Room: Stars at Night Ballroom 4
Purpose: To develop an automated, machine-learning-based method to generate consistent, objective and fast treatment plan assessment for both FB and BH, to prospectively determine, on a patient-specific basis, if a clinically meaningful benefit from using breath-hold exists.
Methods: A knowledge model to estimate heart and lung dose sparing achievable with breath-hold (BH) versus free-breathing (FB) in treating left-sided breast cancer was developed previously utilizing clinical whole breast treatment plans. In this study, we refined and extended our model by including composite single-isocenter 3D conformal plans for the treatment of breast/chestwall and regional nodes (supraclavicular, axillary and internal mammary lymph node). 13 patients had FB and BH scans and were treated BH, and 7 patients had only FB scans and were treated FB. The database used for modelling consisted of: 7 FB clinical plans, 13 BH clinical plans and 13 FB plans created retrospectively offline for the same patients (in order to augment the training data sample so a wide range of anatomical variations are represented in the database). The correlation between plan dose metrics and patient anatomical features was identified by a step-wise regression method.
Results: The correlation between the predicted mean lung and the actual plan value has R2=0.68; for mean heart dose, R2=0.39. The weak correlation coefficient for heart is consistent with our previous study. The likely reason is that, unlike IMRT plans, the isodose lines do not conform to the PTV contour in 3D plans without fluence modulation. They are significantly affected by the field borders.
Conclusion: The composite dose in 3D conformal total nodal breast treatment can be predicted with reasonable accuracy. Further improvement is needed in order to take into account the different dosimetric features in 3D plans.
Not Applicable / None Entered.