Orientador: Prof. Dr. Marcelo Zanchetta do Nascimento / Dissertação (mestrado) - Universidade Federal do ABC, Programa de Pós-Graduação em Engenharia da Informação, 2015. / O câncer de próstata é o segundo que provoca o maior número de vítimas fatais entre os homens, atingindo principalmente a população mundial com idades superiores a 60 anos. Entre os métodos empregados para o diagnóstico médico estão os exames clínicos, laboratoriais e o diagnóstico por imagem, o que pode indicar a necessidade da biópsia da próstata. As biópsias são avaliadas por especialistas para auxiliar na conduta mais adequada de tratamento, desta forma o estudo por imagem histológica é realizado e se destaca como um dos métodos utilizados devido a facilidade de diagnosticar a doença. Porém, ainda existem problemas que precisam ser solucionados para reduzir o número de falsos positivos. Este trabalho apresenta um conjunto de técnicas para identificar e quantificar as regiões de interesse em imagens histológicas da próstata. As análises foram realizadas com dimensão fractal de imagens coloridas e classificadas com SVM com os kernels linear, polinomial e RBF. As regiões de interesses foram segmentadas em núcleos da célula cuboide, lúmens glandulares e tecido estromal e aplicado o cálculo da dimensão fractal. A avaliação de desempenho foi baseada na área sob a curva ROC (AUC) e pela acurácia. Os resultados obtidos com essas ferramentas mostram que o grupo de imagens segmentadas por estroma com magnificação de 100x obtiveram melhores resultados de classificação, obtendo valores de AUC de 92,21% e 86,77% de acurácia para os grupos de tecido normal versus tecido tumoral, obteve 73,53% de acurácia para o grupo tecido normal versus tecido hiperplásico e de 80,00% para o grupo de tecido hiperplásico versus tecido tumoral. O método proposto quantificou tecidos histológicos da próstata com descritores baseados em técnicas não lineares multi-escala. O uso de informações dos canais de cores em conjunto com a segmentação das estruturas foi mais relevante para um sistema de apoio ao diagnóstico. / Prostate cancer is the second type of cancer that causes more deaths between men. It affects mainly the population over the age of 60. Laboratory exams and diagnostic imaging are among the methods used for medical diagnosis, which may indicate the need for a prostrate biopsy. Biopsies are evaluated by experts in order to indicate the most appropriate treatment strategy. Hence, the study of histological images stands out as one of the most used methods as it allows an easier diagnosis. However, there are still problems that need to be addressed to reduce the number of false positives. This work presents a set of techniques to identify and quantify regions of interest in histological images of the prostate. Color and greyscale images were analysed using fractal dimension then classified in SVM with linear, polynomial and RBF kernels. Regions of interest were segmented in basal cell cuboid, glandular lumens and stromal tissue and then a fractal dimension was applied. Performance evaluation was based on the area under the ROC curve(AUC) and accuracy. The results obtained by applying these tools show that images segmented by stroma with a magnification of 100x had better classification results, achieving AUC values of 92.21% and 86.77% accuracy for the normal tissue groups versus tumor tissue. Also, in this group of images a level of accuracy of 73.53% for hyperplastic tissue versus normal tissue and 80.00% for hyperplastic tissue versus tumor tissue. The method quantified histological prostate tissue with multi-scale techniques based on nonlinear descriptors. Therefore, the use of information from color channels together with the segmented structures are most relevant to a diagnostic support system.

In cancer diagnosis there is a goal of having the treatment being tailored to each patient. This in order to increase efficiency and reduce side effects. Using more data on each patient can help in achieving this. One such data source is histological images on tissues, such as lymph nodes. This report sets out to find a method in which such images on lymph nodes can be automatically segmented. This so that they can later be analysed and maybe tell in what stage a cancer is in. Such work is today done by hand, and this makes it a subjective process, that might differ between doctors and institutions. If there was a method done by a computer, the process would be replicable and objective. Also, a lot of time would be saved. The results show that such a method is reachable in this early stage of development. It is also quite efficient when segmenting the lymph node itself. The segmentation of smaller areas of the lymph nodes is not as efficient, but with further work in the area it might improve enough to be useful. Some issues are still had since the method relies in part on a person to decide a parameter in order to get a clean segmentation. The final conclusion is that one model is to prefer compared to the others and that further work on this might make it a useful tool in analysing histological images.