AI triage reduces screening mammography workload by half while maintaining sensitivity

An AI system used for triage in breast cancer screening reduced the number of mammograms requiring human reading by 50% while maintaining a population-level sensitivity of 94.3% compared with double reading (94.8%). Retrospective simulation using a large U.K. cohort.

• Design: Retrospective simulation study using a large U.K. breast screening cohort (n=159,198 exams).
• Key secondary finding: The AI-triage workflow resulted in a 66% reduction in human reads when sensitivity was allowed to drop to 90%. Specificity was similar between AI triage and standard double reading.
• Limitation: Retrospective simulation; results depend on the chosen operating point and require prospective validation in a real screening program.

RadPigeon summaries are original and for information only. They are not clinical advice.