ESTRO 2020 Abstract book

S864 ESTRO 2020

Figures: DSC scores and Median surface distance (SD) for both methods.

Purpose or Objective The purpose of the study is to present the result of comparison of the automatic optimization using the RapidPlan with manual optimization, and to present the configuration and training process of the estimation model, its evaluation of geometric and dose components, and the algorithm evaluation. Material and Methods The calculations were made in the Ecpice planning system ver. 15.6. Dose constraints from the RapidPlan estimation algorithm were used for optimization. The plans were made for the True Beam HD accelerator, for FF 6MV beams. The model consisted of 258 plans covering all types of prostate radiation therapy, i.e. 1 / prostate alone, 2 / with vesicles, 3 / with lymph nodes. The model includes plans with the conventional dosing of 2Gy / 1.8Gy and the hypofractionated 2.5Gy for the prostate area alone. The three targets have been defined: PTV1 – the prostate, PTV2 – the vesicles, and PTV3 – the nodes. 10 OARs have been defined: the Rectum, the Bladder, the Femoral Heads, the Intestines, the Esica, the Rectus, the Penile Bulb, the Marrow, the Large Blood Vessels, and the Bones. The model was trained on the plans after an iterative tuning, i.e. the re-optimization of base plans with a model based on these plans, each time creating a new model, etc. At the training stage, approx. 35% of Outliers were obtained for each of the OARs. Most were the geometric deviations due to an inconsistent contouring and a diverse population. Results The 258 plans were evaluated. For each of them a model plan was created and compared with the corresponding plan created by the physicist. For PTV1 , the average difference between the automatic model and the clinical plan (AM) was V98 -0.20% at STD 0.27, for ΔV2 0.76% at STD -0.17, for ΔSTD 0.26 at STD -0.05. PTV2 ΔV98 -0.22% at STD -1.11, for ΔV2 0.83% at STD 0.08, for ΔSTD 0.18 at STD -0.43. PTV3 -1.24% at STD -0.01, for ΔV2 0.21% at STD 0.25, for ΔSTD 0.37 at STD -0.28. Rectum ΔDmean -2.85%, at STD 0.44, Bladder ΔDmean -5.33%, STD -3.78, Femoral heads ΔDmean = -6.33%, STD -3.14, Intestines ΔDmean (%) = -5.42%, STD -1, 37, Anus ΔDmean = -5.49%, STD -2.96. The number of plans with lower average doses in the population of the analyzed plans (Manual / Automatic) was 71/197 for rectum, 36/232 for bladder, 11/83 for intestines, and 20/81 for anus. Conclusion For all the critical organs, the average population dose is lower in the automatic planning compared to the manual planning. The model generates similar quality (V98 rating) dose distributions for PTV1 and PTV2. For PTV3 a slightly understated dose was observed for the automatic planning but within tolerance ranges. More consistent plans were achieved - population less different, as indicated by a smaller standard deviation. Shorter treatment planning times have been observed. The use of the automatic planning also suggests a reduction in the likelihood of human random errors. The RapidPlan guarantees versatility for all prostates regardless of the contouring range. PO-1510 Cardiopulmonary Sparing During Breast RT with Personalized Breast Holder and Proton Beam Therapy R. Wang 1 , T.H. Chen 1 , M.Y. Chung 2 , D.C. Tien 2 , K.H. Tseng 2 , J.F. Chiou 1,3,4 , L.S. Lu 1,5 1 Taipei Medical University Hospital, Radiation Oncology, Taipei, Taiwan ; 2 National Taipei University of Technology, Electrical Engineering, Taipei, Taiwan ; 3 Taipei Medical University, Radiology- School of Medicine- College of Medicine, Taipei, Taiwan ; 4 Taipei Medical University, Taipei Cancer Center, Taipei, Taiwan ; 5 Taipei Medical University, Graduate Institute of Biomedical Materials and Tissue Engineering, Taipei, Taiwan

Conclusion The preliminary study shows that deformable image registration for data augmentation is comparable or slightly better than affine jittering. Due to the limited test set available, further investigation is needed with a larger test set to further explore the generalisabilty and robustness of the approach and to assess the extent of optimisation needed in training. Such an approach may allow training of high quality DLC models using fewer curated datasets. References [1] Lin, A data augmentation approach to train fully convolutional networks for left ventricle segmentation. Magnetic Resonance Imaging 2019 [2] Simonyan, Very Deep Convolutional Networks for Large-Scale Image Recognition, Computer Science, 2014 PO-1509 Application of RapidPlan in the optimization of planning for patients with prostate cancer M. Raczkowski 1 , T. Siudziński 1 , M. Janiszewska 1 , A. Maciejczyk 2 1 Lower Silesian Oncology Center, Medical Physics Departmen, Wroclaw, Poland ; 2 Lower Silesian Oncology Center, Department of Radiotherapy, Wroclaw, Poland