【作者】 黄晓阳；

【导师】 王博亮；

【作者基本信息】 厦门大学 ， 通信与信息系统， 2009， 博士

【摘要】 肝脏是人体重要的消化器官,肝脏疾病直接影响人的健康和生存。随着医学、计算机技术及生物医学工程技术的发展,医学影像学为临床诊断提供了多种模式的医学图像,这些图像在医学诊断和治疗中的作用显得非常重要。但是单一模式的图像往往不能提供医生所需要的足够的诊断信息。在实际临床应用中,往往需要将不同模式的医学图像有机地结合起来,得到更丰富的信息以便了解病变组织或器官的综合信息,从而做出准确的诊断或制订出合适的治疗方案,这就首先寻找一种(或一系列)空间变换,使两幅图像的对应点达到空间位置和解剖结构上的完全一致即进行图像配准。配准结果应使两幅图像上所有的解剖点,或至少是所有具有诊断意义的点都达到匹配。由于在体肝脏图像是患者在临床检查中获取的,其中包含了患者的个体特征信息,对其进行处理分析对临床诊断具有更加直接的意义。本文以福建省卫生教育联合攻关计划项目“数字化虚拟肝脏及手术计划系统”课题为依托,以肝脏图像配准为背景,系统研究了在体肝脏图像的配准方法,研究实现了肝脏CT多相期扫描图像配准方法,肝脏CT与MR图像配准方法,肝脏CT与DSA图像配准方法,以及肝脏管道系统分支变异与走行分析,为进一步的研究和应用奠定基础。本文研究的创新点体现在以下五个方面:1.提出了一种基于联合直方图的刚性配准算法,该算法分别对刚性配准过程中的四个关键步骤进行了改进,实现了肝脏CT多相期扫描图像刚性配准。分别对图像变换方法、图像插值方法、配准的相似函数以及配准的优化搜索方法是刚性配准过程中四个关键步骤,提出了不同的改进优化算法:1)提出一种图像刚性变换的快速坐标变换方法,大大提高了图像刚性变换的速度;2)提出一种改进的B样条插值方法,改进了B样条插值函数的缺陷,提高了图像插值的效果;3)提出了一种改进的基于联合直方图的计数函数作为配准的相似函数,大大提高了图像配准的速度;4)提出了一种混合优化算法,将PSO优化算法与Powell优化算法相结合,克服了Powell算法的局部最优的缺陷。实现肝脏CT多相期扫描图像刚性配准。2.提出了一种具有自动设定标记点的薄板样条配准方法,该方法能够自动设定图像中的标记点,同时改进了薄板样条配准中存在的空值间隙的问题,实现了肝脏CT多相期扫描图像非刚性配准。标记点的选取及插值函数的确定是非刚性配准方法中最关键的两步。本文采用薄板样条函数实现图像的非刚性配准过程,采用加权插值方法克服了存在的空值间隙的问题。此外,本文提出一种改进的SIFT算法,自动提取图像中的感兴趣点,有效的提高了配准的方便性以及配准的精度。最后利用图像的小波变换算法将配准后的肝脏CT多相期扫描图像进行图像融合。3.针对肝脏DSA图像的特点,提出了一种肝脏血管配准方法,实现了肝脏CT图像与DSA图像的配准。首先将CT图像进行投影处理,把DSA与CT图像变换到一个统一的坐标系中,将3D/2D的配准过程转化为2D/2D的配准过程。在对CT图像中肝脏管道进行血管分割、血管细化、提取血管骨骼线后,以肝脏内血管的结点为基准点,将CT内的血管切分为若干部分,在最大互信息配准算法的基础上,实现了肝脏CT图像与DSA图像的多模配准。。4.提出一种由粗到精的混合配准算法,实现了肝脏CT图像与MR图像的多模配准。算法具有较好的收敛速度,且可以避免由于配准过程陷入局部极值而引起的错误。图像配准通常在开始时使用粗略的快速算法,然后使用精确的慢速算法。力矩主轴配准法是根据图像的形状信息获得图像的质心、主轴,该方法可以获得图像的平移、旋转参数;基于Fourier-Mellin变换的图像配准算法是将图像变换到频域上后进行基于非特征的图像配准。采用以上两种配准方法后,图像自动快速大致配准,进一步采用最大互信息法对图像进行精确配准。5.针对肝脏内部管道的特点,提出一种管道分析算法,实现了对肝脏管道系统分支变异以及管道走行情况分析判断。肝脏内部的血管包括肝动脉,肝静脉以及门静脉系统,根据解剖学上的统计分析以及管道拓扑结构的特点,对肝脏内部的管道进行分析,判断它们类型,在计算管道直径、长度的基础上,对肝内的肝静脉、肝动脉以及门静脉的分支类型进行分析,判定其走行以及拓扑结构,有利于保护重要的血管结构,对于肝脏的手术计划有重要的意义。总之,本文系统地研究了在体肝脏图像配准方法,主要包括肝脏CT多相期扫描图像的刚性配准方法,肝脏CT多相期扫描图像的非刚性配准方法,肝脏CT与MR图像配准方法,肝脏CT与DSA图像配准方法,以及肝脏管道系统分支变异与走行的分析,为数字化虚拟肝脏及手术计划系统的建立奠定了算法和程序基础。更多还原

