ESTRO 2021 Abstract Book

S279

ESTRO 2021

Conclusion The dose to neurovascular structures varied with the tumor location, and SC dose can be used as a surrogate for the CW dose for all tumor locations. The closest distance between the target and the SC was less than 10 mm in the central and nearly all posterior fossa tumors. Overall, a minimal distance of more than 10 mm between the target and the SC indicates a potential for dose sparing. This knowledge might further guide the treatment planning process for sparing of neurovascular structures, in particular for hemispheric tumors. PH-0380 Longitudinal evaluation of Knowledge-based planning for prostate treatments A. Scaggion 1 , M. Fusella 1 , F. Dusi 1 , B. El Khouzai 2 , A. Germani 1 , N. Pivato 1 , A. Roggio 1 , M.A. Rossato 1 , M. Sepulcri 2 , R. Zandonà 1 1 Veneto Institute of Oncology IOV – IRCCS, Medical Physics Department, Padova, Italy; 2 Veneto Institute of Oncology IOV – IRCCS, Radiation Oncology Department, Padova, Italy Purpose or Objective Changes in the management and quality of prostate treatment planning have been tracked and analyzed since the clinical introduction of RapidPlan. Moreover, the periodical check, update and modification of the KBP model has been tracked and analyzed. Materials and Methods The knowledge-based planning RapidPlan (Varian Medical Systems, Palo Alto, CA, USA) was firstly introduced in our department in early 2016. Since January 2018 radical prostate VMAT treatments were planned following the prediction of a KBP model. From then on, approximately 130 new cases have been treated, and a process of continuous improvement has been implemented. During this period, we kept track of: 1) the plan quality , 2) the optimization time and 3) the plan complexity. Plan quality was assessed through a dedicated Plan Quality Metric (PQM%) score algorithm. The initial KBP library contained 80 clinical plans, and was named M0. Approximately every 9 months the KBP model was reviewed and updated adding the more recent plans. In the end we obtained 4 new KBP models with an increased library numerosity.These models were named M1-M4 and contained respectively 103, 128, 160 and 183 plans. Each model has been trained, refined and validated following literature recommendations and previuos published experience. From each of these models another one was obtained including only the 80 highest quality plans (highest PQM%) among the whole set of available treatment cases, namely B1-B4 models. All of these models were benchmarked against manual plans (MP) on approximately 30 patients and cross tested against each other on a fixed set of 30 patients left outside of every model. Results Introducing RapidPlan into the clinical activity resulted in improved plan quality (average 5% increase of PQM% with respect to manual plans) and a net reduction of optimization time (form 45’ for MP down to 27’ for KBP plans) with no significant increase of plan complexity. The cross validation showed a monotonically increasing plan quality for subsequent models with increased library population (M0-M4), see Figure. The selection made to generate B1-B4 models systematically left out most of the oldest plans. In B4 library only 10% of the original M0 library was kept. Despite the population change, the four models at fixed library numerosity (B1-B4) yielded a constant plan quality on the validation set. Conclusion The use of RapidPlan systematically increased the clinical plan quality and reduced the optimization time with no increment in plan complexity. The periodic update and revision of the RapidPlan model was beneficial, at least at the beginning of its use. Nonetheless, a balance has to be found between the size of the library and its quality in order to ensure both high quality outputs and the highest possible accuracy and versatility. PH-0381 Creating and Validating a Distributed RapidPlan Head and Neck ‘Super-Model’ M. Frizzelle 1 , N. Lalli 1 , A. Pediaditaki 2 , C. Thomas 3 , C. South 4 , S. Jagadeesan 4 , E. Adams 4 , R. van der Straeten 5 , W. Wiessler 6 , R. Lynn 7 1 University College London Hospital, Radiotherapy Physics, London, United Kingdom; 2 Northampton General Hospital NHS Trust, Radiotherapy Physics, Northampton, United Kingdom; 3 Guy's and St Thomas NHS Foundation Trust, Radiotherapy Physics, London, United Kingdom; 4 Royal Surrey NHS Foundation Trust, Radiotherapy Physics, Guildford, United Kingdom; 5 Varian Medical Systems, Sr. TPS Product Manager, Brussels, Belgium; 6 Varian Medical Systems, Sr. Technical Product Manager Data & Analytics, Frankfurt, Germany; 7 Varian Medical Systems, TPS Solutions Specialist, London, United Kingdom

Purpose or Objective RapidPlan uses a library of treatment plans to create a knowledge-based model capable of predicting