Comparison of subjective and fully automated methods for measuring mammographic density

Background Breast radiologists of the Norwegian Breast Cancer Screening Program subjectively classified mammographic density using a three-point scale between 1996 and 2012 and changed into the fourth edition of the BI-RADS classification since 2013. In 2015, an automated volumetric breast density assessment software was installed at two screening units. Purpose To compare volumetric breast density measurements from the automated method with two subjective methods: the three-point scale and the BI-RADS density classification. Material and Methods Information on subjective and automated density assessment was obtained from screening examinations of 3635 women recalled for further assessment due to positive screening mammography between 2007 and 2015. The score of the three-point scale (I = fatty; II = medium dense; III = dense) was available for 2310 women. The BI-RADS density score was provided for 1325 women. Mean volumetric breast density was estimated for each category of the subjective classifications. The automated software assigned volumetric breast density to four categories. The agreement between BI-RADS and volumetric breast density categories was assessed using weighted kappa (k w ). Results Mean volumetric breast density was 4.5%, 7.5%, and 13.4% for categories I, II, and III of the three-point scale, respectively, and 4.4%, 7.5%, 9.9%, and 13.9% for the BI-RADS density categories, respectively ( P for trend < 0.001 for both subjective classifications). The agreement between BI-RADS and volumetric breast density categories was k w = 0.5 (95% CI = 0.47–0.53; P < 0.001). Conclusion Mean values of volumetric breast density increased with increasing density category of the subjective classifications. The agreement between BI-RADS and volumetric breast density categories was moderate.