ESTRO 38 Abstract book

S64 ESTRO 38

all the plans that pass these criteria is not exploited. With automated planning combined with rapid feedback to the institutions, this potential can be quantified on an individual patient level. The institutions can be presented with an alternative, improved plan which they might be able to use in their department. For example, for the RTOG 0933 whole brain with hippocampus sparing it was shown that automated planning could lead to better treatment plans than plans that were optimised just to adhere to the trial protocol (krayenbuehl et al. RO 2017). Another example, comparing the plans actually used in the EORTC-1219 Dahanca H&N study plans, showed that the machine learning model was able to generate plans which could frequently improve OAR sparing (Tol et al. RO 2018). For another trial, comparing hypofractionated prostate treatment to conventional treatment, a higher grade 2 toxicity was found in the hypofractionated arm. Looking back at the treatment plans and generating new automated planning plans, it was shown that it was possible to reduce the NTCP of hypofractionated treatment to below the NTCP values of the clinically used standard fractionation schedule. This also indicates that potentially, automated planning could have led to a much lower late toxicity in the experimental arm and thus could have had a mayor influence on the trial outcome (Sharfo et al. RO 2018). In the Trendy trial of primary liver cancer, it was shown in a benchmark exercise that automated planning led to generally better plans with a lower liver NTCP (Habraken et al RO 2017). However, all these studies have been conducted retrospectively. There are still some challenges in order to fully use automated planning in clinical trials prospectively like easy to use software for up- and download of dicom-RT and CRFs including anonymization software which incorporates patient trial numbers and the use of standardized naming conventions for targets and OARs (AAPM TG report 263). Implementation of full prospective ICRs based on automated planning could lead to improved treatment of patients within the trial, a reduction of the number of patients needed in a trial and thus also a reduction of trial costs and more reliable trial outcomes leading to better treatment of future patients. SP-0131 Using automated planning for “bias-free” plan comparison L. Rossi 1 , A.W. Sharfo 1 , S. Breedveld 1 , B.J.M. Heijmen 1 1 Erasmus Mc - Cancer Institute, Department Of Radiation Oncology, Rotterdam, The Netherlands Abstract text Plan quality in current interactive trial-and-error treatment planning (designated ‘manual’ planning) is highly dependent on the planner’s skills and on allotted time. Many treatment planning studies may then suffer from considerable bias e.g. due to differences in planning experience for the investigated techniques, use of different treatment planning systems for the techniques, or a wish that one of the techniques will be superior. Due to the large manual planning workload numbers of plans may be small. Automated planning can play a key role in substantially reducing bias in these planning studies. Erasmus-iCycle is a system for fully automated multi-criterial planning, based on a wish-list. Resulting plans are always Pareto- optimal and have clinically favourable trade-offs between objectives. Due to the involved prioritized optimization, plan generation is highly consistent among different patients. By using the same wish-list for all evaluated treatment techniques in the study, bias in the comparisons can be substantially reduced. Due to the automation excessive workload may be avoided and studies with large plan/patient numbers may be performed.

demonstrated, and specifically, the example of KBP models in the potential for spacing in moderately hypo- fractionated prostate treatments. KBP models were generated using twenty prostate patients, with and without a hydrogel spacer (SpaceOAR®, Augmenix Inc. Waltham, MA, USA). Four models were developed including plans generated for both pre and post gel implantation, and then a further two models, with these original plans re-optimised in combination with Multi- Criteria Optimisation, with focus in reducing the bladder and rectum doses, whilst ensuring adequate tumour coverage. SP-0129 Does automation jeopardise personalised treatment? Are we going back to prêt-à-porter instead of bespoke fashion? R.Moeckli UniversHospital CHUV Institut de Radiophysique(IRA) Lausanne, Switzerland Abstract text Trial radiotherapy quality assurance (RTQA) is an important part of the good conduct of a clinical trial. Failure to perform the radiotherapy according to the trial protocol might lead to either a lower tumour control than expected or a higher morbidity. This in turn might lead to false positive or false negative trial results. The most important part of RTQA is a good protocol definition of what radiation treatment should be given in terms of fractionation, dose per fraction and dose to targets and organs at risk. This is assured by several RTQA steps: a facility questionnaire, standard beam output assessments (BOA) under reference conditions, benchmark planning exercises, Individual case reviews (ICRs) and complex dosimetric checks. Recently, it has been shown that BOA results have improved steadily over the last years and approximately only 0.5% of all BOA results are found to be out of the 5% tolerance level (kerns 2018). This does however not guarantee that complex treatments can be administered with the same accuracy. Complex dosimetry checks have revealed that inaccurate beam modelling in a TPS can be an important reason why phantom dose measurements differ from the calculated dose (Kerns et al. IJROBP 2017). However, in general the current linear accelerators are very stable and individual plan quality assurance measurements show very good results. The most frequent failures to adhere to a trial protocol are found in the ICRs. Unacceptable trial protocol deviation rates between 11 and 48% have been reported and have shown to have a significant impact on the clinical outcome (Weber et al. RO 2012). Mostly, these ICRs were performed retrospectively and thus it seems clear ICRs should be conducted prospectively (Branquinho et al. RO 2018). Prospective feedback also potentially leads to a steeper learning curve on how to adhere to the trial guidelines. In the new EORTC 1735 ADHERE trial investigating the effect of an immune check-point inhibitor after chemoradiation in H&N patients, it will for the first time be prospectively investigated how much impact retrospective vs prospective ICRs will have on the proportion of acceptable patient plans. Until now, the ICR feedback given was usually limited to either per protocol, acceptable variation or unacceptable variation. Once prospective reviews are in place, unacceptable variations can be prevented. However, as protocol plan criteria are set such that most plans will be able to pass the criteria, the enormous potential further improvement in plan quality of Abstract not received SP-0130 The potential of automated treatment planning in clinical trials C. Hurkmans 1 1 Catharina Ziekenhuis, Radiation Oncology, Eindhoven, The Netherlands