DSA image registration method fusing moving target features

Abstract

The invention relates to the field of medical image processing, and aims to provide a DSA image registration method fusing moving target features. Comprising the steps that a DSA image registration model fusing blood vessel information is established, and the model comprises a similarity measurement item, a regularization item of a displacement field and a regularization item of the blood vessel information; taking the DSA angiography image and the background image as the input of a model, and discretizing on a node grid; discretizing the model by using a difference operator to obtain an optimization model; decomposing and optimizing the model through an alternating direction algorithm, iteratively solving a displacement field and updating blood vessel information, and outputting a DSA subtraction image after a stop condition is met. The registration model provided by the invention can effectively cope with artifacts caused by rigid and non-rigid image deformation, more accurately estimate a displacement field, extract a vascular structure and improve the image registration accuracy; the alternating direction algorithm framework enables the solution of the displacement field to utilize the newest image registration algorithm, so that the optimization problem of blood vessel image updating has a closed-form solution, and the solution can be efficiently solved.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and particularly to a DSA image registration method integrating moving target features. Background Art

[0002] Digital Subtraction Angiography (DSA) is a vascular anatomical structure visualization technology, which is widely used in the diagnosis and treatment of vascular diseases. Its basic principle is to use DR (Digital Radiology) technology to continuously image the flow process of the contrast agent injected into the blood vessels (referred to as contrast images) to obtain vascular information. Due to the influence of the depth overlap of DR images and complex background information (such as bones), the quality of vascular imaging is not high. Therefore, by performing a subtraction operation on the contrast image and the DR image before contrast agent injection (referred to as the background image), the influence of background information can be removed, and high-quality vascular images or vascular information can be obtained. However, the inevitable organ movement of the imaging target (such as breathing, heartbeat, swallowing) causes the spatial positions of organs at different time points to be different. Therefore, there are differences and changes between the background information of the contrast image and the background image, and artifacts inevitably exist in the directly subtracted image, resulting in a reduction in the quality of vascular imaging. Therefore, the usual practice is to first register the background image and the contrast image, and then perform a subtraction operation.

[0003] Image registration is a classic problem in the field of image processing, and its purpose is to align images at different times, different modalities, and different perspectives in space. The variational model of image registration can be described as the following optimization problem:

[0004]

[0005] In the formula, u represents the displacement field; R represents the fixed reference image; T represents the deformable template image; T(u) represents the image of T deformed by the displacement field; represents the similarity metric, which measures the similarity between the image R and the image T(u); represents the regularization term of the displacement field, which constrains the smoothness and reversibility of the displacement field. Specifically, this model means that by optimizing the displacement field u, the image T(u) is similar to the image R.

[0006] In fact, the DSA registration problem is different from the traditional image registration problem. The contrast image of DSA imaging not only contains the background image information but also includes the vascular information S after contrast agent injection, that is, the contrast image where is the background information corresponding to the current contrast image. Different from directly aligning two complete images in traditional image registration, the core task of DSA image registration is to estimate the background information of the contrast image starting from the background image T acquired before injecting the contrast agent through registration technology and then extract the vascular image from the contrast image, that is Therefore, DSA image registration is not exactly the same as the traditional image registration problem. If directly using the traditional image registration method to register the DSA contrast image and the background image, it will lead to inaccurate displacement field estimation and loss of small blood vessels, affecting the accuracy of imaging. Therefore, it is necessary to propose a new DSA image registration method based on the vascular information in the image

[0007] Therefore, the present invention intends to propose an innovative technical solution, fuse the vascular information as the moving target feature into the DSA image registration model, and design a corresponding solution algorithm for the newly established model to solve the above problems Summary of the Invention

[0008] The technical problem to be solved by the present invention is to overcome the deficiencies in the prior art and provide a DSA image registration method that fuses moving target features

[0009] To solve the technical problem, the solution of the present invention is as follows

[0010] Provide a DSA image registration method that fuses moving target features, including the following steps

[0011] (1) Establish a DSA image registration model that fuses vascular information, and this model includes a similarity metric term, a regularization term of the displacement field, and a regularization term of vascular information

[0012] (2) Use the DSA contrast image and the background image as the input of the model and discretize them on the node grid; use the difference operator to discretize the model to obtain an optimization model

[0013] (3) Decompose the optimization model through the alternating direction algorithm, iteratively solve the displacement field and update the vascular information, and output the DSA subtracted image after reaching the stopping condition

[0014] As a preferred solution of the present invention, in the step (1), the mathematical model of the DSA image registration model is shown in formula (1):

[0015]

[0016] In the formula, E(u, S) represents the energy function related to the displacement field and vascular information represents the similarity metric term represents the regularization term of the displacement field; ‖S‖ 1The regularization term representing vascular information; α represents the regularization parameter of the displacement field; β represents the regularization parameter of the vascular information.

