AI and Radiology: Experts Parse the Layers and the Complexity

Dec. 2, 2021
At RSNA 2021 in Chicago, a panel of experts parsed the complexities around the actual adoption of artificial intelligence algorithms in radiological practice—and where things are headed from here

If anything was true about RSNA 2021 this year, it was the omnipresence of both the term and the concept: “artificial intelligence” was everywhere, both in terms of educational sessions and in terms of the exhibit floor. A few years in, radiologists, health IT leaders, and other stakeholders are moving forward to create AI algorithms and use them for various purposes—tremendously diverse purposes, as it turns out, with complexities everywhere.

One of the most stimulating panels at the conference—which was held this year as in years past, at Chicago’s vast McCormick Place Convention Center—this year was entitled “The Business of AI in Radiology: A Cost, a Long-term Investment, or an Immediate Business Opportunity?” Held Tuesday afternoon, November 30, the panel was moderated by Paul J. Chang, M.D., a professor of radiology at the University of Chicago health system in Chicago. Dr. Chang’s fellow panelists were Nina Kottler, M.D., M.S., associate medical director at the El Segundo, Calif.-based Radiology Partners national radiology group practice; Hari Trivedi, M.D., assistant professor of radiology and co-director of the HITI Lab at Emory University; Luciano Prevedello, M.D., M.P.H., of The Ohio State University Wexner Medical Center; and Mona G. Flores, M.D., global head of medical AI at the Santa Clara, Calif.-based NVIDIA Corporation. Dr. Flores appeared virtually, while everyone else was present in person at McCormick Place.

Dr. Chang initiated the discussion by making a number of statements. After several years of early adoption of artificial intelligence, he said, AI has moved forward into early stages of the “Gartner Hype Cycle,” showing the classic depiction of the “Gartner Hype Cycle,” including the following stages: “innovation trigger; peak of inflated expectations; trough of disillusionment; slop of enlightenment; plateau of productivity: appropriate consumption.” Right now, AI adoption in radiology, he said, is living through the “trough of disillusionment, after five years.” Indeed, he said, “There’s nothing new under the sun when it comes to new technology. We over type, over promise, and under deliver, and fall into the trough of disillusionment. We eventually learn appropriate consumption.” And, he added, with regard to the trough of disillusionment, signs include “lots of VC [venture capital] action, lots of investment, but a lack of significant consolidation.”

Dr. Flores, who spoke remotely, shared some data on the rise of AI investment. She quoted the “Artificial Intelligence for Medical Imaging 2020 Report” by Yoke Development, which stated a global market size for AI tools and solutions for radiological practice, at reaching $2 billion by 2023 and $1.billion, with a predicted annual compound growth rate of 36 percent. The “inflection point” in that growth trajectory, she said, is expected sometime in 2023. “We are seeing more and more investment every day, and the funders, the venture capitalists, are taking notice,” she said. She also quoted statistics from a 2019 AI study be Definitive Healthcare that found 32 percent of hospitals and 35 percent of diagnostic imaging centers already adopting some form of AI by that year. All that said, she emphasized, “We need to ask four questions: the why, the who, the how, and the when. Why? Just because we have a hammer does not mean that everything is a nail.” In other words, she stressed, AI adoption must be thoughtful and meaningfully improve workflow, diagnostics, and/or operations in radiology practices in order to be worthwhile pursuing.

“Is the AI offloading lower-level functions?” Flores asked. “Are we using AI as a replacement or an aid? Adoption will suffer if we see AI as a threat. For AI to be adopted, it needs many champions, and they all need to be on the same page. The how? How is it being integrated into processes? How is it changing workflow? How is it being developed? Are radiologists involved from inception? I think it’s obvious that AI will not replace physicians,” she said. “Physician practices will replace those physicians who do not.”

A broader, open discussion then ensued.

Asked whether AI has been adopted in actual radiological practice at Emory, Dr. Trivedi said that “Academic medical centers are more willing to take risks on innovation than are medical practices. We’re eager to adopt AI, but at Emory, we’re just in the early stages. We’re not yet using AI in practice; 12 months from now we will be.” Asked by Chang whether he and his colleagues had yet established a governance program or strategic plan around AI adoption, he said that they had not.

Then Dr. Kottler offered that “I am in a private practice. We have a national onsite radiology practice. We have about 3,000 radiologists and do about 10 percent of the radiology in the U.S. We’ve invested in AI and, in one case, we created our own NLP [natural language processing] AI algorithm, in 2017, deployed it in 2018, have deployed it to about 1200 radiologists so far. We’ve developed an NLP platform. These tools are clinical tools. You have to go do training to work with it. Radiologists need feedback on how well they’re using it. The NLP tool has gone through millions of reports. We generally pilot something first. Our pilots are big because our practice is big. We did partner with a vendor, and have seven of their FDA-cleared algorithms. We have millions of exams going through their tool. Piloting another NLP algorithm that provides summary of findings.”

