Large Lung Tumor Segmentation Based On Sparse Model And Random Walk Algorithm

Lung cancer remains one of the leading causes of cancer-related deaths in clinical medicine.Of all the clinical imaging modalities,computed tomography(CT)scans appear to be the most straightforward and are used for feature extraction,disease diagnosis and fore-and-aft therapeutic efficacy assessment of lungs and their lesions.Accurate segmentation of lung tumors based on CT image is of great importance for accurate radiotherapy planning and evaluation of therapeutic response,which is a hot topic in imaging diagnosis of lung cancer.However,it still remains the technical challenges of the segmentation of large tumors with complex structures in lung CT images,including boundary adherent tumors,irregular shapes,heterogeneity,and large tumors with fine burrs.Based on this background,this paper mainly studies the segmentation of large tumors with complex structures in lung CT images,and uses the method of segmenting lung parenchyma before segmenting tumor.The main research contents are as follows:1.Design a method of lung tumor segmentation based on CT image.According to gray-scale information,the lung parenchyma is extracted first,and then the tumor is segmented on the lung parenchyma image.The lung parenchyma image is the image with original gray value on the lung parenchyma region and the rest of the image with black background.As for the extraction processes of lungs,the whole lung can be extracted directly for lungs with solitary tumors,but for lungs with adhesion tumors,the extracted lungs have wrong boundaries caused by the adhesion tumors,which need to be repaired.2.Study on segmentation methods for lungs with boundary adhesive tumors.A sparse model segmentation method for lungs with adhesive tumors is proposed.When tumors adhere to the surrounding tissues of lungs,there is no obvious difference in the gray value between tumors and the surrounding tissues of the lung,which forms a weak boundary between the tumor and the surrounding tissues.Because the gray value of tumors is similar to that of the adhesive tissues,it is difficult to segment the adhesive tumor directly.Moreover,due to the existence of weak boundary,the existing methods can not directly extract the intact lung parenchyma containing the tumor.In order to solve this problem,this paper proposes a lung parenchyma segmentation method based on sparse model,including shape prior model and shape repair model.First,a sparse prior model is used to construct the prior shape of the lungs with adhesive tumors.Then,a sparse deformation model is used to repair the weak boundaryof the pulmonary parenchyma according to the prior shape,and a complete contour of the pulmonary parenchyma is obtained.3.Study on segmentation methods for large tumors with complex structures.The random walk algorithm with a designed way of seed points adaptive generation is used to segment large tumors.The random walk algorithm has a good ability to capture local details of tumor boundary.It is suitable for the segmentation of large heterogeneous tumors with complex local details and irregular shape,which can improve the accuracy of tumor segmentation.However,the determination,the number and distribution of seed points have a great influence on the accuracy,efficiency and reproducibility of segmentation results.In this paper,an image morphology technique is designed to automatically generate the uniformly distributed target seed point and background seed point required by the random walk algorithm in the segmented lung parenchyma image,which achieves automatic tumor segmentation.