A high-stability spine ultrasound and x-ray image registration method

Abstract

The application discloses a kind of high-stability spine ultrasound and X-ray image registration method, set from low to high multi-resolution registration grid, adopts gradient descent method, with the reciprocal of normalized mutual information similarity measure function as objective function, according to the order from low to high resolution, respectively, control point in different resolution registration grid is mutually optimized and calculated, and registration result X-ray image is obtained, registration result X-ray image obtained by registration of preceding low-resolution registration network is used as input X-ray image of subsequent high-resolution registration network, finally, registration result X-ray image generated by highest resolution registration grid is defogged using fuzzy perception attention network, and clear X-ray image after registration is output.The application can improve registration accuracy while ensuring registration stability, and greatly improve the visibility of image.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of medical image processing, and particularly relates to a high-stability spine ultrasound and X-ray image registration method. BACKGROUND

[0002] Currently, in the detection of scoliosis, the standing X-ray film with Cobb angle measurement is the gold standard for evaluating scoliosis. Curve progression monitoring usually involves regular X-ray scanning throughout childhood and adolescence. However, repeated exposure to X-ray radiation can have an impact on the body. Compared with X-ray photography, ultrasound has the advantages of no radiation, real-time and low cost. It is very promising in screening a large number of patients, monitoring progress and evaluating treatment effect. However, since ultrasound cannot penetrate bone structures, it uses surface reflection for imaging and cannot provide bone details. In contrast, X-ray films can present more details. Therefore, registering real-time ultrasound images with previous X-ray films can help doctors better understand the scoliosis of patients, which has important significance in clinical application.

[0003] Image registration is to find the best geometric transformation to best align the floating image with the fixed image. Existing registration methods are divided into deep learning-based registration methods and traditional registration algorithms. In deep learning technology, supervised transformation estimation registration and unsupervised transformation estimation registration have been widely applied. However, deep learning-based algorithms often perform well on a large amount of data, and have low interpretability. In the case of small amount of training data or unbalanced training data, it cannot produce more accurate result images.

[0004] In traditional medical image registration algorithms, feature extraction and feature matching-based registration methods are not universal in medical image registration tasks because the quality of feature points depends on the quality of the image. Image gray-based registration methods are based on image information and do not require complex preprocessing of multi-modal images, and are widely used in medical diagnosis and treatment. However, in the gray-based registration method, the Demons-based registration is not suitable for large deformation image registration tasks because the optical flow model will cause changes in the topological structure. The FFD-based registration algorithm has superior local deformation modeling capability and unlimited freedom characteristics, which can describe complex geometric shapes and better register irregular medical images, but the stability of the registration is poor.

[0005] The spine has the overall characteristics of multi-segment, multi-curved and multi-morphology. The human spine can be divided into four parts: neck, chest, waist and sacrum, specifically including 7 cervical vertebrae, 12 thoracic vertebrae, 5 lumbar vertebrae and 5 sacral vertebrae combined into one, and the basic anatomical structure of the spine also includes spinous process, transverse process, articular process, lamina, pedicle, etc., and there is a big difference in the structure of different segments of the spine, and the stability of the registration is relatively high, and it is still challenging to accurately register the spine ultrasound and X-ray images. The existing traditional registration method cannot solve the problem of bone tissue differentiation in registration, and cannot well display the lesion information, and cannot better adapt to the registration task of the spine ultrasound and X-ray images. SUMMARY

[0006] The purpose of the present application is to provide a high-stability spine ultrasound and X-ray image registration method to overcome the problem that the prior art cannot complete the high-stability registration of the spine ultrasound and X-ray images. Based on the FFD model, the registration grid is preprocessed before registration, each control point is calculated by mutual optimization during the registration process, and then mutual registration is performed. The registration result image is generated after multiple iterations, the fuzzy perception attention network (BANeT) is selected to perform deblurring processing, a clear result image is generated, and the high-stability registration task is completed.

