Automatic identification of biomedical concepts in spanish-language unstructured clinical texts

Abstract

The processing of health information from medical records and, especially, clinical notes is a complex task due to the nature of the texts themselves (i.e., hand-written and containing semistructured or unstructured data) and the diversity of the terminology used. While certain technologies exist to process these types of texts and data in the English language, only a few such initiatives exist for similar texts and data in the Spanish language. This paper presents a new proposal for the semantic annotation of Spanish-language clinical notes, implementing an automated tool similar to the UMLS MetaMap Transfer (MMTx) for the identification of biomedical concepts in the Spanishlanguage SNOMED CT ontology. Moreover, an assessment of the tool using 100 Spanish-language clinical notes is presented. Using the clinical notes manually annotated by specialists of a Spanish hospital as the gold standard, it is concluded that precision scores are sufficiently good for the several types of matching achieved by the automated tool proposed. The research presented in this contribution offers a launching point for the establishment of semantic relationships between concepts and the application of mining techniques to Spanish-language clinical notes.

Automatic Annotation of Medical Records in Spanish with Disease, Drug and Substance Names

“…We did not found any publicly avalaible corpus composed of electronic medical records in Spanish, so after several meetings with the legal advice services of the University and the Hospital, and after signing the corresponding confidentiality agreement, we obtained a corpus of patient records. Having a "private" corpus, our results are not comparable to others, as in other related works [10].…”

“…In [9] the authors present a first simple approach to the Spanish MetaMap, using Google Translator to obtain an English version of the text and then applying English MMTx to extract the concepts. In [10] a system for the automated identification of biomedical concepts in Spanish-language clinical notes is presented.…”

Spanish Named Entity Recognition in the Biomedical Domain

“…They achieve high F1s, but use following extremely loose matching criteria: "two elements are considered to be equivalent if an element given by the system is entirely contained within an extension of a manually tagged element by six positions both to the left and to the right". A tool similar to UMLS MetaMap Transfer (MMTx) 7 has been presented in [7] for the identification of Spanish SNOMED CT 8 terms corresponding to the procedures and disruptions hierarchies in Spanish clinical notes. The tool is tested with 100 clinical notes.…”

Cross-lingual semantic annotation of biomedical literature: experiments in Spanish and English

“…Later, Castro et al (2010) developed an automatic system for the recognition of SNOMED CT concepts by computing a similarity function between sentences in clinical notes and then term normalisation is based on the results obtained by querying an Apache Lucene 3 index of SNOMED CT and re-ranking the candidates with a function of their own. They obtained an average F1 score of 0.11 on their own corpus of 100 manually annotated documents.…”