[0017] As a preferred solution of the present invention, in the step (1), in order to make the background image T and the contrast image R similar in the background image after removing the vascular information S, the displacement field should be minimized, and the following formula is established to represent the similarity metric term: Similar, the displacement field should be minimized, and the following formula is established to represent the similarity metric term:

[0018]

[0019] In the formula, represents the similarity metric term; Ω represents the region of the image; u represents the displacement field, which is a two-dimensional vector function, that is T represents the template image to be registered, that is, the DSA background image; T(u) represents the image of the template image T deformed by the displacement field u; R represents the fixed reference image, that is, the DSA contrast image.

[0020] As a preferred solution of the present invention, in the step (1), in order to estimate the global displacement field and the local displacement field at different scales, an elastic regularization function is selected to constrain the displacement field, and the following formula is established to represent the regularization term of the displacement field:

[0021]

[0022] In the formula, represents the regularization term of the displacement field; λ and μ are settable parameters, representing the bulk modulus and the shear modulus respectively; l represents the component index of the displacement field; ‖…‖ represents the L 2 norm; represents the gradient operator; div represents the divergence operator.

[0023] As a preferred solution of the present invention, in the step (1), the L 1 norm is used to characterize the sparsity of blood vessels, and the following formula is established to represent the regularization term of vascular information:

[0024]

[0025] In the formula, represents the regularization term of vascular information; S represents vascular information, that is, the vascular part in the contrast image; ‖…‖ 1 represents the L 1 norm.

[0026] As a preferred solution of the present invention, the step (2) specifically includes:

[0027] Step 2.1, Take the angiographic image as the reference image R and the background image as the template image T; after inputting into the DSA image registration model, discretize the two images on the node grid;

[0028] Step 2.2, Discretize the mathematical model of the DSA image registration model using the difference operator to obtain the optimization model shown in formula (2):

[0029]

[0030] In the formula, is the discrete form of the similarity metric term; the difference operator includes the gradient operator and the divergence operator.

[0031] As a preferred solution of the present invention, in step (3), decompose the discretized optimization model by the alternating direction algorithm to obtain a decomposition model (3) for iteratively solving the displacement field u and a decomposition model (4) for iteratively solving the vascular information S:

[0032]

[0033] In the formula, k represents the number of iterations; R represents the DSA angiographic image; S (k-1) represents the vascular image obtained in the (k - 1)-th iteration; R - S (k-1) represents the angiographic image after removing the vascular structure obtained after (k - 1) iterations; u represents the displacement field to be solved; T(u) represents the deformed background image to be solved; S represents the vascular image to be solved; T(u (k) ) represents the deformed background image obtained in the k-th iteration.

[0034] As a preferred solution of the present invention, perform the following solving steps for the decomposition model (3) and the decomposition model (4):

[0035] Step 3.1, For the decomposition model (3), downsample the background image T and the image R - S (k-1) to obtain two columns of image pyramids; sequentially estimate the displacement field u (k) from top to bottom on the image pyramids using the Gauss - Newton method, and update the step size factor using the Armijo line search during this process;

[0036] Step 3.2, Apply the displacement field u (k) obtained in step 3.1 to the background image T, and use the cubic spline interpolation method for interpolation to obtain the deformed image T(u (k) );

[0037] Step 3.3, For the decomposition model (4), use the image T(u (k)) Substitute and use the soft-threshold function to quickly update the vascular information S. Denote b = R - T(u (k) ), then

[0038]

[0039] In the formula, soft(…) represents the soft-threshold function; β represents the regularization parameter of the vascular information;

[0040] Step 3.4, check whether the condition of the maximum number of iterations for the preset stop is reached;

[0041] If not reached, re-execute Step 3.1 and Step 3.2 to update the displacement field u (k) and the registered background image T(u (k) ); then execute Step 3.3 to update the vascular information S (k) ;

[0042] Re-check whether the condition of the maximum number of iterations for the preset stop is reached;

[0043] If reached, terminate the calculation of the alternating direction algorithm and output the final DSA subtracted image, which contains the vascular information S with the best effect.

