基于几何形变模型的CT图像肝脏肿瘤分割

发布时间：2018-05-11 18:02

本文选题：肝脏肿瘤分割 + 几何形变模型　；参考：《山东师范大学》2017年硕士论文

【摘要】：CT图像肝脏肿瘤分割,是肝癌等肝脏疾病计算机辅助检测与诊断技术的基础与关键,具有重要的研究意义和应用价值。目前已得到深入研究,并取得大量研究成果。其中,基于形变模型的分割方法得到了广泛的运用。传统几何形变模型一般适用于对比度比较高的图像,对于像CT图像肝脏肿瘤这种具有灰度不均匀和低对比度特性的图像,分割效果不是很好。针对这一问题,在传统几何形变模型的基础上,提出了一种新的CT图像肝脏肿瘤分割方法。所做的具体研究工作如下:(1)相关理论基础的研究和算法的提出。仔细研究了CT图像肝脏肿瘤的特点,包括灰度特性以及几何特性。阅读了大量有关CT图像分割方法的文献,对这些方法进行了仔细研究、分类总结,最终确定了本文的CT图像肝脏肿瘤分割方法。(2)图像预处理方法选择。CT图像由于获取途径的关系,具有一定的噪声。如果直接对原始CT图像进行分割,结果会不尽人意。所以根据去噪结果,确定了适合CT图像肝脏肿瘤分割的预处理方式。(3)几何形变模型改进思路的确定。为了获得更好的分割结果,首先对预处理之后的图像进行偏差估计和纠正,提高图像质量。然后,基于CT图像肝脏肿瘤的灰度不均匀以及周围组织具有低对比度的特性,提出了一个局部强度聚类属性,来说明图像灰度的不均匀等级。在CT图像肝脏肿瘤每个区域的周围,设定一个局部聚类准则函数,作为分割区域周围组织的核心,给定一个统一的分割标准。依据这一标准设置一个能量函数,其与周围各个区域的能量函数以及代表肝脏肿瘤CT图像灰度不均匀特性的偏向量场有关。最后,通过使能量函数最小化,实现对CT图像感兴趣区域的分割以及偏差估计与纠正。(4)分割后图像优化处理。为了获得更好的分割效果,需要对分割后图像进行优化处理。针对分割后肝脏肿瘤的特点,选择了闭运算的优化方式。(5)实验验证。利用软件开发平台VS2010与Matlab R2010a以及辅助性软件,对算法进行了验证实验,并进行了实验结果的对比和量化分析,证明了算法的可行性和有效性。研究的创新之处是,(1)设定了本地强度聚类准则函数,可以更好地处理局部灰度不均匀的情况。(2)提出了双向几何形变模型能量函数,将演化方向设定为两个方向,缩短了处理时间。研究的不足之处是,对于边界变化比较多的图像,迭代次数相对较多,分割时间没有达到理想状态。同时,算法对于对比度的敏感程度,还有待于增强。
[Abstract]:Ct image segmentation of liver tumors is the basis and key of computer aided detection and diagnosis of liver diseases such as liver cancer. It has important research significance and application value. At present, it has been deeply studied, and a large number of research results have been obtained. Among them, the segmentation method based on deformation model has been widely used. The traditional geometric deformation model is generally suitable for images with high contrast, but the segmentation effect is not very good for the images such as liver tumors in CT images, which have the characteristics of uneven grayscale and low contrast. In order to solve this problem, a new method of liver tumor segmentation in CT images is proposed based on the traditional geometric deformation model. The specific research work is as follows: 1) the theoretical basis and the algorithm. The characteristics of liver tumors in CT images, including grayscale and geometric characteristics, are carefully studied. Has read a lot of literature about CT image segmentation method, has carried on the careful research to these methods, classifies the summary, Finally, it is determined that the method of liver tumor segmentation in this paper, I. E. the preprocessing method of CT image, has some noise due to the way of obtaining it. If the original CT image is segmented directly, the result will be unsatisfactory. Therefore, according to the denoising results, the preprocessing method of liver tumor segmentation in CT image is determined. In order to obtain better segmentation results, the image after preprocessing is first estimated and corrected to improve the image quality. Then, based on the heterogeneity of liver tumors in CT images and the low contrast of surrounding tissues, a clustering attribute of local intensity is proposed to show the uneven grayscale of the images. A local clustering criterion function is set up around each region of liver tumor in CT image as the core of the tissue around the segmentation area and a unified segmentation criterion is given. According to this criterion, an energy function is set up, which is related to the energy function of the surrounding regions and the bias field which represents the heterogeneity of the gray level of the liver tumor CT image. Finally, by minimizing the energy function, the segmentation of the region of interest in CT images and the estimation and correction of the deviation are realized. In order to obtain a better segmentation effect, it is necessary to optimize the image processing after segmentation. According to the characteristics of segmented liver tumor, the closed operation optimization method. By using the software development platform VS2010, Matlab R2010a and auxiliary software, the algorithm is validated and compared with the experimental results. The feasibility and effectiveness of the algorithm are proved. The innovation of the study is that the local intensity clustering criterion function is set up, which can better deal with the local grayscale inhomogeneity. The energy function of the bidirectional geometric deformation model is proposed, and the evolution direction is set in two directions. The processing time is shortened. The disadvantage of the study is that the number of iterations is relatively large and the segmentation time is not up to the ideal state for the images with more boundary changes. At the same time, the sensitivity of the algorithm to contrast still needs to be enhanced.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R735.7;TP391.41

【参考文献】