Multiscale registration for noisy medical images

Abstract

a multiscale deformable registration framework is proposed by combining edge preserving scale space with the free form deformation (FFD) for robust registration of noisy medical images. The edge preserving scale space derived from the total variation model with the L1 norm (TV-L1) has two advantages for registration: the first one is removing noise from the image pairs to be registered and the second one is selecting edges and contours of an image according to their geometric sizes. After being denoised, the selected edges and contours are sufficient enough to drive deformation using the FFD grid at each scale. Then the deformation fields are gained accurately in a coarse-to-fine manner. The proposed method is effective where one or both of the images to be registered with significant level of noise. Experiments on both artificially deformed medical images with ground truth and clinical CBCT images demonstrate the accuracy and robustness of our proposed multiscale deformable registration techniques.