【Abstract】 Liver is one of the most important organs in human.With the development of the medicine and computer technique,more and more modals medical images have emerged.Modern medical research usually requires integrated analysis of multiple images to get more information.A fundamental problem in medical image integrated analysis is that the images should be perfectly aligned,one essential aspect thereof is image registration,i.e.,recovering the geometric relationship between corresponding points in multiple images of the same scene.Medical image registration is an important technique in the field of medical image processing,and becoming more and more important for clinical diagnosis and treatment.The living liver images are obtained from the clinical medical examination,which contain the individual characteristic information.Registration of the living liver images will help for clinical diagnosis directly.Based on the registration methods,this work was supported by the National Nature Science Foundation of China(30770561 and 60701022)and the Science Research Foundation of ministry of Health & United Fujian Provincial Health and Education Project for Tacking the Key Research,P.R.China(WKJ2005-2-001). This thesis covered several registration methods of living liver images such as rigid registration of different phases of contrast enhanced liver CT data,nonrigid registration of different phases of contrast enhanced liver CT data,registration of MR and CT images,registration of DSA and CT images and the analysis of the types of branch variation of liver vessels.The main contents and contributions of this thesis are as follows:1)A rigid registration algorithm based on joint distribution histogram was proposed for different phases of contrast enhanced liver CT data.The four key steps of the rigid algorithm have been improved in this algorithm.1) A fast image transformation method was proposed,which speed up the transformation procedure 2)An improved measure function based on joint distribution histogram was proposed,which speed up the registration procedure without losing the accuracy;3)A hybrid interpolation algorithm based on B-spline transform was proposed,which improved the interpolation results;4)A hybrid optimization algorithm based on PSO and Powell was proposed,which had higher convergence speed and settling local extremum preferably.After rigid registration the images of different phases of contrast enhanced liver CT were aligned approximately.2)A nonrigid registration algorithm based on thin-plate splines was proposed which locate the control points automatically.The control points and the interpolation function are the key factors of the nonrigid algorithm.This algorithm adopted weighted interpolation based on thin-plane splines interpolation so that there were no“blank value points”in the images.Because the images were aligned approximately after rigid registration,the control points could be located by an improved SIFT algorithm.After nonrigid registration the images of different phases of contrast enhanced liver CT were aligned precisely.Finally,the post-registration images were fused by a wavelets-based scheme and the different information was merged into a new image.3)An algorithm based on maximum mutual information was proposed for the registration of CT and DSA images.Firstly,projection transformation was applied to CT images so that 2D DSA information and 3D CT information was in the same coordinate system.Then after segment blood vessels and extracting skeleton line of blood vessels,the blood vessels were separated into several parts by some blood vessel nodes in the liver.Finally based on the maximum mutual information registration algorithm,the multimode registration of CT and DSA liver images was achieved.4)A two-step registration algorithm was proposed for the registration of CT and MRI images.Firstly a method based on principle axes and Fourier-Mellin transform was adopted for the coarse registration.Because there are clear contours in CT and MR image,we presented a registration method based on principal axes.Through matching the centroid and principal axes of image,the desired parameters for translation and rotation transform were obtained.After that the scaling parameter was obtained by Fourier-Mellin transform.Secondly,maximum mutual information algorithm was applied to achieve precise registration.Experiments show that the two-step algorithm could avoid erroneous results because of the local extremum.5)An algorithm was proposed to analyze the route and topology configuration of the liver vessels.After registration the types of branch variation of liver vessels were determined, while the diameter and length of vessels were calculated.This algorithm helped the liver surgery planning greatly.In summary,this thesis covered on several registration methods of living liver images including rigid registration of different phases of contrast enhanced liver CT data,nonrigid registration of different phases of contrast enhanced liver CT data, registration of MR and CT images,registration of DSA and CT images and the analysis of the types of branch variation of liver vessels.Several algorithms were proposed in this thesis for solving some problems for liver image registration. Although there is still much work to do,this thesis may help a lot for establishing a real applied digital virtual liver and liver surgery planning system.更多还原