Research On Key Techniques Of Medical Image Segmentation Based On Graph Theory

Medical image segmentation occupies an important position in clinical diagnosis and surgery planning.Accurate segmentation of lung tumor and liver has been widely studied as that lung tumor and liver cancer are common malignant at this stage.CT image is widely used in clinical diagnosis due to its high spatial resolution and high signal-to-noise ratio.Lung tumor area is difficult to identify due to lack of functional information when only CT image is applied in lung tumor diagnosis.PET can reflect the functional information of organs and tissues as a functional imaging technology,however,it is difficult to precisely detect lung tumor region relying solely on PET image due to its low resolution.Liver segmentation based on CT image often suffers from the low contrast between liver and adjacent organs and the liver shape variability between different individuals.This thesis studies the segmentation of lung tumor image and liver image based on the random walk and graph cut to solve those problems.The main work of this study is as follows:(1)This thesis proposes a random walk algorithm with constraint to use the functional information of PET image and the anatomic information of CT image at the same time to solve the problem of CT and PET single-mode image used in lung tumor segmentation such as: lung tumor boundary missing detection and unsatisfied segmentation accuracy.Firstly,the initial contour is obtained by the pre-segmentation of PET using region growth and mathematical morphology.The initial contour can be used to automatically obtain the seed points required for random walk of PET and CT image,at the same time,it can be also used as a constraint in the random walk on CT image.For the reason that CT provides essential details on anatomic structures,the anatomic information of CT can be used to improve the weight of random walk on PET image.Finally,the similarity matrices obtained by random walk on PET and CT image are weighted to obtain an identical result.Experiments have shown that the proposed method has much better performance than other traditional segmentation methods.(2)A novel method for liver segmentation from abdominal CT volumes based on the improved U-Net convolution neural network and graph cut is proposed to solve the problem that the low contrast between liver and adjacent organs and the liver shape variability between different individuals.Firstly,the improved U-Net is trained with the constructed dataset,improved U-Net is used to complete the liver segmentation to obtain the probability distribution map of last-level of network.Secondly,the contextual information of the image sequence and the probability distribution map are used to construct the energy function required for the graph cut algorithm.Finally,the final segmentation is completed by minimizing the energy function.Experiments have shown that the proposed method can achieve much more accurate liver area when faced with fuzzy boundary of liver area caused by a low contrast between the liver area and surrounding organs.