It should come as no surprise that AI was featured heavily in this year’s Radiological Society of North America convention. From booths to sessions to one-on-one conversations around the hall, it seemed like almost everyone had something to say about AI.
Nearly every major company on the show floor was excited to talk about their advancements in the field, showcasing in real time just how well their programs could help radiologists and patients alike. While many technologies were still in the development or approval-gathering stages, many offered real glimpses at the tools radiologists have available right now.
Related article: AI and the Future of Radiology
But even with those available and soon-to-available tools, questions still swarmed the show about AI: How do current AI products fit into practices right now? What will happen in the next 5, 10, 15 years? What is the main goal of AI? Will there be any patient impact? Will AI replace the radiologist?
The questions and answers about AI seemed to all come in three categories: The relationship between AI and the radiologist, the workflow, and improvements to patient care.
AI vs. radiologist?
Seemingly the biggest question—and greatest fear—was a simple one: will radiologists be around in 15 years, or will machines be doing everything?
On the show floor, exhibitors were quick to point out that any currently-available AI technologies, and any near-future ones, aren’t meant to replace the radiologist but to enhance the radiologist’s abilities. So far, nothing can bypass the radiologist’s expert eyes. Instead, most of the products featured at the show—from GE’s Edison to Samsung’s AI to ScreenPoint to Fujifilm’s REiLI—offered deep learning to help triage cases.
With these programs, the computer can detect any abnormalities and bring them to the radiologist first. Several exhibitors described this as a way to give radiologists the most difficult cases first, so as to avoid getting them toward the end of the workload while working with tired eyes.
Even if programs could develop to the point where the computer could bypass a set of human eyes completely, experts say it still probably wouldn’t affect the field.
In a roundtable discussion session titled “Artificial Intelligence: Impact and Implications to Radiology,” one of the presenters, Lawrence Tanenbaum, MD, stressed that regulatory and risk-mitigation will continue to play a role in AI’s future. He pointed to cardiology as an example, where EKG’s can be read by a computer, but liability issues have led to cardiologists still having a significant role in reading and interpreting them.
Other experts echoed those same concerns, asking, “What company is going to be the first one to take the risk of not having a radiologist in the room?”
Another roundtable discussion of AI, “AI After Dark” brought together dozens of AI experts and questioning radiologists. One of the major discussions of the event revolved around who would be the “referee” for AI, if it were to ever approach replacing a human. Who decides when the technology is good enough? How are potential problems going to be addressed?
These questions don’t have an answer yet, but nearly every expert agreed that radiologists would need to play at least some part in designing and confirming any AI.
So, while AI isn’t going to replace radiologists any time soon, some radiologists are excited about the work that can be replaced in their day to day.
According to some reports, radiologists are dealing with 7% more work per patient year over year—not to mention the increasing patient load itself. As a result, many radiologists in forums across RSNA expressed interests in offloading simple cases in favor of being able to spend more time with complicated cases.
Related article: Artificial Intelligence: Own it Now
If AI was the overarching theme of the RSNA trade floor, workflow improvement seemed to be the main goal. While there was talk of more refined images and new screening methods, most of the innovations on the floor related to a more streamlined workflow. Major innovations seemed to all take the form of increasing automation of simple tasks for faster processing, promising greater speed.