Research On Automatic Segmentation Of Lung Lobes Based On CT Images

Lung cancer has become the leading cause of cancer deaths in the world widely. Surgical removal of the cancered lobe is the preferred method of treatment of lung cancer. Compared to developing lung cancer operation plans by reading two-dimensional (2D) CT images directly, analyzing the thoracic three-dimensional (3D) CT images and providing3D visualization of lung cavities with computer has distinct advantages for surgical planning and high practical utility in fundamental research and clinical application. A crucial step for achieving3D visualization of lung cavities is the segmentation of lung lobes by identifying lobar fissures in volumetric CT images.In pulmonary3D CT images, lobar fissures has many features, such as low contrast, noise, fuzzy boundary. In high-resolution CT image the tiny vessels or other structures around the fissure complicate the detection task. In this paper, a novel method is proposed to do lung lobe segmentation. This algorithm applies a three-stage approach:the first stage is preprocessing. Use3D region growing to extract a lung field. After morphological dilation and erosion, do threshold segmentation to the lung field to remove the main part of tracheas and vessels. In the second stage, adaptive fissure sweeping is used to coarsely define fissure regions of lobar fissures based on general lung anatomy. Enhance the fissure region had been found with Hessian matrix, then use uniform cost search to refine the location of fissures. The third stage is implicit surface fitting with feature points selected from fissures found in the last stage and corresponding appending control point calculated after. Get a integrated fissure surface and visualize the segmented lung lobe finally.The algorithm combines the morphological characteristics of lobar fissure fully, includes not only the segmentation of left and right pulmonary oblique fissures as many traditional algorithm, but also detailed descriptions of the segmentation of the three lobes in right lung, has strict logic and clear orderliness and provides great convenience to the designation of program. The algorithm had been tested with10cases. The results show that the proposed algorithm had strong adaptive capacity, can do segmentation for the five lobes of left and right lungs properly with segmentation accuracy exceeding92.83%n average. It shows this algorithm has some significance to the segmentation of lung lobe and surgical treatment planning.