Non-rigid registration between 3D MR and CT images of the liver based on intensity and edge orientation information

The multi-modal image registration is essential to integrate complementary image information from different modalities of the same organ. In order to find a reliable geometric transformation for registration, a proper choice of similarity measure is required. Even though normalized mutual information (NMI) is popular for multi-modal image registration, it limits the registration performance since the NMI measure only takes intensity values into account without considering any useful spatial information. In this paper, we attempt to enhance the non-rigid registration accuracy between MR and CT images of the liver, by jointly using intensity and spatial information. In order to effectively manipulate the relationship of the intensity and spatial information between MR and CT images, we adopt our previously proposed similarity measure based on 3D joint histogram where edge orientation coincidence is utilized as the spatial information. Since we need non-rigid registration due to the easy deformation of liver, the B-spline free-form deformation (FFD) is adopted as the transformation model. The experimental results for clinical datasets show that the proposed method provides more accurate registration results than a conventional NMI-basedregistration.