Purpose: Locoregional breast cancers require volumetric modulated arctherapy (VMAT) to best conform isodoses to target volumes. However, in the case of breast irradiation, many beam setup are proposed throughout the literature. To determine which beam setup best fitted a patient, we hypothesized patient anatomy could be used as a predictor of beam setup.
Methods: From our patient database, we selected 106 computed-tomography scans for which all structures of interest were contoured. Using Eclipse Scripting Application Programing Interface (ESAPI, Varian), we developed a first script to measure distance, volume and angles between the contoured structures. Eleven anatomical indicators such as breast volume, distance between breast and heart or Haller index were measured.Then, we developed a second script to automatically plan five beam setups on each patient. The script acted the same way for every patient. It placed the isocenter with respect to possible collisions, created the beams and launched RapidPlan (Varian) to optimize the dose distribution. After dose calculation, the script collected about 40 dosimetric indicators.
Results: We analysed our results by constituting a final database including anatomical and dosimetric indicators together. Correlation analysis showed that breast volume, rib chord length and the falling shape of the breast particularly influenced dose distribution meanwhile the angle of the controlateral breast and height of the ipsilateral breast along the heart did not.
Conclusion: These preliminary results will allow us to remove certain indicators and train a machine learning algorithm to determine beam setup for future patients.
Breast, Radiation Therapy, Software
TH- External Beam- Photons: Treatment planning using machine learning/Knowledge Based Planning/automation