Navegando por Palavras-chave "expert testimony"

Agora exibindo 1 - 1 de 1
  • Nenhuma Miniatura disponível
    Item
    Somente Metadadados
    Reconhecimento de entidades mencionadas para auxílio na descoberta de conhecimento em laudos de biópsia renal escritos em texto livre
    (Universidade Federal de São Paulo (UNIFESP), 2014-07-30) Nicolas, Flavia Pena [UNIFESP]; Pisa, Ivan Torres Pisa [UNIFESP]; Universidade Federal de São Paulo (UNIFESP)
    Introduction: The health area is currently experiencing a great need for acquisition of knowledge, particularly from patient health records. This demand has caused techniques for natural language processing and text mining become indispensable resources for processing information. Objective: Thus, this study aimed to recognize named entities in renal biopsy reports, using text mining techniques supported by NLP and machine learning. Secondary objectives were to group the terms, characterizing sections of reports and create a specific vocabulary to renal biopsy area, aiming a future establishment of an ontology and support the knowledge discovery. Methods: To achieve the main goal, we used text mining techniques and tools, and we create an automatic terms recognizer based on the four vocabularies in Portuguese in the UMLS: DeCS, MedDRA, WHO and ICPC. In a complementary manner, we use techniques of machine learning and statistical analysis to classify and characterize the sections of the reports in accordance with the terms DeCS automatically recognized. Results: The recognizer was applied to the pre-processed reports, using the four vocabularies in Portuguese, also pre-processed. The best performance was achieved with DeCS while the worst was with ICPC. The number of terms that was automatically recognized was small, which was confirmed in the validation, after manual recognition of terms held for six volunteer doctors. This result is due to the scarcity of vocabularies in Portuguese, neither of which specifically covers the renal area. Conclusion: Thus, we conclude that the text mining techniques and term extraction tools were satisfactory, but because of the lack of vocabularies in Portuguese, in renal area, we couldn’t recognize a lot of terms automatically, generating differences between the terms that were automatically recognized and the terms that were recognized by doctors. Based on the intersection of these two results we create a vocabulary for renal biopsy that will be used to creating ontology and decision support systems, and assist in knowledge discovery. As complementary activities, we grouped the terms DeCS recognized in sections, using ML classifiers and we characterized the sections of reports based on the connections between DeCS terms, using statistical analyses.