基于B样条和互信息的非刚性医学图像配准的研究与应用

发布时间：2018-04-19 09:10

本文选题：非刚性配准 + LBFGS算法　；参考：《太原理工大学》2017年硕士论文

【摘要】：图像配准技术发展到今天也不过三十多年的历史,由于计算机硬件与医学成像技术突飞猛进的发展,不断地涌现出许多信息各异的医学图像,为临床诊断提供了丰富的资料。由于各种医疗仪器的成像原理不同,导致提供的医学信息也就大不相同,然而单模态图像为医生的临床诊断提供的都是零散片面的信息,要想得到更全面完整且互补的图像信息,就必须将携带各种不同类型信息的多种模态图像融合在一起,以便医生做出更准确可靠的诊断。当前国内外在刚性配准领域的技术已经成熟,但事实上人体器官组织通常都存在各种复杂的非线性形变,只有非刚性配准才能满足这方面要求。因此,医学图像非刚性配准在临床治疗上尤为关键,具有十分重要的现实应用意义,必将成为未来医学数字技术的热点。本文主要研究非刚性图像配准关键技术的两大模块,一是反映两幅图像之间空间关系的变换模型,一是度量两幅图像是否达到完全配准的相似性测度。具体的研究内容有以下几个方面:(1)研究了配准的整个框架、流程及具体步骤,详细归纳并整理总结了图像配准中的空间变换、灰度插值、相似性测度和优化算法四大关键技术,重点阐述了样条函数模型、LBFGS优化算法的搜索原理以及基于空间变换和物理模型的非刚性图像配准方法。(2)针对存在非线性形变的图像,采用B样条变换可以很好地拟合图像间的不规则形变,直观上看可以得到较佳的配准效果;针对B样条控制网格间距的选取具有随机性,无法很好地权衡配准精度与效率的问题,提出了基于多层次B样条变换模型的医学图像非刚性配准方法;针对多层次B样条的均匀形变场无法很好地模拟图像局部区域大形变的问题,提出了基于局部区域多层次B样条变换模型的非刚性配准方法。(3)互信息的测度在配准前不用进行任何特征提取、检测、分割等预处理操作,仅需计算两幅图像各自信息熵和联合信息熵,确实简单方便易行,但也正因如此忽略了图像本身存在的任何空间信息,影响了配准结果的精度与鲁棒性,因此本文引进一种把图像本身信息置于很高地位的局部互信息概念,详细地分析了局部互信息的原理、计算方法与步骤,并提出了基于局部互信息的图像配准方法,进行了对比仿真实验,结果表明,提出的方案有效的提高了配准的精确性和鲁棒性,但也存在时间代价大的缺陷。(4)进行基于B样条和局部互信息的前列腺图像非刚性配准的应用研究,从总体上阐述使用的配准算法及配准过程的步骤;然后,介绍配准使用的关键主函数框架以及配准用到的图形用户界面;最后,对比试验结果,提出的方案对存在非线性无规则形变的医学图像有很好的配准效果。
[Abstract]:Image registration technology has been developed for more than 30 years. Because of the rapid development of computer hardware and medical imaging technology, many medical images with different information have been emerging, which provides abundant information for clinical diagnosis.Because of the different imaging principles of various medical instruments, the medical information provided is very different. However, the single mode image provides the doctor's clinical diagnosis with scattered and one-sided information.In order to obtain more complete and complementary image information, it is necessary to fuse a variety of modal images with different types of information, so that doctors can make more accurate and reliable diagnosis.At present, the technology of rigid registration has been mature at home and abroad, but in fact, human organs and tissues usually have a variety of complex nonlinear deformation, only non-rigid registration can meet this requirement.Therefore, non-rigid registration of medical images is very important in clinical treatment and has very important practical significance. It will become a hot spot of medical digital technology in the future.This paper mainly studies two modules of the key technology of non-rigid image registration, one is the transformation model which reflects the spatial relationship between the two images, the other is to measure whether the two images achieve the similarity measure of complete registration.The main contents of this paper are as follows: (1) the whole frame, flow and concrete steps of registration are studied, and the four key techniques of image registration, such as spatial transformation, gray interpolation, similarity measure and optimization algorithm, are summarized and summarized in detail.The search principle of the spline function model LBFGS optimization algorithm and the non-rigid image registration method based on spatial transformation and physical model.The irregular deformation between images can be well fitted by B-spline transform, and the registration effect can be obtained intuitively, and the selection of B-spline control mesh spacing is random.This paper presents a non-rigid medical image registration method based on multi-level B-spline transform model, which is unable to balance the accuracy and efficiency of registration.In view of the problem that the uniform deformation field of multilevel B-spline can not well simulate the large deformation in the local region of the image,A non-rigid registration method based on local region multi-level B-spline transform model is proposed. The measure of mutual information does not need any pre-processing operations such as feature extraction, detection and segmentation before registration.It is simple and convenient to calculate the information entropy and joint information entropy of the two images, but because of this, any spatial information that exists in the image itself is ignored, which affects the accuracy and robustness of the registration results.Therefore, this paper introduces a concept of local mutual information, which places the image itself in a very high position, analyzes in detail the principle, calculation methods and steps of local mutual information, and proposes an image registration method based on local mutual information.The simulation results show that the proposed scheme can effectively improve the accuracy and robustness of registration.However, there is also a time-cost defect. (4) to study the application of non-rigid registration of prostate image based on B-spline and local mutual information, and to describe the registration algorithm and the steps of registration process.The key principal function framework and the graphical user interface used in registration are introduced. Finally, compared with the experimental results, the proposed scheme has a good registration effect for medical images with nonlinear irregular deformation.
【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】