Research On Registration Techniques For Complex Deformation And Nonlinear Gray-scale Differences Issues Of Medical Images

Medical image registration is a critical step in medical image processing and analysis.It plays an important role in completing image fusion,assisting clinical diagnosis,predicting the postoperative compensatory effect,tracking pathological changes,evaluating therapeutic effect and so on.However,the characteristics of medical images,such as the complexity of deformation,the diversity of modalities,and the nonlinearity of gray-scale differences,etc.,make the medical image registration techniques extremely challenging.Most of the existing medical image registration techniques either ignore the rich spatial structure information contained in the image,or do not have enough consideration of the characteristics of the nonlinear gray-scale differences invariance in the image,which leads to medical image registration under complex deformation and nonlinear gray-scale differences having the problems of slow convergence speed,low registration accuracy,and weak robustness.Aiming at these problems,this paper has carried out a research on registration techniques for complex deformation and nonlinear gray-scale differences issues of medical images.This paper has put forward a series of improved algorithms by means of the deep research and analysis of the existing key techniques of medical image registration under complex deformation and nonlinear gray-scale differences.The main research contributions of this paper are as follows:(1)The traditional mutual information only considers the globally consistent gray-scale statistical characteristics of the image,but ignoring the spatial structure information,which is easy to result in the defects like registration errors and so on.This paper puts forward a multi-modal image registration algorithm based on Local Structure Tensor-Mutual Information(LST-MI).In order to introduce the spatial structure information of the image,this paper combines the mutual information with the extracted image structure contribution intensity information based on the local structure tensor.Then it constructs a new similarity measure LST-MI.Subsequently,this paper takes LST-MI as the target energy function based on the rigid transformation registration model,and uses the optimized strategy of steepest descent method to seek the globally optimal model parameters,thus realizing the registration of multi-modality images.Next,this paper explores the problem of non-rigid registration of lung organs with complex deformation from different angles.The non-rigid registration process is all carried out on the rigid registration platform set up by the algorithm.(2)Since there is relatively serious deformation between CT images of chest follow-up,they have problems such as weak registration robustness,low accuracy and so on.In order to solve the above problems,this paper studies the regularization model based on edge-preserving smoothing filter,and proposes an image registration algorithm constrained by new regularization model based on HDCS(Hybrid Diffusion filter with Continuous Switch).Firstly,this paper uses the HDCS smoothing to replace the Gaussian smoothing.The purpose is to achieve the regularization of the deformation displacement field.And a new regularization method is constructed.Then,the paper uses this new regularization method for deformation registration model based on diffeomorphism Demons in order to achieve registration.This new regularization method can avoid the over-smooth phenomenon of the deformation displacement field during the registration process.Consequently,it reduces the risk of the registration process falling into the local extremum and improves the robustness and accuracy of the registration.(3)The Log-Demons registration algorithm and its improved algorithms only used SSD as the similarity item of the registration model,and ignored the similarity measurement of the local structural characteristics,which made their registration performance of large and complex deformed images still not ideal.For this reason,a medical image registration algorithm based on Log-Euclidean covariance matrix descriptor is proposed in this paper.Firstly,based on the Log-Demons registration model,this paper builds the local structure descriptor LECM(Log-Euclidean Covariance Matrices)with rotation,scaling and scale invariance.Then it uses the Euclidean distance between the LECM descriptor logarithm of the reference image and the LECM descriptor logarithm of the floating image as a new match item.Finally,it adds the matching item to the objective function of the Log-Demons registration model to realize image registration.The new match provides structural constraints for the update of the deformation displacement field during the registration process.It also maintains the differentiability of the new objective function,and improves the registration robustness and registration accuracy.(4)For the problems of low registration accuracy and poor robustness caused by larger nonlinear gray-scale differences between multi-modality medical images,this paper proposes a multi-modality medical image registration algorithm based on local phase means and phase congruency values of different orientations in consideration of the invariance of frequency characteristics to nonlinear gray-scale differences.In order to resist the impact of larger nonlinear gray-scale differences,this paper builds LPPCO(Local Phase means and Phase Congruency of different Orientations,LPPCO)characteristic descriptors with the nonlinear gray-scale difference invariance by using local phase means and phase congruency characteristics of different orientations.And the normalized cross Correlation(NCC)between LPPCOs is used as a similarity measure of the fast template matching model to realize registration.The experimental results show that the algorithms proposed in this paper have stronger robustness,higher accuracy and faster convergence speed for image registration with complex deformation and nonlinear gray-scale differences.Therefore,they have a great reference value for perfecting the medical image registration theory,expanding the registration application field,etc.