Purpose

To assess the diagnostic performance of computer-aided detection (CAD) for full-field digital mammography (FFDM) correlated with breast density assessed by an automated breast density measurement system (Hologic, Quantra) in breast cancers and age-matched healthy controls.

Materials and methods

Two hundred breast cancers imaged with FFDM and 200 age-matched healthy controls were evaluated retrospectively using CAD. A CAD mark was scored true-positive if it correctly indicated a malignant lesion. All other CAD marks were considered false. CAD sensitivity and specificity were calculated and correlated with mammographic breast density (%).

Results

CAD correctly identified 157 of the 200 cancers, a sensitivity of 79%. Sensitivity was suggestively but nonsignificantly lower with increased density (P = 0.09). In those cancer cases with density at or below the median of 20%, sensitivity was 82%, compared with 75% in those with density above the median. The presence of one or more false CAD prompts was suggestively but not significantly more likely in controls than cases (87% vs. 80%, P = 0.06). The number of false prompts was significantly higher in controls (average 3.6 vs. 2.6, P <0.001). False prompts were significantly less likely with higher density (P = 0.008). False prompts were present in 86% of cases and controls with density at or below the median, and in 81% of those with density above the median.

Conclusions

Increased breast density is significantly associated with higher specificity of CAD, and there is suggestive evidence that it is also associated with lower sensitivity.