“Is it fair to say in your organization, the motivation was use case, clinical need, rather than hypothesis testing?” Chang asked. “That’s correct,” Kottler responded. “You have to have a need. And we went to a vendor to help us; they ended up joining us, and are now internal to us.”

“My view,” Chang said next, “Is that we have to be polymorphic in what we use and how we use it. Your big ROI was actually NLP?” “Yes,” Kottler replied. “At the time we started, vendors were exploring image interpretation-type tasks, including detection, and maybe diagnosis. But that for us was not a great use case. With NLP and doing things we’re not good at, initiating a workflow, bringing things together, it was a 90-percent ROI.”

Dr. Prevedello said that “I wear three hats. One is associate chief clinical information officer. As an imaging informaticist in radiology, I helped implement four FDA-approved applications. We did have to remove one. As director of the AI lab, we’re developing algorithms. All the algorithms we’ve developed were borne out of necessity. One of the algorithms we’re developing is close to the process of starting to get FDA approval.”

“Nina, how did you get institutional support, or articulate the benefit of an AI solution, in a way that was ultimately successful?” Chang asked. “You want to start small; you don’t want to boil the ocean,” Kottler said. “We had a very specific use case. We wanted to decrease the variability in radiology report., and identify incidental findings. So we ran it as a change management program. We talked to our radiologists; they understood the use case, around how you decrease variability in practice. We decided that we should invest in it to drive transformation. If you’re a very small group, investing $1 million is a lot. But once we created a case and proved that it worked, people got very excited.”

“You focused on what people wanted,” Chang said. “There’s always the quality argument,” he added. “We pay lip service to quality, but we invest in productivity and efficiency” in practice in radiological practice, he added.

“You have to be more comprehensive than quality; quality is a baseline,” Kottler said. “Fewer thyroid biopsies being an example. Several payers now support that algorithm.” “Even though the original motivation might have been quality, you can pitch it as an economic benefit, yes?” Chang asked. “Yes, that’s right,” Kottler replied.

Responding to Chang’s question regarding his organization’s strategy, Dr. Trivedi said that “We’re at the point now where we know that AI ROI depends on what type of organization you’re in. And this discussion focuses on the U.S., where we have a mostly fee-for-service payment system. But it’s important to think bigger, to think about integration” of practice processes.

With regard to obtaining senior management-level support, Prevedello advised the audience, “Make sure to get the finance discussion into the mix, involve the c-suite.”

Further, Kottler said, “AI is really good at adding structure to unstructured data; and radiology has a huge amount of unstructured data. So we’re using AI for naming and categorization purposes. And the AI can go in and say, that study is missing recons or other elements. Let’s have an IT system that can identify missing elements for the rads.”

“All those things make sense from an institutional perspective,” Trivedi said. “Do I want to reduce the need for re-scanning? Absolutely. Do I want something that will decrease missed scanner appointments? Absolutely. The challenge is we have not seen a ton of companies doing this successfully, because it’s difficult to scale.” And, he added with a bit of lightness, “One of my colleagues said, ‘I can track my burrito, but I can’t track my studies.’ That needs to change.” In terms of how effective applications are created, he added, “The reason most of those applications are developed in-house, because your use case is specific to you, your workflow is specific to you. Everyone’s protocols are different.”

Asked by Chang what the barriers are to implementation, Prevedello said flatly, “How to manage all these new tools. Each server I’m adding into the system is a headache. So that ingestion of new solutions has to be streamlined; and there are several ways to do that, but that’s something that’s front and center when considering new applications. Also, it’s not much commented on—everyone talks about AI solving everything and streamlining our workflows, and speeding things up—faster is always there. But actually, with new algorithms, it can take longer to read a case, because there’s a period of time during which you’re dealing with new technology and outputs, and are having to learn.”

“One of the major barriers is actually convincing the decision-makers; the issue is finance,” Trivedi said. “What’s the ROI? The return is for the radiology group but the investment is form the hospital. The other major barriers are time and resources. Our IT guys are working around the clock just to keep our lights on. I can’t really say, could I please have three or four guys for your team just to test this? That’s been a major challenge.”

Sponsored Recommendations

Enhancing Remote Radiology: How Zero Trust Access Revolutionizes Healthcare Connectivity

This content details how a cloud-enabled zero trust architecture ensures high performance, compliance, and scalability, overcoming the limitations of traditional VPN solutions...

Spotlight on Artificial Intelligence

Unlock the potential of AI in our latest series. Discover how AI is revolutionizing clinical decision support, improving workflow efficiency, and transforming medical documentation...

Beyond the VPN: Zero Trust Access for a Healthcare Hybrid Work Environment

This whitepaper explores how a cloud-enabled zero trust architecture ensures secure, least privileged access to applications, meeting regulatory requirements and enhancing user...

Enhancing Remote Radiology: How Zero Trust Access Revolutionizes Healthcare Connectivity

This content details how a cloud-enabled zero trust architecture ensures high performance, compliance, and scalability, overcoming the limitations of traditional VPN solutions...