Dynamic contrast enhancement (DCE) MRI and its quantitative hemodynamic parameters have increasing importance in diagnostic radiology of the liver diseases. The DCE MRI usually takes long time and the shape and position of liver is varied because of respiratory motion. Breath-holding can significantly get rid of the intra-phase motion. However the inter-phase non-rigid deformations still exist.The purpose of this study is to develop a robust 3D deformable registration method that can be employed to define the distortion between phases of DCE MRI and compensate it. Ideally, as a result, the liver will be located at the exact position with the exact shape (include vessels) of the reference image after registration. The proposed registration technique includes five steps (Figure 1): pre-processing, liver segmentation, vessel segmentation, Coherent Point Drift (CPD) match and B-spline based registration. In the pre-processing step, both reference and target data sets were re-sampled to be isotropic and a Gaussian smooth filter was used to de-noise them in 2D level (transverse plane). Fuzzy C-means (FCM) is a clustering technique which divides a given set of data into different groups [1]. In this case, it was used to distinguish liver tissues from other tissues based on their intensity differentiation. The mathematical morphology method was then used three times with different structural elements (shape and size) to remove irrelevant tissues or noises [2, 3]. Vessels were segmented by de-noising the non-liver regions inside of the corresponding liver regions using a thresholding method. The boundaries of segmented livers and vessels in 2D images were reconstructed into 3D point clouds for the reference and target data respectively and the CPD technique was used to find the correspondences between them [4, 5]. A B-spline based free form deformation (FFD) method was then applied based on this match to find the best distortions between them [6].As shown in Figure 2 and 3, with appropriate parameter settings, liver tissues and vessels inside of livers in the target data set (end-expiratory DCE MRI) were registered and deformed based on the information in the reference data set (end-inspiratory DCE MRI) automatically. The SSD of manually defined boundary points of vessels and livers before and after registration was shown in Figure 4 to assess the performance of the proposed method. The SSD for both livers and vessels was significantly reduced through the registration process (39.77% for liver and 54.38% for vessels by average). And the overlap rate of liver volumes was increased from 43.19% before registration to 93.62% after registration.The registered results by the proposed technique were checked in both transverse and coronal plane and it was considered to be clinically acceptable for test data sets. In DCE MRI studies, an accurate knowledge of the vascular signal intensity as a function of time is crucial for the estimation of kinetic parameters. In this study, as shown in Figure 3 and 4, not only liver volume was replaced and deformed accurately, the vessels were also well registered. The proposed registration method can identify both the rigid and non-rigid differences between livers in DCE MR images taken at different time points fully automatically. As a result, it can deform the target data set and track the shape and position variation of livers as well as the vessels inside of them accurately.