A majority of patients appear to support the use of AI in radiology, but they’re not ready for it to act as a standalone reader, according to study results presented at the RSNA meeting.
The finding is from a survey of 1,041 Irish patients who attended the radiology department at Beaumont Hospital in Dublin and filled out a 16-item questionnaire that included questions on basic demographics, general attitudes toward AI, and radiology-specific applications of AI.
“The majority of respondents in this study would not agree to AI as the sole reader of their imaging, even if quicker, more efficient, and more accurate. Over half of respondents wish to be asked for written consent if AI is to be used in the interpretation of their radiological imaging,” noted presenter Hayley Briody, MD, in a December 3 session.
AI models such as machine-learning algorithms have significant potential to improve radiology workflows, reduce the time to notification of significant findings, and possibly improve accuracy, yet their impact on patients and their perceived trust in this “AI revolution” is presently insufficiently studied, Briody said.
To bridge the knowledge gap, Brody and colleagues distributed a survey among patients in the waiting areas of their department over a 10-month period between July 2024 and April 2025. A majority of respondents were between the ages of 50 and 69 (42.3%), were of white Irish ethnicity (85.4%), and were educated beyond high school (47.6%).
According to the results, 55.1% of respondents agreed or strongly agreed that the use of AI in healthcare is a good idea. Participants older than 70 were more likely to respond “I don’t know” when asked about their interest in AI (p < 0.05), and a higher level of education (master's degree or higher) was associated with an increased likelihood of strong interest in AI (p < 0.05), Briody reported.
When asked if they would allow a radiologist and AI to interpret their imaging, over two-thirds (67.3%) of respondents agreed or strongly agreed. However, when asked if they would allow AI to act as the sole reader of their imaging, only 20.4% respondents agreed or strongly agreed. Most patients (67.2%) indicated they would still prefer a radiologist to interpret their imaging even if AI were a more accurate reader.
Responding to the statement, “I think replacement of radiologists by AI could happen in the future,” 40% disagreed or strongly disagreed, and further, only 4% of patients indicated the use of AI for reporting of radiology images did not need to be disclosed, with the majority (54%) wishing to be asked for written consent.
Overall, patients were agreeable to having their images shared for the purposes of research into AI tool development (70.9% agreed or strongly agreed), Briody noted.
“Most patients do have a positive attitude toward AI in radiology, but they favor it being used as a collaborative tool with the radiologist,” she said.
Ultimately, the findings of the study should be considered in clinical practice going forward, Briody concluded.
