【作者】 宋卫卫；

【导师】 欧宗瑛；

【作者基本信息】 大连理工大学 ， 机械设计及理论， 2008， 博士

【摘要】 股骨头是人体髋关节解剖结构中最重要的组成部分之一,股骨头坏死是一种对人体健康危害较为严重的疑难疾病,高发人群多为青壮年,且发病率较高,特别是近年来发病率呈明显上升趋势。因此对股骨头坏死疾病的治疗和研究具有非常重要的意义,其治疗方法也一直是骨科的研究热点。本文针对股骨头坏死修复手术和假体设计的需要,系统研究了人体关节和骨骼的精确建模问题,并对其中涉及到的CT图像预处理、去噪、增强、骨骼分割、缺损骨骼模型的修复与重建以及由CT图像反求建模的理论与方法等相关内容进行研究,主要研究内容和贡献如下:(1)对医学图像的去噪和边缘信息增强算法进行研究,提出了一种基于Contourlet变换的具有边缘信息保护和增强能力的去噪算法。医学图像的去噪和边缘信息增强,对于后续的图像分析与建模具有非常重要的意义。Contourlet变换是继小波变换后发展起来的、在离散域构造的真正的2D数字图像变换方法,具有小波变换所不具备的多方向性和各向异性,能够从多个方向上捕获图像信息中的几何特征,因而比小波变换具有更好的揭示边缘轮廓特征的能力。但一般的Contourlet变换去噪法仍存在图像部分边缘细节弱化或丢失等问题,对噪声严重的图像去噪后还会出现伪影、失真等现象,影响去噪后的图像质量。本文针对以上问题,根据Contourlet变换多方向性和各向异性的特点,利用一种新的变换系数分类方法,通过自适应阈值选取,自动判断噪声、强边缘和弱边缘所对应的变换系数,并对不同性质的系数分别处理,较好地解决了噪声去除、弱边界增强和强边界保护等问题;同时利用递归的循环平移法,使图像在预设位移扰动量下循环平移,从而有效地消除了由于Contourlet变换缺乏平移不变性造成的图像失真和伪影等现象。最终实现了一种具有边缘保护和增强能力的Contourlet变换去噪法。与其它去噪算法相比,本文算法不仅具有较强的去噪能力,而且具有保护和增强图像边缘信息的特性,对信噪比较低的医学图像也能达到良好的去噪和增强效果,提高图像信噪比的同时,也改善了图像的视觉效果。(2)精确的骨骼分割是骨骼建模和诊断的基础,但由于关节病变、骨骼缺损或噪声等因素的影响,从病人的关节CT图像中自动分割骨骼是非常困难的。目前最可靠的分割方法是由具有人体解剖结构知识的医学专业人员手工分割,但手工分割存在耗时、烦琐等问题。本文将手工分割所依据的人体骨骼解剖结构知识应用于关节骨骼的自动分割中,提出一种基于先验知识模型引导的骨骼轮廓自动提取方法。在建立正常人体髋关节骨骼样本集的基础上,构建包含人体髋关节骨骼解剖结构知识的点分布模型,研究了构建先验模型中点的自动选取和对应及主成分参数选定等关键问题,以此为基础实现了面向髋关节骨骼分割的活动形状模型的搜索匹配。对于存在严重缺损或畸变的股骨,则在上述搜索匹配的基础上,应用Snake的自由形变能力实现对缺损轮廓的精确定位。由于先验模型综合了专业人员的解剖结构知识,因而有利于解决复杂结构的不确定性,能增强对噪声和数据残缺的容忍度,较好地解决了髋关节CT图像中髋臼和股骨头的自动分割问题,对解决其它关节中骨骼间的分割问题同样适用。(3)精确的股骨头模型是指导股骨头坏死修复重建手术的关键,为得到缺损股骨头的修复模型,将统计形状模型扩展到三维空间,提出一种基于三维分割的、具有形状预测能力的缺损器官的修复重建方法。该方法首先根据解剖结构知识建立健康股骨近端的三维统计形状模型,然后将模型作用于股骨头发生缺损的股骨模型上,最后在该先验模型的引导下,通过搜索、匹配、变形等操作,得到缺损股骨头的修复模型,为股骨头修复重建手术提供科学的模型支持。该方法也可推广到其它应用领域,对其它缺损器官的修复重建或整形美容等领域的手术模拟或规划具有一定参考价值。(4)对点云数据的规范化和精简方法进行研究。由于离散点云数据间没有显式的几何拓扑关系,点的邻域搜索比较困难,不利于后续处理中对数据点的遍历和邻域搜索操作;并且由于点云中包含大量的噪声和冗余点,不仅占用大量的磁盘空间、影响计算速度,而且对三角剖分、曲面拟合等后续环节的计算精度也有很大影响,因此需要对点云进行有效的规则化表示和数据精简。为此,本文将用于实体模型表达的八叉树结构引入到点云数据的处理中,提出一种基于新型八叉树编码方式的点云数据规则化表示和精简方法,在保证点云几何特征和测量精度的前提下,有效地精简了点云数据量,并且由于八叉树结构的使用,使点的遍历和邻域搜索效率也得到很大提高。(5)对基于点云数据的曲面拟合重建法进行研究,针对人工假体个性化设计与制造的需要,提出一种基于CT图像反求技术的髋关节建模和缺损股骨头修复建模的理论方法及实现的系统框架。将已提取的序列骨骼轮廓转化为三维离散点云数据后,利用本文提出的八叉树结构和曲率精简方法对其进行结构化表示和数据精简,最后根据骨骼的表面形态特征,利用最小二乘法拟合生成其曲面模型。实验表明,本文方法不仅能够建立正常人体髋关节曲面模型,而且能够直接利用股骨头坏死病人的CT数据进行修复重建,恢复其完整的髋关节表面形态,建立模型的同时自动测量其主要形态结构参数。建立的模型还可方便地转化为STL文件,建立与快速原型系统的集成,还可以在曲面模型的基础上进一步建立CAD实体模型和有限元模型,为指导人工假体的个体化设计与快速制造及生物力学有限元分析提供模型支持。更多还原

