Diagnóstico de glaucoma em retinografias utilizando funções geoestatística

• Ano de defesa: 2017
• Autor(a) principal: SOUSA, Jefferson Alves de
• Orientador(a): ALMEIDA, Joao Dallyson Sousa de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Maranhão
• Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM CIÊNCIA DA COMPUTAÇÃO/CCET
• Departamento: DEPARTAMENTO DE INFORMÁTICA/CCET
• País: Brasil
• Palavras-chave em Português: Glaucoma; Diagnostico; Retinografia; Processamento de Imagens
• Palavras-chave em Inglês: Diagnosis; Retinography; Image Processing
• Área do conhecimento CNPq: Processamento Gráfico; Oftalmologia
• Link de acesso: http://tedebc.ufma.br:8080/jspui/handle/tede/1536

Glaucoma is one of the diseases that mopstly causes blindness, according to estimates by the World Health Organization (WHO). The Brazilian Council of Ophthalmology (CBO) estimates that in Brazil there are 985 thousand glaucoma patients with more than 40 years of age. The use of computer aided detection and diagnosis systems (CAD - Computer Aided Detection and CADx - Computer Aided Diagnosis) has contributed to increase the chances of detection and correct diagnoses. They provide a second opinion, assisting the experts in making decisions about the treatment of glaucoma. The main objective of this work is to present a method for automatic diagnosis of glaucoma in retinography images using texture characteristics. The Local Binary Pattern is used to generate a representation of the texture patterns of the image. The geostatistical functions, semivariogram, semimagram, covariogram and correlogram, are used as texture extractors. With the generated characteristics, a step is made to select the best classification model using the genetic algorithm. Then sorting is performed using the Support Vector Machine. The best result was an accuracy of 91%, sensitivity of 95% and specificity of 88%, proving that the characteristics generated by the geostatistical functions for texture extraction generate a satisfactory discriminant set.