【作者】 郭皞岩；

【导师】 王大政； 程远志；

【作者基本信息】 哈尔滨工业大学 ， 机械电子工程， 2016， 博士

【摘要】 医学成像技术和计算机技术紧密结合使计算机医学影像辅助技术在骨关节炎等疾病的诊断和治疗等方面发挥巨大作用。高分辨率和高信噪比的MR及CT骨关节医学图像中含有大量图像的相关信息,包括复杂的骨结构、多变的骨形态、病灶位置和厚度等,然而大量的骨关节图像信息远远不是医生可以通过人工手段来处理的,医生很难从医学图像中构想出骨关节病变位置、软骨厚度变化和邻近组织状态,因此现代临床诊断对骨关节结构的检测提出了高速度和高精度的技术要求,然而现有的骨结构的检测算法还无法达到实际临床应用的程度,尤其在针对紧密相邻或者患有严重关节炎的骨结构进行检测时会更加困难。针对上述问题,本文以膝关节、髋关节和腕骨为研究对象,针对MR图像研究精确的关节软骨的自动分割和厚度测量方法;针对CT图像研究精确的关节骨的自动分割方法,以便提高医生的诊断精度和治疗效果。本文提出了基于B样条DGVF(DGVF-directional gradient vector flow)蛇模型的多层次三维图像自动分割算法,解决了膝、髋关节软骨的分割问题;提出了软骨模型理论模拟方法,验证了零交叉方法不适合测量间隙狭窄的软骨结构,会产生相当大的误差问题;提出了新的基于误差模型的平面内的髋关节软骨边界检测和厚度测量的方法,解决了极其接近骨结构(股骨软骨和髋臼软骨)厚度测量问题;最后提出了基于表面跟踪校正与高斯标准差(SD)σ相结合的多阶段自动分割方法,解决了极其接近的骨结构(髋关节和腕骨)的分割问题。本文主要研究工作和成果如下:针对髋关节软骨(股骨软骨和髋臼软骨)的分割,即解决极其接近和边缘模糊的软骨分割问题,本文提出了基于B样条方向梯度矢量流蛇模型的多层次三维图像自动分割算法。该分割方法主要包括图像预处理、关节软骨的粗分割和精确分割三个阶段:第一阶段使用非线性滤波方法和正弦插值算法解决图像噪声和各向同性问题;第二阶段使用Hessian矩阵三个特征值和最优阈值化方法解决强化关节软骨、确定软骨位置并获得紧邻关节软骨边缘的初始轮廓线问题;第三阶段使用基于B样条DGVF蛇模型的三维图像分割算法提取关节软骨边缘轮廓。该自动分割算法计算量少,可以对关节软骨进行有效的分割。针对目前世界上最常用的零交叉方法对髋关节软骨厚度测量出现的问题,本文提出的软骨模型理论模拟方法验证了零交叉方法不适合测量间隙狭窄的软骨结构,指出了零交叉方法产生厚度测量误差大的原因。针对当前广泛应用的零交叉方法对髋关节股骨软骨和髋臼软骨厚度测量存在的误差问题,本文提出了基于模型的平面内的髋关节软骨边界检测和厚度测量的新方法,简称为基于误差模型方法。该模型方法把软骨厚度测量问题转化为对核磁共振数据中所观察到的预测曲线和实际曲线之间的误差问题。当软骨表面任意边缘点法线方向的模拟信号与实际软骨边缘点的法线方向信号的误差最小的时候,就可获得精确的平面内软骨边缘位置和厚度。基于误差模型的厚度测量方法可以克服相邻薄结构间的距离过小和系统固有的空间分辨率对厚度测量的限制。同时本文提出了一个新的三维软骨厚度校正方法,纠正由于倾斜切片引起的对图像平面厚度的过大估计问题。针对髋关节和腕骨的分割,由于髋关节和腕骨内部的紧密骨结构使分割变得十分困难,本文提出了基于表面跟踪校正与高斯标准差相结合的多阶段自动分割方法,解决了这类紧密相连的骨结构(髋关节和腕骨)的分割问题,能够为外科全关节置换手术计划制定、术中导航以及术后评估提供重要信息。在表面跟踪校正过程中,本文使用当前点的几何信息改善后续点表面法线方向估计的方法,使三维表面跟踪算法能够持续获得后续点的信息,直到先前发现的点被重新访问或者某些条件不再满足为止。因为校正法线方向的同时优化了高斯标准差的值,所以本文的方法对噪声图像和关节严重退化造成的关节间隙狭窄的分割具有鲁棒性。通过实验与目前最先进的方法比较,本文的方法获得了更高的分割精度。本文中的表面跟踪校正与高斯标准差相结合的方法,以及在法线方向校正过程中获取表面点的最佳尺度方法大大的提高了骨分割的效果。更多还原

