The Quantitative Measurement Of Volumetric Breast Density Based On Cone Beam Breast CT

Objective To compare the computer aided breast density(BD)measurement method based on cone beam breast computed tomography(CBBCT)images and the artificial vision evaluation of mammary gland classification.Material and methods A retrospective analysis was made of 195 patients received both CBBCT and mammography(MG)examination at Tianjin Medical University Cancer Institute and Hospital from May 2012 to September 2013.Sixty-four sides breasts were included in this research.According to the breast density classification standard in Breast Imaging Reporting and Data System(BI-RADS),5 radiologists classified 64 sides breasts based on MG images and CBBCT images,and got the majority report.3 radiologists measured the breast density based on CBBCT images using threshold segmentation method,and got the breast density after manual correction.All statistical analysis was performed with SPSS Statistics(version 18.0,IBM).The intraclass correlation coefficient(ICC)was used to compare the consistency between the intra-observer,inter-observer,threshold segmentation method and manual correction,dense and non dense breast density.The Kappa test was used to compared the consistency between the article classification based MG and CBBCT images.The correlation between the breast density measured with threshold segmentation and the BI-RADS density category was estimated by Spearman rank correlation coefficient(ρ).The receiver operating characteristic curve(ROC curve)was used to assess the sensitivity and specificity of the threshold segmentation method in reproducing the BI-RADS category and the optimal cut-off value of the threshold segmentation method in distinguishing dense and non-dense breast.Results According to the consensus rating of BI-RADS breast density categories reported on the mammograms,two breasts(3.13%)were classified as fatty,five breasts(7.81%)were classified as scattered,fifty breasts(78.12%)were classified as heterogenous,and seven breasts(10.94%)were classified as extremely dense.The intraclass correlation coefficient of BI-RADS classification was 0.792(95% CI: 0.719 ~ 0.854)among the five radiologists in four scale classification,0.714(95% CI: 0.625~0.796)in two scale classification.According to the consensus rating of BI-RADS breast density categories reported on the cone beam breast computed tomography,one breast(1.56%)was classified as fatty,three breasts(4.69%)were classified as scattered,fifty breasts(78.13%)were classified as heterogenous,and ten breasts(16.39%)were classified as extremely dense.The intraclass correlation coefficient of BI-RADS classification was 0.789(95% CI: 0.716~0.852)among the five radiologists in four scale classification,0.797(95% CI: 0.726~0.858)in two scale classification.The consistency between the artificial classification based mammograms and cone beam breast tomography was good.The Kappa value was 0.618(P<0.01).The intra-observer consistency of the CBBCT based classification using threshold segmentation method breast density measurement was 0.962(95% CI: 0.939~0.977)and inter-observer consistency was 0.967(95% CI: 0.966~0.979),the intra-observer and inter-observer ICC values were 0.975(95% CI: 0.959~0.985)and 0.978(95% CI: 0.966~0.986)in manual correction measurements.The value of ICC between threshold-based segmentation method and manual correction was 0.996(95% CI: 0.994 ~ 0.998).The ICC values of threshold-based and manual correction measurement in non-dense and dense breasts were 0.950(95% CI: 0.707~0.991)and 0.998(95% CI:0.997~0.999),respectively.A positive correlation was found between the mean breast density estimated by the quantitative method and the BI-RADS categories based on mammograms(ρ=0.645).A positive correlation was found between the mean breast density estimated by the quantitative method and the BI-RADS categories based on cone beam breast tomography(ρ=0.675).The classification of dense and non-dense breast using quantitative measurement achieved a sensitivity of 87.7% and a specificity of 85.7% at a cut-off value of 10.13%.Conclusion Threshold-based segmentation method based on CBBCT is a more stable and accurate computer aided measurement of breast density.There is a positive correlation between the quantitative breast density in CBBCT images and BI-RADS classification in MG.It is expected to be applied to large-scale screening of breast cancer and provide more information to predict the risk of breast cancer.