Identificação e Quantificação de Células Oncocíticas em Imagens Microscópicas

• Autor(a) principal: Tiago Marques Dias da Mota
• Data de Publicação: 2014
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: https://hdl.handle.net/10216/74376
• topic: Engenharia electrotécnica, electrónica e informática; Electrical engineering, Electronic engineering, Information engineering

Nowadays great scientific fields, such as Medicine, have been recurring to technological advances in terms of computational power and storage capacity. Now it is possible to store large quantities of high resolution images in databases, allowing medical images to be saved for posterior analysis by experts. The problem associated with this resides in the task ofmanually analyze the images, which can be exhausting and time consuming, with the probability of having direct influence on the results and conclusions obtained by the pathologists, due to these factors and also their subjectivity. By applying Image Processing techniques and Data Mining methods, many medical images have been successfully analyzed with a computer, by means of automatic procedures showing results with high accuracies, that expert pathologists may use to better support their medical diagnosis decisions. Previous studies show that the presence of oncocytic cells in certain types of diseases, like thyroid tumors, may have direct influence on used treatments, which makes extremelly important for a pathologist to have access to this information, at the time he or she is performing the diagnosis. OncoFinder shows that it is possible to create a software tool totally capable of identifying and classify automatically the oncocyte present in microscopic images of thyroid tumors with high quality and resolution, provided by the National Institute of Health. With the help of OncoFinder, the experts, that worked with us, had automatic access to images with cell nuclei segmented, ready to be classified as oncocyte, non-oncocyte or any other component. They generated data that was used to build appropriate datasets to train and test different learning classifiers. The outcomes show that some classifiers can achieve accuracies around 90% of correctly classified oncocytic cells.