Javier Castell-Díaz

Javier Castell-Díaz

Auto-generated profile, based on publicly available information

Javier Castell-Díaz

Dept. Informatica y Sistemas, Universidad de Murcia, IMIB-Arrixaca, Murcia, Spain

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
31 January 2023 in Journal of Biomedical Informatics
Verified
Journal Article
Open access
JCJavier Castell-Díaz
Javier Castell-Díaz
JGJose Antonio Miñarro Giménez
Jose Antonio Miñarro Giménez
Catalina Martinez-CostaCatalina Martinez-Costa
Catalina Martinez-Costa
SNOMED CT postcoordination is an underused mechanism that can help to implement advanced systems for the automatic extraction and encoding of clinical information from text. It allows defining non-existing SNOMED CT concepts by their relationships with existing ones. Manually building postcoordinated expressions is a difficult task. It requires a deep knowledge of the terminology and the support of specialized tools that barely exist. In order to support the building of postcoordinated expressions, we have implemented KGE4SCT: a method that suggests the corresponding SNOMED CT postcoordinated expression for a given clinical term. We leverage on the SNOMED CT ontology and its graph-like structure and use knowledge graph embeddings (KGEs). The objective of such embeddings is to represent in a vector space knowledge graph components (e.g. entities and relations) in a way that captures the structure of the graph. Then, we use vector similarity and analogies for obtaining the postcoordinated expression of a given clinical term. We obtained a semantic type accuracy of 98%, relationship accuracy of 90%, and analogy accuracy of 60%, with an overall completeness of postcoordination of 52% for the Spanish SNOMED CT version. We have also applied it to the English SNOMED CT version and outperformed state of the art methods in both, corpus generation for language model training for this task (improvement of 6% for analogy accuracy), and automatic postcoordination of SNOMED CT expressions, with an increase of 17% for partial conversion rate.

30 May 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
JCJavier Castell-Díaz
Javier Castell-Díaz
Jesualdo Tomás Fernández-BreisJesualdo Tomás Fernández-Breis
Jesualdo Tomás Fernández-Breis
CMCatalina Martínez-Costa
Catalina Martínez-Costa
Medical entity linking is essential for structuring clinical narratives, allowing to normalize clinical mentions to standardized terminologies. Despite recent advances based on pretrained language models and reranking strategies, generalization to unseen mentions and unseen concepts remains a major challenge. In particular, most retrieval-based pipelines often rely on selecting the top-ranked candidate after reranking, without fully exploiting the contextual information of the clinical span. In this work, we investigate how large language models can be incorporated into a retrieval and reranking pipeline to improve final candidate selection. The proposed pipeline combines bi-encoder retrieval, cross-encoder reranking, and a large language model that selects the final concept from the reranked candidates by leveraging the full textual context of the mention, including both sentencelevel and section-level information. Our pipeline was evaluated on well-known, publicly available clinical entity linking benchmarks that cover different clinical domains: an English dataset that focuses on clinical findings, procedures, and body structures (SNOMED CT Entity Linking Challenge); and three Spanish datasets that include diseases (DisTEMIST), procedures (MedProcNER), and symptoms (SympTEMIST). Our results show that integrating an LLM as part of the pipeline improves the generalization capabilities of the approach. Our approach shows improvements across the various datasets tested, obtaining an increase in performance of up to 15% compared to state of the art methods.