Research On Non-rigid Cardiac Image Registration Based On Statistical Shape Model

With the development of the social economy and the change of lifestyle,especially for the accelerated process of population aging,the epidemic trend of cardiovascular disease is obvious,and the morbidity and mortality have been increasing,which has become the main risk factors threatening human health.The prevention and early diagnosis of cardiovascular disease are of great significance to reduce the mortality in clinical practical,a variety of images to analyze cardiac function parameters,assist in the diagnosis,and treat heart-related diseases have been widely used.Medical image analysis methods have played an important role in this process.As an important part of medical image analysis methods,image registration can post-process images of different phases or modes and effectively fuse the spatial structure and anatomical information between images.It has extensive research and application value in detecting disease changes,guiding clinical diagnosis and disease treatment.Many clinical applications depend on deformation registration,and the choice of similarity measurement is the key factor to determine whether accurate registration can be achieved.According to the similarity measurement,medical image registration is generally divided into intensity information-based and feature-based methods.The registration based on intensity information is to determine the registration quality using the similarity measured by intensity information between images.however,this method takes into account the spatial relationship between images,target structure deformation and other effects less.moreover,mutual information is not a convex function,it is easy to fall into the risk of misalignment caused by local optimization in the process of registration.Feature-based methods mainly use different feature descriptors or similarity measurement criteria to determine the accuracy of registration,which depends on the difference of feature extraction.In this paper,based on the mutual information,the target shape prior feature is integrated to guide the multi-modal registration process of heart image,Moreover,we make a more in-depth study on the specific effect and performance of the proposed algorithm.The main research contents of this paper can be summarized as follows:1.Because the method using mutual information only considers the intensity statistical relationship between images,the registration effect is not ideal in multimodal image registration with intensity inhomogeneities.In this study,a registration method between multimodal cardiac images based on statistical shape model constraints is proposed.First,we build a statistical shape model based on the extraction of target shape feature points from atlas label images.It is completed through the establishment of the label image training set,the extraction of the shape feature points of the template label image,and the automatic propagate of the template shape to the shape feature points to be marked.Then,based on the statistical shape model,we consider the model as a shape prior constraint and construct a new cost function to guide the process of multimodal heart image registration.Finally,the specific experimental results show that the registration accuracy of the proposed method is significantly higher than that of the registration method using mutual information only.2.In order to overcome the high complexity of computation and reduce the runtime of registration,we optimize the statistical shape model algorithm based on the idea of parallel computing,and propose a fast registration algorithm based on statistical shape model constraint.First,we make a detailed analysis of the specific calculation process of the statistical shape model algorithm,and elaborate the specific reasons for the timeconsuming of the statistical shape model algorithm.Then,based on the multi-thread parallel computing method,the gradient calculation process which leads to the timeconsuming of the algorithm is optimized and implemented.Finally,the effectiveness of the method is validated by experiments,and the performance of the algorithm is greatly improved when keeping the accuracy of the registration.