Multi-modal non-rigid image registration based on similarity and dissimilarity with the prior joint intensity distributions

Multi-modal non-rigid image registration is widely used in different areas, including medical image analysis and image processing. In this paper, we introduce a new learning-based method for non-rigid image registration. The proposed method is based on a priori knowledge of the joint intensity distribution of a pre-aligned image pair. The similarity and dissimilarity of the expected and observed joint intensity distributions are measured by two Kullback-Leibler distances (KLD). Free-Form Deformation (FFD) is employed as the transformation model along with the L-BFGS-B optimizer. The derivatives of KLDs are derived to work with the L-BFGS-B optimizer. Moreover, we have tested our method with CT-T1 image pairs and compared the results obtained by using the mutual information based FFD and the conventional KLD based FFD. The experimental results show that our method gives remarkable improvement on the registration quality.