Background and purpose Radiotherapy for breast cancer can increase the risks of heart disease . Patient-specific risk assessment may be improved with the inclusion of doses to cardiac substructures . The purpose of this work was to use automatic segmentation to evaluate substructure doses and develop predictive models for these based on the dose to the whole heart . Material and methods Automatic segmentation was used to delineate cardiac substructures in a Danish breast cancer trial (DBCG HYPO) dataset comprising over 1500 Danish women treated between 2009 and 2014 . Trends in contouring practices and cardiac doses over time were investigated, and models to predict substructure doses from whole heart dose parameters were fit to the data .
Results: Manual contouring consistency improved over the study period when compared with automatic segmentation; systematic differences between automatically and manually defined heart volume decreased from 106 cm to 12.0 cm . Doses to the heart and cardiac substructures also decreased . Mean whole heart doses for left-sided treatments in 2009 and 2014 were 1.94±1.19 Gy and 1.29±0.69 Gy (average ± SD), respectively . Prediction of mean substructure doses is accurate, with R scores in the range 0.45-0.95 (average 0.77), depending on the particular structure . Conclusion This study reports heart and cardiac substructure doses in a large breast cancer cohort . Predictive models generated in this work can be used to estimate mean cardiac substructure doses for datasets where patient imaging and dose distributions are not available, provided the tangential field techniques are consistent with those used in the trial.