International Journal of Data Science and Analytics Информационные системы

Entity extraction in clinical summary of mammary malignancy data using transformer-based models

Patra, Payel, Di Marco, Antinisca, Nguyen, Phuong T.
Springer Nature · 2026-10-31 · DOI 10.1007/s41060-026-01342-z

Аннотация

Clinical summaries in mammary malignancy contain important patient information, but they are often unstructured and difficult to use for automated analysis. Extracting relevant clinical entities from such text is essential for improving data accessibility and supporting clinical research. In this study, we focus on entity extraction from Clinical Summary Mammary Malignancy (CSMM) data using transformer-based models. We manually annotated a domain-specific corpus covering entities such as age, gender, disease, symptoms, medication, dosage, medical history, and cancer stage. We evaluated multiple BERT-based models, including BioBERT, BioClinicalBERT, RoBERTa, PubMedBERT, and BlueBERT, under both zero-shot and fine-tuned settings. Our results show that fine-tuning significantly improves performance compared to pre-trained models. Among the evaluated models, fine-tuned BioBERT achieved the best performance with an F1-score of 97.03. We also observed that certain entity types, such as dosage and cancer stage, remain more challenging due to variability in clinical text. Although the dataset is limited in size, the high annotation agreement supports its reliability. This work highlights the importance of domain-specific annotation and provides insights into applying transformer-based models for structured information extraction in breast cancer clinical data.

Читать у издателя →