ESTRO 37 Abstract book

S340

ESTRO 37

Abstract text Knowledge-based automated planning (KBP) is one of the approaches for automated treatment planning and has the potential to improve plan efficiency, quality and consistency. In KBP, a large number of clinically acceptable treatment plans are used to characterise the relationships between anatomical, geometric and dosimetric features for a given technique and treatment site to build a predictive model. For any prospective patient to be treated using the same technique, this model can be used to predict the likely DVH that could be achieved, accounting for the size and relative positions of targets and organs at risk (OAR) in the new patient. From this predicted DVH, dose objectives can be automatically generated as inputs for the plan optimisation. KBP is implemented in the Varian RapidPlan software which was installed at the Royal Surrey County Hospital NHS Foundation Trust, UK, in June 2014 and is currently in clinical use for prostate cancer and cervical cancer patients. There are various challenges to implementing automated planning including benchmarking and fine- tuning of the models generated, barriers to uptake and training issues particularly to ensure experience is both developed and maintained. For example, a suitable RapidPlan model cannot just be created ‘out-of-the-box’. Putting it together and validating it is a resource intensive process and we have found that the default settings in the model need to be fine-tuned to get an acceptable automated plan (Hussein et al, Radiother Oncol, 2016 vol 120 pp473-479). This lecture will discuss the experience of our institute in implementing automated planning. It will briefly describe our experience of clinical implementation, summarising the successes that we have had so far. The lecture will then focus more specifically on the challenges that we have faced, both those that have been overcome and those that are still ongoing. SP-0647 Challenges for clinical automated treatment planning encountered at Erasmus MC M. Dirkx 1 , J. Penninkhof 1 , A. Sharfo 1 , E. Venema 1 , C. Timmermans 1 , S. Petit 1 , B. Heijmen 1 1 Erasmus Medical Center Rotterdam Daniel den Hoed Cancer Center, Radiation Oncology, Rotterdam, The Netherlands Abstract text In 2012 Erasmus-iCycle/Monaco was clinically introduced at Erasmus MC for automated multi-criterial treatment planning. IMRT and VMAT plans are generated fully automatically, (i.e., without any manual interference) in a two-step process. First, a plan is created using our in- house developed multi-criterial optimizer Erasmus-iCycle [1]. Based on this plan a patient-specific template for our clinical TPS (Monaco, Elekta AB, Stockholm, Sweden) is generated. Next, Monaco automatically generates a clinically deliverable plan that mimickes the Erasmus- iCycle plan. This system is in routine clinical use for IMRT and VMAT plan generation for prostate cancer, head-and- neck cancer, cervical cancer (plan-of-the-day adaptive therapy), and advanced lung cancer patients. In several studies we have demonstrated that the quality of Erasmus-iCycle/Monaco plans is equivalent, and often superior to the quality of manually generated plans [2-5]. Automated plan generation with Erasmus-iCycle is based on a tumour site-specific ‘wishlist’ containing hard constraints to be strictly obeyed, and plan objectives with ascribed priorities. The objectives are used in a multi-criterial optimization to ensure clinically favourable trade-offs between treatment goals. All plans generated with Erasmus-iCycle are Pareto optimal. In case of IMRT, the system can also be used for integrated beam profile optimization and (non-coplanar) beam angle selection. At the moment, generation of tumour site-specific

wishlists for Erasmus-iCylce may be time consuming. It involves an iterative procedure with updates of the wishlist in every iteration step, based on physicians’ feedback on the quality of plans generated with the current wishlist version. Moreover, the design of tumour site-specific approaches for conversion of Erasmus-iCycle plans into patient-specific Monaco templates often requires substantial tuning. To solve these issues, we are currently working on a simplification/automation of the wishlist generation. In addition, Elekta is working on the integration of iCycle in Monaco, making the conversion step superfluous. During time, clinical requirements on treatment plans for specific tumour sites may change. Therefore, it is very important that the tumour site-specific wishlists for Erasmus-iCycle are regularly evaluated to verify whether they are still up to date. Otherwise, there is a growing risk that Erasmus-iCylce/Monaco generates sub-optimal treatment plans and dosimetrists may spend quite some time attempting to manually improve the automatically generated plan. This may seriously undermine the confidence in the automated treatment planning process. For regular evaluation and updating of wishlists sufficient time of well-trained physicists and dosimetrists has to be allocated. Erasmus-iCycle generates one Pareto optimal plan with clinically favourable trade-offs between all treatment objectives as specified for that tumour site. For specific patients, a physician may be interested to evaluate slightly different trade-offs, like the balance between target coverage and organ at risk sparing. To allow this, we intend to implement a Pareto navigation tool in Erasmus-iCycle. 1. Breedveld S, Storchi PR, Voet PW, Heijmen BJM. iCycle: Integrated, multicriterial beam angle, and profile optimization for generation of coplanar and noncoplanar IMRT plans. Med Phys. 2012; 39(2): 951-963. 2. Voet PW, Dirkx ML, Breedveld S, Fransen D, Levendag PC, Heijmen BJM. Toward Fully Automated Multicriterial Plan Generation: A Prospective Clinical Study. Int J Radiat Oncol Biol Phys. 2013; 85(3): 866-72. 3. Voet PW, Dirkx ML, Breedveld S, Al-Mamgani A, Incrocci L, Heijmen BJM Fully Automated Volumetric Modulated Arc Therapy Plan Generation for Prostate Cancer Patients. Int J Radiat Oncol Biol Phys. 2014; 88(5):1175-9 4. Sharfo AW, Breedveld S, Voet PW, Heijkoop ST, Mens JM, Hoogeman MS, Heijmen BJ. Validation of Fully Automated VMAT Plan Generation for Library-Based Plan- of-the-Day Cervical Cancer Radiotherapy. PLoS One. 2016; 11(12): e0169202. 5. Della Gala G, Dirkx ML, Hoekstra N, Fransen D, Lanconelli N, van de Pol M, Heijmen BJM, Petit SF. Fully automated VMAT treatment planning for advanced-stage NSCLC patients. Strahlenther Onkol. 2017;193(5):402-409. SP-0648 Challenges for clinical automated planning encountered at Odense University Hospital C. Hansen 1 1 Odense University Hospital, Laboratory of Radiation Physics, Odense, Denmark Abstract text Treatment planning is often highly complex due to multiple dose prescription levels and numerous organs at risk. Automatic planning can help in the process of creating a deliverable plan, which handles the many issues that constitutes a plan of high quality. Automatic plan generation has been performed at Odense University Hospital (OUH) since 2015 by use of Autoplan from Pinnacle for more than 1000 patients. Autoplan has been used within almost all treatment sites; H&N, prostate, brain, bladder, esophagus, cervix, lymphomas – in descending order of number of patients treated. For most of the treatment sites, there was a clear gain in