ESTRO 2020 Abstract Book

S998 ESTRO 2020

Conclusion We proved that CNN is an excellent tool for labeling data with high accuracy and a low misclassification rate. Even for tiny training sizes (2 patients), we achieved a 100% accuracy and a misclassification rate of about 10%. All misclassification structures can be labeled manually. From 20 patients on, the accuracy was 100% while dropping the misclassification percentage to 0%. Thus, by just manually labeling 20 patients, we could automatically label a bigger dataset, decreasing the amount of time and errors dramatically. The small number of training data makes this CNN suitable for further investigation at any other tumor site. PO-1789 Commissioning experience with the world’s first CT-integrated linear accelerator uRT-linac 506c L. YU 1 , J. Wang 1 , W. Hu 1 1 Fudan University Shanghai Cancer Center, Radiation Oncology, Shanghai, China Purpose or Objective A newly-designed linear accelerator which integrated with helical CT (uRT-linac 506c) was introduced into clinical application by United Imaging Healthcare Co., Ltd (UIH, Shanghai, China). It combines a C-arm medical linear accelerator with a diagnostic-quality 16-slice CT imager, providing seamless workflow from simulation to treatment, and enabling personalized adaptive radiotherapy. Recently, this linac was installed and prepared for clinical operation in our department. In this study, the characteristics of uRT-linac 506c and beam commissioning data were to be reported. Material and Methods The processes of mechanical and imaging system tests, commissioning data collection and Treatment Planning System (TPS) modelling were summarized. The accuracy of Image-Guided Radiotherapy (IGRT) was investigated through CBCT/FBCT image registration and couch correction. Finally, a series of end-to-end 2D/3D plans, including static/dynamic Intensity Modulated Radiation Therapy (s/dIMRT) and UIH ARC therapy (uARC) plans on different treatment sites, were tested to estimate the overall accuracy for a treatment scheme before clinical usage. Results The mechanical RT isocenter diameter was measured as less than 0.6 mm. TPS model parameters were adjusted to reproduce the beam commissioning data within 1% deviation. The overall accuracy of CT-based IGRT was less than 1 mm considering the accuracy of couch motion from CT to RT position, as shown in Figure 1. Absolute point dose agreements for commissioning plans between ion chamber measurements and TPS calculations were within 2% for all cases, as listed in Table 1. The average gamma passing rates of 2D dose distributions measured by EBT3 film were 95.4±3.0% for Collapse Cone (CC) convolution dose calculation algorithm and 95.6±2.9% for Monte Carlo (MC) algorithm (3% DD, 3 mm DTA and 10% threshold), respectively.

Conclusion The commission results show that the uRT-linac 506c platform has a comparable beam performance as the linacs from other vendors. As the first clinical model type, its long-term reproducibility and stability are still under inspection. The integrated CT system, as a highlight, allows a diagnostic-quality visualization of internal patient anatomical structures for more accurate image guidance, and paves the way towards advanced and adaptive radiotherapy. PO-1790 A 3D model of proton treatment rooms for collision prevention T. Henry 1 1 SkandionKliniken, SkandionKliniken, Uppsala, Sweden Purpose or Objective The current proton therapy workflow in Sweden is based on the collaboration between the seven University Hospitals, where patients are diagnosed and treatment plans are made, and the Skandion Clinic in Uppsala, where patients receive proton treatments. The workflow can cause issues in the dose-planning process due to the dose- planners not being able to easily check for mechanical limitations of a plan ( e.g. collisions between parts of the nozzle/snout/couch or with the patient) during this process. In some cases, it can lead to a need of a replan after the first fraction could not be delivered because of such collision. The aim of this project was to provide a tool to give the dose-planners at the University Hospitals the possibility to check any plan for collision risks during the dose-planning process and therefore considerably improve the latter. Material and Methods The treatment rooms at the Skandion Clinic were modelled in a CAD software (Fusion 360, Figure 1). Each relevant mechanical part of the room was recreated as accurately as possible, including movement freedom and constraints. The following parts were included: gantry, nozzle, snout, range shifter, couch, couch robot arms, mask plates for brain irradiations & imaging panels. Geometry inputs such as gantry angle, couch rotation or range shifter extension can be entered easily through a script to quickly recreate a specific geometry from e.g. a patient treatment plan/field. Validation was performed by comparing distances measured in the rooms and in the model for multiple scenarios that resulted in two or more mechanical