【作者】 王瑜；

【导师】 郑津津；

【作者基本信息】 中国科学技术大学 ， 精密仪器及机械， 2009， 博士

【摘要】 医学图像数据(如计算机断层扫描(CT)数据及核磁共振图像(MRI)数据)的可视化技术经过二十多年的发展如今已经作为医学图像处理中研究的热点问题之一。三维重建是医学图像可视化的最直观的方法。要实现医学图像的三维重建,首先就要对图像数据进行处理,以便从中提取出感兴趣的器官、组织或病变体的外轮廓,从而实现这些被提取出的器官、组织或病变体的三维重建,重建出模型,达到辅助治疗、手术规划、教学模型和假肢设计等目的。在医疗诊断和外科手术策划中,应用逆向工程技术生成病人组织的三维数字模型,进而利用快速成型技术制作手术模型,医生在此基础上进行诊断及手术,能有效地提高诊断和手术水平,缩短时间,节省费用,提高手术的安全性。本文的核心研究问题是断层医学图像的三维重建。本文首先回顾了逆向工程和医学图像的三维重建的研究背景,并对断层数据的采集方法做了说明。介绍了数字图像处理、医学图像重建和CAD模型的转换方法和研究现状。本文以此为背景,主要对断层医学图像上的轮廓线的提取和曲面的几何重建及其相关技术进行了研究和实现,主要包括断层图像的图像增强和降噪处理,边缘检测和轮廓线提取,曲面重建算法和三维表面模型转换,以及曲面重建过程中的轮廓匹配和分支问题的解决方法。针对医学图像对比度低、噪声较大、灰度缓慢变化的特点,本文首先利用灰度均衡算法进行预处理,以提高图像的对比度,然后利用中值滤波保持边缘信息的平滑效果和小波变换良好的局部化特性和其固有的多尺度特性,通过中值滤波和小波变换结合的方法对图像进行增强处理。实验表明该算法取得了较好的效果,而且可以处理低质量或边缘模糊的医学图像。边缘轮廓提取是三维重建的一个主要的前期处理过程。利用Canny算子和边界跟踪在提取单像素轮廓方面有较好的效果,具体针对医学图像边缘轮廓提取的特点,提出了一种利用迭代的改进Canny边缘检测和边界跟踪来提取图像的轮廓线的方法,提取出单像素和连续封闭的边缘轮廓。文中第四章介绍了文中曲面重建的理论基础——B样条曲线曲面的知识。介绍了曲线曲面的定义、性质、正算和反算方法。文中接着对轮廓线表面重建方法——基于B样条曲面的表面重建进行了研究。首先分析了型值点获取中的主要关键技术,包括特征点的选取和采样点的确立。依据曲率特征首先提取各层特征点,对其重采样使每行(列)获得统一的采样点数。再对采样点插值得到非均匀双三次B样条曲面。最后,在一定控制精度下对曲面依据距离特征进行节点插入,通过最小二乘逼近法算出新的控制顶点,从而得到误差在容许范围内的逼近曲面。根据断层轮廓的特点,本算法综合运用了周期B样条和非周期B样条,讨论了封闭曲面和非封闭曲面的计算方法。另外插值和逼近的结合应用使该算法考虑到了曲面重建的精度,减少了计算量。除此之外,文中还对重建表面模型向CAD实体模型的转换过程进行了研究。文中接着研究了复杂曲面重建中遇到的两类关键技术问题——轮廓匹配和分支问题。文中给出了多轮廓曲面重建和含有复杂嵌套问题的多轮廓曲面重建时的匹配方法:通过建立轮廓树,确定轮廓的内外属性以及轮廓之间的相互嵌套关系,并缩小轮廓匹配的搜索范围;在轮廓树的基础上,利用阈值半径寻找轮廓,确定相邻切片上轮廓的对应关系。另外本文对分支问题可能遇到的情况和解决方法进行了研究。利用加权补分法,进行母轮廓分割,并结合中间层的插入,解决分支问题,从而进行曲面重构。更多还原

【Abstract】 At present,image data visualization,especially visualization of medical image (such as computed tomography(CT) and magnetic resonance imaging(MRI)),has become one of the hotspots of image processing research after more than two decades development.Three-dimensional(3D) reconstruction is the most intuitive method of the medical image visualization.In order to achieve 3D reconstruction of the medical images,we must process images first.So we can detect the contours of the interested templates,organize and pathological area easily.Then,we can achieve the 3D reconstruction of the extracted organs,tissues or body lesions,and get the purpose of adjuvant therapy,surgical planning,teaching model and prosthetic design. In medical diagosis and surgical planning,3D figure model of the sick organization is generated by the application of reverse engineering.Further,operation model is producted by the use of rapid prototyping technology.On the basis of the model, doctor achieves diagosis and surgery,which effectively improve the level of diagnosis and surgery,shorten the time,save the cost,and improve the safety of surgery.The center issue of this thesis is the 3D reconstruction of medical slice-images.In this paper,we begin with an introduction to reverse engineering and 3D reconstruction technique of medical images,and we introduce the collection methods of slice-images data.Then,we introduce the methods and research status of digital image processing,medical image reconstruction and CAD models conversion.We mainly research and implement the methods and related technologies on the extraction of contour lines of medical slice-images and geometric surface reconstruction,which includes image enhancement and noise reduction processing of slice-images,edge detection and contour extraction,surface reconstruction algorithm,and the conversion of 3D model surface.According to the low contrast,noise and slow changes of medical images,a gray balance algorithm is used for pre-processing,that can improve image contrast. Then,combining median filtering with wavelet transform to enhance and denoise image.The results show that the algorithm achieves good effects,and can deals with low quality or marginal vague images.Contour extraction is one of the main pretreatment process of the 3D reconstruction.Using of improved Canny edge detection and boundary tracking,the paper proposed a image contours extracted method,thus propose a single pixel and closed contours.In Chapter 4 of the paper,we introduce the basical theorey of surface reconstruction-the knowledge of B-spline curves and surfaces,and the definition, properties,calculation and backcalculation methods of curves and surfaces.Then, the surface reconstruction method-based on the B-spline surface reconstruction were studied.Fist,the paper analyse the key technologies of data points accessed, including the selection of feature points and the establishment of sampling points. Besides,we extracted feature points based on curvature feature,and resampled feature points in order to get a unification sampling points in each line(column). Then,we interpolated sampling points to get a bi-cubic non-uniform B-spline surface.Finally,we inserted nodes on the surface based on distance feature under the certain control accuracy,and calculated the new control points through the least-squares approximation method,to get approximate surface within the permissible range error.Based on the characteristics of slice contours,we integratively used B-spline cycle and non-cycle B-spline,and discussed how to calculate closed and non-closed surface.In addition,the combination of interpolation and approximation reduces the computational complexity.Finally,the paper studied the conversion process from the surface model to the CAD model.Then,the paper studies the two key technical issues of surface reconstruction—matching and branch.First,the paper gives the methods of many surfaces reconstruction and nested surfaces reconstruction:Through the establishment of tree contour,we recognize the in-out features of contours and the outline of the relationship between the contours,and reduce the searching range.Based on tree contour,we find marching contour by use of the threshold radius to determine the correspondence contour on the adjacent sections.Second,the possible problem of branch was studied.Using weight segmentation method,combined with interpolation,we separated the mother contour to solve the branch problem and reconstruct surface.更多还原