S‐HAMMER: Hierarchical attribute‐guided, symmetric diffeomorphic registration for MR brain images

Abstract

Deformable registration has been widely used in neuroscience studies for spatial normalization of brain images onto the standard space. Because of possible large anatomical differences across different individual brains, registration performance could be limited when trying to estimate a single directed deformation pathway, i.e., either from template to subject or from subject to template. Symmetric image registration, however, offers an effective way to simultaneously deform template and subject images toward each other until they meet at the middle point. Although some intensity-based registration algorithms have nicely incorporated this concept of symmetric deformation, the pointwise intensity matching between two images may not necessarily imply the matching of correct anatomical correspondences. Based on HAMMER registration algorithm (Shen and Davatzikos, [2002]: IEEE Trans Med Imaging 21:1421–1439), we integrate the strategies of hierarchical attribute matching and symmetric diffeomorphic deformation to build a new symmetric-diffeomorphic HAMMER registration algorithm, called as S-HAMMER. The performance of S-HAMMER has been extensively compared with 14 state-of-the-art nonrigid registration algorithms evaluated in (Klein et al., [2009]: NeuroImage 46:786–802) by using real brain images in LPBA40, IBSR18, CUMC12, and MGH10 datasets. In addition, the registration performance of S-HAMMER, by comparison with other methods, is also demonstrated on both elderly MR brain images (>70 years old) and the simulated brain images with ground-truth deformation fields. In all experiments, our proposed method achieves the best registration performance over all other registration methods, indicating the high applicability of our method in future neuroscience and clinical applications.

Random walker image registration with inverse consistency

“…In summary, our method 1) achieved a Jaccard Index of 0.324, which was an improvement over the original framework [3] by 0.01, and more importantly, 2) ensured inverse consistency in the solution, while [3] does not. We also note that our method achieved a target overlap of 0.485±0.041, which, based on the figures reported in [14], ranks 6th out of the 17 methods compared. However, only 2 of the 5 more superior methods ensured symmetric solutions, and that these two these methods employ more advanced similarity measures, unlike the computationally more efficient SD measure that we have employed in this work.…”

“…This approach is adopted in many registration algorithms, including those proposed in recent literature [9], [11]- [14]. Performing two independent optimizations per iteration, however, at least requires twice as much resource as required by standard, inverse nonconsistent algorithms [9].…”

Scalable joint segmentation and registration framework for infant brain images

“…Given the segmentation image for the subject image (and also assuming the tissue segmentation of template image is already obtained), we can deploy the state-of-the-art registration method, i.e., symmetric HAMMER [53], to simultaneously estimate the deformation pathways ϕ 1 (from subject image to the hidden common space) and ϕ 2 (from template image to the hidden common space). Since geometric invariant moment (GMI) features are extracted from the segmented images, image registration is free of dynamic appearance changes in the original intensity images.…”

“…In the beginning of registration ( k = 0), $M^{false(0false)}=M_{t_1}\left({\varphi }_1^{(0)}\right)$ and $F^{false(0false)}=F_{t_2}\left({\varphi }_2^{(0)}\right)$, along with the initial deformation pathways ${\varphi }_1^{false(0false)}$ and ${\varphi }_2^{false(0false)}$ estimated above. Also, only a small number of key points [53] with distinctive features are selected from M (0) and F (0) to establish anatomical correspondences by matching the GMI features. In the following, we adopt the hierarchical deformation mechanism for establishing the correspondence between the deformed subject image $M^{false(kfalse)}=M_{t_1}\left({\varphi }_1^{(k)}\right)$ and the deformed template image $F^{false(kfalse)}=F_{t_2}\left({\varphi }_2^{(k)}\right)$.…”

A Case Analysis on Different Registration Methods on Multi-modal Brain Images

“…It can refer for example to [46] for the combination of curves and surfaces.  The combination of different information from the gray levels, as is the case in [47] where the gradient of the image and the information of the gray levels are used together [48], for their part, construct an attribute vector for each voxel in the image. This vector contains both the intensity of the voxel in question, the invariant geometric moment characteristic of the vicinity of the voxel, and information resulting from the segmentation into 3 classes of the image.…”