[0044] The present invention further provides a computing device, including:

[0045] A memory configured to store instructions; and

[0046] A processor configured to call the instructions from the memory and be able to implement the foregoing DSA image registration method for fusing moving target features when executing the instructions.

[0047] The present invention also provides a computer-readable storage medium, on which instructions are stored, and the instructions are used to cause a computer to execute the foregoing DSA image registration method for fusing moving target features.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] 1. The important innovation of the present invention is to propose a new DSA image registration model framework. This model fully considers the particularity of the DSA image registration problem and fuses vascular information in the model. Specifically, the present invention improves the similarity measurement term of the traditional registration model, focuses on considering the sparsity characteristics of blood vessels, and introduces a sparsity characterization method; this model can effectively handle the artifacts brought by rigid and non-rigid image deformations, more accurately estimate the displacement field, more precisely extract the vascular structure, and improve the accuracy of DSA image registration.

[0050] 2. For the model of the present invention, the present invention proposes an alternating direction algorithm framework, enabling the displacement field solving algorithm to utilize the latest existing image registration algorithms.

[0051] 3. The present invention uses the alternating direction algorithm to make the optimization problem of vascular image update have a closed-form solution, which can be efficiently solved without affecting the overall execution efficiency of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is the execution framework of the DSA registration algorithm proposed by the present invention.

[0053] Figure 2 is the calculation example of the registration model proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0054] First of all, it should be noted that the present invention relates to image processing technology, which is an application of computer technology in the field of medical image recognition and processing. In the implementation process of the present invention, the application of multiple software function modules will be involved. The applicant believes that after carefully reading the application documents and accurately understanding the implementation principle and invention purpose of the present invention, and in combination with the existing well-known technologies, those skilled in the art can fully implement the present invention by using their mastered software programming skills. All that is mentioned in the application documents of the present invention belongs to this category, and the applicant will not list them one by one.

[0055] The DSA image registration method integrating moving target features of the present invention will be described in detail below.

[0056] Step 1: Establish a DSA image registration model integrating vascular information.

[0057] (1) Similarity metric term

[0058] The present invention proposes that DSA image registration has single-modal characteristics and unique task requirements, that is, separating vascular information. On the one hand, as a single-modal image sequence, the pixel value distributions of the background image T and the angiographic image R have inherent consistency, and the gray values are comparable. Therefore, it is reasonable to use the "sum of squares of pixel intensity residuals". On the other hand, as mentioned above, the essence of DSA registration is not to directly align T and R, but to align the background image T to the background information in the angiographic image through the displacement field u Therefore, the background image T and the background image of the angiographic image R after removing the vascular information S are similar, that is, the displacement field should be minimized; thus, the following formula is established to evaluate the similarity metric term:

[0059]

[0060] In the formula, Represents the similarity metric term, which is used to measure the similarity between the angiographic image (R - S) after removing vascular information and the image T(u). Ω represents the region of the image; u represents the displacement field, which is a two - dimensional vector function, that is T represents the template image to be registered (i.e., the DSA background image); T(u) represents the image obtained by deforming the template image T through the displacement field u; R represents the fixed reference image (i.e., the DSA angiographic image).

[0061] (2) Regularization term of the displacement field:

[0062] During the DSA imaging process, there is not only the relative motion between the human body and the DSA device (global), but also the local motion of human organs. In order to improve the well - posedness of the optimization problem, it is necessary to estimate the global displacement field and the local displacement field at different scales and add regularization constraints to the displacement field. For the above reasons, the regularization function for constraining the displacement field in the present invention selects the elastic regularization term:

[0063]

[0064] In the formula, represents the regularization term of the displacement field; u represents the displacement field; λ and μ are adjustable parameters, representing the bulk modulus and the shear modulus respectively; l represents the component index of the displacement field; ‖…‖ represents the L 2 norm; represents the gradient operator; div represents the divergence operator.

