Breast arterial calcifications on mammography: intra- and inter-observer reproducibility of a semi-automatic quantification tool

A strong association between breast arterial calcifications (BAC) and cardiovascular disease has been demonstrated. However, BAC quantification tools are lacking. We evaluated the intra- and inter-observer reproducibility of a semi-automatic tool for BAC quantification on digital mammograms. A multivendor image dataset of 212 mammographic views, 106 cranio-caudal (CC) and 106 medio-lateral oblique (MLO), were retrospectively selected from 53 subjects if BAC were seen in at least one view. Images were segmented twice by two intensively trained residents in Radiodiagnostics with > 6-month experience in mammography using a semi-automatic software. The two observers (O1, O2) independently positioned rectangular ROIs where they recognized BAC on both CC and MLO views, separately. The adaptive thresholding algorithm automatically provided the BAC amount in mm2. Number, size, and position of the ROIs were observer-dependent. Total BAC amount was calculated for each patient. Bland–Altman analysis was used. Total BAC amount was 56.6 (IQR 18.1–91.1) and 41.0 (IQR 18.8–90.9) for O1 and O2, respectively. Intra-observer Bland–Altman analysis showed a bias of 11.9 mm2, a coefficient of repeatability of 32.7 mm2, an average measurement of 72.8 mm2, for a 55% reproducibility; the same data were − 7.0, 61.4, 63.4 mm2, and only 3%, respectively, for the inter-observer analysis. Our semi-automatic tool for BAC quantification showed a poor reproducibility. These results pointed out that the human identification of BAC represents the main source of variability. Further research is needed to translate BAC quantification into clinical practice.