Two-dimensional-three-dimensional image rigid registration method for deformed image

Abstract

The invention discloses a two-dimensional-three-dimensional image rigid registration method for a deformed image, and relates to the technical field of image registration, and the method comprises the steps: rotating a reference CT image to obtain a plurality of out-of-plane DRR images, and then carrying out the registration of each out-of-plane DRR image with an X-ray image to be registered, in the registration process, the skeleton texture similarity value is independently calculated in a blocking mode, then the global similarity is counted and evaluated, influences caused by cavity areas and the like can be reduced, the overall skeleton texture similarity can be more accurately measured through the obtained global similarity, and then the registration results of all out-of-plane DRR images are synthesized to obtain the final registration result. According to the similarity calculation method based on block statistics, the part which can be rigidly registered can be mined to the greatest extent, so that the influence of a deformed area on global similarity calculation is small, and the stability of the overall registration result and the registration effect are improved.

Description

Technical Field

[0001] The present application relates to the technical field of image registration, and in particular to a two-dimensional-three-dimensional image rigid registration method for deformable images. Background Art

[0002] During radiotherapy, respiratory movement or patient movement may cause unexpected movement of the irradiated target area, which may lead to "off-target" irradiation, causing normal tissues to receive large doses and increasing the chance of complications. The emergence of image-guided positioning technology has solved the problem of precise positioning during treatment. Through image-guided positioning technology, the errors caused by the movement and placement of tumors (target areas) and important organs caused by patient breathing and organ movement are corrected, thereby achieving more accurate treatment. The kV-level X-ray imaging system is currently a commonly used image-guided positioning device for radiotherapy. During the positioning process, the system collects a pair of kV-level two-dimensional images of the patient and performs 2D-3D registration and positioning with the treatment plan CT image to determine the position of the patient or tumor, thereby achieving image-guided positioning.

[0003] Clinical treatment and positioning of patients may usually occur in any part of the body, such as the head, neck, chest, abdomen, and pelvis. The repeatability of positioning in different parts is different. When the treatment positioning part is the head, the head is relatively easy to fix, and the positioning of the head in the treatment positioning is easy to reproduce the positioning state during the CT scan. For the neck, chest, abdomen, and pelvic parts, in the second positioning (the second positioning is the positioning during treatment positioning, and the first positioning is the positioning during the CT scan), there may be different degrees of twisting and deformation compared with the first positioning, especially in the chest and abdomen. Because the chest and abdomen are affected by breathing, the thorax and ribs will fluctuate at all times, and the liver and other organs and tissues in the abdomen will also change position with breathing. Due to the deformation between the reference image and the image to be registered, rigid registration cannot obtain a solution that can perfectly match the reference image and the image to be registered, and when the deformation between the reference image and the image to be registered is large, the result of rigid registration will be very unstable, which will affect the accuracy of image-guided positioning. Summary of the invention

[0004] In view of the above problems and technical requirements, this application proposes a 2D-3D image rigid registration method for deformable images. The technical solution of this application is as follows:

[0005] A two-dimensional to three-dimensional image rigid registration method for deformable images, the two-dimensional to three-dimensional image rigid registration method comprising:

[0006] A first X-ray image of a three-dimensional imaged object to be registered is acquired through a first imaging component in an X-ray imaging device, and a second X-ray image of the three-dimensional imaged object to be registered is acquired through a second imaging component in the X-ray imaging device, wherein the two imaging components are arranged crosswise and each imaging component includes a ray generator and a two-dimensional imaging flat panel arranged opposite to each other on both sides of the three-dimensional imaged object;

[0007] Acquire a reference CT image of a three-dimensional imaged body, and rotate the reference CT image by a plurality of angles around the x-axis and the y-axis of the imaging plane coordinate system of the two-dimensional imaging flat panel in the first imaging component with the positioning center point as the center to obtain N first out-of-plane DRR images, and rotate the reference CT image by a plurality of angles around the x-axis and the y-axis of the imaging plane coordinate system of the two-dimensional imaging flat panel in the second imaging component with the positioning center point as the center to obtain N second out-of-plane DRR images, where the integer parameter N≥2;

