Diagnóstico automático de ovos de parasitos intestinais humanos a partir de imagens microscópicas utilizando redes neurais convolucionais

• Ano de defesa: 2020
• Autor(a) principal: Bruno Alberto Soares Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ENG - DEPARTAMENTO DE ENGENHARIA ELÉTRICA; Programa de Pós-Graduação em Engenharia Elétrica; UFMG
• Palavras-chave em Português: Diagnóstico por imagens médicas; Doenças parasitárias; Imagem microscópica óptica; Ovos de parasitos intestinais; Redes neurais convolucionais; Engenharia elétrica; Diagnóstico por imagem; Intestinos - Parasitos
• Link de acesso: http://hdl.handle.net/1843/40064

One of the biggest concerns in the area of public health is caused by human intestinal parasites, which are found largely in tropical countries. The diagnosis of these parasitic diseases is done through physiological symptoms and fecal examination. Often, few professionals are available and able to perform this type of examination, which is considered slow, difficult, prone to errors, and can cause eye strain in the specialist. The area of pattern recognition in images presents itself as a promising alternative as support and automation of exams based on images. Also, deep learning techniques have been successfully applied for this purpose. Therefore, the objective of this work is to use convolutional neural networks to classify eggs from intestinal parasites, being a system to aid decision making in the diagnosis of a stool test. A real database was built with 66 images of different species of parasite eggs (Ancylostoma duodenale, Necator americanus, Ascaris lumbricoids, Enterobius vermicularis, Schistosoma mansoni, and Trichuris trichiura). Data augmentation techniques were used to obtain a larger number of samples, with a total of 48 thousand images at the end. Empirical experiments were carried out to define a specific network architecture for each problem. The greatest difficulty of the specialist's diagnosis is in locating the eggs mixed with impurities and dirt contained in the slide, for this reason, a Convolutional Neural Network architecture was implemented to solve a binary classification problem and another for a multiclass classification problem. The results obtained demonstrated a recognition rate of 99.9%, for all metrics evaluated. The developed application will be an essential part of a future system that will be fully automated.