基于医学图像的关节软骨分布测量及骨自动分割关键技术

发布时间：2018-05-21 02:01

本文选题：多层次自动分割 + 法线方向校正　；参考：《哈尔滨工业大学》2016年博士论文

【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。
[Abstract]:The combination of medical imaging technology and computer technology makes computer medical imaging aided technology play a great role in the diagnosis and treatment of diseases such as osteoarthritis. The MR and CT bone joint medical images with high resolution and high signal-to-noise ratio contain a large number of related information, including complex bone structure, variable bone morphology, and focus However, a large number of bone and joint image information is far from the doctor can handle by artificial means. It is difficult for doctors to conceive of the position of bone and joint lesions, the change of cartilage thickness and the state of adjacent tissues from medical images. Therefore, the modern clinical diagnosis of bone and joint structure detection puts forward high speed and high precision technique. However, the existing bone structure detection algorithms are still unable to achieve the actual clinical application, especially in the detection of bone structures which are closely adjacent to or suffering from severe arthritis. In this paper, the knee joint, hip joint and carpal bone are used as the research object to study the precise joint soft of the MR image. Automatic segmentation and thickness measurement of bone; an automatic segmentation method of joint bone for CT images is studied in order to improve the diagnostic accuracy and therapeutic effect of doctors. This paper proposes a multi-layer three-dimensional image segmentation algorithm based on the B spline DGVF (DGVF-directional gradient vector flow) snake model, which solves the knee and hip joint soft. It is proved that the zero crossing method is not suitable for measuring the narrow gap of the cartilage structure and produces considerable error problems. A new method for the detection of the cartilage boundary and the thickness measurement in the plane of the hip joint based on the error model is proposed, which solves the extremely close to the bone structure (femur). The thickness measurement problem of cartilage and acetabular cartilage); finally, a multi stage automatic segmentation method based on the combination of surface tracking correction and Gauss standard deviation (SD) Sigma was proposed to solve the extremely close bone structure (hip joint and carpal bone) segmentation problem. The main research work and results are as follows: the cartilage of the hip joint (femur cartilage and acetabulum soft) The segmentation of bone, that is, solves the problem of extremely close and blurred cartilage segmentation. In this paper, a multi-layer three-dimensional image segmentation algorithm based on the B spline direction gradient vector flow snake model is proposed. This segmentation method mainly includes three stages: image preprocessing, coarse segmentation and precise segmentation of articular cartilage: the first stage uses nonlinear filtering. Wave method and sinusoidal interpolation algorithm solve the problem of image noise and isotropy; the second stage uses three eigenvalues of Hessian matrix and optimal threshold method to solve the problem of strengthening the articular cartilage, determining the position of cartilage and obtaining the initial contour of the adjacent cartilage edge, and the third stage uses the 3D map of the B spline DGVF snake model. The segmentation algorithm is used to extract the contour of the articular cartilage edge. This automatic segmentation algorithm is less calculated and can effectively segment the articular cartilage. In view of the problem that the most commonly used zero crossing method in the world appears in the measurement of the thickness of the cartilage of the hip joint, the theory simulation method of cartilage model proposed in this paper proves that the zero crossing method is not suitable for measurement. The reason for the large error in measuring the thickness of the thickness of the zero crossing method is pointed out. In this paper, a new method for measuring the thickness of the cartilage and acetabular cartilage in the hip joint is proposed. A new method of measuring the cartilage boundary and measuring the thickness of the hip joint in the plane is proposed. The method, for short, is based on the error model method, which transforms the problem of the measurement of cartilage thickness to the error between the predicted curve and the actual curve observed in the magnetic resonance data. The error of the simulated signal in the normal direction of the arbitrary edge point of the cartilage surface and the normal direction signal of the actual cartilage edge point is minimal. At the same time, the precise edge position and thickness of the cartilage in the plane can be obtained. The thickness measurement method based on the error model can overcome the limitation of the distance between the adjacent thin structures and the inherent spatial resolution of the system to the thickness measurement. At the same time, a new method for correcting the thickness of the three-dimensional soft bone is proposed to correct the slant slice. Due to the large estimation of the thickness of the image plane, the segmentation of the hip and carpal bone is very difficult because of the tight bone structure in the hip and carpal bones. This paper proposes a multi stage automatic cutting method based on the combination of surface tracking correction and Gauss standard deviation, which solves the closely connected bone structure. The segmentation of hip and carpal bone can provide important information for the surgical planning of total joint replacement surgery, navigation and postoperative evaluation. In the process of surface tracking correction, this paper uses the geometric information of the current point to improve the square method of the direction estimation of the following point surface, so that the 3D surface tracking algorithm can be continuously obtained. The information of the continuation point, until the previously discovered point is revisited or some conditions are no longer satisfied. Since the normals are corrected while optimizing the values of the Gauss standard deviation, this method is robust to the segmentation of the joint gap narrowing caused by the noise image and the severe joint degradation. Compared with the method, the method obtained in this paper has higher segmentation accuracy. In this paper, the method of combining surface tracking correction with Gauss standard deviation and the optimal scaling method to obtain surface points in the normal direction correction process greatly improve the effect of bone segmentation.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：R816.8;TP391.41

【相似文献】