Leveraging Semantic Web Technologies in Healthcare: Development of OntoDrug for Medication Ontology, Prescription Recognition Systems, and Cancer Staging Applications

• Ano de defesa: 2024
• Autor(a) principal: Miranda, Nelson Julio de Oliveira
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Cancer staging; Estadiamento do câncer; Gerenciamento de medicamentos; Large language models (LLMs); Medication management; Modelos de linguagem de grande escala (LLMs); Ontologias; Ontologies; Semantic web; Web semântica
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/55/55134/tde-25042025-085137/

This PhD thesis, structured as a collection of articles, investigates the application of Semantic Web technologies, ontologies, and Large Language Models (LLMs) to tackle important issues in healthcare, specifically in medication management and cancer staging. Every article within this thesis brings forward distinct advancements in these fields. The initial article offers an extensive examination of Semantic Web technologies in the healthcare sector, emphasizing their contribution to improving data interoperability in Electronic Health Records (EHRs). This scoping review highlights progress made with technologies such as RDF, OWL, and SPARQL, as well as difficulties that impede wider acceptance. The second article focuses on cancer staging and introduces an ontology-based TNM classifier tailored for liver cancer. This classifier, based on the AIM4-O ontology, standardizes image annotations to automate cancer staging. With a precision of 85.7% and recall of 81.0%, the TNM classifier aligns closely with physician-assigned stages, demonstrating the efficacy of an automated, ontology-based approach in oncology. The third article examines the application of a novel ontology, OntoDrug, in regulatory compliance, specifically within the field of otorhinolaryngology. OntoDrug was developed to enhance medication management in Brazil. By standardizing regulatory lists and supporting data interoperability within EHR systems, OntoDrug enables safer and more accurate medication management. This study demonstrates high accuracy in identifying medications, achieving a 94.5% success rate, with 61% full recognition and 33.5% partial recognition. These findings highlight the effectiveness of combining Natural Language Processing (NLP) with ontological frameworks to ensure medication safety and accuracy. In the final article, we describe the OntoDrug in more detail and examine how OntoDrug can be integrated with LLMs. OntoDrug with LLMs achieved exceptional results in identifying medications from free-text prescriptions, with GPT-4 turbo achieving 100% recall and 97% precision. This performance underscores OntoDrugs potential to advance medication recognition and enhance patient safety within clinical environments. Together, these articles illustrate how machines can more effectively understand and reason about clinical data using these technologies to offer better cancer staging and more accurate medicine identification.