基于多模板弹性配准的鼻咽癌危及器官自动分割

发布时间：2018-01-02 13:30

本文关键词：基于多模板弹性配准的鼻咽癌危及器官自动分割　出处：《福州大学》2013年硕士论文　论文类型：学位论文

【摘要】：随着计算机技术的发展,放射治疗技术也从二维治疗进入到了三维精确治疗阶段。在三维精确治疗中,对患者危及器官及肿瘤的勾画要求也越来越高。实践中手工勾画存在耗时长和专家勾画互相之间,专家自身在不同时间勾画结果不同等问题。因此需要快速的而准确的勾画出危及器官。鼻咽癌作为我国东南沿海地区的高发肿瘤,肿瘤附近区域是人体中一个结构极为复杂且十分重要部位,因此鼻咽癌危机器官的自动勾画的实现具有十分重大的临床意义。本文首先概要的介绍了国内外鼻咽癌放射治疗的现状,并讨论了临床中自动分割的必要性。同时给出了本文研究的基本路线。其次用三步法实现鼻咽癌危及器官的自动分割。在第一步中主要介绍基本的线性形变配准和相关的评价方案,并简介了图像配准基本框架的四个要素：几何变换,图像插值,相似性测度以及优化算法。同时介绍了EMMA算法步骤。最后在基于ITK软件包的基础上实现了线性配准。第二步中介绍基于线性配准的多边形平滑非线性配准。包括欧氏对数多边形仿射变换框架和块配准的方法,然后说明如何用其对多边形平滑配准的评估。最后一步介绍并实现形状限制的密度形变配准。主要对于高斯窗口局部相关因子(Local correlation coefficients, LCC)进行详细介绍。并用ITK工具箱实现了全自由度形变配准的编程。本文的最后一部分介绍了STAPLE(Simultaneous Truth and Performance Level Estimation)方法及如何使用STAPE方法进行多模板自动勾画的合成。同时介绍本文提出的模板选取的方法,及临床自动分割结果的三种评价方法,包括DICE相似度方法、变形距离法、体积差异法。实践测试中,使用两个不同的医院的病例数据,这些都经过专家用手工勾画好了危及器官,分别选取了7例病例来作为模板数据库。以此对于15例待勾画病例进行了自动勾画。通过专家手工勾画的15例的结果与自动勾画结果、自动勾画修正的结果进行三种方法的评价。实验证明,本文提出了的方法可以进一步提高自动勾画的准确性。同时大幅度的减少了临床医生勾画的时间。
[Abstract]:With the development of computer technology, radiotherapy technology has also entered the stage of three-dimensional precise treatment from two-dimensional treatment. The requirements for patients with dangerous organs and tumor sketching are also increasing. In practice, manual sketching takes a long time and expert sketching each other. Experts themselves draw different results at different times. Therefore, it is necessary to quickly and accurately delineate the dangerous organs. Nasopharyngeal carcinoma (NPC) is a high incidence tumor in the southeast coastal area of China. The area near the tumor is a very complicated and important part of human body. Therefore, the realization of automatic mapping of nasopharyngeal carcinoma crisis organs is of great clinical significance. Firstly, the present situation of radiotherapy for nasopharyngeal carcinoma at home and abroad is briefly introduced in this paper. The necessity of automatic segmentation in clinic is also discussed. The basic route of this paper is also given. Secondly, the three-step method is used to realize automatic segmentation of nasopharyngeal carcinoma (NPC) endangered organs. In the first step, the basic linear deformation registration is introduced. And related evaluation programmes. The four main elements of the basic frame of image registration are introduced: geometric transformation and image interpolation. Similarity measure and optimization algorithm. At the same time, the steps of EMMA algorithm are introduced. Finally, linear registration is realized on the basis of ITK software package. In the second step, the smooth nonlinearity of polygon based on linear registration is introduced. Registration. Includes the Euclidean logarithmic polygon affine transformation frame and block registration methods. Then it explains how to evaluate the smooth registration of polygons. The last step is to introduce and realize the shape limited density deformation registration. Local correlation coefficients. The programming of full degree of freedom deformation registration is realized with ITK toolbox. The last part of this paper introduces Staple (. Simultaneous Truth and Performance Level estimation). Methods and how to use STAPE method to synthesize multi-template automatic sketch. At the same time, the method of template selection proposed in this paper is introduced. And three evaluation methods of clinical automatic segmentation results, including DICE similarity method, deformation distance method, volume difference method. In practice, two different hospital case data were used. These are all hand-drawn by experts, endangering organs. Seven cases were selected as template database, and 15 cases were drawn automatically. The results of 15 cases were drawn by hand by experts and the results of automatic sketching were obtained. The results of automatic sketch correction are evaluated by three methods. Experiments show that the proposed method can further improve the accuracy of automatic sketching, and greatly reduce the time of drawing by clinicians.
【学位授予单位】：福州大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：R739.63;TP391.41

【相似文献】