[0007] In order to achieve the above-mentioned purpose, the technical scheme of the present application is as follows:

[0008] A high-stability spine ultrasound and X-ray image registration method, comprising:

[0009] A multi-resolution registration grid from low to high is set, the spine X-ray image is embedded into the lowest resolution registration grid, and the ultrasound image is embedded into each resolution registration grid;

[0010] A gradient descent method is used, the reciprocal of the normalized mutual information similarity measure function is used as the objective function, the control points in the registration grid of different resolutions are calculated by mutual optimization in order from low to high resolution, and the registration result X-ray image is obtained. The registration result X-ray image obtained by the low-resolution registration network is used as the input X-ray image of the subsequent high-resolution registration network;

[0011] The registration result X-ray image generated by the highest resolution registration grid is deblurred by the fuzzy perception attention network, and the clear X-ray image after registration is output.

[0012] Further, the high-stability spine ultrasound and X-ray image registration method further comprises:

[0013] The length-width ratio of the registration grid is adjusted based on the length-width ratio of the input image, and finally the input image is embedded into the registration grid.

[0014] Further, the mutual optimization calculation of the control points in the registration grid obtains a registration result X-ray image, comprising:

[0015] Performing forward registration:

[0016] Using the gradient descent method, taking the reciprocal of the normalized mutual information similarity measure function as the objective function, and taking the ultrasound image as the reference image to calculate the control point best result of the X-ray image to the ultrasound image, and taking the X-ray image as the reference image to calculate the control point best result of the ultrasound image to the X-ray image;

[0017] Performing weighted operation on the two control point best results to generate a control point weighted optimization result;

[0018] Constraining the control point weighted optimization result by using a piecewise function based on Tanh(X) to obtain a control point final optimization result set;

[0019] Using the final optimization result set to resample the X-ray image to obtain a forward registration result X-ray image after one registration;

[0020] Performing reverse registration:

[0021] Using the gradient descent method, taking the reciprocal of the normalized mutual information similarity measure function as the objective function, and taking the X-ray image as the reference image to calculate the control point best result of the ultrasound image to the X-ray image, and taking the ultrasound image as the reference image to calculate the control point best result of the X-ray image to the ultrasound image;

[0022] Performing weighted operation on the two control point best results to generate a control point weighted optimization result;

[0023] Constraining the control point weighted optimization result by using a piecewise function based on Tanh(X) to obtain a control point final optimization result set;

[0024] Taking the inverse operation on the final optimization result set, and then using the result set obtained by the inverse operation to resample the X-ray image to obtain a reverse registration result X-ray image after one registration.

[0025] Using the normalized mutual information similarity measure evaluation index to evaluate the performance of the forward registration result X-ray image and the reverse registration result X-ray image after one registration, and selecting an excellent registration result X-ray image as the input X-ray image for the next iteration;

[0026] Iterating registration in turn until the iteration termination condition is reached, ending iteration and outputting a final registration result X-ray image.

[0027] Further, the iteration registration in turn until the iteration termination condition is reached, ending iteration, comprising:

[0028] When the set number of iterations is reached or the absolute difference between the similarity measure indicators of the registered X-ray images and the ultrasound images of two adjacent registrations is less than the preset value for a preset number of consecutive times, the iteration is ended, and the registered X-ray image is output.

[0029] The application provides a high-stability spine ultrasound and X-ray image registration method, improves the FFD registration model, and combines a fuzzy perception attention network (BANet) to provide a high-stability spine ultrasound / X-ray image registration method, which achieves good results in experiments. In the registration task preprocessing of the spine image, a mutual optimization and mutual registration method is introduced to improve the registration accuracy and ensure the stability of the registration, and solve the problem of conical dissimilation deformation caused by overfitting. The fuzzy perception attention network (BANet) is combined to perform defogging processing on the above result image, which greatly improves the visibility of the image, and makes the result image have good clinical value. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 A high-stability spine ultrasound and X-ray image registration method flowchart is provided. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical scheme and advantages of the application clearer, the application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application and not to limit the application.

[0032] The application provides a high-stability spine ultrasound and X-ray image registration method, which mainly solves the four potential problems of low registration accuracy, poor stability, blurred result image and slow running speed of the traditional registration method. The spine ultrasound image refers to the human spine bone quality ultrasound image obtained by scanning with an ultrasound probe with spatial positioning function, and the volume projection is generated by reconstruction according to the spatial position information. The ultrasound probe with spatial positioning function refers to a two-dimensional ultrasound linear array probe with magnetic positioning markers or a two-dimensional ultrasound linear array probe fixed on a six-degree-of-freedom robot; the different depths refer to the depths from the body surface of the back, and the image contents contained in the ultrasound images at different depths are different, the imaging shallowest surface is the human back skin contour, and the imaging deepest is the spine bone information.

