Robust non-rigid image registration using incomplete information, adaptive normalized convolution and similarity measure combination

A novel algorithm within block matching framework is presented for robust non-rigid registration of medical images with incomplete information. After getting the deformation field computed from block-matching, we introduce robust and structure-adaptive normalized convolution in spatial regularization of deformation field, using cross-correlation (CC) for low resolution images and mutual information (MI) for high resolution images. In order to reduce deformation diffusion across discontinuities caused by incomplete image information and boundary condition, the applicability function of structure-adaptive normalized convolution based on an anisotropic Gaussian kernel is adapted to local linear or edge structures in the images to register. A robust signal certainty is also iteratively adapted to each displacement vector in the deformation field to measure its accuracy and robustness to local incomplete information. Moreover, we show that similarity measure combination changing CC into MI under hierarchical framework was better behaved than that changing MI into CC, or that using any of these similarity measures alone for multi-resolution images with incomplete information. The results show that the method is sufficiently accurate and robust to incomplete information for non-rigid brain image registration.