基于三维块匹配的超声纹理评估分割方法

发布时间：2018-04-12 14:02

本文选题：图像分割 + 超声图像　；参考：《西南石油大学》2017年硕士论文

【摘要】：医学成像作为现在医学的重要组成部分在医学理论和临床诊断上都有重要的意义。超声图像由于其无损、廉价、实时运行、安全的特点在众多医学图像中有着重要的地位。与其他医学成像技术相比,超声在临床诊断上的应用有着显著的普及度,所以对超声图像的处理技术的研究有着重要的意义。图像分割作为图像分析和信息医学的关键,一直都受到研究者们的重视。然而由于超声图像在成像精度上很难和CT、MRI技术抗衡,这造成了超声图像分割的一大难点。超声图像特有的散斑纹理增加了超声分割的难度。在超声图像处理领域有着大量的文献研究如何去除散斑。此外,超声散斑也被指出和组织性质有着一定的的关系。如何应用散斑为组织分割提供信息的同时减小散斑对超声图像的影响,是超声分割研究的重点。除此之外,超声散斑纹理在不同的超声图像中有着较大的差异,除开某些统计模型,并没有一个较好的固定的纹理模型对其进行描述的随机性也是超声图像分割的难点之一。本文针对超声散斑纹理的随机性和二义性提出了两种基于非局部思想的算法。二义性是指,散斑影响了超声的成像质量的同时于组织分布有密切的联系。第一,本文针对超声散斑纹理随机性提出了非局部块聚类特征提取算法,解决了超声散斑纹理随机性较强时,难以提取统一特征值的问题。第二,本文针对超声散斑纹理的二义性提出了自适应的特征提取算法,解决了一般去噪预处理时容易将散斑携带的组织特征信息一并去掉的问题。并通过对比实验证明了这两种方法的有效性。最后,本文综合应用这两种方法,提出了基于三维块匹配的超声纹理提取方法,并对超声图像进行了分割。此外,将该方法做出了改进,解决了该方法在超声弱回声处分割不稳定的问题。并通过对比实验对这两种方法进行了有效性分析。
[Abstract]:As an important part of current medicine, medical imaging plays an important role in medical theory and clinical diagnosis.Ultrasonic images play an important role in many medical images because of their nondestructive, inexpensive, real-time operation and safety.Compared with other medical imaging techniques, ultrasound is widely used in clinical diagnosis, so the research of ultrasonic image processing technology is of great significance.As the key of image analysis and information medicine, image segmentation has been paid attention by researchers all the time.However, the imaging accuracy of ultrasonic image is very difficult to compete with CT MRI technology, which leads to a big difficulty in ultrasonic image segmentation.The special speckle texture of ultrasonic image increases the difficulty of ultrasonic segmentation.In the field of ultrasonic image processing, there is a lot of literature on how to remove speckle.In addition, ultrasonic speckles were also noted to have a certain relationship with tissue properties.How to use speckle to provide information for tissue segmentation while reducing the influence of speckle on ultrasonic image is the focus of ultrasonic segmentation.In addition, ultrasonic speckle texture has great differences in different ultrasound images. Apart from some statistical models, it is also one of the difficulties in ultrasonic image segmentation that there is not a good randomness of the description of ultrasonic speckle texture model.In this paper, two algorithms based on nonlocal idea are proposed for randomness and ambiguity of ultrasonic speckle texture.Ambiguity means that speckle affects the imaging quality of ultrasound and is closely related to tissue distribution.Firstly, a nonlocal block clustering feature extraction algorithm is proposed for the randomness of ultrasonic speckle texture, which solves the problem that it is difficult to extract uniform eigenvalues when ultrasonic speckle texture is stochastic.Secondly, an adaptive feature extraction algorithm is proposed for the ambiguity of ultrasonic speckle texture, which solves the problem that the information of tissue features carried by speckle can be removed simultaneously during the general denoising preprocessing.The effectiveness of the two methods is proved by comparative experiments.Finally, this paper proposes an ultrasonic texture extraction method based on 3D block matching, and segments the ultrasonic image by using these two methods.In addition, the method is improved to solve the problem of unstable segmentation at the weak echo of ultrasound.The effectiveness of the two methods is analyzed through comparative experiments.
【学位授予单位】：西南石油大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R445.1;TP391.41

【参考文献】