Deformable Image Registration based on Similarity-Steered CNN Regression

Abstract

Existing deformable registration methods require exhaustively iterative optimization, along with careful parameter tuning, to estimate the deformation field between images. Although some learning-based methods have been proposed for initiating deformation estimation, they are often template-specific and not flexible in practical use. In this paper, we propose a convolutional neural network (CNN) based regression model to directly learn the complex mapping from the input image pair (i.e., a pair of template and subject) to their corresponding deformation field. Specifically, our CNN architecture is designed in a patch-based manner to learn the complex mapping from the input patch pairs to their respective deformation field. First, the equalized active-points guided sampling strategy is introduced to facilitate accurate CNN model learning upon a limited image dataset. Then, the similarity-steered CNN architecture is designed, where we propose to add the auxiliary contextual cue, i.e., the similarity between input patches, to more directly guide the learning process. Experiments on different brain image datasets demonstrate promising registration performance based on our CNN model. Furthermore, it is found that the trained CNN model from one dataset can be successfully transferred to another dataset, although brain appearances across datasets are quite variable.

Fig. 1

The framework of the proposed similarity-steered CNN regression for deformable image registration. The input image pair has already been linearly aligned. Here, we use 2D examples for easy illustration, while our implementations are in 3D.

Fig. 2

Illustration of the equalized sampling strategy. The displacement magnitudes distribute unevenly as shown in (a). The deformation field is thus underestimated in (b) if following the conventional sampling strategy, and much improved in (c) when using our proposed strategy.

Fig. 3

Similarity maps of one sample with different kernel size 2r + 1. Solid and dashed circle indicate correct guidance and incorrect guidance, respectively.

Fig. 4

Mean DSC of each of 54 ROIs based on 10 testing subjects from LONI dataset, after deformable registration by Demons, SyN, and our proposed method. “*” indicates statistically significant improvement by our proposed method, compared with other two methods (p<0.05).

Fig. 5

Visualized registration results by Demons, SyN and our proposed method. Obvious improvements by the proposed method on ventricle region (first row), and the central sulcus and postcentral gyrus in 3D rending view (second row).