Skip to main content

How is Artificial Intelligence Impacting Radiology?

Apr 6th 2020

What hot topics and clinical trends in radiology can you expect to hear more about in 2020 and beyond? We asked two experts for their take. David Naeger, MD, is Director of Radiology at Denver Health and Professor and Vice Chair of Radiology at University of Colorado School of Medicine. Elissa Price, MD, is Associate Professor of Clinical Radiology and Director of Clinical Operations in Breast Imaging for UCSF Department of Radiology and Biomedical Imaging.

Q. The most discussed topic in radiology these days seems to be artificial intelligence (AI). Do you see this much in your practice?

A. Dr. Naeger: I do not personally see much of this in daily practice, and I don’t think there are many widespread applications. Some sites across the country are demonstrating various AI products, but it is mostly in the experimental stage.

The big question is how to apply algorithms to modern medicine. For example, what is the work flow? The AI algorithm must be checked, so it becomes a question of efficiency. Also, “mostly correct” does not work well in medicine. When people or algorithms are incorrect, there needs to be accountability, and this has not been worked out for AI.

Dr. Price: In a very basic form, AI has existed in mammography for a long time. Specifically, computer-aided detection. Many practices use CAD as part of their standard workflow. Aside from that, although AI is a topic for frequent discussion in the breast imaging world, it is not having an impact on most clinical practices. There is a lot of research evaluating the use of AI to improve mammographic performance metrics. Results have been mixed.

Q. Where do you think AI in radiology will be in 5 to 10 years?

A.  Dr. Naeger: I think the accuracy of AI in targeted uses will improve dramatically. Some of the workflow and accountability issues will be resolved. Still, it will be expensive and a work in progress.

Dr. Price: There will be some form of AI in clinical practice, though I only see its realistic application in screening mammography. If a computer system could safely help us find more breast cancer, I am in favor of it.

Q.  Do you think breast imagers should be worried that AI may someday replace them?

A.  Dr. Price: I have no concerns whatsoever about that. So much of what we do involves patient communication and interaction. The diagnostic work up is an iterative process of problem solving and getting clinical context from the patient. No AI program can do all that.

Q.  Dr. Naeger, what about AI in thoracic imaging?

A.  Dr. Naeger: AI will likely help in screening chest radiographs for specific findings, as well as with pulmonary nodules on a chest CT. I think of radiographs the same way I do about EKGs, where a computer does a preliminary read of an EKG, which is then checked by a human. This could speed up preliminary interpretation and improve final accuracy.

For chest CTs, finding, measuring, and comparing nodules is tedious and won’t require a human once computers are good enough. The results will always need to be checked, however. I also can imagine the day when AI will help assess patterns of interstitial lung disease.

Explore Radiology CME