[0008] The N first out-of-plane DRR images are rigidly registered with the first X-ray image to be registered, and the N second out-of-plane DRR images are rigidly registered with the second X-ray image to be registered; the rigid registration of the N out-of-plane DRR images with the corresponding X-ray images to be registered includes:

[0009] For any i-th out-of-plane DRR image, based on the two-dimensional transformation parameters (dx i ,dy i ,θ i ) performs image transformation on the i-th out-of-plane DRR image to obtain the i-th transformed DRR image, divides the i-th transformed DRR image and the region of interest in the to-be-registered X-ray image to obtain a plurality of blocks, calculates the bone texture similarity value between the i-th transformed DRR image and the corresponding to-be-registered X-ray image in each block region, and statistically calculates the bone texture similarity values ​​in all block regions to obtain the global similarity S between the i-th out-of-plane DRR image and the to-be-registered X-ray image under the current two-dimensional transformation parameters i ; Adjust the two-dimensional transformation parameters until the maximum global similarity that the i-th out-of-plane DRR image can achieve with the X-ray image to be registered is determined; where the integer parameter 1≤i≤N, dx i is the translation parameter along the x-axis of the imaging plane coordinate system of the two-dimensional imaging plate, dy i is the translation parameter along the y-axis of the imaging plane coordinate system of the two-dimensional imaging plate, θ i is the in-plane rotation angle of the imaging plane coordinate system of the two-dimensional imaging flat panel around the origin of the coordinate system; determine the rotation angle θ of the out-of-plane DRR image with the greatest global similarity to the X-ray image to be registered among all out-of-plane DRR images relative to the imaging plane coordinate system 3D, and combined with the two-dimensional transformation parameters when the maximum global similarity is achieved, the final registration result of the X-ray image to be registered is obtained;

[0010] According to the final registration result (dx A ,dy A ,θ A ,θ 3D_A ) and the final registration result (dx B ,dy B ,θ B ,θ 3D_B ) to obtain the final registration result in the three-dimensional space coordinate system.

[0011] A further technical solution is to adjust the two-dimensional transformation parameters until the maximum global similarity that the i-th out-of-plane DRR image can achieve with the X-ray image to be registered is determined, including:

[0012] Initialize the two-dimensional transformation parameters (dx i ,dy i ,θ i ) is (0,0,0), and the global similarity S between the i-th out-of-plane DRR image and the X-ray image to be registered under the two-dimensional transformation parameters is i As the cost function, the gradient descent method is used to update the two-dimensional transformation parameters and obtain the global similarity S i The maximum value of and the two-dimensional transformation parameters when the maximum value is reached.

[0013] Its further technical solution is to use the gradient descent method to update the two-dimensional transformation parameters and obtain the global similarity S i The maximum values ​​include:

[0014] Downsampling the i-th out-of-plane DRR image and the X-ray image to be registered multiple times to different degrees to obtain Q sampling image groups at different resolutions. The sampling image group at each resolution includes the i-th out-of-plane DRR image and the X-ray image to be registered at the resolution obtained by downsampling. The integer parameter Q ≥ 2;

[0015] Initialize the initial values ​​of the two-dimensional transformation parameters of the first iteration to (0,0,0), and initialize the number of iterations q = 1;

[0016] Based on the initial values ​​of the two-dimensional transformation parameters of the qth iteration, the maximum global similarity between the i-th out-of-plane DRR image in the qth sampling image group and the X-ray image to be registered in the same sampling image group is calculated by using the gradient descent method; wherein the qth sampling image group is the sampling image group at the qth resolution from small to large;

[0017] The two-dimensional transformation parameters that reach the maximum global similarity in the qth iteration are used as the initial values ​​of the two-dimensional transformation parameters in the q+1th iteration, and the q+1th iteration is entered until the number of iterations q=Q.

[0018] A further technical solution is that the bone texture similarity value between the i-th transformed DRR image and the corresponding X-ray image to be registered in each block area is calculated, including:

[0019] The information entropy of the block area is calculated according to the gradient value of the X-ray image to be registered in each block area, and the block area containing the bone texture is screened out as the valid block area according to the information entropy of each block area;

[0020] The similarity between the i-th transformed DRR image and the X-ray image to be registered in each valid block area is calculated as the bone texture similarity value in the valid block area.

