Proposta de sistema de baixo custo para triagem de pacientes com retinopatia diabética

• Ano de defesa: 2019
• Autor(a) principal: Werner, Lucas Vieira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Curitiba; Brasil; Programa de Pós-Graduação em Engenharia Biomédica; UTFPR
• Palavras-chave em Português: Retinopatia diabética - Diagnóstico; Rastreamento médico; Oftalmoscopia; Aplicativos móveis - Desenvolvimento; Aprendizado do computador; Processamento de imagens - Técnicas digitais; Lentes - Avaliação; Imagens e fantasmas (Radiologia) -Testes; Redes neurais (Computação); Interface de programas aplicativos (Software); Inteligência artificial; Software - Validação; Diabetic retinopathy - Diagnosis; Medical screening; Ophthalmoscopy; Mobile apps - Development; Machine learning; Image processing - Digital technique; Lenses - Evaluation; Phantoms (Radiology) - Testing; Neural networks (Computer science); Application Program Interfaces (Computer software); Artificial intelligence; Computer software - Validation; CNPQ::ENGENHARIAS::ENGENHARIA BIOMEDICA; Engenharia Biomédica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/4838

Diabetes is an important worldwide disease, with impacts mainly on primary health care, which can present several complications, among them diabetic retinopathy. Diabetic retinopathy is one of the leading causes of blindness worldwide. Diagnosis of diabetic retinopathy is done through fundus examination, but the equipment to perform such an examination is difficult to access, especially in less developed countries. Therefore, this study aims to analyze and propose methods for better screening for diabetic retinopathy, using background images taken by a cell phone in conjunction with lenses. Lenses and methods for obtaining the fundus image were analyzed, as well as the analysis of machine learning techniques and neural networks for image processing. As a result, the most suitable lens was the unattached VOLK 20 diopter lens. The proposed method for image acquisition, consisting of a stable and movable physical structure on all axes, with varying presence (eye, lens and cell), was developed by 3D modeling and printing. For processing, the Tensorflow was chosen because of its very active community and its good results for two-dimensional images. For the system integration, a mobile application was developed in order to interface between the image obtainment and the neural network. Phantom tests were performed using the images from the EyePacs-1 database. After processing, the integrated system returns to the user, through the application, the likelihood of the existence of diabetic retinopathy. An f-score of 0.64 was obtained. Thus, it is possible to develop a screening method for diabetic retinopathy, for use in primary care in locations with little access to eye care.