ESTRO 38 Abstract book

S53 ESTRO 38

PV-102 A prediction of intrinsic uncertainties in radiotherapy treatment planning systems K. Kiers 1 , P. Van Horssen 1 , J.T. Trinks 1 , P.R. Pronk 1 , A. Mans 1 , C.J. Schneider 1 , E.M.F. Damen 1 , T.M. Janssen 1 1 The Netherlands Cancer Institute, Department of Radiation Oncology, Amsterdam, The Netherlands Purpose or Objective In radiotherapy treatment planning, differences between the simulated and delivered dose can arise from intrinsic limitations of the treatment planning system (TPS) beam model. Among these limitations, we recognize two types: 1) approximating a continuous VMAT delivery with discrete control points and 2) uncertainties in the beam model. It is expected that increasing plan complexity leads to higher dosimetric uncertainty in both aspects. The aim of this work is to predict for Pinnacle 3 v9.10 (Philips, Fitchburg, USA) the impact of these two intrinsic TPS problems which could lead independently to dosimetric errors, using complexity metrics (CMs) derived from the plan. Material and Methods Regarding arc discretization, 80 clinical non‐SBRT VMAT plans were selected. Each was computed with a control point gantry angle spacing of 4° and 2°. Plans which differed ≥1% in PTV‐mean dose were classified as too complex. 9 CMs (Tab. 1) were calculated for all plans. A model to predict ∆D mean ≥1% (Prediction 1) was created using leave‐one‐out cross validated logistic regression and a combination of CMs, and validated on a stratified split train‐ and test set (80/20). To test intrinsic beam model uncertainty, the dose distribution was recalculated for 77 non‐SBRT VMAT plans comprising of three ‘machines’ (MLCi, 6 and 10MV; Agility, 10MV (Elekta, Stockholm, Sweden)), using two clinically commissioned beam models based on the same set of measurements. From the commissioning point of view both beam models per machine appear equivalent, only differing marginally in parameters like leaf offset correction, MLC transmission and penumbra blurring. Beam models of different machines were independent. Using CMs (Tab. 1), a linear regression model was trained to predict the ∆D mean between beam models for all MLCi 10MV plans (Prediction 2) and validated against the remaining plans. This model was then applied to 1412 independent plans and compared to their dosimetric outcome using the isocentre gamma value (3%/3mm) derived from in‐vivo EPID dosimetry. Results were binned on Prediction 2 and per bin the 90 th percentile of the isoc‐ gamma was used as a measure of dosimetric uncertainty. Results Prediction 1: The ∆D mean ≥1% could be predicted with a specificity of 0.98 and a sensitivity of 1.0 by linear combination of three CMs (Tab. 1, Fig. 1). Prediction 2: The predicting value was found using a linear regression of 7 CMs (Tab. 1, Fig. 1B1) with a fit of R 2 =0.6 (p<0.001) for MLCi‐10MV. The model was validated for Agility‐10MV and MLCi‐6MV with an R 2 of 0.64 (p<0.001) and 0.55 (p<0.001) respectively. The predicting value has a strong linear correlation with the 90% CI isoc‐gamma of the 1412 plans (R 2 =0.6, p<0.001, Fig. 1B2). The spread in the data in Figure 1B2 is due to the noisy nature of in vivo data.

Conclusion A dummy‐run accreditation as applied in EMBRACE II based on defined dose volume parameters for targets, OARs, conformity indices, V43Gy and including training and examination cases enables to test institutes’ treatment planning capabilities and improves plan quality considerably whilst reducing inter‐center variations. [1] Defined as (V43Gy)/(PTV45Gy volume) PV-101 Clinical implementation of a dedicated brain treatment planning optimizer for stereotactic treatment T. Gevaert 1 , A. Girardi 1 , B. Engels 1 , M. Boussaer 1 , C. El Aisati 1 , M. De Ridder 1 1 Universitair Ziekenhuis Brussel, Radiotherapy, Brussels, Belgium Purpose or Objective Clinical implementation of a novel dedicated and automated treatment‐planning solution for cranial indications, Elements. Single lesions can be targeted with an inversely optimized VMAT approach using automated arc trajectory optimization (Cranial SRS Element, Brainlab, München, Gernmany) while up to fifteen metastastic brain tumors can be automatically targeted with a single isocenter and multiple inversely‐optimized dynamic conformal arcs (Multiple Brain Mets SRS Element, Brainlab, München, Gernmany). Material and Methods The very first 25 treated patients were analyzed, each representing a variable number of lesions (1‐12). Depending on the number and location of the lesions a dedicated Element was selected and used in order to achieve the specific planning constraints. The plans were evaluated by means of Paddick conformity (CI) and gradient index (GI). Patient specific quality assurance (QA) was performed with gafchromic EBT3 film and portal The Elements software tools generated plans with CI of 0.71±0.09 and a gradient index of 3.9±1.4. All plans achieved the organ at risk constraints. A gamma of 3%/3mm was used for the QA. A 98 % and 98,2 % passing rate was found for the EBT3 film and portal imager, respectively. This shows also the good concordance between film and EPID, suggesting that patient specific QA can be performed with the portal imager rather than the time‐consuming films. Conclusion The automated dose planning Elements revive dynamic conformal arcs as the paradigm for linac‐based stereotactic radiosurgery of multiple brain metastases and at the same time implements an improved VMAT approach for single lesions with the use of automated arc trajectory optimization. This study shows the implementation of this technique in the routine clinical environment with an improved planning and treatment efficiency. imager. Results