Medical image registration has been applied to the diagnosis of breast cancer, cardiac studies, and a variety of neurological disorders including brain tumors. A new approach called maximization of mutual information (MMI), quantifying the similarity between corresponding voxel intensities of two images, has been demonstrated to be a very powerful criterion for three-dimensional medical image registration. MMI can be thought of as a measure of how well one image explains the other, it is maximized at the optimal alignment of the images. Though MMI has been widely used, many implementation issues, such as multi-resolution, interpolation, probability distribution estimation, capture ranges, and optimization methods, are still under investigation.; In this thesis, we present three new implementation approaches for multimodal brain image registrations based on normalized mutual information. Each of these approaches gives improved accuracy or/and speed. We illustrate these approaches by presenting the results of tests from the data of 9 patients. Registration accuracy is evaluated by J. M. Fitzpatrick, at Vanderbilt University, U.S.A., as part of the Retrospective Registration Evaluation Project (http://www.vuse.vanderbilt.edu/∼image/registration/). We also compare the performance of each approach to results available in the literature.; The first chapter in this thesis gives an overview of medical image registration by maximization of mutual information. The thesis then continues with a chapter describing the implementation used. In chapter 3 we describe our binarization/multi-resolution scheme for CT-MR brain image registration. This method is designed to improve the accuracy of registration. Next, in chapter 4, we present another approach, a nonlinear binning method for CT-MR image registration based on background segmentation and a k-means clustering algorithm, aiming at improving accuracy and speed. Lastly, in chapter 5, we develop a wavelet-based multi-resolution approach for multimodal brain image registration. This approach performs a registration between CT and MR images or a registration between PET and MR images. This approach gives promising results for both CT-MR and PET-MR registrations.
展开▼
机译：医学图像配准已被应用于诊断乳腺癌，心脏研究以及包括脑肿瘤在内的各种神经系统疾病。已经证明了一种称为互信息最大化（MMI）的新方法，该方法可量化两个图像的相应体素强度之间的相似性，这是三维医学图像配准的非常有效的标准。可以将MMI视为衡量一幅图像对另一幅图像的解释程度的一种度量，它可以在图像的最佳对齐方式下最大化。尽管MMI已被广泛使用，但是许多实现问题，例如多分辨率，插值，概率分布估计，捕获范围和优化方法仍在研究中。本文提出了三种基于归一化互信息的多模态脑图像配准的新实现方法。这些方法中的每一种都提高了准确性或/和速度。我们通过展示来自9位患者的数据的测试结果来说明这些方法。作为追溯追溯评估项目（http://www.vuse.vanderbilt.edu/〜image / registration /）的一部分，美国范德比尔特大学的J. M. Fitzpatrick评估了注册准确性。我们还将比较每种方法的性能与文献中可获得的结果。本文的第一章通过互信息最大化来概述医学图像配准。然后，本文将继续介绍所使用的实现的一章。在第3章中，我们描述了CT-MR脑图像配准的二值化/多分辨率方案。此方法旨在提高注册的准确性。接下来，在第4章中，我们提出了另一种方法，即基于背景分割和k均值聚类算法的CT-MR图像非线性配准方法，旨在提高准确性和速度。最后，在第5章中，我们开发了基于小波的多分辨率方法来进行多模态脑图像配准。这种方法执行CT和MR图像之间的配准，或PET和MR图像之间的配准。对于CT-MR和PET-MR配准，此方法都可提供令人鼓舞的结果。
展开▼
