ESTRO 2021 Abstract Book

S745

ESTRO 2021

Figure 2: PRD profiles in the film plane for the three patients.

Conclusion The presented method is a new measurement-based evaluation of dose calculation on in-room image data for adaptive proton radiotherapy. Additionally, both the measurement and the simulation approach include the possibility to evaluate the effect of daily image data on the optimization itself. The proposed method can serve as quality assurance tool for adaptive radiotherapy. PD-0904 Limitation of van Herk’s recipe in robust optimization of clinical IMPT for head and neck cancer J. Rojo Santiago 1,2 , S.J. Habraken 1,2 , D. Lathouwers 3 , Z. Perkó 3 , M.S. Hoogeman 1,2 1 Erasmus MC Cancer Institute, Radiotherapy, Rotterdam, The Netherlands; 2 HollandPTC, Medical Physics & Informatics, Delft, The Netherlands; 3 Delft University of Technology, Radiation Science & Technology, Delft, The Netherlands Purpose or Objective In proton therapy (PT) with pencil beam scanning, scenario-based robust optimization and evaluation are commonly used to ensure adequate dose to the CTV. However, the setup robustness setting (SRS) is often still based on photon margin recipes assuming dose invariance under small shifts, which does not apply to PT. In this study, robustness of CTV dose is accurately evaluated using polynomial chaos expansion (PCE) in a clinical cohort of head and neck cancer (HNC) patients, so as to investigate the adequacy of photon margin recipe- based SRS. Materials and Methods A cohort of 18 HNC patients planned in RayStation TPS according to the Dutch robustness evaluation protocol were selected. Prescribed dose (D Pres ) was 70 Gy(RBE) to the primary CTV and 54.25 Gy(RBE) to the elective CTV, both in 35 fractions using clinical setup/range robustness settings of 5mm/3% for treatment planning. Prescription was based on a voxel-wise minimum dose distribution computed from 28 scenarios. SRS of 5 mm is consistent with van Herk´s margin recipe SRS = 2.5Σ + 0.7σ and systematic and random setup errors of 1.65 mm and 1.25 mm (1 SD) respectively. In addition to patient setup errors, these included beam-alignment errors, registration errors and (some) anatomical variation. A proton relative range error of 1.5% (1 SD) was also assumed. PCE was applied to evaluate the consistency between the SRS and errors, which was used to generate a fast, patient- and plan-specific model of the dependence of voxel-doses on setup and range uncertainties. For PCE validation, dose distributions and dose-volume histograms were compared against the clinical MC dose engine. Then, PCE-based robustness evaluations were performed by simulating 100,000 fractionated treatments per patient, allowing to obtain accurate statistics on clinically relevant CTV dosimetric parameters. Population dose histograms were calculated as the average of all sampled treatments for all patients. Results PCE validation results are shown in Figure 1. For a 5 mm error, PCE model is in good agreement with the clinical TPS, with an average absolute CTV voxel-dose error of less than 0.5%. Figure 2 displays population D 98% dose histograms for both primary (blue) and elective (red) CTVs. At a 90% population coverage, the D 98% is well above of the 95% of D Pres (treatment intent of van Herk recipe) for both CTVs: D 98% is 98%/68.8 Gy(RBE) and 99%/53.8 Gy(RBE) for the primary and elective CTV respectively. Even at 98% population coverage, the average CTV dose is conservative with 98%/68.3 Gy(RBE) and 98%/53.3 Gy(RBE) for the primary and elective CTVs respectively.