Tuesday, December 1 | 9:30 a.m. to 10:30 p.m. | T3-SSBR05-6 | Room S406A
In this presentation, researchers will report whether AI-assisted mammography improves diagnostic performance differently for junior and senior readers and whether it narrows experience-related performance gaps.
The retrospective multi-reader, multi-case study analyzed 2,430 screening mammograms, including 33 cancers, with a histopathologic or follow-up reference standard of at least 12 months.
Four radiologists were grouped by experience: two junior readers with five years or less and two senior readers with at least 10 years. Each reader interpreted every case twice, once unaided and once with AI assistance (Lunit Insight MMG; Lunit), separated by a four-week washout period.
The researchers reported success from the junior radiologists using AI, though the senior radiologists also improved with AI assistance. Sensitivity rose from 78.8% to 87.9% for junior radiologists and from 90.9% to 97.0% for senior radiologists.
False positive rates fell 45% for juniors (3.21% to 1.75%) and 25% for seniors (1.96% to 1.46%), and the difference in reduction between groups was significant (p = 0.004). Positive predictive value increased from 25.2% to 40.8% for juniors and from 39.0% to 47.8% for seniors. F1 scores improved by about 46% for juniors and by 17% for seniors. Reading time dropped by 34% for juniors and by 27% for seniors. Performance gaps between experience levels narrowed across all metrics.
The findings support wider use of AI as a decision-support tool to improve consistency and efficiency in breast cancer screening, the study authors highlighted.
Attend this presentation to learn more.
