Click here to


Are you sure ?

Yes, do it No, cancel

Dosimetric Validation of Automatic Cardiac and Coronary Artery Segmentation in Breast Radiotherapy

R Finnegan1,2*, E Lorenzen3 , J Dowling1,4,5, L Holloway1,2,6,7,8, D Thwaites1 , C Brink3,9 (1) University of Sydney, School of Physics, Institute of Medical Physics, Sydney, Australia, (2) Ingham Institute for Applied Medical Research, Medical Physics Research, Liverpool, Australia, (3) Laboratory of Radiation Physics, Odense University Hospital, Odense, Denmark, (4) University of Newcastle, School of Mathematical and Physical Sciences, Newcastle, Australia, (5) CSIRO Health and Biosecurity, The Australian e-Health and Research Centre, Herston, Australia, (6) Liverpool and Macarthur Cancer Therapy Centers, Liverpool, Australia, (7) University of Wollongong, Centre for Medical Radiation Physics, Wollongong, Australia, (8) University of New South Wales, South Western Sydney Clinical School, Sydney, Australia, (9) Institute of Clinical Research, University of Southern Denmark, Odense, Denmark


(Sunday, 7/14/2019) 3:00 PM - 3:30 PM

Room: Exhibit Hall | Forum 6

Purpose: To validate the dosimetric accuracy and consistency of automatic segmentations of the whole heart and left anterior descending coronary artery (LADCA) for retrospective analysis of breast cancer radiotherapy patients, within the context of baseline uncertainty due to inter-observer variability.

Methods: An imaging dataset comprising 15 patients who were treated with external beam radiotherapy (EBRT) for breast cancer was contoured by 9 independent observers, both before and after adherence to common guidelines. Two separate atlas sets were used to generate automatic segmentations of the whole heart and LADCA, where a tube splining algorithm allows generation of automatic LADCA volumes. Segmentation accuracy was evaluated by comparing automatic and manual delineations, using overlap (Dice Similarity Coefficient, DSC) and surface (Mean Absolute Surface Distance, MASD) metrics. Dosimetric consistency of automatic segmentations was assessed by comparing DVHs for automatically segmented structures with manually contoured structures, as well as common dosimetric parameters.

Results: For the whole heart the segmentation accuracy was at a similar level to the inter-observer variability, with an average (standard deviation) DSC of 0.931 (0.016) and MASD of 2.12 (0.64). Dosimetric assessment of the generated automatic segmentations indicates suitability of automatic segmentations when considering the variability in dose as a result of differences in manual contouring, with consistent mean, median and maximum doses.

Conclusion: This work demonstrates the feasibility of automatic segmentation for use in delineating the whole heart and LADCA in radiotherapy planning images of patients receiving EBRT for breast cancer. Using a multi-observer dataset the differences between automatic and manual delineations are judged in the context of the uncertainty due to baseline inter-observer variability. This validation serves as an important step in the development of a framework for the analysis of large retrospective studies.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email