AI Detection vs Risk Models in Mammography: Best Performance in Intended Settings

In a Swedish mammography cohort, Transpara Detection and Risk models achieved AUCs of 0.92 for detection and 0.81 for short-term risk, each performing best in their intended settings.

• Transpara Detection and Transpara Risk had similar AUC for detection (both 0.92; 95% CI 0.91–0.93; P=.93), outperforming other models (all P<.001).
• At double-reading sensitivity, Transpara Detection specificity was 97.1% (95% CI 96.8–97.3), the highest among models.
• For interval cancer risk prediction, Transpara Risk achieved 49.8% sensitivity at 90% specificity.

Automated summary

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