Local Elastic Registration of Multimodal Medical Image Using Robust Point Matching and Compact Support RBF

Abstract

A novel local elastic registration of multimodal medical image method is proposed in this paper. At first, local deformation regions are detected by evaluating the variation of mutual information in re-quantified gray space of images. The re-quantified image retains anatomical structure of the organ well and reduces the gray levels greatly. Mutual information performs better in the quantification space and can be used to detect whether the deformation happens in small sampling images. Next, edges of the local deformation regions are detected. Fuzzy clustering method is performed on edge points and the clustering centers are chosen as candidate landmarks. Robust point matching is used to estimate landmarks correspondence in the local deformation regions. Finally, a new compact support radial basis function CSTPF has been adopted to deform image, which cost less bending energy than other RBFs. Local registration experiments of multimodal medical images show the feasibility of our method.