Research On Key Technologies Of Three-dimensional Brain Magnetic Image Registration

The main purpose of medical image registration is to carry out appropriate spatial transformation for a series of images in the same scene shot by the same equipment or the same scene shot by different equipment,so that the corresponding points between two or more images are aligned in the spatial position.The application of medical image registration technology to medical image analysis can assist in disease diagnosis and surgical treatment.But because of the different imaging techniques,many different modes of medical images are generated.If the imaging advantages of multiple modal images can be combined effectively,more accurate and complementary information can be provided,thus providing a more comprehensive basis for doctors in disease diagnosis and surgical treatment.The best way to solve the above problem is medical image registration technology.However,most image registration methods still have problems such as low accuracy and poor stability in the case of large offset.Therefore,it is of great significance to study how to realize accurate and stable multi-mode image registration.In order to solve the problem of poor pairwise registration in large offset due to relatively single feature point extraction,this paper proposes a pairwise registration method for 3D NMR brain images,which includes a process from coarse to fine.First,the global coarse registration is carried out,and then the local fine registration is carried out.In the overall rough registration stage,feature points are extracted from the image to form a set of feature points,then the feature point set is transformed by Principal Component Analysis.Then the Iterative Closest Point algorithm is used for rough registration to reduce the initial offset of the image.In the local fine registration stage,the pixels of the Region of Interest were modeled by the joint gray histogram,and the registration accuracy was further improved by the mutual information maximization criterion.The experimental results show that the proposed method can be effectively applied to the paired registration of multi-modal 3D brain MR images,and the accuracy and accuracy of the registration are improved significantly.In view of the limitation of paired registration in some specific scenes,collective registration is needed to meet the application requirements.In this paper,a new multimode 3D rigid medical image collective registration method combining feature points and voxel information is proposed.The method using Gaussian mixture model as well as the image intensity and distribution of the image feature point information modeling,it is the integration of these models to a likelihood function,and through the Expectation Maximization Algorithm estimated registration parameters,through the three medical image data set on the experimental results show that the method has good performance in small offset scope.In addition,this method has higher success rate and accuracy than other methods at large offset.