Research On Medical Image Segmentation Based On Deep-Learning And Level Set

With the rapid development of medical imaging technology,computer aided diagnosis system has become an important auxiliary tool to help doctors do their judgement.Medical image segmentation is the key technology of computer aided diagnosis system design.Its purpose is to has some special meaning part of medical image segmentation,and extract the feature parameters,To provide a reliable basis for clinical diagnosis and treatment and pathological research,and assist doctors to make a more accurate diagnosis.However,due to the complexity of medical images,organs and complex problems such as low contrast and weak edge,adjacent organs,although at present with the development of convolution neural network in image processing,deep learning has been used in medical image segmentation,as well as the traditional image segmentation method also has contribution to the medical image segmentation.However,medical image segmentation is still has a considerable challenge.In this paper,the method of medical image segmentation in deep learning,the basic theory and calculation of active contour model and other knowledge are introduced.Through the combing of the above knowledge points,it is found that the deep learning segmentation method lays more emphasis on the extraction of advanced features and pays attention to semantic information.It can identify the category and location of the relevant organs better,but it cannot segment the edges of medical images well.On the contrary,the segmentation method based on active contour focuses more on the extraction of low-level features,which can make better use of the local information of the image(such as boundary,position,etc.),but is easily limited by the selection of the initial contour.The main contributions of this paper are as follows: Firstly,for deep learning focus on high-level semantic information extraction and the advantages of level set method emphasizing the division of the edge detail,the advantages of the two proposed a fusion method,namely using the depth study of target detection in the abdominal CT image in the liver area from a host of organs and tissues,and to locate it.The location information is used as the initial contour of the level set method,and then the organs are segmented by the improved active contour model.Secondly,In order to solve the problem of lack of positioning information data set in abdominal CT image,the Bounding Box was used to generate and calculate the maximum inner junction of liver image,and the Label Img was used to label it.In view of the noise generated in abdominal CT image and the problems of weak edge and low contrast between liver and adjacent organs and tissues,the image format transformation,image smoothing and image enhancement are carried out.The experiments show that the image preprocessing has good effect on the following image segmentation.Finally,the edge indicator function of DRLSE model is changed.