[0065] (3) Regularization term of the vascular information:

[0066] The distribution of blood vessels in the human body has the characteristic of sparsity. The present invention uses the L 1 norm to characterize the sparsity, that is:

[0067]

[0068] In the formula, represents the regularization term of the vascular information; S represents the vascular information, that is, the vascular part in the angiographic image; ‖…‖ 1 represents the L 1 norm.

[0069] Based on the above three main considerations, the present invention establishes a DSA image registration model for vascular image information; the mathematical model of this model is shown in formula (1):

[0070]

[0071] In the formula, E(u, S) represents the energy function related to the displacement field u and the vascular information S; represents the similarity metric term; The regularization term representing the displacement field; ‖S‖ 1 The regularization term representing the vessel information; α represents the regularization parameter of the displacement field; β represents the regularization parameter of the vessel information.

[0072] The above DSA registration model constructed based on the variational framework can solve the two variables of the vessel information S and the displacement field u using the alternating direction algorithm after discretization.

[0073] Step 2: Discretize the image and the DSA registration model respectively.

[0074] Input the DSA contrast image R and the background image T into the DSA registration model, and discretize the two images on the node grid; use the difference operator to discretize the DSA registration model to obtain the optimization model; specifically:

[0075] Discretize the image region Ω using a grid with a step size of h, and discretize the images R and T on the node grid. Assume there are n 1 grid points in the horizontal direction and n 2 grid points in the vertical direction.

[0076] Use the difference operator to discretize the DSA registration model represented by formula (1) in Step 1. For convenience, the same symbols as in the previous text are used in the following formulas, but they represent the discrete form.

[0077] First, discretize the gradient operator at each pixel (i, j) position:

[0078]

[0079] where

[0080]

[0081] In the formula, l = 1, 2 represents the index of the displacement field component;

[0082]

[0083] Based on the conjugacy of the gradient operator and the divergence operator, the discretization of the divergence operator is:

[0084]

[0085] The discretization of the elastic regularization term is:

[0086]

[0087] The discrete form of the similarity metric term is:

[0088]

[0089] Based on the above processing, after discretizing the DSA registration model represented by formula (1), the optimized model shown in formula (2) is obtained:

[0090]

[0091] In the formula, is the discrete form of the similarity metric term; the difference operator includes a gradient operator and a divergence operator.

[0092] Step 3: Decompose the optimized model through the alternating direction algorithm, iteratively solve the displacement field, update the vascular information, and then output the DSA subtracted image.

[0093] For the optimized model shown in formula (2), the present invention uses the alternating direction algorithm framework to decompose it into the following two decomposed models for iterative solution:

[0094]

[0095] In the decomposition model (3), k represents the number of iterations; R represents the DSA angiography image; S (k-1) represents the vascular image obtained in the (k - 1)-th iteration; R - S (k-1) represents the angiography image after removing the vascular structure obtained after (k - 1) iterations; u represents the displacement field to be solved; T(u) represents the deformed background image to be solved; in the decomposition model (4), S represents the vascular image to be solved; T(u (k) ) represents the deformed background image obtained in the k-th iteration.

[0096] The alternating direction algorithm framework adopted by the present invention can be summarized as Algorithm 1:

[0097] (1) Initialize S (0) = 0, set the maximum number of iterations and other stopping conditions;

[0098] (2) Solve model (3) in sequence to update the displacement field u (k) , solve model (4) to update the vascular information S (k) ;

[0099] (3) Check the stopping conditions. If satisfied, go to step (4); if not satisfied, go to step (2);

[0100] (4) Output the DSA subtracted image.

[0101] Next, describe the specific solution methods for each decomposition model:

[0102] Step 3.1, for the decomposition model (3), for the background image T, the image R - S (k-1)Perform downsampling to obtain a two-column image pyramid; estimate the displacement field u from top to bottom on the image pyramid using the Gauss-Newton method in sequence. (k) During this process, the Armijo line search is used to update the step size factor.

[0103] Use the Gauss-Newton method combined with the multi-scale method to solve the decomposition model (3), and calculate the differential forms of the similarity term and the displacement field regularization term.

