【作者】 田沄；

【导师】 郝重阳；

【作者基本信息】 西北工业大学 ， 信息与通信工程， 2007， 博士

【摘要】 医学体数据场可视化是科学可视化与生物医学信息工程领域中的热点技术，近年来发展非常迅速。人体组织器官的准确提取、实时的人机交互和高质量的绘制是医学体数据场可视化的关键技术。人体组织器官的准确提取是研究解剖结构、组织器官定量化测定、病灶定位和疾病诊断的重要基础：实时的交互性是医学可视化技术在临床诊断和治疗中近一步推广应用的必要条件；高质量的绘制图像是处理海量医学体数据场的基本要求。本文深入研究了上述三种／类关键技术的具体内容，包括医学体数据场的获取与处理、医学体数据场层间插值、基于模糊理论和粗糙集理论的医学体数据场增强、基于体绘制的人体感兴趣组织器官的提取及医学体数据场绘制等技术。同时研究了以上内容在医疗辅助诊断中的应用及其实现技术。论文的创新性成果有：(1)深入研究了Cubic卷积插值方法。系统推理了多种Cubic卷积插值方法的联系与区别。在此基础上，提出了一种基于最优Cubic卷积核的医学体数据场层间插值方法，该方法充分结合原始断层图像的局部特征，利用迭代运算确定出最优的形状控制参数；然后利用该参数下的Cubic卷积核对体数据场进行一次性插值，避免了传统Cubic卷积插值方法中误差的传递；从而有效提高了插值的精度。(2)提出了一种基于像素分类的医学体数据场层间插值方法。首先对插值图像像素进行分类，然后对不同类别的像素采用不同方法进行插值。该方法能够有效克服插值图像边界信息的丢失，尤其适用于层间距较大的断层图像插值。同时，提出了一种基于物体边缘信息的医学体数据场层间插值方法。首先运用基于形状轮廓插值方法确定出中间插值图像物体的边缘，然后对插值图像边缘采用高精度的基于对应点匹配的方法进行插值。实验结果表明该方法不仅有效提高了插值的精度，同时兼顾了插值的效率。(3)改进了一种基于双线性变换的医学体数据场快速模糊增强算法。由于采用的隶属函数和模糊增强算子均为线性的，不仅能够提高增强的速度，而且也有效避免体数据场细节信息的丢失。在此基础上，结合粗糙集理论，提出了一种基于粗糙模糊集理论的医学体数据场模糊增强算法。首先应用粗糙集对体数据场进行分类，在分类的同时，去除了噪声。然后针对分类后的子数据场，采用不同方法进行模糊增强，使得在三维人体组织器官的显示过程中，能够有效避免背景部分的“遮挡”。(4)改进了一种窗口调节函数，并在此基础上提出了一种人体组织器官动态提取方法，该方法通过调节各参数，能够有效提取出人体的多种组织器官。(5)提出了一种适合医学体数据场的体绘制算法。该算法不仅克服了传统Ray Casting算法绘制模糊的缺陷，而且生成的图像更加符合人的视觉特征。利用上述技术，设计并实现了一个医学体数据场可视化医疗辅助诊断实验系统——VolMTDSys。该系统是一种基于微机的全视野、全时空视觉导航型医疗辅助诊断系统，比传统的医学平面影像辅助诊断系统有更好的立体直观可视性；比一般视觉漫游系统有更好的漫游实时性和身临其境的沉浸感。更多还原

【Abstract】 Volume visualization in medicine is an advanced topic in the field of Scientific Visualization and the Biomedical Information Engineering. This technology has rapidly developed greatly in recent years, and it has widely been applied to the clinical diagnosis. Accurate extraction of the tissues or organs, real-time interactive reaction of the manipulation and high quality of the rendering image are the key techologys of the whole visualization of the medical volume data field. Accurate extraction of the tissues or organs is the important basis of the study of anatomical structure, quantification of tissue volumes, location of pathology, diagnosis, treatment planning, and computer aided surgery, etc; Real-time interactive reaction of the manipulation is the requirement of the further extend application of medical visualization in the clinical diagnosis and treatment; Due to the increasing amount of medical image data, it is necessary to visualize the date with high quality image rendering.The thesis deeply studies the key techniques mentioned above, including obtaining and preprocessing of the medical data, interpolation between slices, contrast enhancement based on the fuzzy sets and rough sets theories for the medical field, extraction of the tissues or organs of interest based on the volume rendering and the rendering techniques for the medical volume data. In addition, some research on the application and realization of these technologies in the medical assistant diagnosis and treatment are also involved in this dissertation.The main contents and contributions of this dissertation are as follows:(1) The Cubic convolution interpolation is deeply investigated. The interpolation principle of Cubic convolution is formulated by the numbers, including the differences and relations among the Cubic interpolation methods with the dissimilar sharp control parameters. Based on these, an interpolation method for the medical volume data field based on the optimal Cubic kemel is presented. The method makes full use of the local characteristic of the original cross-sections, and then the optimal parameter is determined by iterative operation. Finally, the volume data field is interpolated by the Cubic convolution with the optimal parameter in one operation, which avoids the error transfer of the conventional methods. So the interpolationprecision is improved effectively.(2) A method of the interpolation between slices based on the pixel classification is presented. The method classes the pixels into two groups, and then the pixels of the different groups are evaluated by the different ways, which can effectively avoid the loss of the edge information. This method is especially effective for the volume data with large distance between slices. In addition, a method for interpolation of the volume data field is presented; it is based on edge information. Firstly, the boundary of the interpolated image is determined using weighted mean of the profile, and then the value of the boundary pixels is evaluated by the best matching corresponding points. The experimental results show that, the approach not only improves the interpolation precision as well as efficiency.(3) A fast algorithm is improved for the fuzzy contrast enhancement, and it is based on the dual-linear transform. Due to the linear subjectional function and the linear fuzzy enhancement operator, the algorithm not only speed the process, but also voids the losses of particular information. Based on these, an algorithm combining the rough sets is presented for the contrast enhancement of volume data field. In this algorithm, the volume data is classified using the rough sets theory, and the noise is removed simultaneously, then the classified sub-volume data is enhanced in different ways. The algorithm can effectively avoid the cover of background when the interested tissues and organs are displayed.(4) A windows function is improved. Based on the function, an approach based on volume rendering is developed for extracting the VOI (Volume of Interest) from a medical dataset. The most tissues and organs of human can be extracted by adjusting the parameters.(5) A volume rendering algorithm which suits the medical volume data field is presented. The algorithm not only solves the problem of image blur of the classical Ray Casting algorithm, but also the displayed image accords with the vision characteristic of human perfectly.Taking advantage of the above-mentioned technologies, an experimental system named VolMTDSys is designed and realized to assist diagnosis and assist doctors. The system is that kind of interactive assistant diagnosis system which has a new full filed vision and full space vision navigation. In comparison with traditional image aided diagnosing system, the new system has the comparative advantages of high real-time, strong feeling of immersion and good visibility.更多还原