Measuring mammographic density: comparing a fully automated volumetric assessment versus European radiologists’ qualitative classification

Abstract

ObjectivesBreast Imaging-Reporting and Data System (BI-RADS) mammographic density categories are associated with considerable interobserver variability. Automated methods of measuring volumetric breast density may reduce variability and be valuable in risk and mammographic screening stratification. Our objective was to assess agreement of mammographic density by a volumetric method with the radiologists’ classification.MethodsEight thousand seven hundred and eighty-two examinations from the Malmö Breast Tomosynthesis Screening Trial were classified according to BI-RADS, 4th Edition. Volumetric breast density was assessed using automated software for 8433 examinations. Agreement between volumetric breast density and BI-RADS was descriptively analyzed. Agreement between radiologists and between categorical volumetric density and BI-RADS was calculated, rendering kappa values.ResultsThe observed agreement between BI-RADS scores of different radiologists was 80.9 % [kappa 0.77 (0.76–0.79)]. A spread of volumetric breast density for each BI-RADS category was seen. The observed agreement between categorical volumetric density and BI-RADS scores was 57.1 % [kappa 0.55 (0.53-0.56)].ConclusionsThere was moderate agreement between volumetric density and BI-RADS scores from European radiologists indicating that radiologists evaluate mammographic density differently than software. The automated method may be a robust and valuable tool; however, differences in interpretation between radiologists and software require further investigation.Key Points• Agreement between qualitative and software density measurements has not been frequently studied. • There was substantial agreement between different radiologists´ qualitative density assessments. • There was moderate agreement between software and radiologists’ density assessments. • Differences in interpretation between software and radiologists require further investigation.

Comparison between software volumetric breast density estimates in breast tomosynthesis and digital mammography images in a large public screening cohort

“…density scoring in DM images was performed prospectively as part of the reading protocol in the MBTST by one of five radiologists, all with more than ten years of experience in breast radiology. The observed agreement between BI-RADS scores of different radiologists was 80.9% [linear weighted ĸ = 0.77 (95% CI: 0.76–0.79)] based on a previous study on a subset of the MBTST population with the same readers as in the present study [24]. The MBTST interim results (n = 7500) have been described in detail previously [4].…”

“…In a recent study on approximately half of the MBTST population, the agreement between BI-RADS 4 th ed. density scores and a non-masking software (Volpara) in DM images only was shown to be equally moderate (observed agreement 57.1%, ĸ = 0.55) [24]. The difference between BI-RADS 4 th ed.…”

“…A subset of this population (n = 348) was analysed with VBDA prototype software and reported as a conference proceeding [18]. Further, subsets of the MBTST population were published previously on other topics [4, 24, 26, 27] however, none of them including present software version.…”

Positive predictive values by mammographic density and screening mode in the Norwegian Breast Cancer Screening Program

“…No studies have previously examined the three-point scale in relation to estimates of volumetric breast density obtained by the fully automated software. However, several studies have compared fully automated methods for breast density assessment to BI-RADS (85,96,102,103,105,(156)(157)(158)(159)(160)(161)(162), indicating a clear increase in volumetric breast density as the BI-RADS density category increased (103) or a strong relationship between volumetric breast density and the BI-RADS categories (96). Similar trends have been reported in other studies (102,(157)(158)(159)162).…”

Volumetric breast density measurement for personalized screening: accuracy, reproducibility, consistency, and agreement with visual assessment

“…Ng and Lau 7 have identified six requirements (denoted in their paper as "sanity checks") that should be fulfilled by an automated breast density measurement software: Some studies exist that validate existing software applications for automated breast density assessment. [14][15][16][17][18][19] Typical aspects assessed by these studies are accuracy (comparing measured breast density to an objective ground truth), reproducibility (see Ng and Lau's requirements 1 to 5), consistency (see Ng and Lau's requirement 6), and agreement with visual assessment (comparing classified breast density to a subjective reference).…”