AI breast cancer tools disappoint radiologists in real-world use

A survey published in Clinical Imaging examined the real-world performance of FDA-approved AI tools for breast cancer detection among 215 members of the Society of Breast Imaging. The findings paint a sobering picture: while about half of respondents already use such tools and 11 percent plan to, the technology is delivering far fewer benefits than anticipated.

When asked about specific outcomes, the gap between expectation and reality became stark. Radiologists had expected 59 percent to report lower recall rates (the rate at which cancers are missed), but only 35 percent actually reported this improvement. They had hoped 36 percent would see fewer unnecessary biopsies, yet only 9 percent experienced this reduction. On burnout, which was seen as a major potential benefit, 56 percent of radiologists had expected relief, but only 29 percent reported less burnout in practice. Most radiologists view the tools mainly as a second opinion rather than as a transformative technology.

The survey, led by Joud Almogati at UC San Diego Health, identified costs and lack of institutional support as the biggest barriers to adoption. These obstacles suggest that even where AI tools do work, institutional and financial constraints may limit their deployment. The findings reflect a broader tension in AI adoption: the technology industry has a history of making bold claims about labor displacement. About ten years ago, prominent AI researchers predicted that radiologists would soon be out of work. Similar forecasts are being made today about knowledge workers who use computers. Nvidia CEO Jensen Huang has criticized such predictions, calling them a "God complex" among those who prophesy about AI-driven job loss. Almogati's research suggests a more complicated reality—one where AI is useful but limited, and where institutional factors matter as much as the technology itself.