[0021] Its further technical solution is that the bone texture similarity value s of the i-th transformed DRR image and the X-ray image to be registered in each valid block area is:

[0022]

[0023] Among them, x i is the gradient value of the i-th pixel in the effective block area of ​​the i-th transformed DRR image, y i It is the gradient value of the i-th pixel point of the X-ray image to be registered in the effective block area, and the i-th pixel point of the i-th transformed DRR image in the effective block area corresponds to the i-th pixel point of the X-ray image to be registered in the effective block area; n is the number of pixels in the effective block area.

[0024] A further technical solution is to obtain the global similarity S between the i-th out-of-plane DRR image and the X-ray image to be registered under the current two-dimensional transformation parameters. i include:

[0025] The bone texture similarity values ​​in all valid block areas are normalized respectively to obtain the normalized bone texture similarity value in each valid block area, and the number m of valid block areas whose normalized bone texture similarity value is greater than the similarity threshold σ is determined to obtain the global similarity between the i-th out-of-plane DRR image and the X-ray image to be registered under the current two-dimensional transformation parameters. M is the total number of valid block areas, and the similarity threshold is 0.5<σ<1.

[0026] Its further technical solution is to calculate the information entropy of each block area and screen out the effective block areas including:

[0027] Calculate the information entropy of each block area Among them, p k is the ratio of the number of pixels with the kth gradient value to the total number of pixels in the block area, and K is the number of gradient values ​​in the block area;

[0028] The information entropy of all block regions is normalized to obtain the normalized information entropy of each block region, and the block region whose normalized information entropy is greater than the information entropy threshold is taken as a valid block region.

[0029] A further technical solution is that determining the information entropy threshold includes:

[0030] Calculate the average and minimum values ​​of the normalized information entropy of all block areas, and take the median of the average and minimum values ​​of the normalized information entropy as the information entropy threshold.

[0031] A further technical solution is that the two-dimensional-three-dimensional image rigid registration method further includes:

[0032] Performing image transformation on the reference CT image according to the final registration result in the three-dimensional space coordinate system, and obtaining a DRR image that is completely registered with the X-ray image to be registered as a target DRR image;

[0033] The target DRR image and the corresponding X-ray image to be registered are divided into multiple display blocks, and the similarity value between the target DRR image and the corresponding X-ray image to be registered in each display block is calculated, and the target DRR image is superimposed and visualized according to the display format corresponding to the similarity value in each display block.

[0034] A further technical solution is that, according to the final registration result of the first X-ray image to be registered and the final registration result of the second X-ray image to be registered, a final registration result in the three-dimensional space coordinate system is obtained, which includes:

[0035] The final registration result (dx A ,dy A ,θ A ,θ 3D_A ) and the final registration result (dx B ,dy B ,θ B ,θ 3D_B ) performs 2D-3D back-projection geometric transformation to obtain the registration result in the three-dimensional space coordinate system.

[0036] The beneficial technical effects of this application are:

[0037] The present application discloses a method for rigid registration of two-dimensional to three-dimensional images for deformable images. The method rotates a reference CT image based on an imaging plane coordinate system of a two-dimensional imaging plate in an X-ray imaging device to obtain a plurality of out-of-plane DRR images, and then registers each out-of-plane DRR image with the X-ray image to be registered. When performing the registration, the bone texture similarity value is first calculated in blocks and then the global similarity is statistically evaluated, which can reduce the influence of the cavity area and the like, so that the obtained global similarity can more accurately measure the overall bone texture similarity, and then the registration results of all out-of-plane DRR images are combined to obtain the final registration result. The block-by-block statistical similarity calculation method can maximize the mining of the parts that can be rigidly registered, and the similarity between blocks is calculated independently, so that the deformation area has less influence on the calculation of the global similarity, thereby improving the stability of the overall registration result, and achieving a good rigid registration effect even when the X-ray image to be registered and the reference CT image are deformed.

[0038] When evaluating the global similarity, we first use information entropy to filter out the block area where the bone texture is located, thereby filtering out the cavity area. On the one hand, this can reduce the error caused by the cavity area to the global similarity, and on the other hand, it can reduce the amount of calculation and further improve the stability of the overall registration result.