[0033] In one embodiment, as shown in Figure 1 A high-stability spine ultrasound and X-ray image registration method, comprising:

[0034] Step S1, set a multi-resolution registration grid from low to high, embed the spine X-ray image into the lowest resolution registration grid, and embed the ultrasound image into each resolution registration grid.

[0035] In the non-rigid registration process of the spine ultrasound and X-ray images, a multi-resolution registration grid from low to high is set, i.e., a multi-layer registration grid with different resolutions from low to high is set.

[0036] For example, the multi-resolution registration grid from low to high is represented as Grid(H, E), where H represents the number of layers, and E represents the grid that should be set for each layer, i.e., the intersection point of the grid lines in the registration grid. For a multi-resolution registration grid with 3 layers of different resolutions, each layer of the registration grid has a grid number of 4, 16, and 64, respectively, and the multi-resolution registration grid is represented as Grid(3, {4, 16, 64}).

[0037] In this application, for the first layer of the registration grid with the lowest resolution, the input is the spine ultrasound image and the X-ray image, and the sizes of the two are the same; for the subsequent layers of the registration network, the input is the ultrasound image and the X-ray image of the registration result of the previous layer. That is, the ultrasound image does not change throughout the process, and only the X-ray image is deformed to obtain the registration result X-ray image of each layer. In this way, fewer X-ray images can be used for registration with the ultrasound image, reducing the number of X-ray images taken.

[0038] In order to adapt to input images of various sizes, the multi-resolution registration grid of the present application also includes:

[0039] The aspect ratio of the registration grid is adjusted based on the aspect ratio of the input image, and finally the input image is embedded into the registration grid.

[0040] That is, for each layer of the registration grid, the aspect ratio of the registration grid is adjusted based on the aspect ratio of the input image, so that the aspect ratio of the registration grid is consistent with that of the input image, and each grid intersection point (control point) can be more scientifically radiated to each area of the input image.

[0041] In this embodiment, the control points of the registration grid with different resolutions change from less to more, and the subsequent registration is performed on the effect of the previous registration. Through multiple iterations of registration, the registration results from low to high are obtained, which prevents overfitting of the registration image, improves the stability of the registration, and greatly improves the registration efficiency.

[0042] Step S2, using gradient descent method, taking the inverse of the normalized mutual information similarity measure function as the objective function, respectively optimizing the control points in the different resolution registration grids in order from low to high resolution, obtaining the registration result X-ray image, and taking the registration result X-ray image obtained by the previous low resolution registration network registration as the input X-ray image of the subsequent high resolution registration network.

[0043] The embodiment starts from the first layer with the lowest resolution in the multi-layer registration grid and performs iterative registration. After the registration of one layer registration grid is completed, the output registration result X-ray image is taken as the input X-ray image of the next layer registration grid, and then the image registration of the next layer registration grid is performed, until the image registration of the last layer is completed, and the final registration result X-ray image is output.

[0044] For the image registration of each layer registration grid, one image is usually taken as the floating image and the other image is taken as the reference image. For example, the forward registration is taken as the X-ray image as the floating image, the ultrasound image as the reference image, and the X-ray image is registered to the ultrasound image. The reverse registration is taken as the ultrasound image as the floating image, and the X-ray image as the reference image, and the ultrasound image is registered to the X-ray image.

[0045] However, in a registration process, simply studying the registration of the floating image to the reference image may cause problems such as instability of the registration result and overfitting. The present application performs mutual optimization calculation on the control points in the registration grid. When each control point is optimized for the spine ultrasound image to the X-ray image, the optimization calculation of the X-ray image to the ultrasound image is added, two sets of results are obtained, the weighted operation is performed on the two sets of results, the above weighted calculation result is constrained by the piecewise function based on Tanh(X), and the final result is taken as the effective parameter for the calculation of the subsequent steps, so that the above problems are overcome.

[0046] The present application uses gradient descent method to take the inverse of the normalized mutual information similarity measure function as the objective function to perform mutual optimization calculation on the control points in the registration grid to obtain the registration result X-ray image, including:

[0047] Forward registration is performed:

[0048] Gradient descent method is used to take the inverse of the normalized mutual information similarity measure function as the objective function, and the control point best result of the X-ray image to the ultrasound image is calculated with the ultrasound image as the reference image, and the control point best result of the ultrasound image to the X-ray image is calculated with the X-ray image as the reference image.

