Studies On Multi-modal Medical Image Registration Based On Structure Information

With the continuous development of medical technologies,image fusion in different modalities becomes an important issue to help doctors make scientific diagnosis.However,image fusion can only be performed after aligning different modal images to each other upon implementing registration procedure.Medical image registration is a process of seeking one or multiple spatial transformation so that any anatomical appears on the same spatial location in different-modal images.However,due to its intensity inconsistency over different-modal images,this technique is still a challenging task.In this thesis,we focus ourselves on studying different algorithms for multi-modal medical image registration.The main work and contributions are as follows:Firstly,due to the negligence of the structural information and the influence of the noise on the target tissues in most current mutual information based or local entropy descriptors based multi-modal image registration methods,local misalignment often happens.To resolve such a problem,we propose in this thesis a novel robust and adaptive binary pattern(Rabp)descriptor to extract the texture feature and convert multi-modal images into a unified image description.Based on this description,the deformation displacement field between multi-modal images is estimated in the framework of optical flow filed.Deformation and clinical registration experiments were performed on Brain Web and the Whole Brain Altas dataset.Experiment results indicate that the proposed algorithm outperforms the state-of-the-art algorithms in terms of both accuracy and robustness.Secondly,in order to improve both the accuracy and efficiency of the registration,this thesis proposes a novel method to extract image structure information.This method combines the gradient magnitude(extracted by Canny operator)and phase consistency to convert different modal-images into a common space so that image registration can be performed in the optical flow framework.As compared with the Rabp descriptor method,the FGP method needs a smaller amount of calculation and therefore can further improve the efficiency of registration without affecting the quality of registration.Experiments show that the FGP method can achieve better accuracy and speed than the method that combined local entropy descriptors.