[0039] When registering each out-of-plane DRR image with the X-ray image to be registered, a better registration effect can be achieved by downsampling multiple times and iteratively registering from low to high resolution.

[0040] This method can also visualize the similarity of bone textures in different block areas, thereby more intuitively showing the alignment of different block areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a schematic diagram of the installation imaging geometry of the X-ray imaging equipment.

[0042] Figure 2 It is a schematic diagram of a process of rigidly registering N out-of-plane DRR images with corresponding X-ray images to be registered in one embodiment of the present application.

[0043] Figure 3 It is a schematic diagram of a process for obtaining the bone texture similarity value between the i-th transformed DRR image and the corresponding X-ray image to be registered in each block area in one embodiment of the present application.

[0044] Figure 4 It is a schematic diagram of a process for obtaining the maximum global similarity between the i-th out-of-plane DRR image and the X-ray image to be registered in one embodiment of the present application. DETAILED DESCRIPTION

[0045] The specific implementation of the present application is further described below in conjunction with the accompanying drawings.

[0046] This application discloses a method for rigid registration of deformed 2D-3D images, which is used to rigidly register deformed X-ray images to be registered and reference CT images of the same 3D imaged object. First, the installation imaging geometry of the X-ray imaging device is introduced as follows, please refer to Figure 2 The present application is directed to an X-ray imaging device with a dual-plate structure, wherein the X-ray imaging device includes two imaging components, which are respectively recorded as a first imaging component and a second imaging component, and the two imaging components are arranged crosswise. Each imaging component includes a ray generator and a two-dimensional imaging flat plate arranged opposite to each other on both sides of a three-dimensional imaged object. The X-rays emitted by the ray generator irradiate the three-dimensional imaged object, and an X-ray image of the three-dimensional imaged object to be registered can be obtained on the two-dimensional imaging flat plate on the other side.

[0047] like Figure 1 As shown, the first imaging assembly includes a ray generator 11 and a two-dimensional imaging flat panel 12 arranged opposite to each other, and the second imaging assembly includes a ray generator 21 and a two-dimensional imaging flat panel 22 arranged opposite to each other. In the operation process of the X-ray imaging device, a first X-ray image to be registered of a three-dimensional imaged object is obtained through the first imaging assembly, and a second X-ray image to be registered of the three-dimensional imaged object is obtained through the second imaging assembly.

[0048] A three-dimensional space coordinate system OXYZ is established with the position O of the three-dimensional imaged object as the origin. The Y axis passes through the coordinate origin O vertically downward. The X axis passes through the origin O and is parallel to the horizontal plane and points to the direction of the second imaging component in the imaging plane where the center lines of the beams emitted by the ray generators in the two imaging components intersect. The Z axis is determined accordingly according to the right-hand coordinate system rule.

[0049] For each imaging component of the X-ray imaging device, the line connecting the ray generator in the imaging component and the position O of the three-dimensional imaged object is perpendicular to the imaging plane of the corresponding two-dimensional imaging flat panel, and an imaging plane coordinate system is established in the imaging plane of the two-dimensional imaging flat panel with the intersection of the line and the imaging plane of the two-dimensional imaging flat panel as the origin, the x-axis of the imaging plane coordinate system is along the width direction of the imaging plane of the two-dimensional imaging flat panel, and the y-axis of the imaging plane coordinate system is along the length direction of the imaging plane of the two-dimensional imaging flat panel. Figure 1 In the example of FIG. 1 , an imaging plane coordinate system of the two-dimensional imaging flat panel 12 in the first imaging assembly is established. A x A y A , and an imaging plane coordinate system o of the two-dimensional imaging flat panel 22 in the second imaging assembly is established B x B y B .

[0050] After obtaining a first X-ray image to be registered and a second X-ray image to be registered of a three-dimensional imaged object through an X-ray imaging device, and obtaining a reference CT image of the same three-dimensional imaged object, the two-dimensional-three-dimensional image rigid registration method of the present application can be used to rigidly register the X-ray image to be registered with respect to the reference CT image, comprising the following steps:

[0051] Step 1: An imaging plane coordinate system o is formed around the two-dimensional imaging flat panel 12 in the first imaging assembly with the positioning center point as the center of the reference CT image. A x A y A The x-axis (x A axis) and y axis (y A The reference CT image is rotated by the positioning center point to obtain N first out-of-plane DRR images. The imaging plane coordinate system o of the reference CT image is centered around the two-dimensional imaging flat panel 22 in the second imaging assembly. B x B y B The x-axis (x B axis) and y axis (y B The first image is rotated about the x-axis and the second image is rotated about the y-axis to obtain N second out-of-plane DRR images. The integer parameter N ≥ 2. The positioning center point can be predetermined, and the rotation angles around the x-axis and the y-axis of the imaging plane coordinate system can be customized.

