ESTRO 38 Abstract book

S94 ESTRO 38

using a train-test split approach and validated on the remaining 20% . The performance of the predictions is scored using the rms difference of the actual versus the predicted rectum Dmean for given homogeneity and conformity. Results Observed rectum-Dmean values ranged 5-28 Gy over all plans and patients. Ranges for the homogeneity and conformity were 97-108%, 69-94% of the prescribed dose, indicating that we sampled a large part of the PF. The PFs could be described well using the 3 parameter parametrization (rms difference of 0.9 Gy in the rectum- Dmean direction), see Figure 1. The model led to an excellent prediction of the patient specific PF, the rms difference between the fit and the prediction was 0.5 Gy. The rms difference to the actual value was 1.1 Gy.

previously been treated in their respective hospitals. The Model C1 plans were compared to the clinical plans using the respective local plan acceptance; for C2, the clinical plans had been generated using their own validated in- house RP model; for C3 they were generated using template-based optimisation. Based on the feedback and learning experience of the model comparisons, further refinements to the Model C1 parameters were tested. Model C1 and the refined models were evaluated for ten patients from C1 using a single optimisation without user interaction. The resultant plans were compared using a consensus set of plan acceptance criteria which incorporated the different metrics used in each of the three centres. Results Initial testing of Model C1 compared to clinical plans from centres C2 & C3 showed improvements in OAR sparing at higher dose levels, but increased doses at lower dose levels, with higher mean doses for bladder and rectum in Model C1 compared to clinical plans. Two different approaches (Model UKRC1 and Model UKRC2 ) to addressing the potential areas for improvement were taken. Both gave statistically significant improvements in OAR sparing, but for Model UKRC1 , this was at the expense of PTV coverage and homogeneity (see Table 1). However, Model UKRC2 was able to achieve similar OAR dose reductions with only minimal impact on PTV doses (see Fig 1).

Conclusion We successfully demonstrated that that patient specific PF could be predicted based on patient anatomy only, for a relatively simple site as the prostate. OC-0185 A multi-centre knowledge-based treatment planning model for radiotherapy of cervical cancer E. Adams 1 , M. Hussein 2 , S. Currie 3 , C. Thomas 4 , C. South 1 , A. Greener 4 , G. Currie 3 , A. Nisbet 1 1 Royal Surrey County Hospital NHS Foundation Trust, Radiotherapy Physics, Guildford, United Kingdom; 2 National Physical Laboratory, Medical Radiation Physics, Teddington, United Kingdom ; 3 NHS Greater Glasgow and Clyde, Radiotherapy Physics, Glasgow, United Kingdom ; 4 Guy's and St Thomas' NHS Foundation Trust, Radiotherapy Physics, London, United Kingdom Purpose or Objective There can be significant resource requirements in the initial setting up of Knowledge Based Treatment Planning (KBP) models for different treatment sites, particularly as the models depend on a sufficiently sized library of clinically acceptable plans. This can be a barrier for centres with a small number of patients and in rarer clinical sites. Sharing between cancer centres may overcome these issues. The purpose of this work was to address two key questions; whether a KBP model developed in one institution could be successfully used in another; and whether combining the expertise of the different institutions could lead to an improved model. These questions were addressed for the case of cervical cancer VMAT planning in the UK RapidPlan consortium (UKRC) of centres using Varian RapidPlan KBP. Material and Methods An RP model from one of the UKRC centres, C1, was selected for multi-centre testing. This model had been fully developed and benchmarked for planning cervical cancer cases using VMAT, including both post- hysterectomy and intact uterus cases, with prescribed doses of 45-50.4Gy in 1.8Gy/#. This model (Model C1 ) was shared with two other UKRC centres, C2 and C3, who used the model to generate plans for fourteen patients who had

Conclusion Combining experience from multiple centres allowed the generation of a RapidPlan model for cervical cancer patients which gave improved results compared to single- institution models.

Proffered Papers: PH 4: Proffered paper: New technologies

OC-0186 A system of materials capable of mimicking soft tissues and bone with both CT and MR imaging.