[Impact of mammographic breast density on computer-assisted detection (CAD) in a breast imaging department]. / Impacto de la densidad mamaria mamográfica en el funcionamiento de un sistema de detección asistido por ordenador en una unidad de patología mamaria.

OBJECTIVE: To evaluate whether breast density influences the sensitivity of a computer-assisted detection (CAD) system for the detection of breast cancer. MATERIAL AND METHODS: We prospectively studied 8750 digital mammograms with an associated CAD system. We used BI-RADS criteria to classify breast density. We calculated the overall sensitivity of the radiologist and of the CAD system, as well as the sensitivity for each projection and type of finding in relation to the mammographic density of the breast. Finally, we analyzed the interval carcinomas. We used SPSS 11 for all statistical analyses. RESULTS: The overall sensitivity of the CAD system was 88.5% (95% CI: 83.2-92.7%), and the overall sensitivity of the radiologist was 93.5% (95% CI: 84.4%-95.5%). The sensitivity of the craniocaudal view was 81.6% (95% CI: 76.5-90.7%) vs 76.5% (95% CI: 69.3-89.3%) for the mediolateral oblique view. The sensitivity for microcalcifications was 98.6% (95% CI: 96.5-99.7%), and the sensitivity for masses 83.4% (95% CI: 81.2-91.7%). We detected discrepancies smaller than 20% both for microcalcifications present in the four types of densities and for masses with densities 1 and 2. In masses with density 3 the discrepancy was 20.8% and in those with density 4 it was 55%. The CAD system failed to mark only 9.1% (9/94) of the cancers presenting as masses. Half of the interval carcinomas were found in type 4 density and 75% manifested as masses, asymmetries, and distortions. The CAD system had marked 35.7% of the carcinomas. CONCLUSIONS: The craniocaudal view was more sensitive, although this difference was not statistically significant. The sensitivity of CAD was high for microcalcifications in all four density types; however, CAD's sensitivity for masses was low in density types 3 and 4. The CAD system only failed to mark 9.1% of the cancers presenting as masses but was not sensitive for the other two radiological findings included in this marking. Half of the interval carcinomas occurred in type 4 densities and 35.7% had been marked by the CAD system.