ESTRO 2023 - Abstract Book

S1690

Digital Posters

ESTRO 2023

Figure 2. Model predictions on training (left) and holdout set (right) for rectum (top), prostate (middle) and bladder (bottom) Conclusion We propose a neural network capable of predicting DSC of deformably registered OAR contours using vCT and CBCT scans as inputs. The model gives reasonably low RMSE and can be used to evaluate accuracy of deformed contours and their eligibility for plan adaptation.

PO-1935 Clinical experience with robust VMAT optimization for breast and lymph node radiotherapy

C. Papalazarou 1 , K. Crama 1 , R. van der Slot 1 , A. van Eijk 2 , E. van Reij 1 , M. Immink 1 , E. Astreinidou 1

1 Leiden University Medical Center, Radiotherapy, Leiden, The Netherlands; 2 Reinier de Graaf Gasthuis, Radiotherapy, Delft, The Netherlands Purpose or Objective To quantify the effectiveness and evaluate the parameters of a robust planning technique using clinical IGRT data. Materials and Methods In January 2022 our department introduced a robust VMAT planning technique for breast/chest wall and lymph nodes irradiation, to a total dose of 40.05 Gy/53.4 Gy (SIB). The optimization includes a set of simulated CT’s that model breast deformation, as in [1]. We analyzed all cases in which a breast contour deviation was observed following local IGRT protocol. For consistency, we analyzed the original robust plan for all fractions, even in the cases when the plan was adapted during the series. CTV and PTV coverage of 95% of the elective dose was evaluated on the planning CT and on one of the simulated deformed CT’s (defCT’s) with a deviation of 8 mm in the lateral, inferior and anterior directions, corresponding to the action limit of the IGRT protocol. For each plan, 3 CBCT’s sampled from the treatment series (including those with maximum deviation) were imported in the TPS to evaluate CTV and PTV coverage during treatment, by deforming all relevant structures and recomputing the dose on an extended field-of-view CBCT. The dose of the missing fractions was estimated by applying random shifts to the closest available fraction image, based on locally collected intrafraction variation data ( σ =2/3/4 mm for LR/IS/AP direction respectively). In contrast to Dunlop et al [2], our goal is not to establish the superiority of robust planning (here considered a given), but to quantify the extent to which the chosen robustness parameters are