AI applications are projected to result in an up to 49% decrease in radiologist hours worked in the next five years, according to a study by a team at Stanford University in Stanford, CA.
The main reasons for this are its use in radiology report drafting for all modalities and study delegation for radiography and mammography, noted lead author Curtis Langlotz, MD, PhD, of the school’s Institute for Human-Centered Artificial Intelligence.
“Given the relatively static radiology workforce and the continued growth in imaging volumes, radiologist job loss is unlikely for the foreseeable future,” the group wrote. The study was published December 22 on medRxiv.org.
About 10 years ago, many experts were predicting that machine learning would displace much of the work of radiologists. Since then, a substantial radiology AI industry has arisen, and there is now sufficient published evidence to support a quantitative task-based analysis to predict the effect of AI on the radiology workforce, Langlotz and colleagues wrote.
To that end, the group reviewed the literature to establish the tasks on which radiologists spend their time. These included performing and interpreting studies (66.7%), protocoling (5.5%), communicating with technologists or nurses (3.9%), communicating with other radiologists (3.7%), communicating with providers (3.3%), answering order questions (0.7%), communicating with patients (0.2%), other (6.8%), and personal (9.1%).
Next, the team developed categories of AI applications that could affect these tasks. They used published evidence to estimate the effect of each AI application on each radiology task using a five-year time horizon. Two areas that emerged with the strongest positive ROIs for radiologists were report drafting (a 15% productivity increase for radiography) and study delegation strategies that reduce volume by placing radiologists out of the loop, the researchers reported.
“Large productivity gains could be achieved if reports are drafted automatically for review and signature. These tools can serve a function like a trainee in practices without trainees,” the group wrote.
Regarding study delegation strategies, several studies have shown that a subset of radiology studies can be identified by AI that are almost certainly normal and require no further human review, including up to 50% of screening mammograms and up to 40% of outpatient radiographs, according to the findings.
Overall, the analysis shows that AI will likely cause an up to 33% reduction in the need for radiologists over the next five years, with a range of 14% to 49%.
Ultimately, the findings should be interpreted in the context of the steady growth in imaging volumes, with a near doubling of the volume of cross-sectional studies over the past decade, the researchers noted. As this trend is likely to continue, the projected reduction in hours worked by radiologists is likely to be more than offset by the growth in imaging volumes over the same period, they wrote.
“This suggests AI may cause shifts among radiologist tasks, rather than a reduction in the need for radiologists,” Langlotz and colleagues concluded.
The full study is available here.