Research On Segmentation Method Of Medical Images Based On Active Contour Model

Image segmentation is a fundamental work in the field of image processing and computer vision,and its main purpose is to extract the region of interest from the image for analysis and recognition.Medical image segmentation is to extract the boundary of specific tissue or lesion from the medical image collected by special medical devices,which is used to assist doctors in diagnosing diseases,formulating a treatment plan,and operating the navigation during surgery.The imaging principle of the device,the setting of parameters,and the movement of human organs during the imaging process can all affect the quality of the collected medical images.Therefore,for the same anatomical structure,medical images of different modalities will be different.Even at the same modality,the images acquired at different times may differ.Moreover,the problems of intensity inhomogeneity,artifacts,noise,etc.in the medical image tend to cause the target boundary to be blurred,which adds difficulty to the segmentation of local target of the medical image.The active contour model is an image segmentation method based on curve deformation.The evolution of the closed curve is controlled by a speed function derived by minimizing the energy function.In this dissertation,the image segmentation method based on active contour model is mainly studied and the proposed models are applied to the object segmentation of inhomogeneous medical images.The main work can be summarized as follows.(1)Research on active contour model based on local signed pressure functionAiming at the problem of selective segmentation of inhomogeneous images,an improved segmentation model based on local signed pressure function is proposed.The local term of the speed function is designed using the local signed pressure function of the pixels on the evolution curve,which is weighted by the homogeneity coefficient of the local neighborhood.When the curve is located in the local gray uniform region of the image,the local signed pressure function is 0,and then the local item of speed function is 0.In order to prevent the curve from stopping evolution at this time,an adaptive weighted global term is introduced into the speed function to drive the curve evolution.Experimental results on inhomogeneous synthetic and medical images demonstrate that the proposed model can adaptively achieve global or local object segmentation according to the initial contour.(2)Research on active contour model based on saliency map and width-variable narrowbandWhen using the active contour model for image segmentation,a closed curve needs to be given as the initial contour.The position and size of the initial contour will affect the segmentation result.In order to reduce the subjectivity of artificially given initial contour,saliency detection and adaptive threshold method are used to extract the initial contour close to the object.Then,a narrowband with variable width is constructed using local means and variances of the image,and a simplified speed function based on local information is designed to control the curve evolution in the narrowband.Experiments show that the active contour model based on saliency map and width-variable narrowband can better achieve automatic segmentation of the image and improve the operation efficiency to some extent.Moreover,the proposed model can achieve local segmentation by adding a bounding box to define the extraction range of the initial contour.(3)Research on active contour model based on average fuzzy energy for local segmentationIn order to achieve robust local segmentation,the fuzzy membership function is introduced into the energy function of the active contour model,and an average fuzzy energy function is designed.A contrast constraint is introduced in the evolution process of the curve to help judge whether the pixels on the curve reach the edge.Experimental results on synthetic images and medical images show that the active contour model based on average fuzzy energy has strong local segmentation ability,and the contrast constraint can effectively prevent the curve from stopping evolution due to falling into local minimum.The average fuzzy energy function can be divided into two forms:weighted average and arithmetic average functions.By combining weighted average and arithmetic average fuzzy energy function linearly and adopting adaptive contrast constraint threshold,the generalized active contour segmentation model based on average fuzzy energy can be obtained.Experimental results show that for medical images with complex background and intensity inhomogeneity,the introduction of arithmetic average energy term increases the instability of the segmentation model,and there is no improvement in segmentation accuracy,but the local segmentation ability is improved,which extends the application range of the active contour model only based on weighted average fuzzy energy.(4)Research on active contour model based on graph cuts and local statistical informationIn order to improve the running efficiency of the active contour segmentation model,an improved active contour model based on graph cuts is proposed,which uses graph cuts instead of gradient descent flow to minimize the energy function.The global optimal solution can be obtained and the running speed is faster.In the iterative process,a dynamic narrowband is constructed according to the evolution curve and the preset outer boundary,and the pixels in the narrowband are mapped to the nodes of the network graph.The gray means and Kullback-Leibler distances of local neighborhoods are used to assign weights to n-links and t-links of the graph,and then the maximum flow/minimum cut algorithm is used to cut the graph to achieve image segmentation in the narrowband.Experimental results on synthetic and medical images with intensity inhomogeneity show that the proposed model can make full use of local features of the image and achieve local segmentation more efficiently.The innovations of this dissertation are mainly reflected in the following aspects.(1)In order to reduce the subjectivity of artificially given initial contour,an improved automatic extraction method of initial contour based on saliency map and threshold is proposed.And the design of width-variable narrowband and simplified speed function effectively improves the efficiency of the segmentation model.(2)Introducing the fuzzy membership function into the energy function of the active contour model,an active contour local segmentation model based on average fuzzy energy function and contrast constraint is proposed,and an adaptive contrast constraint threshold is defined,which can effectively prevent the curve from falling into local minimum and improve the robustness of the segmentation model.(3)In order to improve the running efficiency,an improved active contour local segmentation model based on graph cuts is proposed.The statistical information of local regions is used to weight the edges of narrowband network graph,which is more conducive to extract the feature of inhomogeneous images and achieve efficient and accurate local segmentation.