A Study of Pulmonary Nodule Detection in Three-Dimensional Thoracic CT Scans

Early detection of pulmonary nodules on multi-slice computed tomography (MSCT) is an important clinical indication for early-stage lung cancer diagnosis. Currently, support vector machines (SVMs) have been widely used in pattern recognition. We are developing a nodule detection system by using the combination of SVMs. For training and testing the SVM classifier, a lung nodule simulation method in three-dimensional (3D) image space is developed to provide a better test environment. The detection method is multi-step, including 3D nodule simulation and insertion, lung fields segmentation, initial nodule candidates extraction, 3D geometry and intensity features calculation and false positives (FPs) reduction based on SVM. The proposed scheme is applied on three mixed chest MSCT datasets and evaluated using receiver operating characteristic (ROC) curve analysis. The experimental results illustrate the efficiency of the proposed method and the sensitivity of SVM classifier is found to be 86%.