• Maciej A Mazurowski.
• Departments of Radiology, Electrical and Computer Engineering, and Biostatistics and Bioinformatics, Duke University, Durham, North Carolina. Electronic address: maciej.mazurowski@duke.edu.
• J Am Coll Radiol. 2019 Aug 1; 16 (8): 1077-1082.

AbstractThe increasingly realistic prospect of artificial intelligence (AI) playing an important role in radiology has been welcomed with a mixture of enthusiasm and anxiousness. A consensus has arisen that AI will support radiologists in the interpretation of less challenging cases, which will give the radiologists more time to focus on the challenging tasks as well as interactions with patients and other clinicians. The possibility of AI replacing a large number of radiologists is generally dismissed by the radiology community. The common arguments include the following: (1) AI will never be able to match radiologists' performance; (2) radiologists do more than interpret images; (3) even if AI takes over a large portion of the reading tasks, the radiologists' effort will be shifted toward interactions with patients and other physicians; (4) the FDA would never agree to let machines do the work of radiologist; (5) the issues of legal liability would be insurmountable; and (6) patients would never put complete trust in computer algorithms. In this article, I analyze these arguments in detail. I find a certain level of validity to some of them. However, I conclude that none of the arguments provide sufficient support for the claim that AI will not create a significant disruption in the radiology workforce. Such disruption is a real possibility. Although the radiology specialty has shown an astonishing ability to adapt to the changing technology, the future is uncertain, and an honest, in-depth discussion is needed to guide development of the field.Copyright © 2019. Published by Elsevier Inc.

9 keywords...

hide…