EUVIMEDOdkrijte EUVIMED
Open access · CC BY 4.0

Enhancing unstructured clinical note summarization with biomedical entity awareness and heterogeneous graph representation learning

Meena Elumalai, Sivashankari Rajadurai

Array · 2026-11 · 101477

DOI: 10.1016/j.array.2026.101477 ↗

At a glance

Journal article in Array, 2026-11. Written by Meena Elumalai and 1 co-authors. 38 references. The full text is freely available (open access, CC BY 4.0).

PovzetekBibliografski podatkiCiteRelated studies

Abstract (original language)

Purpose

Automatically creating summaries from patient electronic medical records is a common feature of medical information systems. The medical records are long and messy. Clinical notes should be summarized to facilitate access to and use of electronic health records (EHRs).

Methods

This paper suggests a hybrid clinical text summarization pipeline that combines graph-based structural scoring, biomedical language modeling, and conventional natural language processing. The method emphasizes clinically significant entities by combining UMLS-based fuzzy logic, BioLinkBERT for biomedical semantic scoring, and TF-IDF for sentence importance estimation. Sentence ranking incorporates graph-based structural information from a heterogeneous sentence-UMLS concept graph, where sentence-node connectivity is used as the graph-based ranking signal. The MTSamples dataset is used to test the model.

Results

Strong summarization performance is demonstrated by the suggested approach, which achieves ROUGE-1 of 61.11, ROUGE-2 of 59.58, and ROUGE-L of 43.30, suggesting efficient lexical and structural information capture. A BERTScore F1 of 0.9102 demonstrates strong contextual semantic similarity with the corresponding source transcriptions.

Conclusion

The findings demonstrate the potential of combining biomedical language models with knowledge-graph based structural scoring for extractive clinical note summarization.

Cite

APA

Elumalai, M., & Rajadurai, S. (2026). Enhancing unstructured clinical note summarization with biomedical entity awareness and heterogeneous graph representation learning. Array, 101477. https://doi.org/10.1016/j.array.2026.101477

Vancouver

Elumalai M, Rajadurai S. Enhancing unstructured clinical note summarization with biomedical entity awareness and heterogeneous graph representation learning. Array. 2026:101477. doi:10.1016/j.array.2026.101477

BibTeX
@article{Elumalai2026,
  author = {Elumalai, Meena and Rajadurai, Sivashankari},
  title = {Enhancing unstructured clinical note summarization with biomedical entity awareness and heterogeneous graph representation learning},
  journal = {Array},
  year = {2026},
  pages = {101477},
  doi = {10.1016/j.array.2026.101477}
}
Start for free →