ESTRO 2021 Abstract Book

S372

ESTRO 2021

small, breath-hold techniques should be used to ascertain target coverage and minimize motion. However, the size of the target volume, specifically the depth of it, will affect the time it takes to deliver the dose to the target, in turn setting the number of breaths the patient must take for each beam. Using protons for treating Hodgkin’s lymphoma is absolutely feasible but may not be for every patient [3]. Target size, target localization as well as the ability to comply with breath-hold techniques must be considered. 1. Global, regional, and national burden of Hodgkin lymphoma from 1990 to 2017: estimates from the 2017 Global Burden of Disease study. Zhou L. et al. Journal of Hematology & Oncology (2019) 12:107 https://doi.org/10.1186/s13045-019-0799-1 2. Comparative treatment planning study for mediastinal Hodgkin’s lymphoma: impact on normal tissue dose using deep inspiration breath hold proton and photon therapy, Edvardsson A. et al Acta Oncologica, (2019) 58:1, 95-104 https://doi.org/10.1080/0284186X.2018.1512153 3. Pencil beam scanning proton therapy of Hodgkin’s lymphoma in deep inspiration breath-hold: A case series report. Andersson K, Edvardsson A, Hall A, Enmark M, Kristensen I. tipsRO (2020) 13, 6–10 https://doi.org/10.1016/j.tipsro.2019.11.006

Symposium: Mobility/Grant reports

SP-0484 Implementation of MR-guided radiotherapy to treat oligometastases H. Saxby 1 1 Royal Surrey County Hospital, Radiotherapy, Guildford, United Kingdom

Abstract Text Background

Magnetic resonance-guided radiotherapy (MRgRT) is novel to clinical practice and its use is expected to expand. It has many potential benefits in the treatment of patients with intra-abdominal soft tissue oligometastases. These include better soft tissue delineation leading to more accurate radiotherapy delivery to the target and smaller margins required from the gross tumour volume (GTV) to the planned target volume (PTV). MRgRT also enables daily adaptive replanning which can reduce radiation to the organs at risk in patients where great variation in intra-abdominal organs is observed, and further improve target delivery. Objective: To explore the implementation of MRgRT to treat intra-abdominal soft tissue oligometastases at Rigshospitalet, Copenhagen Denmark. Methods: During my visit to Rigshospitalet I will study the methodologies involved during each step of the treatment pathway required for treating patients with intra-abdominal soft tissue oligometastases with MRgRT. 1. Pre-treatment: patient selection, preparation pre- planning scan 2. Treatment planning: GTV and OAR contouring, generation of PTV margins 3. Treatment: daily adaptive radiotherapy planning 4. Quality assurance: routine QA related to MRgRT 5. Patient follow up: management and surveillance of patients post treatment I will discuss specific challenges with MRgRT and which patients to prioritise treatment using MRgRT with the multi-disciplinary team. Conclusion: MRgRT is an exciting new technology in the field of clinical oncology. It requires a lot of resources in terms of excess costs, time, staffing and expertise. Despite the many potential advantages in the accuracy of delivery of radiotherapy there is still a need for evidence based implementation which the MOMENTUM study sets out to achieve. Medical imaging data are acquired during the routine process of radiotherapy treatment. It was hypothesized that these data could play role in tumor characterization and treatment prognosis by utilizing the data for machine learning algorithms. There are two main ways to extract valuable latent components (features) from imaging data: automated (deep learning based) feature extraction and hand-crafted (radiomic) feature extraction. In this project, we focus on the latter approach as easy-to-train and more intuitive feature extraction method. Increasing sample size for machine learning models Many radiomic models are built in a centralized one-center (one-hospital) manner. The main limitation of such approach is that it rarely can fit a model, which is generalizable on a broader cancer patient population. One- center data can seldom ideally represent the general population, therefore, the sample size must be increased. There are two main approaches to address the issue of the sample size increase: the first method is to centralize a multi-center federated patient data in one center and build radiomic model locally; the second approach is to build the radiomic model in a decentralized (federated) manner without data sharing. Both approaches have their drawbacks: the centralized method can take long time due to data transfer agreements, on the other hand, the federated approach requires building a trustworthy platform and infrastructure. We propose our federated radiomics solution in this project. Federated radiomics network infrastructure SP-0485 Federated (distributed) radiomics across three European Radiation Oncology centers I. Zhovannik 1 1 Radboudumc, Radiation Oncology, Nijmegen, The Netherlands Abstract Text Imaging data and machine learning