CASIA OpenIR  > 毕业生  > 硕士学位论文
Alternative TitleResearch of Fast Medical Image Registration Algorithms Based On CUDA
Thesis Advisor戴汝为
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Discipline控制理论与控制工程
Keyword医学图像配准 块匹配 Sift Cuda Medical Image Registraiton Block-matching Sift Cuda
Abstract医学图像配准是医学图像处理的一个重要组成部分,在现代医疗当中正发挥着越来越大的作用。图像配准技术可以用来进行多模态数据融合,为医生的诊断提供更多的疾病信息;研究者还可以通过配准一系列连续的生物组织剖面来进行三维重建;也可以使用图像配准来帮助制定医疗计划和引导手术进行,这已经成为现在很多医院的一个常规过程。 一方面,经过多年的发展,医学图像配准领域已产生众多的研究方法,这些方法有的是来源于本领域的研究,有的则是来自于其他相关领域的研究,如计算机视觉、模式识别等。另一方面,随着人类寿命的延长,人们对医疗的要求也越来越高,医学影像学作为现代医疗的一个重要组成部分,其需求与日俱增,包括对医学图像配准的需求,而传统的医学图像配准由于其耗时性,难以满足这种日益增长的需求。 本文正是着眼于这种需求,通过采用现在流行的并行编程方法CUDA来对传统的医学图像配准算法进行加速。现在主要有两种医学图像配准算法:一种是基于灰度的配准方法,另一种是基于特征的点集匹配算法。在本文中,我们选取这两个类别中各一种算法,即块匹配算法与本文提出的基于尺度不变特征变换和使用高斯混合模型的点集匹配算法(GMMREG)的图像配准算法(SIFT-GMMREG)作为研究对象,根据相关文献,这几种方法都有很好的鲁棒性。我们通过分析算法本身的特点, 结合CUDA对它们进行并行加速和相关改进。实验证明,本文所采用的编程方法和改进方法对于这两种算法的速度都有明显的改善。 本文首先简单的介绍医学图像配准问题的定义和特点,目前主要的研究方法和类别,以及医学图像配准的意义以及现在所面临的挑战。然后分别详细介绍两种本文重点研究的图像配准方法,以及我们对于这两种方法加速方面所做的改进。最后对本文的研究做简单的总结和展望。
Other AbstractMedical Image registration is an important component of Medical Image Processing, and it is playing a key role in current medical treatment and diagnose. The technique of medical image registration has various applications. It can be used to fuse multi-modal image data to offer doctors more comprehensive information about their patients’ diseases. It can also be applied to implement 3D reconstruction by registering a successive series of sections of biological tissue. And furthermore it can facilitate to make plans for clinical treatment and instruct operation, which has now become a routine of some hospitals. On the one hand, with the development of research on medical image registration over these years, this field has been enriched by a plenty of new algorithms and ideas. Some of them are originated from medical image processing, while others are introduced from correlative realms, such as computer vision and pattern recognition. On the other hand, in the wake of the extension of humans’ life span, our demand for medical service is higher than ever before, especially the demand for high-quality medical image processing, including medical image registration. However, traditional medical image registration algorithms are quite time-consuming and hence it becomes hard to meet the demand. To improve the efficiency of image registration, we combine traditional registration methods with the state-of-the-art parallel computing platform, CUDA(Compute Unified Device Architecture), which can accelerate the speed of these algorithms effectively. Our attention is focused on the speed-up of two algorithms, block-matching and a new registration algorithm based on Scale Invariant Feature Transform(SIFT) and point sets registration using Gaussian mixture model algorithm(GMMREG), which are robust registration methods, according to relevant references, but are respectively from two different registration sub-domains, namely intensity-based registration and feature-based registration. Our experiment’s results demonstrate the effectiveness of our methods. In this thesis, we firstly make a brief introduction to medical image registration, including the definition and features of image registration, current main methods addressing on it and its classification, and the challenges of this field. Then, we elaborate our work in this thesis. In this part, we will make clear the process of the two algorithms and our improvement to them. Lastly, we make a review to our w...
Other Identifier2010E8014669001
Document Type学位论文
Recommended Citation
GB/T 7714
朱明. 基于CUDA的快速医学图像配准算法研究[D]. 中国科学院自动化研究所. 中国科学院大学,2013.
Files in This Item:
File Name/Size DocType Version Access License
CASIA_2010E801466900(2229KB) 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[朱明]'s Articles
Baidu academic
Similar articles in Baidu academic
[朱明]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[朱明]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.