Comparison of breast percent density estimated from digital mammograms and central reconstructed tomosynthesis slice images

We analyzed breast percent density (PD) estimated from 35 women with existing or recently detected abnormalities. Analyzed were digital mammograms (DM) and the central digital breast tomosynthesis (DBT) reconstructed slices. PD was estimated from the breast contralateral to cancer. We have examined the effect of modifying the DBT reconstruction filter on the accuracy of dense tissue segmentation; we selected the filter that effectively reduced image intensity overshoot near the breast edge, which interfered with segmentation. The Pearson correlation coefficient between PD estimates from DM and the central reconstructed DBT slices was r = 0.90. The corresponding quadratic-weighted kappa coefficient was kappa = 0.78, including substantial agreement. The observed results are comparable with the agreement between PD estimates from DM and the central DBT projection images (r = 0.89 and kappa = 0.74). This suggests that PD is robust to variations in acquisition conditions.