Research On Medical Image Segmentation Method Based On Variational Level Set Active Contour Model

Since the application of medical imaging to clinical medicine,the contents and ways of obtaining information from medical images in different modalities have also changed with the development of digital image processing technology.That is from the simple observing of the tissue structure and anatomical structure to using image processing technologies such as denoising,segmentation,registration,analysis,and reconstruction to locate,segment,analyze and extract features of organs or lesions.Image segmentation is the key in a variety of medical image processing technologies,from the lower level to higher level.The active contour model based on the variational level set method has the advantages of easy handling of topology changes.high calculation accuracy,and strong stability.This paper studies the main problems of medical image segmentation based on variational level set active contour model.The major research contents and productions are shown in the following areas:(1)An active contour model based on multiple descriptors is proposed with attention to address the problem of accuracy segmentation of medical images with intensity inhomogeneity and noise.The description of the local intensity distribution of medical images is studied.Local entropy is introduced on the basis of mean and variance to improve the accuracy of image intensity distribution.The problem of noise robustness is studied.The local entropy determined by multiple pixels is insensitive to a single noise point,so it has a certain filtering effect.The bias field is studied.The bias field factor is introduced to characterize the global intensity inhomogeneity.Algorithm performance is tested on noisy medical images.The experimental results show that the segmentation accuracy is higher than some traditional algorithms,which reflects the effectiveness of the multiple descriptors.And the algorithm completes segmentation while correcting the image bias field so that the visual quality of the image is improved,and at the same time,the segmentation accuracy is improved.(2)A novel active contour model based on image global and local information is proposed with attention to address the initial sensitivity in medical image segmentation.The problem of image global information fitting is studied.For the limitation of the existing global active contour model,the partition entropy is defined and a global model based on partition entropy is designed.The initialization sensitivity problem of the local active contour model is studied.The global information based on the partition entropy is used as a supplement to improve the initial contour sensitivity of the local model.The weight of global information and local information in energy minimization is studied.An adaptive parameter is used.When the intensity inhomogeneity is serious,choose a smaller one.Instead,choose a larger one.In the early stage of evolution,global information dominates,accelerating convergence and preventing initial contour sensitivity.And in the late stage of evolution,local information dominates to ensure segmentation accuracy.The validity and initialization robustness of the algorithm are tested on medical images,artificial images and natural scene images.The experimental results show that the new algorithm has stronger initialization robustness and segmentation accuracy.It accelerates the convergence speed while solving the initial contour sensitivity problem.(3)A fast segmentation algorithm for medical ultrasound images based on parameter level set active contour model is proposed to effectively solve the problem of real-time segmentation of medical ultrasound images.The problem of large computational and time-consuming calculations is studied.A parameterized level set function is designed to overcome the disadvantages of traditional level set functions that require additional regularization term or reinitialization procedure.The problem of high computational complexity of the level set method is studied.Using the parameterized level set function,the topology change is naturally integrated into the curve evolution process without increasing the dimension.The initialization problem of the algorithm is studied.A dense initialization way is designed to automatically acquire complex shapes,detect small region,and accelerate convergence.In addition,the algorithm in this chapter is not restricted by the CFL condition and can choose larger step.The experimental results show that when segmenting medical ultrasound images,the proposed algorithm has faster convergence rate than the classical active contour model and composite algorithm.(4)A brain MRI multiphase segmentation algorithm based on variational level set method is proposed to effectively solve the multiphase segmentation problem of brain issues.The multiphase expression of brain MRI is studied.The member functions are designed to represent the white matter,gray matter,cerebrospinal fluid,and background four regions to ensure multiphase segmentation.The segmentation accuracy and robustness to noise of the algorithm are studied.A multi-angle local intensity description method is designed.The problem of image visual effects is studied.The bias factor is introduced in the energy functional to correct the image bias field.Experimental results show that the new algorithm successfully realize multiphase segmentation of brain MRI.The segmentation accuracy and robustness to noise are superior to the traditional methods.The algorithm satisfies four clinically necessary conditions for multiphase segmentation,strong robustness to noise,overcoming intensity inhomogeneity to ensure high accuracy,and implementing bias field estimation.