[0049] The two kinds of control point best results are weighted and operated to generate the control point weighted optimization result.

[0050] The control point final optimization result set is obtained by constraining the control point weighted optimization result based on a segmented function of Tanh(X);

[0051] The X-ray image is resampled by using the final optimization result set to obtain a forward registration result X-ray image after one registration;

[0052] The reverse registration is performed:

[0053] The gradient descent method is adopted to calculate the control point best result of the ultrasound image to the X-ray image by taking the reciprocal of the normalized mutual information similarity measure function as the objective function and taking the X-ray image as the reference image, and the control point best result of the X-ray image to the ultrasound image is calculated by taking the ultrasound image as the reference image;

[0054] The two kinds of control point best results are weighted and operated to generate a control point weighted optimization result;

[0055] The control point final optimization result set is obtained by constraining the control point weighted optimization result based on a segmented function of Tanh(X);

[0056] The reciprocal operation is performed on the final optimization result set, and then the X-ray image is resampled by using the result set obtained by the reciprocal operation to obtain a reverse registration result X-ray image after one registration.

[0057] The normalized mutual information similarity measure evaluation index is used to evaluate the performance of the forward registration result X-ray image and the reverse registration result X-ray image after one registration, and an excellent registration result X-ray image is selected as the input X-ray image for the next iteration;

[0058] The registration is iterated in turn until the iteration termination condition is reached, and the iteration is ended and the final registration result X-ray image is output.

[0059] It should be noted that in the forward registration and reverse registration processes, the results obtained in their own calculation processes are used for calculation, and the forward registration and reverse registration are two separate calculation processes and do not affect each other.

[0060] Specifically, when optimizing and calculating the X-ray image to the ultrasound image for each control point, the ultrasound image to the X-ray image optimization calculation is added, two groups of results are obtained, the two groups of results are weighted and operated, and the above weighted calculation result is constrained based on a segmented function of Tanh(X). The above steps are repeated to finally obtain an optimization result set φ of all control points. Based on φ, the spine X-ray image is resampled to generate a registration result X-ray image after one registration, denoted as D.

[0061] Suppose that any control point K(m, n), the forward and reverse calculation control point best results are (α′m ,β′ n ), (α″ m ,β″ n ), where α and β represent the optimal results in the horizontal and vertical directions of control point K, respectively. The weighted average of the two sets of results is calculated as follows:

[0062] α m =tα′ m +vα″ m

[0063] β n =tβ′ n +vβ″ n

[0064] Where t and v represent weight parameters, in this embodiment, the optimal weights can be set to t = 1 and v = 0.1. The weighted optimization result at the generated K control points is (α m ,β n Repeat the above steps to calculate the optimization results for all control points and map them into a one-dimensional array to obtain the weighted optimization result set of control points, denoted as Grad(α1,α2,α3,...,α). i-1 ,α i ,β1,β2,β3,...,β i-1 ,β i ).

[0065] By using a piecewise function based on Tanh(X), Grad is mapped to constrain the horizontal and vertical optimization results of each control point, as shown in the following formula:

[0066]

[0067] in:

[0068] T=Min(|Max(Grad)|,|Min(Grad)|)

[0069] Grad represents the set of results after optimizing all control points during a single registration process. Here, 'a' represents the minimum value in the Grad result set, 'b' represents the maximum value in the Grad result set, and 'θ' is a user-defined control parameter; in one embodiment, 'θ' is set to 0.85.

[0070] After processing using a piecewise function based on Tanh(X), the final optimization result set for all control points is obtained:

[0071] Grad(y(α1),y(α2),y(α3),...,y(α i-1 ),y(α i), y(β1), y(β2), y(β3),..., y(β i-1 ), y(β i ));

[0072] denoted as based on Resample the spine X-ray image to generate the registration result X-ray image after the first registration, denoted as D.

[0073] This embodiment performs forward and reverse registration twice, and after data processing of the two result sets, the X-ray image is resampled to generate the result images D M , D F . The performance of D M , D F is evaluated by the similarity measure function, and the result image with higher performance is selected as the X-ray image for the next iteration registration.

