2019 Volume 6 Issue 1 Pages 37-43
Objective: In Japan, to achieve sensitivity in mammography, a current approach is to use ultrasonography together with mammography for dense breasts. Although more objective and reproducible determination of dense breasts is necessary, few studies have shown correlations between radiologists’ assessments and fully automated volumetric measurements. This study was undertaken to assess the level of agreement between the two evaluation methods and to identify factors associated with disagreement between them.
Methods: Between April and June 2017, 1,456 women were included in the study. Agreement in breast density assessments between two radiologists and automated software was evaluated using kappa (κ) values. We also evaluated factors that could have contributed to any differences using univariate and multivariate analysis.
Results: Inter-observer agreement was good (κ=0.701). Breast density assessments made by radiologists versus software exhibited fair agreement (κ=0.224). Age, menopausal stage, difference in bilateral breast density, fibroglandular tissue volume, breast tissue volume and volumetric breast density (VBD) were significantly different between the two methods in univariate analysis. In multivariate analysis, VBD was an independent correlative factor for greater possibility of significant disagreement (Odds ratio for agreement = 0.755; 95% CI, 0.730–0.782).
Conclusions: Automated volumetric breast density measurements showed fair agreement with radiologists’ visual assessments. Low VBD was a contributing factor in disagreement. Although the automated method may have value, further research is necessary in Japan before it is put into widespread use.
