ESTRO 38 Abstract book

S510 ESTRO 38

planned DVH. Thus, the majority of the predicted plans are close to the planned ones or even suggests that further dose reduction is possible for those plans. 68% (27 out of 40) of the organs-at-risk are within the ± 3 % dose difference with the planned values. Figure 1 presents the comparison between the planned dose values and the one predicted by our SFA model. The model is well adapted to predict different dosimetric sparing for the OARs and for different dose prescriptions. Considering there is no mathematical relationship between the predicted dosimetric indices the obtained dose-volume curves are quite smooth.

Conclusion Accounting for the variable RBE effect in PT planning with only posterior fields for esophageal patients increases the dose to the heart. The heart exposure is, however, still considerably decreased compared to IMRT. Heart V25Gy and D2ccm were significantly increased when using the two-field configuration compared to the one-field configuration. PO-0945 Stochastic Frontier Analysis to predict sparing of organs-at-risk for VMAT-treated prostate cancer A. Kroshko 1 , O. Morin 2 , L. Archambault 1 1 Université Laval, Physique- génie physique et optique, Quebec, Canada ; 2 University of California San Francisco, Department of Radiation Oncology, San Francisco, USA Purpose or Objective Treatment planning of advanced radiation therapy techniques often require compromises between the delivery of the prescribed dose to the volume of interest and the sparing of the neighbouring organs-at-risk (OAR). A novel knowledge-based model was developed based on Stochastic Frontier Analysis (SFA) to predict achievable dosimetric indices for the bladder and the rectum for prostate cancer treated with VMAT. Material and Methods SFA is modelized as an optimum frontier with dosimetric parameters degradation following a normal-half normal distribution. Model can converge into two extremes, a regression or an absolute frontier, otherwise a stochastic frontier is obtained. Seven parameters were used to describe the geometry between each OARs and PTV (e.g. the overlap volume and the Hausdorff distance). They are extracted with an automatized python routine within the 3D Slicer platform. Estimators of the model are determined using a maximum likelihood technique to obtain stochastic frontiers for 9 dosimetric indices (ex. V60Gy) only in terms of the geometric parameters. The model was developed with a database of 152 patients treated between October 2014 and May 2017 with prescribed dose to the PTV between 60 and 78 Gy with the treated volume being prostatic bed or prostate with seminal vesicles. A validation cohort of 20 random patients treated between December 2017 and April 2018 was then used to compare the planned dose to the predicted dose to the OAR. A metric quantifying the similarity between the predicted and the clinical plan is then calculated. It is defined as the sum of the dose difference between predicted and planned divided by the number of dosimetric indices predicted for an OAR. A negative value means that the predicted dose is lower than the one planned. The overall score is the sum of the score for the bladder and the rectum. Results 75% of the validation plans have an overall score lower than 5% dose difference between the predicted and the

Conclusion SFA is an adequate model to built a predictive model based on the morphological parameters of a retrospective database. The model is validated with a 20 patients cohort with different dose distribution. We wish to implement this technique into the clinical process of radiation therapy planning in order to maximise the sparing of the OARs. PO-0946 Inter-fraction robustness of DECT-based head and neck proton therapy with reduced range uncertainty margins S. O'Reilly 1 , C. Cheng 1 , A. Lalonde 2 , B. Burgdorf 1 , W.J. Zou 1 , L. Yin 1 , S. Swisher-McClure 1 , A. Fotouhi Ghiam 1 , J.N. Lukens 1 , A. Lin 1 , L. Dong 1 , B.K.K. Teo 1 1 University of Pennsylvania, Radiation Oncology, Philadelphia, USA ; 2 Centre hospitalier de l'Université de Montréal, Radiation Oncology, Montreal, Canada Purpose or Objective The Hounsfield unit (HU) to proton stopping power ratio (SPR) relative to water conversion is a source of uncertainty in proton therapy, with the generally accepted clinically applied range uncertainty for single energy computed tomography (SECT) being 3.5%. However, studies have shown that a range uncertainty of 2% is feasible when using SPR calculated with dual energy computed tomography (DECT). This decrease in range uncertainty leads to a reduction in margins, leading to decreased doses to organs at risk (OAR). This dose reduction can be substantial for head and neck cancer patients, who experience significant acute and late