[0104] Denote the Jacobian matrix of the image T with respect to the displacement field u as T u , and the first-order differential of the similarity metric term is:

[0105]

[0106] The second-order differential is approximated in the following form:

[0107]

[0108] The first-order differential of the displacement field regularization term is in the following form:

[0109]

[0110] In the formula, represents the Laplace operator in discrete form;

[0111] Therefore, the Jacobian matrix and the approximate Hessian matrix of the objective function in the decomposition model (3) are respectively:

[0112]

[0113] In the formula, J(u) represents the Jacobian matrix of the objective function E(u); H(u) represents the approximate Hessian matrix of the objective function E(u).

[0114] Then, by solving the following algebraic equations:

[0115] H(u)δu = -J(u)

[0116] Obtain the descent direction δu, and use the Armijo line search to determine the descent step size γ.

[0117] Finally, obtain the update formula for the displacement field:

[0118] u (t+1) = u (t) + γδu (t)

[0119] During this process, the Gauss-Newton method on the fixed layer can be summarized as Algorithm 2:

[0120] (1) Calculate E(u(t) ), J(u (t) ) and H(u (t) );

[0121] (2) Test stop condition;

[0122] (3) Solve the linear equation system H(u (t) )δu (t) = -J(u (t) );

[0123] (4) Update the step size factor γ using Armijo line search, let u (t) = u (t) + γδu (t) ;

[0124] (5) Update the iteration value, u (t+1) = u (t) .

[0125] The stop conditions of Algorithm 2 are as follows:

[0126] (1) Stop(1) = |(E old - E curr )| ≤ 10 -3 (1 + abs(E stop ))

[0127] (2) Stop(2) = |(u old - u curr )| ≤ 10 -2 (1 + u 0 )

[0128] (3) Stop(3) = |(dE curr )| ≤ 10 -2 (1 + abs(E stop ))

[0129] (4) Stop(3) = |(dE curr )| ≤ eps

[0130] (5) Stop(5) = Exceed the maximum number of iterations

[0131] where, E old refers to the objective function value of the previous step, E curr is the current function value; E stop the value of the objective function at u = 0, i.e., when the image has not deformed; u old is the iteration value of the previous step, u curr is the current iteration value; dE curr is the current Jacobian matrix of the objective function; eps represents the machine precision, abs(…) absolute value.

[0132] In the present invention, the Armijo line search algorithm can be summarized as Algorithm 3:

[0133] (1) Input the initial values u (t) and δu (t) , and calculate E(u (t) ) and J(u (t) ) respectively; let γ = 1, the maximum number of iterations be set to 10 times, and η = 10 -4 ;

[0134] (2) Let u (t+1) = u (t) + γδu (t) , and calculate E(u (t+1) );

[0135] (3) Check the termination condition. If , then stop the iteration, otherwise go to step (4);

[0136] (4) Let γ = γ / 2, and go to step (2).

[0137] Downsample the background image T and the contrast image R to obtain two columns of image pyramids. Use the Gauss-Newton method with Armijo line search to solve the displacement field on each layer of the pyramid; then bilinearly interpolate the displacement field to a finer layer as the initial value, and continuously iterate until the stop condition is reached to obtain the displacement field u.

[0138] The calculation process of this step can be understood as a multi-scale calculation process. First, guess an initial value on the coarsest layer, or the initial value can also be set to zero; then use the Gauss-Newton method with Armijo line search to solve model (3), bilinearly interpolate the solution on the coarse layer to the fine layer as the initial value, and then use the Gauss-Newton method to solve; repeat this process until reaching the finest layer and satisfying the stop condition.

[0139] The multi-scale calculation method of the image pyramid can be summarized as Algorithm 4:

[0140] (1) Given the initial input, downsample the input images (R - S (k-1) ) and the image T to obtain the image pyramid.

[0141] (2) Guess the initial value u (0) on the coarsest layer.

[0142] (3) If not on the coarsest layer, use Algorithm 2 to solve for u.

[0143] (4) Interpolate u to a finer layer using bilinear interpolation.

[0144] (5) If already at the finest layer, stop.

[0145] Solve the optimization model (3) by Algorithm 4 to obtain the displacement field u (k) .

[0146] Step 3.2: Apply the displacement field u obtained in Step 3.1 (k) to the background image T and perform interpolation using the cubic spline interpolation method to obtain the deformed image T(u (k) ).

