ESTRO 2023 - Abstract Book

S278

Sunday 14 May 2023

ESTRO 2023

Introduction: The incorporation of publicly available data is not only useful for providing additional data for model training/evaluation, but also the fundamental base of comparison with any past or future work. Despite this, finding data, or converting the data into a usable form for the research being performed, can be particularly time consuming. Furthermore, the evaluation and curation of the data is often difficult, leaving many disheartened or disinterested. Methods: In this talk, we will provide a brief background on the importance of incorporating publicly available data, a list of resources available (TCIA, WHO, etc.) for data acquisition, and examples of best practices/pit falls. To illustrate the necessary steps in this process, we will include simple, real-world examples of best practices that can be implemented by the viewer: from data searching, to final curation. Hopeful takeaways: The viewer should leave with an understanding of the importance of using publicly available data, an ability to confidently find data matching their own desired use cases and have the tools to quickly and efficiently curate the acquired data. Abstract Text Radiotherapy (RT) planning generates a lot of data. DICOM data, in particular, offers a wealth of information that can be used to advance our understanding of the effectiveness of our treatment approaches. To take advantage of this potential, it is essential that we do large-scale RT studies and that we move beyond highly condensed registration of RT data such as yes/no or the prescribed dose and fractionation only, and instead aim to include the full exposure data. This is important when we design new prospective studies, but also if we want to learn from the data that we generate on a daily basis when we treat patients who are not part of a clinical protocol. Working with large datasets can be challenging in a single institution and even more so when moving to a national setup. In this talk, we will discuss the main challenges of dealing with big national DICOM datasets and how to overcome them. These include: - Automated collection of large DICOM data sets from multiple centres using different treatment planning systems. - Data standardisation for non-protocol treatments in a setup where data is not formatted consistently across different institutions. - The need for tools for storage, organisation, and analysis of large national dataset. - Quality assurance and data curation. We will present results and lessons learned from the national Danish Breast Cancer Group (DBCG) RT Nation study, where we collected and analysed DICOM data for 8000 consecutively treated high-risk breast cancer patients (2009-2016). Furthermore, we will highlight some of the possibilities and spin-off applications that came from collecting data for the study. Abstract Text The use of international datasets in radiation oncology is becoming increasingly important in improving treatment outcome and advancing the field. By pooling data from multiple centers and countries, researchers can gain a more comprehensive understanding of the factors that affect treatment outcomes and develop more effective treatment strategies. However, working with international datasets presents unique challenges, such as ensuring data quality, consistency, and privacy, as well as addressing institutional differences. This talk briefly illustrates these concepts based on the example of an international study for radiotherapy treatment of cervical cancer (EMBRACE). An initial barrier when collecting international data is the need to establish common data definitions and protocols. This involves identifying and resolving discrepancies in the way data is collected, recorded, and reported, as well as developing standardized forms and procedures for data entry and analysis. In radiation oncology, the epitome of this challenge are naming conventions for targets and organs at risk, which can be hard to control even within individual centers. While accruing data, centralized data collection can help to overcome this issue. Another important challenge is dealing with data privacy and security, as well as ethical considerations, such as obtaining informed consent from patients. When working with international datasets, it's important to consider the potential impact of unobserved bias on the results of the analysis, i.e., systematic differences between centers. One way to assess and account for unobserved bias in these datasets is by using random effect models. Random effects models allow taking into account the clustered structure of the data when estimating the effects of different variables on an outcome of interest. This type of model allows to account for both observed and unobserved sources of variation between centers by including a random effect at the center level in the model, leading to more robust estimation of center-specific effects. Finally, international research does not happen in isolation. Because an international dataset is likely used by multiple working groups, it should also be noted that effective communication and organization are key to successful collaboration in this area. Regular meetings or communications between collaborators should be implemented to discuss progress, address any issues, and plan future direction for the research. In addition, there are several tools that can be used to collaborate on a dataset in an international setting. Cloud storage platforms allow researchers to securely store and share data files with others. Furthermore, version control systems such as Git can be used to keep track of changes and updates to the dataset. SP-0369 Dealing with national datasets L. Refsgaard 1 1 Aarhus University Hospital, Department of Experimental Clinical Oncology , Aarhus, Denmark SP-0370 Dealing with reconstructed dose data C. Owens USA Abstract not available SP-0371 Dealing with international datasets S. Ecker 1 1 Medical University of Vienna, Department of Radiation Oncology, Vienna, Austria