Abstract
BACKGROUND: Accurate mean glandular dose (MGD) estimation is important in breast cancer screening programs to balance diagnostic benefit with radiation risk. PURPOSE: This study aimed to compare the performance of the Dance and AAPM/EFOMP Task Group 282 (TG282) breast dosimetry methodologies using model versus image-derived breast density metrics. METHODS: This study analyzed over 80,000 digital mammography images acquired in 2023 from BreastScreen Queensland (BSQ). Data were obtained from Siemens and Hologic systems and included 2D cranio-caudal and mediolateral oblique views. Images with compressed breast thickness (CBT) between 20 and 100 mm were included. Volumetric breast density (VBD) and glandularity were extracted using Volpara software. MGD was estimated using both Dance and AAPM/EFOMP TG282 models, employing model-based median and image-measured breast density metrics. The ratios of MGD estimated using model medians to those using measured values ( RDance and RTG282 ) were analyzed across CBT, and Pearson correlations (r) were computed. RESULTS: The Dance model median glandularity overestimates population-derived glandularity for most CBT, resulting in RDance > 1 at low CBT, with the trend reversed for CBT > 80 mm. RDance showed moderate positive correlation with CBT (r = 0.57 Hologic; r = 0.63 Siemens, p < 0.001). RTG282 remained close to unity across CBT, with weak negative correlations (r = -0.17 Hologic; r = -0.04 Siemens, p < 0.001), indicating consistency between model and measured VBD. CONCLUSIONS: The AAPM/EFOMP TG282 dosimetry model exhibited stronger agreement between median model-predicted and population-specific measured breast density metrics than the Dance model. This resulted in improved consistency in ratios of estimated MGD values based on median model-to-measured breast density metrics across the full range of CBT, when using the AAPM/EFOMP TG282 methodology.