【Abstract】 To meet the need of modeling for rehabilitation of femoral head necrosis and manufacture of artificial joint,modeling from CT images of hip joint is systematically studied in this dissertation.Some key technologies related to the research of this dissertation,such as medical image denoising,image enhancement,bone segmentation,the methods of modeling for rehabilitation of collapsing bone and the theory of modeling in reverse engineering etc.are studied and some novel methods have been presented too.The main contents and contributions are as follows:(1) A novel multi-scale denoising-enhancing mehod for CT image based on the Contourlet transform is presented.Contourlet is a new "true" two-dimensional representation for images that can capture the intrinsic geometrical structure of image information. Contourlet transform has better performance in representing edges than wavelets for its anisotropy and directionality,and is therefore well-suited for multi-scale edge enhancement. By analyzing the drawbacks of conventional method of contourlet transform,we present a new method named Enhancing Edge Contourlet Transform(EECT) denoising,which improves the denoising ability of ordinary contourlet transform method in automatic selecting thresholding value,eliminating the effects of moving,enhancing the detail information of image contour and improving resolution,etc.Compared with other denoising methods based on the Wavelet transform and conventional contourlet transform,experiment results show that the proposed algorithm has more prominent effects of denoising and enhancing the weak edges than above methods.The application of the algorithm on CT images of patients’ hip joint shows that images have an encouraging improvement,and might be helpful for segmentation.(2) Accurate segmentation is necessary for followed modeling and diagnosing,ect proceeddings.However,it is very difficult to automatically segment bone precisely, especially at joints,due to injuries,bone’s inhomogeneous structure,and the limitation and resolution of CT images.Under this circumstance,we present a new method that well-suited for bone segmentation based on ASM,which have already incorporated the statistic attributes of femur bone.In this dissertation,a complete solving scheme for building ASM is described in detail.The problem of automatic selecting landmarks is solved by curve simplification.The problem of point correspondence is solved by shape comparability.Then,in order to further upgrade the the segmentation quality,a neighbor gray statistical feature of landmark and Snake model are introduced and a new searching scheme is presented,which can improve the performance of both ASM and Snake.(3) Accurate model of femoral head is the key to surgical simulating,pre-operative planning and surgical navigation of osteonecrosis of femoral head.In order to reconstruct complete model from partial information extracted from hip joint CT images,a modeling for rehabilitation approach based on 3D segmentation and shape prediction using a 3D statistical shape model is presented.The method is the 3D extension of 2D ASM described in the chapter 4.Under the guidance of prior knowledge model,the patient-specific femur bone with the complete femoral head can be constructed from the patient’s own hip joint CT images.(4) The theory and methods of reconstruction of surface based on point cloud data is also studied in this dissertation.On the one hand,the point cloud data usually contains large numbers of noise and redundancy,which affects the calculating precision of triangulation, surfacing fitting,and calculating speed ect.On the other hand,the point in point cloud data has no obvious topology relationship,which will affect the searching efficiency in followed reconstruction.Therefore,a new octree coding method is introduced and applied to express and compress the point cloud data and a curvaturer rule based reduction scheme is put to use then.The proposed method not only reduces the amount of point,but also regularizes the point cloud data,which makes it easy to traversing and searching for point data.(5) The accurate model of femoral head plays an important role in a total hip replacement. An innovative 3D modeling method for the hip joint and modeling for rehabilitation of femoral head based on point cloud data is proposed.After being reduced,the part point data of femoral head and acetabulum were extracted respectively and fitted by the way of the least square.Experiments show that the new method can reconstruct hip joint model and femoral head of patients perfectly.It can be applied to other joints or bone tissues too.The method affords theoretical model for accurate operation position fixing in orthopedics clinic and biomechanics analyses,and it also provides more effective ways for the individually manufacturing of artificial hip joint.更多还原