Welcome to the Trough of Disillusionment.
That's where we are when it comes to artificial intelligence (AI) in radiology, according to Paul J. Chang, MD, professor of radiology and vice chairman of Radiology Informatics at the University of Chicago.
Dr. Chang kicked off a Tuesday session on AI at RSNA 2021 by citing the Gartner Hype Cycle (see image), a widely used construct that describes the adoption of disruptive technologies. In the Gartner model, an escalating hype cycle is followed by a trough of disillusionment. For radiology, that means lots of companies and venture capital money but suboptimal adoption.
(Left to right) Prevedello, Kottler, Trivedi, Chang
"We're right in the middle of this phase where people are throwing millions of dollars at companies and yet when you look at the numbers there's not a lot of great adoption going on," Dr. Chang said. "People are still kind of dipping their toes in the water, just testing a few algorithms here and there."
To get out of the trough and speed the adoption of AI, radiology leaders and other health care stakeholders must convince the people in the C suite that the technology makes financial sense, an area with which the three panelists at the session have experience.
Nina Kottler, MD, radiologist and associate chief medical officer at Radiology Partners, the largest physician-owned and physician-led radiology practice in the country, described how she and her colleagues developed AI-based algorithms for initiating workflow that offered a far greater return on investment than algorithms related to image interpretation.
"We wanted to create something to help the radiologist as they work in their systems," she recalled. "And when we created that case and proved that it worked, the organization got very excited."
Organizations also get excited by applications that differentiate them from their peers, according to Luciano Prevedello, MD, associate chief clinical information officer at Ohio State University Wexner Medical Center. Dr. Prevedello, who has helped implement four FDA-approved AI-based applications at his facility, said that the expected financial return is central to pitching AI-based solutions.
"You have to ask, 'will the quality that this application is bringing to the table be a business advantage of your organization?'" he said. "If we are one of the few organizations that can do that, then it's something we can advertise and get more patients as a result."
Panelist Harvi Trivedi, MD, emergency radiologist at Emory University in Atlanta and a leading researcher in AI, interviewed more than two dozen stakeholders in preparation for the session. His conversations revealed a surprising finding: Applications that improve radiologist efficiency were not nearly as likely to attract investment as those that help organizations capture more patients. An AI application that helps bring more patients back for follow-up imaging, for instance, has the potential to add tens of thousands of dollars to an organization's bottom line.
Dr. Trivedi also said that single-solution applications, like an algorithm for lung nodule detection, are increasingly less attractive to health care organizations.
"We've got to think bigger, we've got to think about integration," he said.
The stakes are high, Dr. Chang said, as the profession faces burnout and staffing shortages amid increasing complexity and demands from clinical colleagues for precision radiology.
"We barely get through the worklist as it is, and now there is this extra requirement to add more characterization of image datasets," Dr. Chang said. "We can't handle all this alone. We need some help right now because what we're doing is not sustainable."
In a video presentation taped before the session, Mona Flores, MD, the global head of medical AI for NVIDIA, offered a sanguine outlook, noting that the AI market is expanding and investment is increasing. Adoption remains low, however, with only one-third of organizations currently using AI in their practices.
"The time for AI is not now, it's yesterday," Dr. Chang said.
Access the presentation, "The Business of Artificial Intelligence in Radiology: A Cost, a Long-Term Investment or an Immediate Business Opportunity?" (T4-CIN12) on demand at Meeting.RSNA.org.
