Caracterização de nódulos pulmonares em imagens de tomografia computadorizada para fins de auxílio ao diagnóstico

• Ano de defesa: 2015
• Autor(a) principal: Froner, Ana Paula Pastre
• Orientador(a): Silva, Ana Maria Marques da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Elétrica
• Departamento: Faculdade de Engenharia
• País: Brasil
• Palavras-chave em Português: DIAGNÓSTICO POR COMPUTADOR; ENGENHARIA ELÉTRICA; ENGENHARIA ELETRÔNICA; PROCESSAMENTO DE IMAGENS - TÉCNICAS DIGITAIS; PROCESSAMENTO DE SINAIS - TÉCNICAS DIGITAIS; TOMOGRAFIA COMPUTADORIZADA; PULMÕES - DOENÇAS
• Área do conhecimento CNPq: ENGENHARIAS::ENGENHARIA ELETRICA
• Link de acesso: http://tede2.pucrs.br/tede2/handle/tede/6385

Morphological attributes, intensity and texture, are essential to aid the diagnosis of pulmonary nodules. To improve the accuracy of diagnosis, as well as the interpretation of radiological imaging, computer-aided diagnostic systems are used, which help to reduce the interpretation variability. The aim of this study was to evaluate the use of patient data and quantitative attributes of pulmonary nodules in lung computed tomography to build a classification model in terms of malignancy. The study was based on the analysis of 51 patients computed tomography images of lung, 33 patients diagnosed with malignant lesions and 18 patients with benign lesions, all confirmed through anatomo-pathological report of lung tissue. The study included a diagnostic interpretation stage made from a blind study with radiologists and the use of logistic regression to analyze the predictive power of qualitative and quantitative variables extracted from computed tomography lung images. The quantitative attributes of the nodules consisted of twelve morphological attributes (volume, area, perimeter, compactness, roughness and invariant moments of order 1 to 7), five intensity attributes (mass, density, CT number average, fat and calcification indexes) and three texture attributes (homogeneity, entropy and variance). The results showed that in the visual interpretation of radiologists, only the most experienced doctor showed a correlation with the pathology report (64.5%), when excluding the hits due to chance. The model which better predicts the nodules malignancy included a quantitative attribute of localization, the intensity attribute ‘calcification index’ developed in this work, and the compactness, a morphological attribute related to the nodule form. The predictive value of the classification model (86.3%) was much higher than the predictive value based on the visual assessment of the most experienced doctor (65.3%), regardless of the chance, being close to the accuracy of the same doctor (85.1%). This emphasizes the importance of this type of research in the search for a quantitative model that allows the classification of pulmonary nodules.