【Abstract】 The medical imaging technology has been closely integrated with the computer technology,which enables the assistive technology of computer and medical imaging to exert a huge role in diagnosis and treatment of the diseases like osteoarthritis.There is a large quantity of image-related information contained in the high-resolution and high-SNR MR as well as CT medical images of bones and joints,including the complicated bone structures,the changeable bone shapes as well as the positions and thickness of lesions etc.However,the large quantity of information on bone and joint images is far less capable for doctors to process manually.As it is very difficult for doctors to conceive the positions of lesions,variation in thickness of cartilages and state of adjacent tissues from the medical images,so the contemporary clinical diagnosis has brought forward the technical requirements of high speed and high precision on the structures of bones and joints.Nevertheless,the existing detection algorithm on bone structures has been unable to achieve the practical clinical application.In particular,it will be even more difficult to detect the bone structures in close vicinity or with severe osteoarthritis(OA).Aiming at the abovementioned problems,the paper,by taking knee joints,hip joints and carpus as research objects,has realized the accurate automatic segmentation and thickness measurement of articular cartilages speaking of MR images,and realized the accurate automatic segmentation of articular cartilages speaking of CT images,which have greatly improved doctors’ diagnosis accuracy and treatment effect.The paper has put forward a multilevel automatic segmentation method of threedimensional images based on B-spline DGVF snake model,which has solved the segmentation problem of knee and hip articular cartilages;it has brought forward a theoretical simulation method of cartilage model,which has verified that the zero-crossing method is unsuitable to measure the cartilage structures with narrow gaps and that it will cause a considerably big deviation;it has come up with a new method based on boundary detection and thickness measurement of hip articular cartilages within the plane of the deviation model,which has solved the thickness measurement problem of extremely adjacent bone structures(femoral cartilages and acetabular cartilages);and eventually,it has proposed a multistage automatic segmentation method based on the combination of surface tracking correction and Gaussian standard deviation SD σ,which has solved the segmentation problem of extremely adjacent bone structures(hip joint and carpus).The primary research work and achievements of the paper are as follows:(1)As for the segmentation of hip articular cartilages(femoral cartilages and acetabular cartilages),namely,solving the problem of extremely adjacent and edge-blurred cartilages,the paper has put forward a multilevel automatic segmentation algorithm of three-dimensional images based on B-spline DGVF(Directional Gradient Vector Flow)snake model.This segmentation method is primarily comprised of image preprocessing,rough segmentation and accurate segmentation of articular cartilages.In the first phase,the non-linear filtering method and the sinc interpolation algorithm have been utilized to solve the problems of image noises and isotropy;in the second phase,the three eigenvalues of Hessian matrix and the method of optimal threshold have been adopted to solve the problems of strengthening articular cartilages,confirming cartilage positions and acquiring the initial contour of closely adjoined articular cartilage edge;in the third phase,a segmentation algorithm of three-dimensional images based on B-spline DGVF snake model has been taken advantage of to extract the edge contour of articular cartilages.This automatic segmentation algorithm has less calculated amount,which can conduct effective segmentation on articular cartilages.(2)With regards to the problems arising from the most frequently used zero-crossing method measuring the thickness of hip articular cartilages,the theoretical simulation method of cartilage model proposed in this paper has verified that the zero-crossing method is not suitable for measuring cartilage structures with narrow gaps and that it will cause a considerably big deviation.The theoretical simulation method of cartilage model has also pointed out the reasons why the zero-crossing method will cause a big deviation in thickness measurement.The theoretical simulation analysis method of cartilage model proposed in this paper is the first proposed new method.(3)As for the problem of deviation existing in the commonly used zero-crossing method at present measuring the thickness of hip articular cartilages and acetabular cartilages,the paper has put forward a new method based on boundary detection and thickness measurement of hip articular cartilages within the two-dimensional plane of the model,which is called the deviation method based on the model for short.This deviation method has translated the problem of cartilage thickness measurement into the problem of the deviation between the prediction curve and the actual curve observed in NMR data.When there is the smallest deviation between the analog signal of an arbitrary edge point on the cartilage surface along the normal direction and the signal of the actual cartilage edge point along the normal direction,the accurate cartilage edge position and thickness within the plane can be obtained.Based on the fact that the thickness measurement method of deviation model can overcome the limitations of the excessively short distance between mutually adjoined thin structures and the system’s inherent spatial resolution to thickness measurement.Meanwhile,the paper has proposed a new three-dimensional cartilage thickness correction method,which rectifies the excessively estimated plane thickness of images caused by inclined sections.Those based on deviation model method and threedimensional cartilage thickness correction method put forward in this paper are the first proposed new methods.(4)In terms of the difficulty while segmenting the hip joint and the carpus due to their internal closely-integrated bone structures,the paper has put forward a multistage automatic segmentation method based on the combination of surface tracking correction and Gaussian standard deviation,which has solved the problem of segmenting this type of closely integrated bone structures(hip joint and carpus)and is able to provide important information for the plan formulation of total articular replacement arthroplasty,the navigation during operation and post-operation assessment.In the process of surface tracking correction,the paper has utilized the estimation method of geometrical information of the current point improving the normal direction of the follow-up point,which enables the three-dimensional surface tracking calculation to continue to acquire information of the follow-up points until the previously discovered points are re-visited or certain conditions are no longer satisfied.As the value of Gaussian standard deviation has been optimized at the same time of correcting normal direction,so the methods in this paper have robustness on the segmentation of joint space narrowing caused by noise images and severe degeneration of joints.Comparing the experiment with the current most advanced method,the methods in this paper have obtained higher segmentation accuracy.The method of combining surface tracking correction and Gaussian standard deviation,and the method of acquiring the optimum scale of surface points in the correction process of normal directions have been proposed for the first time.更多还原