[0052] Each out-of-plane DRR image obtained has its own rotation angle relative to the imaging plane coordinate system, and the rotation angle is the angle at which the reference CT image is rotated around the coordinate axis of the imaging plane coordinate system to obtain the out-of-plane DRR image. In the present application, each out-of-plane DRR image is rotated only around the x-axis of the imaging plane coordinate system, or only around the y-axis of the imaging plane coordinate system. The N first out-of-plane DRR images obtained include a plurality of images that are rotated only around the x-axis. A The first out-of-plane DRR images with different rotation angles around the y axis and the A The first out-of-plane DRR images are rotated at different angles around the x-axis. Similarly, the N second out-of-plane DRR images obtained include multiple images that are rotated only around the x-axis. B The second out-of-plane DRR images with different rotation angles around the y axis and the B Second out-of-plane DRR images with different axis rotation angles.

[0053] Then, the N first out-of-plane DRR images are rigidly registered with the first X-ray image to be registered, and the N second out-of-plane DRR images are rigidly registered with the second X-ray image to be registered. The rigid registration methods of these two parts are the same, and this application describes them uniformly. Then, the rigid registration of the N out-of-plane DRR images with the corresponding X-ray images to be registered includes the following steps 2 and 3. Please combine Figure 2Flowchart:

[0054] Step 2, for any i-th out-of-plane DRR image, rigidly register the i-th out-of-plane DRR image with the X-ray image to be registered, including:

[0055] (1) Based on the two-dimensional transformation parameters (dx i ,dy i ,θ i ) performs image transformation on the i-th out-of-plane DRR image to obtain the i-th transformed DRR image. Wherein, the integer parameter 1≤i≤N, dx i is the translation parameter along the x-axis of the imaging plane coordinate system of the two-dimensional imaging plate, dy i is the translation parameter along the y-axis of the imaging plane coordinate system of the two-dimensional imaging plate, θ i It is the in-plane rotation angle of the imaging plane coordinate system of the two-dimensional imaging plate around the origin of the coordinate system.

[0056] (2) The region of interest in the i-th transformed DRR image and the X-ray image to be registered is divided into a plurality of blocks. The region of interest is divided and specified by the user. Then, the region of interest is divided into a plurality of block regions, and the specifications of each block region can be equal or unequal. In general, in actual operation, the determined region of interest is a rectangular region, and the region of interest is grid-divided to obtain block regions with the same rectangular structure.

[0057] (3) Calculate the bone texture similarity value between the i-th transformed DRR image and the corresponding X-ray image to be registered in each block area.

[0058] This step calculates the bone texture similarity value between the i-th transformed DRR image and the X-ray image to be registered in the block area, mainly to eliminate the influence of the cavity area on the similarity. This step includes the following steps, please refer to Figure 3 :

[0059] (a) The information entropy of each block region is calculated based on the gradient value of the X-ray image to be registered in each block region. Information entropy of each block region Among them, p k is the ratio of the number of pixels with the kth gradient value to the total number of pixels in the block area, and K is the number of gradient values ​​in the block area.

[0060] (b) The information entropy of all the block areas is normalized to obtain the normalized information entropy of each block area, and the block area whose normalized information entropy is greater than the information entropy threshold is taken as the valid block area. The information entropy threshold here can be an empirical preset value, or in order to improve the accuracy of the information entropy threshold, in one embodiment, the average value and minimum value of the normalized information entropy of all the block areas are calculated, and the median value of the average value and the minimum value of the normalized information entropy is taken as the information entropy threshold.

