When I reflect back on the RSNA Annual Conferences that I’ve attended, beginning in 1991—I haven’t attended every single year, but most years, so I know for certain that I’ve attended at least 25—it really is astonishing how much the face of the conference has changed, along a number of different dimensions.
Twenty-eight years ago, the conference was mostly focused on the modalities—CT, MR, nuclear medicine, PET, mammography, etc.—and the exhibit floor was nearly totally focused on the practicing radiologists in hospitals and large medical groups. It was a clinical and medical-technological conference back then, and informatics had at most a very marginal place. Every year, the radiology chairmen (and then were virtually all men then) came to the show and checked out diagnostic imaging machines, and went back to their home hospitals, in the U.S. and worldwide, and told their c-suite executives what they wanted.
All that’s changed dramatically; RSNA really is a different conference altogether now. Hospitals and imaging centers are saturated with robust, very contemporary diagnostic imaging equipment, and hospital organizations are under tremendous pressure to control costs, both in the U.S. and abroad. And even the transformation of radiology created through the shift from film to digital imaging, and the adoption of PACS (picture archiving and communication systems) systems, is mature now, meaning that PACS, as a technology, has become absolutely commoditized now. Practically the only purchases of PACS now are upgrades, plus the occasional switch of vendors brought on by organizational consolidation or by dissatisfaction with a particular product. So while PACS was on the exhibit floor this year, it was virtually invisible in the large scheme of things.
The fact that this year’s conference was totally dominated by discussions of artificial intelligence (AI) was not coincidental. The immense surge in interest in AI in radiology right now reflects the perfect confluence of policy, payment, operational, and clinical factors. Put very simply, as the U.S. healthcare system marches steadily towards a total cost cliff, from the current $3.6 trillion per year in annual total healthcare expenditures, to $6 trillion in the next six to seven years, the need to make all medicine more efficient, cost-effective, and better in terms of patient outcomes, means that the old ways of doing things simply aren’t going to cut it any longer, and everyone—policy leaders, purchasers, payers, and providers—knows it. Thus, not only did the term “AI” absolutely dominate the exhibit floor this year; it was also an element in literally dozens of presentations on the formal educational program this year, as well as in lectures sponsored by vendors and other organizations.
As Whitney Palmer, Senior Editor at Diagnostic Imaging, said in a video blog on Wednesday, “This year, we’re finding that AI is no longer a buzzword; instead, it’s being integrated into a host of products that have been unveiled. From patient experience to workflow to equipment monitoring to productivity, a host of companies are placing their fingerprints on AI tools.”
I myself attended several different presentations on the leveraging of AI tools to support the prioritization of radiologists’ workflows.
As I wrote last Monday, “On Monday, December 2, at Chicago’s McCormick Place Convention Center, the annual RSNA Conference included more sessions than ever that were focused on artificial intelligence (AI), and its application to radiological practice. Among numerous sessions today that involved AI was a session held at 3 PM local time, and entitled ‘Informatics (Artificial Intelligence: Triage, Screening, Quality).’” Inn that session, S.S. Jayadeepa, M.D., a practicing radiologist in Bangalore, India, shared the results of a study that she conducted in her local hospital, in which it was determined that incorporating AI into the process of evaluating potential midline brain shift on head CT exams through the use of a specific algorithm to flag an alert when a midline shift of more than 3 millimeters was detected, has led to a significant advance in prioritizing those studies among the radiologists in her organization.
Meanwhile, later in the session, Melissa A. Davis, M.D., medical director, clinical redesign at Yale New Haven Health and chief of emergency radiology at the Yale University School of Medicine, shared the results of a study executed in that organization that found that the use of an AI algorithm there to help flag cases in which the possibility of intercranial hemorrhage existed, significantly decreased the turnaround time involved in radiologists addressing such cases within their workflow.
Earlier that same day, Terence A. Matalon, M.D., chairman of the Department of Radiology at Einstein Healthcare Network in Philadelphia, had shared his perspectives on the experience of integrating AI algorithms into radiology workflow, during a lunch-and-learn sponsored by the Lexington, Mass.-based FUJIFILM Medical Systems. “Among the many use cases” for AI, Lacy said, “is the example of worklist priority notification. How can we take the worklist and re-prioritize it, based on AI results?” he asked. “How can I communicate directly with the physicians who must most urgently be communicated with? How do we incorporate AI findings into the report? How do we do broader analysis of archived reports?” His organization’s experience with the incorporation of AI alerts into workflow has been overwhelming positive, Dr. Matalon noted.
