Post-processing Segmentation Of The Pulmonary Fissure In 3D CT Images

Lung is one of the most important organs of the human body, but also with a high incidence rate of organs, the common lung diseases emphysema, lung cancer, pulmonary bullae, pneumonia, etc. And the lung cancer is one of the highest mortality of cancer. Clinical doctors to diagnose lung diseases and make treatment decision through the interpretation of CT images. But reading CT images is very tedious and boring, but lobe segmentation can effectively alleviate the doctor's burden, and have great significance for dissease prevention and control.The human lungs can be divided into five independent functional units, called as lung lobe. As physical boundaries between lung lobes, fissure is the foundation of lung region division, lung disease location and quantitative evalution. The segmentation of pulmonary fissure from CT images plays a key role in early disease examination and post surgical navigation. But it is not an easy thing to segment the pulmonary fissure form CT images automatically. First of all, fissures displayed a structure of thin film in 3D space, and it is difficult to deal with 3D image directly; secondly, pulmonary fissure showed a slight curve in 2D plane, the contrast is very low and the distribution of density is uneven; lastly, the tubular structure of blood vessels and trachea in lungs will have a great impact on fissure segmentation. Many experts and scholars at home and abroad have done a lot of work with fissure enhancements, and proposed many effective enhancement method. This paper have put forward several method of the post-processing segmentation of the pulmonary fissure based on derivative of stick filter and the characteristics of the enhancement method, adopt the method of semi-automatic and fully automatic respectively on the 2D and 3D realizing the segmentation of the pulmonary fissure. The main contents of this paper include:1. According to the characteristic of linear structure of the fissure in 2D slices, the open active contour model was adopted to realize the semi-automatic segmentation of pulmonary fissure on 2D slices. Firstly, placing an initial contour line near the fissure. Then, under the internal force in the contour, functions of image force and tension at both ends of the contour line, and make the initialization line near to fissure through iteration. Finally, under the condition of internal and external force balance, the initial line no longer evolution, so as to achieve the purpose of fissure segmentation on 2D slices.2. According to the characteristics of DoS filter, this paper proposed a fissure post-processing method based on skeletonization, realizing fully automated segmentation of the enhanced fissure in 3D CT image. Firstly, using parallel thinning algorithm to extract the skeleton of enhanced image; then retaining the branch point in the skeletonization image and inflate it properly, and then perform arithmetic subtraction operation with the original image, so as to remove branch point in the enhanced image; Finally we take connected component analysis with the image which its branch point has been removed, which can obtain relatively complete fissure through the largest connected component analysis, and then using the characteristic of that fissures and impurities are not directly connected, combining multiple structural elements and multiple connected component analysis, so as to get a complete segmentation of the pulmonary fissure. This method is directly dealing with 3D image, and can keep the original shape of the pulmonary fissure effectively.3. In order to verify the validity of the algotithm, the 3D segmentation results are compared with those normal reference which drawn by hands. Experiments were carried out using typical data in Glucold database. Using the Dice similarity coefficient to determine the correct rate of the alrorithm. At the same time, in order to distinguish whether the algorithm is over or less segmentation, the False Discovery Rate and False Negative Rate are to be used respectively, and using the several indicators as evaluation criteria.The experimental results show that:although the idea of the proposed algorithm in this paper is simple, the segmentation algorithm can get satisfactory result and is stable, and has low time complexity and it has a certain practical significance.