[0074] The forward registration generates the final optimized result set, denoted as based on Resample the X-ray image in this iteration as the forward registration result image D M ; the reverse registration generates the final optimized result set, denoted as The inverse operation is performed on to generate based on Resample the X-ray image in this iteration to generate the reverse registration result image D F .

[0075] The normalized mutual information similarity measure evaluation index is used to evaluate the performance of the forward registration result image D M and the reverse result image D F , and the excellent result image is selected as the input X-ray image for the next iteration.

[0076] It is easy to understand that in the iteration registration process, an iteration termination condition is usually set. The iteration registration described in this application is performed in turn until the iteration termination condition is reached, and the iteration is ended, including:

[0077] When the set number of iterations is reached or the absolute difference between the similarity measure index of the registration result X-ray image and the ultrasound image of the two adjacent registrations for a continuous preset number of times is less than a preset value, the iteration is ended and the registration result X-ray image is output.

[0078] Specifically, when the set number of iterations is reached (for example, the number of iterations exceeds 300), the iteration is ended. Or the absolute difference between the similarity measure index of the registration result X-ray image and the ultrasound image of the two adjacent registrations for a continuous 3 times is less than 0.0001, the iteration is ended, otherwise the iteration registration is continued.

[0079] For example, a1 represents the similarity measure index of the registered X-ray image and the ultrasound image after the first iteration registration, a2 represents the similarity measure index of the registered X-ray image and the ultrasound image after the second iteration registration, a3 represents the similarity measure index of the registered X-ray image and the ultrasound image after the third iteration registration, and a4 represents the similarity measure index of the registered X-ray image and the ultrasound image after the fourth iteration registration. When (|a1-a2|<0.0001&&|a2-a3|<0.0001&&|a3-a4|<0.0001), it is considered that the termination condition is reached, and the iteration is ended.

[0080] It should be noted that the gradient descent method is used in the present application to calculate the optimal result of the control points, and the reciprocal of the normalized mutual information similarity measure function is used as the objective function, which is a relatively mature technology in the art and will not be described here. For two images A and B, image A is similar in shape to image B, but has a positional offset: translation, rotation, scaling, local distortion, etc. If image A is used as the reference image and image B is used as the floating image, image B is registered to align with image A, then the FFD (Free Form Deform) free deformation model can be used as the coordinate transformation model to deform image B, and the similarity between image A and the deformed image B is calculated. The optimal control parameters of FFD deformation are solved to maximize the similarity between the two images, so that the optimal control parameters are used to deform image B to achieve registration. The similarity between image A and the deformed image B is used as the objective function, and then an optimization algorithm (such as the gradient descent method) is used to solve the optimal solution of the objective function.

[0081] In the iteration registration process of the present embodiment, the floating image is registered to the reference image, and the calculated optimal parameters generate a deformation field, while the reference image is registered to the floating image, and the deformation field is calculated. This improves the accuracy and effectively improves the stability of the registration and reduces the overfitting problem.

[0082] Step S3: The registered X-ray image generated by the highest resolution registration grid is subjected to a deblurring operation using the blur-aware attention network, and a clear registered X-ray image is output.

[0083] The present application uses the registered X-ray image generated by the multi-resolution registration grid as the final registration result, and then performs a deblurring operation on the result image using the blur-aware attention network (BANeT).

[0084] In order to better display the lesion position and provide doctors with clearer result images, the clarity of the result image needs to be improved through a deblurring algorithm. Traditional deblurring algorithms usually use prior information of the image to simulate the blur characteristics. They cannot well summarize complex image examples. The self-loop module based on deep learning deblurring method has achieved great success in gradually restoring clear images in different resolutions, different time characteristics and different fields of view. However, because the self-loop model has a long self-reasoning time and cannot well solve the problem of uneven blur in the image, it is not suitable for medical images to solve the problems of edge information loss and image blur caused by linear interpolation.

