RadLink: Linking Clinical Entities from Radiology Reports

Radiology reports are a critical source of information for patient diagnosis and treatment in the medical domain. However, the vast amount of data contained in these reports is often unstructured, making it challenging to extract and normalize relevant clinical entities. Named Entity Normalization (NEN) is essential for mapping these entities to a standard ontology, facilitating better data integration, retrieval, and analysis. In this paper, we introduce RadLink, a benchmark for NEN in radiology. RadLink builds upon 425 expert-annotated radiology reports from the RadGraph dataset, extending it for NEN by mapping entities to the Unified Medical Language System (UMLS) ontology. We employ a combination of morphological and semantic matching approaches to generate normalization annotations, followed by human review for validation. We aim to set a standard with our benchmark for evaluating NEN methods in the radiology domain, that facilitate interoperability across healthcare systems and accelerate medical research by providing structured, standardized data.

Citation: