Random Walk And Graph Cut For Co-Segmentation Of Lung Tumor On PET-CT Images

Today the environment is worsening, lung cancer has become a major threat to human health. Accurate lung tumor delineation plays an important role in radiotherapy treatment planning and has become a popular research task. PET-CT, as a quantitative molecular-anatomic modality, is widely used in radiation treatment planning. We proposed a co-segmentation theory on PET-CT images.In this study, we effectively integrate the two modalities by making fully use of the superior contrast of PET images and superior spatial resolution of CT images. Random walk and graph cut method are integrated to solve the segmentation problem, in which random walk is utilized as an initialization tool to obtain initial contours. The co-segmentation problem is solved by graph cut algorithm using the initial contours as object seed points. Graph cut formulates the segmentation problem as an energy minimization problem which can be solved by max-flow/min-cut method. In this paper, the novelty is the graph construction and the energy representation. The graph includes two sub-graphs, one for PET and another for CT. To fully utilize the characteristics of PET and CT images, a novel energy representation is devised. For PET, a cost based on SUV distribution, a downhill cost and a 3D derivative cost are proposed. For CT, a shape penalty cost is integrated into the traditional data energy term and boundary energy term. The shape penalty cost helps to constrain the tumor region during the segmentation. We validate our algorithm on a dataset which consists of 18 PET-CT images which are from the patients with NSCLC(Non-small cell lung cancer). We compare our method with random walk and traditional graph cut method and so on. The comparative experimental results indicate that our proposed method is superior to some other segmentation algorithms.