Detecção de câncer de próstata em imagens de microscopia utilizando grafos de contexto glandular

Abstract

Every year a large part of the male population is affected by Prostate Cancer (PCa), with many cases of deaths occurring especially in men over 40 years old. PCa is a pervasive condition that diffuses and manifests itself in a wide range of histological patterns, which can be visualized with details in histological images acquire through biopsy or prostatectomy. Early detection of PCa can improve the prognosis and reduce the risk of death. Currently, the main methodology for the diagnosis of PCa consists of a qualitative analysis carried out by specialists to define the degree of the disease in the so-called Gleason Grading System(GGS), originally defined by Donald Gleason and refined by the International Society of Urological Pathology. Given the importance of identifying abnormal prostate tissue (staging) to improve the prognosis, many computerized methodologies have been developed to assist pathologists in a systematic way for the diagnosis. It is often argued that an improved diagnosis of a tissue region can be obtained by considering measures that take into account various properties of the tissue surroundings, henceforth referred as the context of the tissue. Such a context is considered an important biological factor in staging. This work proposes a new methodology that can be used to systematically define contextual features related to prostate glands. The Gland Context Network (GCN) structure is defined, which is a representation of the prostate sample containing information about the spatial relationship between the glands as well as the similarity between their appearance. It is shown that the GCN can be used to establish contextual features at any spatial scale. Therefore, information that is not easily defined from traditional features can be easily extracted using the proposed approach. In addition, it is identified that even basic properties derived from a GCN can lead to state-of-the-art classification performance in relation to PCa. All in all, GCNs can assist in defining the most effective approaches for PCa detection.