Non-rigid registration of medical images with scale-space corner detection and thin-plate spline

Image registration is an important enabling technology with many clinical applications, and has received considerable research interest. This paper presents a new non-rigid registration method by using nonlinear complex diffusion corner detection and thin-plate spline model. Compared with general multiple scale space process which often blurs edges in the image and causes loss of useful information, nonlinear diffusion process has many favorite properties because it can provide a hierarchical representation of images in different scales and preserve image edges well. Furthermore, the imaginary part of the image in the complex diffusion process can serve as an edge detector. In this study, corner detection is performed by combining the nonlinear complex diffusion process and global and local curvature properties. The detected corners are served as landmarks, and thin-plate spline (TPS) is employed for non-rigid registration. The proposed registration method is evaluated on CT and CBCT images, and the experimental results demonstrate the effectiveness of the new method.

► By applying complex nonlinear diffusion, different scale-space images can be got. ► More corners can be detected by applying global and local curvature properties. ► Ideal results can be got by applying thin-plate spline registration method.