Citation

da Silva DP, da Rosa Fröhlich W, de Mello BH, Vieira R, Rigo SJ. Inform. Med. Unlocked 2023; 43: e101381.

Copyright

(Copyright © 2023, Elsevier Publishing)

DOI

10.1016/j.imu.2023.101381

PMID

unavailable

Abstract

The available natural language data in electronic health records is of noteworthy interest to health research and development. Nevertheless, their manual analysis is not feasible and poses a challenge to accessing valuable information in these records. This paper presents an approach to automatically extract information from these unstructured medical records using Domain Entity Recognition and Relation Extraction, structuring the results through a domain ontology. We developed our work in the oncology domain, an attention-demanding field. The main contribution of this work lies in integrating multiple resources in a complete methodology to accomplish this task. We developed a new entity and relation annotated dataset of medical evolutions in Brazilian Portuguese, containing 1622 documents, 146,769 entities, and 111,716 relations. We attained 78.24 % accuracy for entity and relation extraction in the exams domain. Healthcare specialists evaluated the approach regarding entity recognition and relation extraction positively and considered the methodology valuable to health professionals.

Language: en

Keywords

Deep learning; Electronic health record; Named entity recognition; Natural language processing; Ontology; Relation extraction