Research On Lung And Lung Tumor Segmentation For CT Images Based On Random Walk Algorithm

Segmentation of lung and lung tumor is still a challenging task. To deal with the difficulty of lung segmentation and lung tumor segmentation problem under complex conditions in chest CT images, this paper proposed an accurate and effective segmentation method for lung segmentation and lung tumor segmentation, which is based on the improved random walk algorithm.Random walk algorithm is namely image as connected weighted directed graph composed of fixed vertices and edges. Given a small number of pixels with user-defined labels, one can analytically and quickly determine the probability that a random walker starting at each unlabeled pixel will first reach one of the pre-labeled pixels. By assigning each pixel to the label for which the greatest probability is calculated, the image segmentation may be obtained. Wherein, the placement and the quantity of seeds can affect the final segmentation results. According to the imaging characteristics of the chest CT images, and the properties of random walk algorithm, random walk algorithm was improved and the edge weights in this algorithm were redefined. Furthermore, this paper proposed an innovative way for seeds placement.Aim at the case like juxta-pleural tumor and high density blood vessels in the hilum, left and right lung close to each other, lung area disconnection, and fuzzy boundary of lung in the CT images, this paper proposed the method which is "seeds selection twice—random walk segmentation twice" for lung segmentation, wherein use the improved random walk algorithm as the main segmentation method, and utilize threshold and mathematical morphological operation to select seeds automatically. The segmentation results can achieve the goal of the integrity of the lung segmentation. Afterwards, the improved random walk algorithm is extended to 3D field. Image sequences of segmented 2D lung regions were utilized to segment the 3D lung tumor, and utilize a small amount of interaction and region growing to select seeds semi-automatically. Image data in the spatial position and gray correlation were considered fully. The segmentation results can ensure the accuracy of the lung tumor segmentation.Finally, this paper conducted a large number of related experiments, and analyzed and discussed the experimental results qualitatively and quantitatively. Compared with the gold standard, results show that the proposed method can ensure the accuracy and robustness of the segmentation. Meanwhile, the results of our proposed method can meet the requirements of clinical diagnosis and treatment and pathological analysis and research.