Research And Implementation Of Lung Nodule Segmentation Algorithm Based On Thoracic CT Images

In recent years, the morbidity and mortality of the lung cancer have been arising all over the world, which are both the highest among all types of cancers. The lung cancer cannot be cured nowadays, so early diagnosis and treatment is the main way to reduce the mortality of the patients. Lung cancer manifests itself as nodule with little volume in their early stage, which is easily overlooked by the doctor and patient. Computed tomography (CT) is superior to other imaging techniques for detecting lung nodules for its high resolution, three-dimensional character and sensitiveness, which can demonstrate the nodules’position, shape, size, inner structure, density, boundary features and attachment to other tissues. The size, growth rate and shape are the most important indicators for judging the malignancy of the nodules. Precise segmentation of pulmonary nodules is an important premise to quantitatively analyze nodules and identifying the benign and malignant nodules, therefore, the segmentation of nodules has become one of the most important contents in lung CAD.Due to the characteristics of CT imaging and the complexity of the nodule itself, different types of nodules show dramatic differences in the grayscale value, shape and adhesion of the surrounding tissue, which make the segmentation of nodules become very challenging. In this paper, three segmentation algorithms are proposed including two designing for specific type of nodules and one for all types of nodules.(1) Segmentation of the juxta-vascular nodules based on local shape analysis. First, utilize the three dimensional distance transform method to extract the core of the nodule; then carry out the region growing process based on the geodesic distance from the core; finally detect the position where the nodule and the blood vessel connect with each other and remove the vessels.(2) Segmentation of the juxta-pleural nodules based on three dimensional ray casting. First, use the ray casting method to detect the nodule boundary adjacent to the parenchyma. Then reconstruct the nodule boundary with the convex hull operation.(3) Segmentation of lung nodules based on steepest descending test. First, binarize the image with competition diffusion system and extract the nodule core with the three dimensional distance transform method; then find the nodule boundary utilizing the steepest descending test; finally extract different types of nodules with region growing method. In the experiments, the three segmentation algorithms all gain good results with high accuracy and efficiency. The overlapped ratio compared to the ground truth reached85.16%,93.03%and85.35%and the average segmentation time was2s,2.7s and1.5s respectively, which shows that these algorithms can be used to quantitatively analyze the nodule and provide helpful information for the nodule diagnosis and treatment.