ESTRO 2020 Abstract Book

S135 ESTRO 2020

Abstract text Introduction

Concluding remarks The actual impact of the applied TPS depends on the purpose of the treatment planning study. The results in a study designed to investigate effects that are in the same order of magnitude as effects caused by differences between different TPSs will not be generalizable to be valid for another TPS than the one used in the study. Although, it might still be possible to draw a conclusion that is specific to the TPS used, provided that there is no other systematic bias that can affect the conclusion, for example differences between different treatment- planners. Smaller studies including case studies based on comparisons of treatment plans can be of great value at the local radiotherapy departments without necessarily being a scientifically designed treatment planning study. To make it valid as a treatment planning study with some general conclusion places higher demands on the design of the study in relation to its aim and also on the reporting of the study, taking into account the uncertainties in the TPS- estimations of the dose distributions to be delivered.

The calculated dose distribution in the treatment planning system (TPS) is an estimation of the actual dose distribution that will be delivered to the patient for a specific treatment plan. The underlying goal of a treatment planning study is to provide information on probable differences in treatment outcome by providing information on the estimated dose distributions to be delivered. Treatment planning studies are often designed as inductive studies where data, i.e. dose distributions of treatment plans, are collected for a finite number of cases, i.e. patients, to build observations that are generalized to some larger population. If the results are to be valid in other cases than those used in the study and possible to reproduce, the data collection needs to be done in a structured and well defined way, and should be independent of for example who is performing the treatment planning, the TPS used, or any other concern that can cause a systematic bias. Systematic differences between treatment plan dose distributions related to the use of different TPSs can be due to for example differences in the beam models or the dose calculation algorithms or differences in the optimization procedures. Examples of such differences will be presented in this talk. Differences from using different treatment planning systems It is well known that there are uncertainties in dose calculations that originates from limitations in the dose calculation algorithm or in the beam model. Those uncertainties can be TPS specific and generate systematic differences between dose distributions calculated in different TPSs. Systematic differences between different TPSs can also be due to differences in the optimization procedures. The delivery design of a treatment plan, i.e. the treatment machine settings used for treatment delivery to create a desired dose distribution, is determined during the optimization procedure in the TPS. Because of physical limitations of the treatment machine it is not possible to deliver all kinds of conceivable dose distributions in exact form. There are many different solutions for machine settings that will deliver similar dose distributions. Depending on the optimization algorithm and the available tools that can be used for the optimization, different TPSs can create different treatment machine settings for one and the same case that will result in different calculated dose distributions that are more or less optimal, i.e. more or less close to the desired dose distribution (the Pareto-optimal treatment plan). The treatment plan delivery design will furthermore have impact on how robust the treatment delivery will be to different parameters, for example variations in patient geometry, patient set-up variations or variations in machine parameters within their tolerances, e.g. multi leaf collimator-, jaw- or gantry-position. In fact, two treatment plans with calculated dose distributions that are evaluated as similar might have a difference in robustness that makes the actual dose distributions to be delivered to the patient considerably different from each other. The calculated dose distribution of a less robust treatment plan, i.e. a treatment plan with larger uncertainty in the TPS-estimation of the dose distribution to be delivered, might have a weaker association to the treatment outcome compared to a more robust treatment plan.

Symposium: Application of machine learning to CTV definition

SP-0270 Lessons learned about tumor invasion from microscopic imaging P. Friedl UMC St Radboud Nijmegen, The Netherlands

Abstract not received

SP-0271 Deep learning for target volume segmentation in the head and neck region M. Aristophanous 1 1 Memorial Sloan Kettering Cancer Center, Medical Physics, New York, USA Abstract text Target volume delineation in head and neck (HN) is one of the largest sources of uncertainty in radiotherapy. Large inter-observer and inter-institutional variability is present even when delineation guidelines are provided. Automatic segmentation methods have been shown to reduce that variability, however, determining ground truth remains a difficult task and has hindered adoption of more clinical auto-segmentation workflows that involve the target. During the talk I will discuss research efforts that have been focused on obtaining ground truth, as well as describing some of the issues with determining the clinical target volume (CTV) and how auto-segmentation can provide guidance. The focus of the talk will be auto- segmentation algorithms aimed at defining the target. I will first touch on previous efforts for auto-delineation of the clinical and gross disease in HN before getting into more recent approaches that involve deep learning algorithms. A significant amount of work has also been performed in auto-delineation of organs at risk in the HN region and we will be briefly discussing some recent methods involving deep learning algorithms as well as workflows that aim at introducing them in clinical practice. SP-0272 The role of tumor growth models for CTV definition J. Unkelbach 1 , B. Pouymayou 1 , R. Ludwig 1 , M. Guckenberger 1 , P. Balermpas 1