because we had a specific use case and wanted to reduce variability in radiology reports, we needed a way to use technology to improve that.”

As an investment that uses technology, AI must provide a compelling story to top executives that offer a measurable benefit, said Luciano Prevedello, MD, associate chief clinical information officer and vice chair of medical informatics and augmented intelligence in imaging at The Ohio State University Wexner Medical Center.

“We start with the question, ‘Can I do that consistently across my department?’ Algorithms can provide benefits by enabling you to do things consistently. The other thing is money – making sure that you have a good story that makes sense. You have to know how to pitch this to the C-suite, and that often involves a financial discussion.”

More for you

Loading data for hdm_tax_topic #better-outcomes...