These case-study examples and many more that were presented this year at RSNA reflect an interesting fact, as Joe Marion, one of the imaging informatics field’s leading luminaries noted in my interview with him on Tuesday afternoon. “There are really two phenomena taking place right now. What you’ve just mentioned—leveraging AI in order to improve radiological workflow process and prioritize important clinical phenomena—that’s one track, and that’s moving forward in place,” Joe told me. “The broader use of AI that’s been predicted for years—the use of algorithms for diagnosis—that is still lagging at this point; and that’s the use that had been widely predicted.”
Indeed, some early broad efforts have proven to be over-broad and not sufficiently focused. But with new initiatives springing up all over, including one sponsored by the American College of Radiology (ACR), that’s set to change. As Keith Dryer, D.O., chief data science officer at the Boston-based Partners HealthCare health system, and vice-chair of radiology at Massachusetts General Hospital and Brigham & Women’s Hospital, told me in an interview last month about the ACR-sponsored AI initiative involving radiologists from several leading academic medical centers co-developing AI algorithms and testing them across the collaborative of hospitals, “This is the wave of the future. We’re using this network to share information and insights. The problem in taking so long to build these models—there have been missteps. And then small companies that have no access to data grab general data and build a model. So at Partners, Mass General and Brigham, we have 20 billion images, increasing by 1 billion a year, so we can build these models to improve patient care. That’s what’s new and transformational” about this initiative, he said. “These tools are on the verge of being in the hands of radiologists and clinicians. They understand what’s needed and the data, and the environments involved; they just need the models.” That initiative is attempting to take on those diagnostic algorithms whose application until now has not proven clinically successful.
So, two tracks are moving forward simultaneously, one tied directly to worklist optimization, and the other connected to broader diagnostically focused algorithms, whose use really could change how radiology is practiced in the future.
Still, even though all these tools, everyone has been reassuring everyone else, will never replace radiologists, only support their work, RSNA’s press release last Tuesday was fascinating, in terms of AI’s potential to at least potentially disrupt some diagnostic radiological work. As the press release published on Tuesday morning noted, “A sophisticated type of artificial intelligence (AI) can detect clinically meaningful chest X-ray findings as effectively as experienced radiologists, according to a study published in the journal Radiology. Researchers said their findings, based on a type of AI called deep learning, could provide a valuable resource for the future development of AI chest radiography models.” That study result should give nearly everyone at least a bit of pause in terms of the accuracy of diagnostic radiological work in present radiological practice. As study co-author Shravya Shetty, an engineering lead at Google Health in Palo Alto California, noted in a statement contained in the press release: "We've found that there is a lot of subjectivity in chest X-ray interpretation. Significant inter-reader variability and suboptimal sensitivity for the detection of important clinical findings can limit its effectiveness.”
Meanwhile, what of the vendor landscape around AI? “The fact that there’s a whole AI Showcase presents an interesting paradox, James Whitfill, M.D. told me on Tuesday. Dr. Whitfill, a practicing family physician, is chief transformation officer at Honor Health in Phoenix, and has been involved in radiology practice issues as a medical group and health system executive for many years now. “AI continues to grow,” he noted. “I listen to people like Kurt Langlotz, M.D. of Stanford, who believe there will be a coming shakeout among AI vendors. You’ve had valuations based on the AI might replace radiologists, but now that that won’t be the case, you have to sell a product valued at pennies per exam, not tens of dollars per exam. The other thing with AI, because it’s so dominant, is that the data scientists will often focus on the area under the curve, which reflects how accurate a model is. And they’ll get very obsessed with that. The problem is, no matter how good the algorithm is, if it doesn’t have clinical relevance, it’s not useful. And there continues to be work to be done there.”
So we are in the early stages of what inevitably will be quite a long, complex, and many-detoured journey. And future RSNA Conferences will reflect that reality. What’s fascinating, though, is how all the policy, payment, clinical, and operational trends are now converging on AI as a support to both workflow optimization and diagnostics, in a medical specialty—radiology—that is under more pressure than ever to produce precise, reliable results more quickly and efficiently than ever, in a nationwide healthcare system groaning under the weight of dramatically accelerating costs and demographic changes, including an aging population and an explosion in chronic illness.
AI couldn’t have come along at a more propitious time, really; and it speaks to the profound transformation that is finally beginning to take place in U.S. healthcare, in its shift from volume to value, a shift that inevitably is encompassing all the medical specialties in one way or another, radiology included. Just as RSNA has shifted dramatically away from the modalities-focused showcase it was 28 years ago, so too, are radiologist leaders recognizing that they will necessarily play a part in shifting their corner of U.S. healthcare forward into value, along a number of dimensions—clinical, practice, operational, and policy—going forward. I will be eager to see what all this looks like at the RSNA Conference a year from now. See you all in December 2020!
