高阶乘幂法应用于基于马尔可夫随机场的肺部4D CT图像配准方法研究

发布时间：2018-09-10 11:56

【摘要】：放疗是治疗肺癌的常用手段,为了减少放疗的剂量,由图像引导的放疗需要估计病人的呼吸运动模型,而肺部4D CT图像配准是运动建模的核心技术。本文研究了基于马尔可夫随机场的4D CT图像配准技术,提出了一种基于高阶乘幂法的肺部4D CT图像配准算法,主要内容如下:本文首先介绍了基于马尔可夫随机场的图像配准模型,然后基于团分解模型分析了各个势函数对约束形变场所起到的作用,并针对2D与3D图像,将拓扑约束势函数转化为各向同性的形式,最后给出了图像配准算法的基本框架。其次,针对图割及消息传播类算法中存在的不能有效处理多元团约束的问题,提出了应用于图像配准的张量模型,并采用对称张量高阶乘幂法来求解相应的最优化问题,提出的算法适用于各向同性马尔可夫随机场中的图像配准问题。为了降低算法的空间复杂度,本文提出了一种张量稀疏表示方法,并说明了对于任意(复合)同质性马尔可夫随机场中的势函数,皆可利用此方法来构造张量。此外,还针对2D图像改进了用于弱拓扑保持的势函数,仿真结果表明同时施加平滑约束与弱拓扑约束能够有效约束形变场的拓扑结构。最后,针对肺部4D CT图像配准,给出了两种多尺度处理策略,并在DIR-Lab公开数据集上进行了仿真实验。主要研究了辅助函数对算法性能的影响,并且以目标配准误差(TRE)作为评价准则,将本文提出的算法与MLS、deeds和isoPTV算法做了对比,实验结果验证了算法的正确性和有效性。
[Abstract]:Radiotherapy is a commonly used method in the treatment of lung cancer. In order to reduce the dose of radiotherapy, image-guided radiotherapy needs to estimate the respiratory motion model of patients, and the lung 4D CT image registration is the core technology of motion modeling. In this paper, 4D CT image registration technology based on Markov random field is studied, and a lung 4D CT image registration algorithm based on higher order power method is proposed. The main contents are as follows: firstly, the image registration model based on Markov random field is introduced. Then, based on the cluster decomposition model, the effects of potential functions on constrained deformation sites are analyzed, and 2D and 3D images are analyzed. The topological constrained potential function is transformed into isotropic form. Finally, the basic frame of image registration algorithm is given. Secondly, aiming at the problem that graph cutting and message propagation algorithms can not deal with multi-cluster constraints effectively, a Zhang Liang model applied to image registration is put forward, and the corresponding optimization problem is solved by using the high order power method of symmetric Zhang Liang. The proposed algorithm is suitable for image registration in isotropic Markov random fields. In order to reduce the space complexity of the algorithm, a Zhang Liang sparse representation method is proposed, and it is shown that the potential functions in arbitrary (composite) homogeneous Markov random fields can be constructed by this method. In addition, the potential function used to preserve the weak topology is improved for 2D images. The simulation results show that both smoothing and weak topological constraints can effectively constrain the topological structure of the deformation field. Finally, two kinds of multi-scale processing strategies are presented for lung 4D CT image registration, and the simulation experiments are carried out on the DIR-Lab open data set. This paper mainly studies the influence of auxiliary function on the performance of the algorithm, and takes the target registration error (TRE) as the evaluation criterion, and compares the proposed algorithm with MLS,deeds and isoPTV algorithms. The experimental results show that the algorithm is correct and effective.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R734.2;TP391.41

【相似文献】