本文为美国爱荷华大学（作者：Joo Hyun Song）的博士论文，共166页。

在医学成像领域，图像配准方法对于许多应用都是有用的，例如主体间和主体内的形态学比较、群体图谱的创建、精确治疗的交付等。用户可能想知道哪种配准算法最适合预期应用，但医学图像配准应用的广泛性，使得图像配准性能的评估和比较成为一项非常重要的任务。一般来说，评估图像配准性能并不简单，因为在大多数图像配准应用中，没有“金标准”或真值对应图可供比较。因此，本文的主要目标是提供一种方法来推荐一个给定任务的最合适的配准算法。

本文的一个贡献就是在组件级检验图像配准算法的性能。本文的另一个贡献是列举了许多最常用的图像配准评估方法的优点和局限性。本文的一个增量式贡献是演示如何在中点坐标系中应用现有的评估方法来评估一些对称图像配准算法，例如SyN配准算法。最后，本文的一个主要贡献是开发实用工具来评估和可视化二维、三维图像配准形状塌陷。本文论证了目前许多微分同胚图像配准算法都存在塌陷问题，为简单形状和真实人脑MR图像的三维塌陷问题提供了初步的可视化方法，并提供了第一个实验，证明了调整图像配准参数可以在一定程度上缓解图像的塌陷问题。

In the field of medical imaging, imageregistration methods are useful for many applications such as inter- andintra-subject morphological comparisons, creation of population atlases,delivery of precision therapies, etc. A user may want to know which is the mostsuitable registration algorithm that would work best for the intendedapplication, but the vastness of medical image registration applications makesevaluation and comparison of image registration performance a non-trivial task.In general, evaluating image registration performance is not straightforwardbecause in most image registration applications there is an absence of “GoldStandard” or ground truth correspondence map to compare against. It istherefore the primary goal of this thesis work to provide a means forrecommending the most appropriate registration algorithm for a given task. Oneof the contributions of this thesis is to examine image registration algorithmperformance at the component level. Another contribution of this thesis is tocatalog the benefits and limitations of many of the most commonly used imageregistration evaluation approaches. One incremental contribution of this thesiswas to demonstrate how existing evaluation methods can be applied in themidpoint coordinate system to evaluate some symmetric image registrationalgorithms such as the SyN registration algorithm. Finally, a majorcontribution of this thesis was to develop tools to evaluate and visualize 2Dand 3D image registration shape collapse. This thesis demonstrates that manycurrent diffeomorphic image registration algorithms suffer from the collapseproblem, provides the first visualizations of the collapse problem in 3D forsimple shapes and real human brain MR images, and provides the firstexperiments that demonstrate how adjusting image registration parameters canmitigate the collapse problem to some extent.

1. 引言

2. 图像配准

3. 图像配准的评估方法

4. 实验

5. 总结与结论

更多精彩文章请关注公众号：