【作者】 陈北京；

【导师】 董光昌；

【作者基本信息】 浙江大学 ， 应用数学， 2006， 硕士

【摘要】 医学影像己成为现代医学的一个重要组成部分,而医学图像处理主要是对已获得的图像进行加工、处理,便于临床诊断。本文主要讨论了两类医学图像处理技术:基础的医学图像对比度增强技术和当前研究热点的医学图像配准技术。 基于灰度变换的医学图像对比度增强技术,是一种简单却比较有效的对比度增强法。在对常见的灰度变换概述之后,针对当前一些算法存在的图像对比度增强和边缘细节保持之间的矛盾,我们提出了两种新的MR图像对比度增强算法:基于阈值分割和3次B样条插值的MR增强算法,首先定义了基于对比度和细节的新的质量评价参数,然后利用Otsu阈值算法选取的多个阈值寻求该质量评价参数最佳的三次B样条插值变换,得到理想的增强效果;基于分割和累积指数变换的MR增强算法,先采用多阈值分割将图像分割为不同的区域,然后统计出各区域的灰度均值和方差,并由它们构造出各区域的累积指数非线性变换作用于各区域进行增强。实验表明,这两种算法都能比较好的解决对比度增强和边缘细节保持两方面的冲突,而且速度也不亚于一些传统的算法。 基于灰度的图像配准技术以其较高的精度、不需要预处理而能实现自动配准被广泛采用。而相似性测度是决定配准准确性、鲁棒性和实时性的最主要因素,因此在介绍完配准技术及其涉及的相关技术后,针对一阶互信息的鲁棒性不强的问题进行了相关分析,并介绍了几种互信息的改进测度,其中特别针对二阶互信息进行了重点讨论,分析了灰阶、相关信息和噪声对其影响,得到了二阶互信息的最佳参数取值,然后分析了分辨率对一阶和二阶互信息的影响,改进了二级分辨率策略的配准技术,在分辨率不同的各个级上采用不同的相似性测度。实验表明,相比采用单一测度的算法,多测度结合的算法在配准精度和速度两方面都能得到更为理想的效果。更多还原

【Abstract】 Medical image has been an important part of modern medical sciences. And, medical image processing focuses on image post-processing in order to improve image quality and facilitate clinical diagnosis. In this thesis, we discuss two classes of technologies on medical image processing, including the basic contrast enhancement technology and registration under intense research.Medical image contrast enhancement based on gray transformation is a simple but effective technology. After the survey of some usual gray transformation, we present two new algorithms for MR, which try to cope with the contradiction between contrast enhancement and edge detail preserving for some traditional algorithms. One is MR image enhancement algorithm based on threshold segmentation and B-Spline interpolation. We first define a new assessment criterion based on luminance contrast and detail, then search the best cubic B-Spline interpolation transformation for ideal enhancement effect by some threhold values chosen by Otsu threshold algorithm. The other is MR image enhancement algorithm based on image segmentation and accumulating index transformation. This method divides the whole image into different regions using multilevel segmentation. For every region, the mean and variance of gray value are calculated to contract the accumulating index transformation for enhancement. Experimental results show that our algorithms can solve the contradition referred above effectively without much speed cost.Voxel intensity based medical image registration has been widely used for its high precision and automatism with no pre-processing. Similarity measure is the most important factor for determining the accuracy, robustness and real-time property of registration. After the description of the registration technology and its related ones, we analyze the robustness of mutual information(MI) and introduce some improved similarity measures of MI. Especially, we discuss the effects of gray level, neighborhood relationship and noise on the second-order information and get its best parameters. Then, after studying the effects of resolution on first-order and second-order information, we improve the multi-resolution registration approach by adopting different similarity measure for different resolution. Experiment results verify that, compared with single measure, multimeasure registration is more perfect not only in precision but also in speed.更多还原