Adapting to AI: 4 key takeaways from a survey of attending radiologists, trainees

As artificial intelligence (AI) and machine learning (ML) technologies continue to evolve, more and more attention is being paid to their potential impact on radiology. Will these technologies make radiologists better at their job? Or will they just replace radiologists altogether?

To gain a better understanding of how the industry perceives radiology’s maturing relationship with AI, researchers surveyed attending radiologists and trainees at a single diagnostic radiology (DR) residency program in August 2017, publishing their findings in the Journal of the American College of Radiology.

Overall, 35 attending radiologists and 34 radiology trainees completed the survey. These are four key takeaways from their responses:

1. Respondents reported low exposure to scientific medical articles about AI.

Thirty-six percent of respondents reported not reading a single scientific medical article on the subject in the last year, and there was no difference in the answers of attending radiologists and trainees. If this statistic seems surprising, the authors were quick to emphasize that the number of articles about AI in “major” radiology journals was still relatively low in the summer of 2017.

“The relatively low exposure to AI and ML in the scientific literature is unlikely to be the only reason for our results,” wrote lead author Fernando Collado-Mesa, MD, department of radiology at the University of Miami Miller School of Medicine, and colleagues. “Additional reasons likely include the facts that advances in these fields and their application to DR are relatively recent and that informatics in general and AI and ML in particular, as they pertain to DR, are topics not yet formally taught at our program.”

2. Respondents are eager to learn more about these technologies, interact with them.

A large majority of attending radiologists and trainees reported that they plan to learn more about AI and ML technologies and would be willing to “create or train” a ML algorithm so it can do tasks typically completed by a radiologist.

3. Some trainees are concerned about what advances in AI could mean for their future.

When asked how AI and ML will influence their job in the next 10-20 years, a large majority of attending radiologists and trainees—said their job “will be dramatically different.” Two trainees (and no attending radiologists) believed their job will be obsolete in 10-20 years.

When asked if they would have chosen a different line of work had they known more about AI and ML when they chose a career in radiology, two trainees said “yes” and another 15 said “maybe.” Eight attending radiologists answered “maybe,” but none of them answered “yes.”

“The facts that trainees are more likely to be concerned with the implications AI and ML may have in their future jobs and that they are more likely to want to learn more about these topics is, at least in part, likely due to the early stage of their careers and the potential long-lasting implications these advances will have on their jobs,” the authors wrote. “Overall, we consider it encouraging that none of the attending radiologists and only a small proportion of the trainees in our program who responded to this survey think that their jobs as radiologists will be obsolete because of AI and ML and that most of them are willing to help in the development of AI and ML applications.”

In addition, more than 22 percent of attending radiologists and more than 11 percent of trainees said it will have either no influence or a minimum influence.

4. Survey responses led the authors to propose significant changes.

Collado-Mesa et al. concluded that they now plan to focus more on promoting scientific literature, as well as resources from groups such as the American College of Radiology, to attending radiologists and trainees.

“Furthermore, plans are under way to develop our own series of talks and lectures on these topics to ensure that our trainees and attending radiologists gain a better understanding of AI and ML and to eventually help them assume a fully aware and active role in the appropriate development, implementation, and use of AI and ML tools in DR,” they wrote.