Segmentation Of Pulmonary Nodules In CT Images Based On Three-dimensional Active Contour

Lung cancer is the leading cause of deaths among all cancers, causing more than 1 million deaths annually in the world. Early detection and treatment of lung cancer are the most important ways of improving the survival rate for lung cancer patients. Computed tomography (CT) has been proven to be a sensitive modality for pulmonary nodule detection and has been widely used for lung cancer detection and diagnosis. Computer-aided detection and diagnosis (CAD) techniques could help to improve the diagnostic accuracy , objectivity and reduce the radiologist's workload. In a Computer-aided detection and diagnosis (CAD) scheme, segmentation of nodule is a crucial and challenging step. Many segmentation methods for pulmonary nodules have been reported, however, automated nodule segmentation in CT scans is still considered to be a very challenging problem because of the variability in shape, texture, and connection of the nodule to normal anatomic structures.Segmentation of pulmonary nodules in images based on three-dimensional active contour has three steps: data preprocessing, three-dimensional active contour segmentation, reconstruction of nodule region.Image Segmentation was performed with a three-dimensional active contour in a local volume of interest (VOI). The three-dimensional active contour segmentation is the extent of classical active contour model, a set of radial lines evenly originating from the center of the VOI was created, each radial line has a point which can move along the radial line, all the points on the radial lines represent a three-dimensional active contour. The contour has continuity energy, gradient energy and balloon energy. The energy function is equal to the sum of all points'weighted sum of three energy term. The final contour which represented the disperse surface of nodule was obtained by minimizing the energy function.A parse point cloud was obtained after active contour segmentation, to obtain a segmentation result in a three-dimensional VOI, a 3D reconstruction method can be used to get all the voxels in a nodule, each voxel was supposed to have a virtual radial line originating from the center of the VOI, existing radial lines can be used to determine that whether the voxel was inside or outside a nodule. An overlap and percentage volume error between nodule regions provided by computer's segmentation and by the radiologist was employed as a performance metric for evaluating the segmentation method. The CT scans from lung image database consortium (LIDC) were used for evaluating the segmentation method. In the evaluation, the coincident rate which is calculated with both the computerized segmented region of nodule and the matching probability map images provided by LIDC was used as performance metric. The two dataset contained 23 and 86 CT scans, respectively, with 23 and 73 nodules that were 3mm or larger in diameter. The mean overlap values were 69% and 63% for the first and second data sets, respectively, the mean percentage volume error value is 21%. The segmentation method is better than other segmentation methods. The preliminary results indicate that the three-dimensional segmentation method is a promising method and would be useful for lung nodule detection and diagnosis scheme.