ParticipantsYuan Tian, Beijing, China (Presenter) Nothing to Disclose
Jing Li, MD, PhD, Beijing, China (Abstract Co-Author) Nothing to Disclose
Erni Li, MD, Beijing, China (Abstract Co-Author) Nothing to Disclose
Zhang Renzhi, MD, Beijing, China (Abstract Co-Author) Nothing to Disclose
Ning Guo, Beijing, China (Abstract Co-Author) Nothing to Disclose
Shunan Che, MD, Beijing, China (Abstract Co-Author) Nothing to Disclose
Chenglu Ke, Beijing, China (Abstract Co-Author) Nothing to Disclose
Xiaohong R. Yang, Rockville, MD (Abstract Co-Author) Nothing to Disclose
Gretchen Gierach, Bethesda, MD (Abstract Co-Author) Nothing to Disclose
tianyuanlaura@hotmail.com
PURPOSEThe purpose of our study was to assess the agreement of automated volumetric mammographic breast density with the radiologists' classification by using the BI-RADS density category (5th edition), and to analyze the clinical-radiologic factors that may influence the discordance between the two density measurements.
METHOD AND MATERIALSA total of 7971 full-field digital mammograms with standard views were retrospectively analyzed. Breast density measurements evaluated by radiologists according to BI-RADS category (5th edition) and by an automated volumetric breast density method (VBDM) which was used to measure VBD (% breast density) and VDG (density grade) were compared. A weighted kappa value was calculated to assess the degree of agreement among the visual and volumetric assessments of the density category and each subject was classified into an concordance or discordance group. A number of clinical-radiologic factors including age, history of breast surgery, indication for mammography, volumes of fibroglandular tissue or total breast and the percentage breast density were compared between the two groups.
RESULTSThe agreement between breast density evaluations by radiologists and VDG was fair (kappa = 0.346) when using the four-grade scale (A/B/C/D), and moderate (k value = 0.597) when using the two-grade scale (A-B/C-D). VBD showed a significant positive correlation with visual assessment by radiologist (Spearman's p=0.691, p < 0.01), but the distribution of density category was statistically significantly different among visual and volumetric measurements (p < 0.001). Category D was more frequently assigned by VBDM (43.9%) than by the radiologists (17.1%). Discordant subjects were more likely to be younger (p < 0.001), had undergone mammography for diagnostic purposes (p = 0.024), and have higher volumetric breast density (p < 0.001) compared with concordant subjects.
CONCLUSIONMore mammograms were classified as dense breast tissue using VBDM, as compared with visual assessments according to the BI-RADS fifth edition. And age, indication for mammography, and volumetric breast density may contribute to the differences between assessments by radiologists and by VBDM.
CLINICAL RELEVANCE/APPLICATIONConsidering the significant positive correlation between VBD and radiologists' classification, the automated method may be used in the future to evaluate the quantitative breast density data.