[0147] Step 3.3: For the decomposition model (4), substitute the image T(u (k) ) obtained in Step 3.2 and use the soft threshold function to quickly update the vascular information S

[0148] Let b = R - T(u (k) ), then

[0149]

[0150] where soft(…) represents the soft threshold function; β represents the regularization parameter of the vascular information;

[0151] The soft threshold function is a closed-form solution and will not affect the overall solution efficiency of the algorithm in terms of time complexity.

[0152] Through the above calculations, the vascular information S is obtained (k) .

[0153] Step 3.4: Check whether the condition of the preset maximum number of iterations for stopping is reached.

[0154] If not, re-execute Step 3.1 and Step 3.2 to update the displacement field u (k) and the registered background image T(u (k) ); then execute Step 3.3 to update the vascular information S (k) ;

[0155] Re-check whether the condition of the preset maximum number of iterations for stopping is reached;

[0156] If reached, terminate the calculation of the alternating direction algorithm and output the final DSA subtracted image, which contains the vascular information S with the best effect.

[0157] Based on a newly designed DSA registration model and variational framework, the present invention effectively solves the problem of image distortion caused by rigid and non-rigid deformations through an iterative optimization strategy. Experimental results show that this method can significantly improve the visualization quality of vascular structures in DSA images, enabling the clear presentation of vascular details originally masked by artifacts. Figure 2The calculation example of the registration model proposed by the present invention is shown, including a background image, a contrast image, a direct subtraction image, a subtraction image obtained by using the existing registration method, and a subtraction image obtained by the registration method proposed by the present invention. It can be seen from the figure that compared with the processing result of the prior art, the subtraction image finally obtained by the solution of the present invention has significantly less noise while the blood vessels are clearly visible, and the structure of the blood vessels is better retained.

[0158] From the above, it can be seen that the present invention proposes a brand-new DSA image registration model that fuses the features of moving objects. This model improves the similarity measurement term of the traditional image registration model and introduces the prior information (sparsity) of blood vessels. By adopting the alternating direction algorithm, the two key variables in the model, "displacement field u and blood vessel information S", are effectively decoupled. The solution of the displacement field can be flexibly combined with the existing DSA image registration algorithm, and there is a fast solution algorithm for the solution of blood vessel information, which will not affect the overall execution efficiency of the algorithm, thus simplifying the solution process of the optimization problem. The image registration model and algorithm proposed by the present invention are not only innovative in theory, but also show strong flexibility and wide applicability in practical applications.

[0159] In addition, as common knowledge that can be understood by those skilled in the art, a computing device is required to implement the method of the present invention. The computing device includes: a memory configured to store instructions; and a processor configured to call the instructions from the memory and be able to implement the aforementioned DSA image registration method that fuses the features of moving objects when executing the instructions. At the same time, a computer-readable storage medium is also required to implement the method of the present invention. Instructions are stored on the computer-readable storage medium, and the instructions are used to cause the computer to execute the method. The computing device can be a personal computer, a server, or a network device, etc. The computer storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0160] The specific embodiments described above further elaborate on the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. On the contrary, within the spirit and principles of the present invention, any modifications, equivalent replacements, improvements, etc. should be included in the protection scope of the present invention. The specific embodiments of the present invention are described above. It should be understood that the present invention is not limited to the above specific implementation manners, and those skilled in the art can make various deformations or modifications within the scope of the claims, which do not affect the essence of the present invention.

Claims

1. A DSA image registration method integrating moving target features, characterized in that: The following steps are involved: (1) Establishing a DSA image registration model integrating vascular information, which includes a similarity measurement term, a regularization term of the displacement field, and a regularization term of the vascular information; (2) The DSA angiography image and the background image are used as the input of the model and discretized on the node grid; the model is discretized using a difference operator to obtain an optimized model; (3) The model is decomposed and optimized through the alternating direction algorithm, the displacement field is iteratively solved and the vascular information is updated, and the DSA subtraction image is output after the stopping condition is reached.

