Tez özetleri Astronomi ve Uzay Bilimleri Anabilim Dalı

Computer Aıded Dıagnosıs System For Pulmonary Nodules
Early detection of pulmonary nodules is extremely important for the diagnosis and treatment of lung cancer. With the rapid improvement of technology in recent years, Multi-detector computer tomography (CT) systems, providing to be viewed the small pulmonary nodules by high resolution, have been used. This case is long and tiring process for radiologists. Recently, Computer-Aided Detection (CAD) systems that having an efficient usage area in the medicine have widely been utilized for the detection of abnormalities on medical images. In this respect, CAD systems are extremely important in the diagnosis of disease in terms of offering a second interpretation opportunity for physician, to be a quick decision making ability and to reduce the diagnosis role of human error by using advanced image processing and pattern recognition techniques on the medical images.
In this thesis study, two different CAD systems have newly been proposed for pulmonary nodule detection from computer tomography images as an alternative system for the literature. One is a system aiming at the detection of pulmonary nodule patterns on CT images. The other is a novel computer-aided diagnosis system that deciding the type of the pulmonary nodule patterns as malign or benign from the CT images. In the study, a new classification approach for pulmonary nodules from CT imagery is presented by using hybrid features. Four different feature extraction methods are used on the detection of pulmonary nodules. The overall detection performance is evaluated using various classifiers. The results are compared to similar techniques in the literature by using standard measures. Malign and benign differentiation of pulmonary nodules has been provided by using the risk factors of the patients and morphological image processing approaches. Classification performance measurements are obtained for each method proposed by using the kernel functions of support vector machines.
For pulmonary nodule detection, the proposed approach gives 78.7 % classification accuracy, 78.8 % sensitivity and 76.4 % specificity values with the hybrid features. On the differentiating of malign and benign nodules, classification performance values are calculated as 94.7 % sensitivity and 0.975 AUROC for benign class; 80.0 % sensitivity and 0.889 AUROC for malign class; 77.8 % sensitivity and 0.862 AUROC for uncertain class by 86.8 % accuracy of the classifier.