[0085] In the embodiment, the existing BANeT network is utilized, a series of stacked blur perception modules (BAM) are used to effectively separate different blur modes, global and local multi-scale blur features are derived in a learnable manner, blur is removed according to the blur features, and finally a deblurred image is output. The BANet network is composed of a series of stacked blur perception modules (BAM), the BAM can effectively separate different blur modes and remove blur according to the blur features. The input image is down-sampled to half resolution by two convolutional layers, and the features are up-sampled to the original size by a transposed convolutional layer. A set of BAMs are stacked in between to associate regions with similar blur and extract multi-scale content features. A BAM is composed of two parts of blur perception attention (BA) and cascaded parallel dilated convolution (CPDC), wherein the BA extracts global and local blur information, and the CPDC captures multi-scale blur features. In combination with the BA and the CPDC, the BAM is a residual-like architecture to derive global and local multi-scale blur features in a learnable manner, remove blur according to the blur features, and finally output a clear image.

[0086] The present application introduces the idea of mutual optimization and mutual registration by increasing an adaptive multi-resolution grid, borrows the existing BANeT deblurring network model to improve the registration model, so as to ensure the stability of image registration, improve the registration accuracy, and improve the clarity and readability of the spine image.

[0087] The above-described embodiments only express several embodiments of the present application, and the description is more specific and detailed, but it cannot be understood as limiting the scope of the patent. It should be noted that for ordinary skilled persons in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of the present application. Therefore, the protection scope of the patent of the present application should be subject to the appended claims.

Claims

1. A method of high stability registration of ultrasound and X-ray images of a spine, characterized in that, The high-stability spine ultrasound and X-ray image registration method comprises the following steps: A multi-resolution registration grid from low to high is set, an X-ray image is embedded into the lowest resolution registration grid, and an ultrasound image is embedded into each resolution registration grid; A gradient descent method is used to take the reciprocal of a normalized mutual information similarity measure function as an objective function, to respectively optimize and calculate the control points in different resolution registration grids in order from low to high resolution, to obtain a registration result X-ray image, and to take the registration result X-ray image obtained by the previous low-resolution registration grid as the input X-ray image of the subsequent high-resolution registration grid; The registration result X-ray image generated by the highest resolution registration grid is subjected to a deblurring operation by a fuzzy perception attention network to output a clear X-ray image after registration.

2. The method of claim 1, wherein, The high-stability spine ultrasound and X-ray image registration method further comprises the following steps: The aspect ratio of the registration grid is adjusted adaptively based on the aspect ratio of the input image, and finally the input image is embedded into the registration grid.

3. The method of claim 1, wherein, The optimization and calculation of the control points in different resolution registration grids to obtain a registration result X-ray image comprises the following steps: Forward registration is performed: A gradient descent method is used to take the reciprocal of a normalized mutual information similarity measure function as an objective function, to calculate the best results of the control points of the X-ray image on the ultrasound image as a reference image, and to calculate the best results of the control points of the ultrasound image on the X-ray image as a reference image; Weighted operation is performed on the two kinds of control point best results to generate a control point weighted optimization result; The control point weighted optimization result is constrained by a piecewise function based on Tanh(X) to obtain a control point final optimization result set; The X-ray image is resampled by using the control point final optimization result set to obtain a forward registration result X-ray image after one-time registration; Reverse registration is performed: A gradient descent method is used to take the reciprocal of a normalized mutual information similarity measure function as an objective function, to calculate the best results of the control points of the ultrasound image on the X-ray image as a reference image, and to calculate the best results of the control points of the X-ray image on the ultrasound image as a reference image; Weighted operation is performed on the two kinds of control point best results to generate a control point weighted optimization result; The control point weighted optimization result is constrained by a piecewise function based on Tanh(X) to obtain a control point final optimization result set; The control point final optimization result set is subjected to an inverse operation, and then the X-ray image is resampled by using the result set obtained by the inverse operation to obtain a reverse registration result X-ray image after one-time registration; The normalized mutual information similarity measure evaluation index is used to respectively evaluate the performance of the forward registration result X-ray image and the reverse registration result X-ray image after one-time registration, and an excellent registration result X-ray image is selected as the input X-ray image for the next iteration; Sequential iteration registration is performed until the iteration termination condition is reached, the iteration is ended, and a final registration result X-ray image is output.

4. The method of claim 3, wherein the method is characterized by, The sequential iteration registration until the iteration termination condition is reached and the iteration is ended comprises the following steps: When the set number of iterations is reached or the absolute difference between the similarity measure indicators of the registered X-ray image and the ultrasound image of two adjacent registrations in succession is less than the preset value for a preset number of times in succession, the iterations are ended and the final registered X-ray image is output.

Images