Uso da inteligência artificial aplicada ao diagnóstico bucal

• Ano de defesa: 2024
• Autor(a) principal: Araújo, Eduarda Gomes Onofre de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal da Paraíba; Brasil; Odontologia; Programa de Pós-Graduação em Odontologia; UFPB
• Palavras-chave em Português: Inteligência artificial; Diagnóstico bucal; ChatGPT; Artificial intelligence; Oral diagnosis; CNPQ::CIENCIAS DA SAUDE::ODONTOLOGIA
• Link de acesso: https://repositorio.ufpb.br/jspui/handle/123456789/36767

The objective of this dissertation was to address and elucidate the use of Artificial Intelligence in the diagnosis of oral lesions and, for this purpose, two work plans were developed. The first plan consisted of preparing and developing a bibliometric review with the objective of quantifying, analyzing and evaluating the scientific academic production on the use of Artificial Intelligence for the diagnosis of oral lesions. The databases chosen to conduct the review were: Medline via PubMed, Scopus, Web of Science and Cochrane Library. For all databases, a search strategy was defined based on the terms of the Medical Subject Headings, synonyms and relevant free terms, combining them with Boolean operators. The results of the searches in the databases were exported for bibliometric analysis through the R and RStudio software, using the Bibliometrix package. In total, 3,858 studies were obtained in the databases. 902 duplicate files were removed, leaving 2,956 publications for content evaluation. After applying the eligibility criteria and excluding studies that were not in line with the topic of interest, 334 articles remained for bibliometric analysis. In the second work plan, a study was carried out to evaluate the performance of the AI Oral Diagnosis Helper (AODH) virtual assistant, developed using the ChatGPT™ platform (versions 4 and 4o). Thirty clinical cases were selected for evaluation by two specialists in Stomatology and the AODH. The diagnoses and treatment suggestions were compared to a gold standard specialist. Agreement rates and accuracy were calculated using Fleiss's Kappa with a 95% confidence interval, using RStudio. The AODH with version 4 correctly diagnosed 22 of 30 cases (73.3%), while the AODH based on version 4o correctly identified 24 cases (80%). Fleiss's kappa indicated substantial reliability between the AODH and the specialists (K = 0.79). With the development of work plans, it is possible to conclude that the integration of Artificial Intelligence in the area of oral diagnosis can enrich learning and clinical practice, as long as it is used cautiously and consciously, serving as support for students and professionals.