Visão computacional aplicada em imagens Speckle para o diagnósticco de lesões de cárie

• Ano de defesa: 2019
• Autor(a) principal: Gavinho, Luciano Gillieron
• Orientador(a): Deana, Alessandro Melo
• Banca de defesa: Deana, Alessandro Melo , Daghastanli, Nasser Ali , Araújo, Sidnei Alves de , Alves, Wonder Alexandre Luz
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Nove de Julho
• Programa de Pós-Graduação: Programa de Pós-Graduação em Informática e Gestão do Conhecimento
• Departamento: Informática
• País: Brasil
• Palavras-chave em Português: Speckle; qui-quadrado; geometria fractal; lesões de cárie; cárie dentária; dimensão fractal; redes neurais artificiais
• Palavras-chave em Inglês: Speckle; chi-squared; fractal; oral lesions; dental caries; fractal dimension; redes neurais
• Área do conhecimento CNPq: CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO
• Link de acesso: http://bibliotecatede.uninove.br/handle/tede/3047

Computer vision technologies for diagnostic imaging applied to oral lesions, specifically, carious lesions of the teeth, are in their early years of development. The relevance of this public problem, dental caries, worries countries around the world, as it affects almost the entire population. The present work is focused on the creation of new methods for the extraction of information and segmentation in special images, denominated emph "laser speckle images", or LSI. The emph laser speckle image has important features that have information about the illuminated surface, but few works have been done in the application of computer vision methods to LSI. These new methods are quite effective in the diagnosis of early-stage lesions, often imperceptible by traditional diagnostic methods in a clinical setting. The first proposed method is based on the distance of the chi-square ( 2) between the histograms extracted of the image, bringing a significant improvement in the accuracy of the diagnosis in relation to the previous techniques, while a second method introduces the use of the fractal geometry, demonstrating, through the fractal dimension, the difference between lesioned areas and sound areas of a tooth. The third method makes the information of the first two methods the input of an artificial neural network, achieving greater precision in the carious lesions diagnostic. Several software were developed for the execution of the models and demonstration of the practical applicability of these techniques. The presented results showed excellent accuracy and precision in the segmentation of the images (>98%), in addition to showing an evolution between them, as expected.