Room: AAPM ePoster Library
Purpose: Prediction of relative biological effectiveness (RBE) for proton therapy is subject to large uncertainties. This includes the constant RBE=1.1 model. To mitigate this, a robust optimization method considering geometrical and biological uncertainties is proposed.
Methods: IMPT plans of 30 fractions for a prostate (2.18 Gy/fraction) and an intracranial case (1.82 Gy/fraction) were robustly optimized against setup and range uncertainties for the CTV (3%, 3mm, 21 scenarios) using minimax optimization in a research version of RayStation v9A. Identical RBE-weighted dose (RWD) objectives were used for three RBE model scenarios: (1) RBE=1.1, (2) the Wedenberg RBE model, and (3) where scenarios (1) and (2) were combined to consider the RBE uncertainty. In this study, the robust uniform RWD CTV objective from scenario (1) had doubled the weight compared to the corresponding one from scenario (2), whereas the opposite weight-relation was used for all organs-at-risk (OARs) objectives. The assumed a/ß was 1.5, 3, 10 and 2 Gy for CTV-prostate, OARs-prostate, CTV-intracranial and OARs-intracranial, respectively.
Results: The CTV RWD was robust against setup and range errors. The nominal CTV-prostate near-minimum and near-maximum RWD (RWD98%/RWD2%) in percentage of the prescribed dose was 99%/102% (RBE=1.1) and 105%/110% (Wedenberg) for scenario (1), whereas scenario (2) gave 88%/97% and 95%/109%, respectively. Scenario (3) mitigated the RBE model dependence resulting in 95%/99% (RBE=1.1) and 101%/106% (Wedenberg) with similar or lower OAR doses than scenarios (1) and (2).
The RWD for CTV-intracranial was comparable for all scenarios and both RBE models (RWD98%=95%, RWD2%=105%). Scenario (3) lowered the brainstem RWD2% with approximately 3 Gy (RBE) compared to scenario (1) for both RBE models.
Conclusions: The proposed method simultaneously mitigates the RBE model dependence while ensuring robustness against geometrical uncertainties. The method is flexible, where the user defines model weightings, and works for multiple RBE models with multiple model parameter settings.