In the July edition of the American Journal of Roentgenology, Ronald Summers, MD, PhD, senior investigator in the NIH Clinical Center’s laboratory for imaging biomarkers and computer-aided diagnosis, updated radiology watchers on the state of the art in fully automated abdominal CT interpretation. On July 7, he took questions on the material from HealthImaging.
Deep learning, he stressed, is “the particular technology in artificial intelligence that’s making a difference.” Here are excerpts from the rest of the telephone interview.
HealthImaging: Along with the July edition of AJR, you presented this material at the last two RSNA conferences. Does the timing reflect a perception on your part that radiologists need to hear about fully automated image interpretation as sort of a wakeup call?
Summers: I don’t know that I would call it a wakeup call as much as [a recognition of the fact] that there’s a lot going on that a lot of people are not familiar with. It’s like there are different communities of people working in parallel, and their opportunities for communication are sometimes kind of limited. So what I’m trying to do is bridge the two communities so that the people who take care of the patients, the clinicians, are aware of what’s possible.
It’s happened to me on more than one occasion that I’ve given a talk and, afterward, a physician will come up to me and say, “You know, I didn’t know that was possible. I would like to be able to do that.” Those sorts of things don’t happen when the communities remain parallel. They have to intersect at some point.
So the desire to facilitate communication and collaboration between radiologists and referring physicians was also a motivator for you.
Yes, and it was also a great opportunity for me to synthesize the field. I have in my head all these things that I’m aware are going on and sometimes writing an article like this gives me an opportunity to bring those strands together so that I get a sense of where the field lies currently. I can see where the gaps in knowledge are so that I can attempt to fill them. Part of my job is to be at the forefront, to do novel research, and this is an excellent way for me to identify where the gaps are. The gaps are where the novel research can happen.
What impact are you seeing these technologies make beyond abdominal applications and into radiology as a whole?
In the article I focused on abdominal because I’m a body radiologist with a focus on abdominal imaging. I worked for quite a while on CT colonography, so a lot of my work has been in abdominal image analysis. And I saw there was a knowledge gap. There’s been a tremendous amount of work in applying these sorts of advanced image analyses to the brain, for example, and to cardiac, lung and breast imaging. Those are areas that have seen tremendous efforts. They’re the dominant areas in which automated image analysis has been successful. I felt we needed more insight into what was happening in the abdomen.
Are some subspecialties quicker to embrace automated image analysis while others tend to feel more threatened by the technology than excited about it?
I think society as a whole is starting to think about those kinds of issues. There have been articles in the popular press about what effects these kinds of advances in artificial intelligence may have on people. What I think is going to happen is that there will be incremental improvements in the quality of care that we can provide. And the ways in which we deliver that care will incrementally change.
I don’t see massive changes, because I think the hype is outpacing the reality. But I definitely think that there is great potential in these artificial intelligence technologies to make a positive impact on patient care.
So that’s kind of an indirect response. You were getting at the fear and the sense of feeling threatened. And I’m trying to suggest that’s simply because people in general—including healthcare providers and even people working in the [AI] field—are not fully aware of where this is all going. But I see the future as bright for practitioners and patients.
In your article you wrote: “As radiology practices consolidate into larger hospital-led groups, it will be more feasible to implement such systems.” How will consolidation encourage greater implementation—or vice versa?
I know people in private groups who are now employees of larger medical centers. I don’t think these AI advancements are what’s driving that. Other