Early adoption of breast cancer detection AI tools is rising, but fewer AI users say these tools have meaningful clinical benefit in practice, according to research published August 11 in Clinical Imaging.
AI users reported “more modest” perceived clinical impact than anticipated, with few noting improvements in callback rate, biopsy rate, or burnout, wrote a team led by Azadeh Elmi, MD, from the University of California, San Diego Health.
“Theses preliminary observations suggest the need for continued refinement and further validation of breast cancer detection AI tools to achieve a meaningful impact on clinical practice,” the Elmi team wrote.
Over three-fourths of AI algorithms approved by the U.S. Food and Drug Administration (FDA) are radiology-related AI tools. Recent studies have shown that AI could be a good breast cancer detection support tool in high-volume screening mammography. Some reported benefits in these studies include lower recall rates and lessened workloads for radiologists.
Elmi and colleagues studied trends in early AI adoption, perceptions, and clinical impact that AI tools have for breast cancer detection. The study included survey responses from 215 members of the Society of Breast Imaging.
About 47% of respondents said they implemented breast cancer detection AI tools. And 41.8% said they had not implemented these tools, 11.2% said they planned to.
The respondents also indicated that implementation cost was the most reported barrier to AI adoption (53.3%). Other reported barriers included concerns about accuracy and reliability, limited supporting evidence of AI tools, and possible workflow disruption.
The researchers also reported the following findings:
-
Non-AI users more often anticipated reductions in recall rates (59.3%) compared with AI users reporting reductions (34.7%; p = 0.003).
-
Non-AI users more often expected reductions in biopsy rates (36.4% vs. 9.1%; p
-
Non-AI users also more often expected reduced burnout than AI users (56.0% vs. 29.4%; p
-
Both AI users and non-AI users had similar perceptions of patient outcomes (p = 0.87).
The results are in line with other studies measuring radiologist satisfaction and workflow benefits from AI tools, the study authors wrote.
This finding implies that radiologists continue to rely on clinical judgment for characterizing mammographic findings, “using AI primarily as a decision-support rather than a directive tool.”
Read the full study here.