2. The method according to claim 1, characterized in that In the step (1), the mathematical model of the DSA image registration model is shown in formula (1): Where E(u,S) represents the energy function related to the displacement field and vascular information; represents the similarity measure; represents the regularization term of the displacement field; ‖S‖1 represents the regularization term of the vascular information; α represents the regularization parameter of the displacement field; β represents the regularization parameter of the vascular information.

3. The method according to claim 1, characterized in that In the step (1), the background image T and the contrast image R are made to be the background image after the vascular information S is removed. Similarity, the displacement field should be minimized, and the following formula is established to express the similarity measurement item: In the formula, represents the similarity measure; Ω represents the area of ​​the image; u represents the displacement field, which is a two-dimensional vector function, that is, u=(u1,u2) μ ; T represents the template image to be registered, that is, the DSA background image; T(u) represents the image after the template image T is deformed by the displacement field u; R represents the fixed reference image, that is, the DSA angiography image.

4. The method according to claim 1, characterized in that: In the step (1), in order to estimate the global displacement field and the local displacement field at different scales, an elastic regularization function is selected to constrain the displacement field, and the following formula is established to represent the regularization term of the displacement field: In the formula, represents the regularization term of the displacement field; λ and μ are configurable parameters, representing the bulk modulus and shear modulus respectively; l represents the component index of the displacement field; ‖…‖ represents the L2 norm; represents the gradient operator; div represents the divergence operator.

5. The method according to claim 1, characterized in that In step (1), the L1 norm is used to characterize the sparsity of blood vessels, and the following formula is established to represent the regularization term of blood vessel information: In the formula, represents the regularization term of vascular information; S represents the vascular information, that is, the vascular part in the angiography image; ‖…‖1 represents the L1 norm.

6. The method according to claim 1, characterized in that The step (2) specifically includes: Step 2.1, the contrast image is used as the reference image R, and the background image is used as the template image T; after inputting into the DSA image registration model, the two images are discretized on the node grid; Step 2.2, use the differential operator to discretize the mathematical model of the DSA image registration model to obtain the optimization model shown in formula (2): In the formula, is a discrete form of the similarity measurement item; the difference operator includes a gradient operator and a divergence operator.

7. The method according to claim 1, characterized in that In the step (3), the optimization model obtained by discretization is decomposed by an alternating direction algorithm to obtain a decomposition model (3) for iteratively solving the displacement field u and a decomposition model (4) for iteratively solving the vascular information S: Where, k represents the number of iterations; R represents the DSA angiography image; S (k-1) represents the vascular image obtained at the k-1th iteration; RS (k-1) represents the angiographic image after the vascular structure is eliminated after k-1 iterations; u represents the displacement field to be solved; T(u) represents the deformed background image to be solved; S represents the vascular image to be solved; T(u (k) ) represents the deformed background image obtained at the kth iteration.

8. The method according to claim 7, characterized in that The following solution steps are performed for decomposition model (3) and decomposition model (4): Step 3.1: for the decomposition model (3), the background image T and the image RS (k-1) Downsampling is performed to obtain two columns of image pyramids; the Gauss-Newton method is used from top to bottom on the image pyramid to estimate the displacement field u (k) ,In this process, the step size factor is updated using Armijo line search; Step 3.2: Substitute the displacement field u obtained in step 3.1 into (k) Applied to the background image T, the cubic spline interpolation method is used to interpolate and obtain the deformed image T(u (k) ); Step 3.3: for the decomposition model (4), the image T(u (k) ) is substituted into the vascular information S using the soft threshold function, and b = RT(u (k) ),but Where soft(…) represents the soft threshold function; β represents the vascular information regularization parameter; Step 3.4, check whether the preset maximum number of iterations is reached; If it is not reached, re-execute steps 3.1 and 3.2 to update the displacement field u (k) and the registered background image T(u (k) ); then execute step 3.3 to update the blood vessel information S (k) ; Recheck whether the preset maximum number of iterations is reached; If it is reached, the calculation of the alternating direction algorithm is terminated, and the final DSA subtraction image is output, which contains the vascular information S with the best effect.

9. A computing device, characterized in that include: a memory configured to store instructions; as well as A processor is configured to call the instructions from the memory and implement the DSA image registration method for fusing motion target features in any one of claims 1 to 8 when executing the instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions for causing a computer to execute the DSA image registration method for fusing motion target features in any one of claims 1 to 8.

Images