ONTOLOGIAS PARA REPRESENTAÇÃO DE CONHECIMENTO EXTRAÍDO DE LAUDOS DE RADIOLOGIA TORÁCICA

Abstract

This work comprises the design and evaluationof a method for the extraction of knowledge from free-textchest X-ray reports. Our final goal is to be able toautomatically transform the free-text reports into ontologyusing OWL with Protégé documents conveying relevantinformation. To achieve that a set of 2000 reports, written inPortuguese, formed the basis from which a corpus wasderived. A classical approach was taken to analyzing thedata: tokens were found and n-grams were identified, theirfrequency and relevance computed. All relevant 160 ngramswere mapped into SNOMED, LOINC and/or DICOMclasses and into an HL7/CDA template. This approach hasmade it possible to convert the free-text reports into CDAdocuments.Index Terms ⎯ Health Information Standards, HealthInformation Ontology, Health Information Systems, NaturalLanguage Processing.