ESTRO 2025 - Abstract Book

S2393

Interdisciplinary – Other

ESTRO 2025

3356

Digital Poster a personal health knowledge graph for clinical data harmonization in breast cancer radiotherapy patients Wenjie Liang 1 , Rutger van Mierlo 1 , Remzi Celebi 2 , Ensar E. Erol 2 , Katerina Serafimova 3 , Todor Primov 3 , Svetla Boytcheva 3 , Andre Dekker 1 , Aiara L. Gomes 1 , Petros Kalendralis 1 1 Department of Radiation Oncology (Maastro), GROW School for Oncology and Reproduction, Maastricht University Medical Centre, Maastricht, Netherlands. 2 Department of Advanced Computing Sciences, Institute of Data Science, Maastricht University, Maastricht, Netherlands. 3 Ontotext, Ontotext AD, Sofia, Bulgaria Purpose/Objective Personal health data contain valuable information, but often remain underutilized. New approaches, including Artificial Intelligence (AI) techniques and Knowledge Graphs, are being implemented to facilitate curation of heterogenous data and improve patient autonomy about their health information. In this study, we aim to harmonize identifiable breast cancer patient data by means of an innovative workflow powered by an AI toolkit and represent the data in the form of personal health knowledge graph, as a use case of the AI-powered Data Curation & Publishing Virtual Assistant (AIDAVA) project. Material/Methods A workflow was proposed to effectively curate breast cancer-related data from the hospital Electronic Health Records (EHR) systems and construct a knowledge graph for data representation. First, a virtual assistant embedding multiple AI techniques (AIDAVA system), including Natural Language Processing and Deep Learning, was developed to facilitate automated data curation. Next, the breast cancer-related data were extracted from the local EHR systems and ingested into the AIDAVA system. Data transformation was conducted to unify formats and make the data mappable (see Figure 1), while entity linking was done with the data items that could be represented by SNOMED CT codes. Data mapping was performed using an internally developed mapping tool (see Figure 2), and driven by applying the Swiss Personalized Health Network (SPHN) Resource Description Framework (RDF) schema based ontology, which was developed based on existing medical terminologies, including SNOMED CT and ICD-10. The data items were annotated into classes within the ontology while relevant predicates were created to represent their relationships, forming a graph-like semantic network. Finally, the graph was assessed by using a validator based on SHACL rules to evaluate its quality, then standardize the mapping process. Figure 1 Example of data transformation