本文为美国肯特州立大学（作者：Yujun Guo）的博士论文，共213页。

医学图像分析中的一个基本问题是同一对象的多幅图像中的信息集成，这些图像是使用相同或不同的成像方式获得的，并且可能是在不同时间获得的。这一问题的一个重要方面是图像配准，即在同一场景的多幅图像中寻找对应点之间的几何关系。近年来，人们提出了多种配准方法，其中基于互信息最大化的配准策略被证明是一种很有前景的配准方法，已被广泛应用于多模态图像配准。然而，将互信息（MI）应用于原始强度时，只考虑了统计信息，而忽略了空间信息。

本文第一部分提出了一种通过梯度向量流（GVF）将空间信息融合到MI中的新方法。使用这种方法，MI现在是根据原始图像的GVF强度（GVFI）图而不是它们的强度值来计算的。将该算法应用于多模态脑图像配准，验证了该方法的准确性和鲁棒性。实验结果表明，该方法的成功率高于传统的基于MI的配准方法。在许多应用中，刚性变换不足以描述两幅图像之间的空间关系。因此，在图像配准中经常需要弹性变换或非刚性变换。在本文的第二部分中，我们提出了一种广义梯度引导的非刚性配准策略。推导过程与Lucas和Kanade的类似，但更具一般性。在实验中，我们将所提出的方法与文献中的其他梯度引导方法进行比较，同时使用了合成图像和真实图像进行测试。结果表明，结合源图像和目标图像梯度的方法通常表现得更好。在第三部分中，我们将先前描述的配准方法应用于基于atlas的脑磁共振（MR）图像分割。首先将预先标记的图像或图谱配准到要分割的被摄体图像，并导出每个体素的变形场。然后，通过将变形场应用于atlas掩模，atlas描绘的结构投影到被摄体图像上。我们使用来自IBSR的数据集来验证我们的结果。使用各种标准进行的定量比较表明，所提出的方法优于或等同于已发表的方法。 

A fundamental problem in medical imageanalysis is the integration of information from multiple images of the samesubject, acquired using the same or different imaging modalities and possiblyat different time. An essential aspect of this problem is image registration,which is the task of finding geometric relationships between correspondingpoints in multiple images of the same scene. Various registration methods havebeen proposed over recent years, among which registration strategies based onmaximization of mutual information have been proved to be promising methods andhave been widely used in multi-modality image registration. However, applyingmutual information (MI) to original intensities only takes statisticalinformation into consideration, while spatial information is completelyneglected. In the first part of this dissertation, a novel approach is proposedto incorporate spatial information into MI through gradient vector flow (GVF).With this approach, MI now is calculated from the GVF-intensity (GVFI) map ofthe original images instead of their intensity values. The algorithm isimplemented and applied to multi-modality brain image registration to test theaccuracy and robustness of the proposed method. Experimental results show thatthe success rate of our method is higher than that of traditional MI-basedregistration. In many applications, a rigid transformation is insufficient todescribe the spatial relationship between two images. Thus, elastictransformations, or non-rigid transformations are often required in imageregistration. In the second part of this dissertation, we present a generalizedgradient-guided non-rigid registration strategy. The derivation procedure issimilar to that by Lucas and Kanade, but in a more general manner. Inexperiments, we compare the proposed method and other gradient-guided methodsin the literature, using both synthetic and real images. It is shown thatmethods combining gradients from both source and target images usually performbetter. In the third part, we apply previously described registration methodsto atlas-based brain magnetic resonance (MR) image segmentation. A pre-labelledimage or atlas is first registered to the subject image to be segmented, andthe deformation field for each voxel is derived. Then the structures delineatedin the atlas are projected onto the subject image by applying the deformationfield to the atlas mask. We validate our results using the datasets from IBSR.Quantitative comparisons using various criteria show that the proposed methodis better than or comparable to published methods.

1. 引言

2. 研究背景

3. 基于梯度矢量流强度互信息最大化的刚性配准

4. 梯度引导的非刚性配准

5. 基于Atlas的脑图像分割

6. 未来工作展望

附录 验证研究的输出文件

更多精彩文章请关注公众号：