Integrating Heterogeneous Real-World Cancer Data for Semantic Interoperability in Oncology and Medical Imaging: Development and Validation of the Cancer Image Europe Hyperontology
TL;DR
A FAIR hyperontology is developed that semantically integrates and harmonizes clinical, biological, and imaging metadata and data spread across disparate sources, and captures and accurately represents oncology and medical imaging.
Abstract
Abstract Background Semantic interoperability in health care, essential for seamless integration of information systems, is partially achieved through the use of terminologies and common data standards that define the semantic structure of data. Various complexities arise when using real-world health care data, including different interpretations of terms and concepts and gaps in domain coverage in standard terminologies. However, ensuring compatibility becomes increasingly challenging when big data are distributed across diverse repositories that use heterogeneous health care standards and overlapping terminologies. Ontologies are key solutions to bridge these gaps, enabling consistent semantic interoperability and data harmonization. Objective We aim to develop and validate a hyperontology within the EUCAIM (Cancer Image Europe) project to semantically integrate and harmonize clinical, biological, and imaging metadata, along with associated data from heterogeneous, disparate cancer image data models, to achieve semantic interoperability in oncology and medical imaging. The hyperontology will be used to support several EUCAIM components, including the extract, transform, and load process; federated query; image annotation and segmentation; and ultimately, AI-federated processing. Methods The ontology development process combines real-world data from a network of European projects on cancer imaging (AI for Health Imaging) with their semantic mappings, as well as conceptual unpacking and modeling of the Minimal Common Oncology Data Elements (mCODE) specifications. The mCODE is a core set of structured data elements for oncology electronic health records. The building process is supported by ontology grounding, layering, and modularization. We adopted this hybrid approach to simplify ontology design, semantically reflect oncology’s essential entities and their interactions, and enhance the extensibility and reusability of the hyperontology. We initiated ontology development with a set of competency questions derived from the provided knowledge, which helped clarify the ontology’s scope and requirements and identify inconsistencies or incomplete information. We also assessed whether the requirements were fulfilled by formalizing the competency questions using SPARQL. Results We developed a FAIR hyperontology that semantically integrates and harmonizes clinical, biological, and imaging metadata and data spread across disparate sources. The ontology also captures and accurately represents oncology and medical imaging. The hyperontology, which covers various cancer types, is rich in axiomatizations and patterns, supporting the semantic understanding and harmonization of heterogeneous data. Additionally, semantic mappings are established across data models and standards, ensuring the efficient and meaningful sharing and integration of health care data. Finally, we evaluated the ontology model and demonstrated its applicability using real-world prostate and breast cancer use cases. Conclusions EUCAIM’s hyperontology is a valuable effort that provides a unifying framework for the essentials of oncology and medical imaging, facilitating communication among disparate and heterogeneous cancer image data models. The ontology model is evaluated and validated using multiple methods, demonstrating compliance with the specified ontological requirements. Challenges include ensuring that the ontology is scalable, extensible, and applicable, given the complexity and dynamic nature of the application domain.