Medidas de centralidade para classificação automática de doenças pulmonares intersticiais em imagens de tomografia computadorizada

Abstract

Interstitial lung disease (ILD), also known as diffuse parenchymal disease, form a group with more than 150 different pathologies. For the diagnosis of ILD, the computed tomography scan is the most widely used image examination because it is more sensitive than chest thoracic radiography, provides more detailed images of the disease, and with that assists physicians in the your diagnosis. In addition to the computed tomography scan, Computational Vision hás been used to aid in the diagnosis, considering that this is an area that has grown considerably in the last years, especially regarding applications that use medical images. The state of the art in this area has as main characteristic the non existence of a generic algorithm that simulates the vision process, however, it is an area with very promising research involving several theories and methods, such as Complex Networks. Based on this context, presents here an automatic method for computed tomography image classification with ILD, where the images are modeled as Complex Networks and the closeness centrality measure is used to generate a vector of characteristics along with characteristics extracted by the Haralick and LBP descriptors. To validate the proposed method, a K-Nearest Neighbors (KNN) classifier was trained and compared the results obtained with traditional and current literature methods. From a dataset of 3:258 ROIs, it was possible to achieve a rating rate of 89:81%.