J Am Coll Radiol. 2018 Mar;15(3 Pt B):509-511. doi: 10.1016/j.jacr.2017.12.025. Jha S1, Topol EJ2.
Predictions about the impact of artificial intelligence (AI) on radiology dwell on AI’s ability to perform the task of radiologists, including pattern recognition such as detecting fractures, finding nodules, flagging pulmonary emboli, fetching and comparing to old studies, and bringing urgent studies to the top of the reading list, among others. In such predictions, AI is an adjunct to radiologists, a rad-helper, whose primary purpose is to scale the productivity of radiologists [1]. We believe this vision, though accurate, limits imagination and shortchanges AI, the radiologist, and the patient. Rather, we believe AI’s true impact on radiologists will be its ability to integrate information (ie, information management) rather than hasten information extraction.
The distinction between the two visions is profound. With AI as a rad-helper, radiologists in 20 years’ time will still be reading films, albeit much faster and with more time to attend tumor boards and interact directly with patients. There need be no change in pedagogy. However, the radiologist as information manager needs a fundamental change in education, training, basic pedagogy, and way of thinking. Radiology residency will need pedagogic deconstruction and reconstruction. Even medical school curricula will need revamping.