[0061] (c) According to the information entropy of each block area, the block area containing bone texture is selected as the valid block area, thereby filtering out the cavity area and no longer participating in the subsequent similarity value calculation. This can not only reduce the interference of the cavity area on the registration, but also reduce the amount of calculation.

[0062] (d) Calculate the similarity between the i-th transformed DRR image and the X-ray image to be registered in each valid block area as the bone texture similarity value in the valid block area. The bone texture similarity value s between the i-th transformed DRR image and the X-ray image to be registered in each valid block area is:

[0063]

[0064] Among them, x i is the gradient value of the i-th pixel in the effective block area of ​​the i-th transformed DRR image, y i is the gradient value of the i-th pixel of the X-ray image to be registered in the effective block area, and the i-th pixel of the i-th transformed DRR image in the effective block area corresponds to the i-th pixel of the X-ray image to be registered in the effective block area. n is the number of pixels in the effective block area.

[0065] (4) The bone texture similarity values ​​in all block regions are statistically analyzed to obtain the global similarity S between the i-th out-of-plane DRR image and the X-ray image to be registered under the current two-dimensional transformation parameters. i .

[0066] In one embodiment, the global similarity S is obtained by using the idea that the minority obeys the majority. i , including: normalizing the bone texture similarity values ​​in all valid block regions respectively, obtaining the normalized bone texture similarity value in each valid block region, and determining the number m of valid block regions whose normalized bone texture similarity value is greater than the similarity threshold σ, and obtaining the global similarity between the i-th out-of-plane DRR image and the X-ray image to be registered under the current two-dimensional transformation parameters M is the total number of valid block areas, and the similarity threshold is 0.5<σ<1.

[0067] Among them, the larger the value of the similarity threshold σ is, the higher the overall registration accuracy is, which can be determined according to the actual accuracy requirements. The actual registration accuracy requirements for different parts are also different. Generally, the registration accuracy requirements for the head are relatively low, while the registration accuracy requirements for the chest and abdomen are relatively high.

[0068] (5) Adjust the two-dimensional transformation parameters until the maximum global similarity between the i-th out-of-plane DRR image and the X-ray image to be registered is determined. In order to improve the registration efficiency, in one embodiment, the two-dimensional transformation parameters (dx i ,dy i ,θ i ) is (0,0,0) and the global similarity S is calculated according to the above method i Then, the global similarity S between the i-th out-of-plane DRR image and the X-ray image to be registered under the two-dimensional transformation parameters is used i As the cost function, the gradient descent method is used to update the two-dimensional transformation parameters until the global similarity S is obtained. i The maximum value of and the two-dimensional transformation parameters when the maximum value is reached.

[0069] Furthermore, in order to avoid the situation where direct pairwise image registration is prone to fall into the local optimal value, and to achieve better registration effect and faster registration calculation speed, in another embodiment, the gradient descent method is used to update the two-dimensional transformation parameters and obtain the global similarity S i Including the following process, please refer to Figure 4 Flowchart:

[0070] S301, downsampling the i-th out-of-plane DRR image and the X-ray image to be registered multiple times at different degrees to obtain Q sampling image groups at different resolutions, each sampling image group at each resolution includes the i-th out-of-plane DRR image and the X-ray image to be registered at the resolution obtained by downsampling, and the integer parameter Q≥2.

[0071] S302, initializing the initial values ​​of the two-dimensional transformation parameters of the first iteration to (0, 0, 0), and initializing the number of iterations q=1.

[0072] S303, using the gradient descent method based on the initial values ​​of the two-dimensional transformation parameters of the qth iteration, calculate the maximum global similarity that can be achieved between the i-th out-of-plane DRR image in the q-th sampling image group and the X-ray image to be registered in the same sampling image group, wherein the q-th sampling image group is the sampling image group at the q-th resolution from small to large.

[0073] S304, the two-dimensional transformation parameters when the maximum global similarity is reached in the qth iteration are used as the initial values ​​of the two-dimensional transformation parameters of the q+1th iteration, and the q+1th iteration is entered until the number of iterations q=Q, and the maximum global similarity between the i-th out-of-plane DRR image and the X-ray image to be registered and the two-dimensional transformation parameters when the global similarity is reached are obtained. The Qth group of sampled images includes the i-th out-of-plane DRR image of the original resolution and the X-ray image to be registered.

[0074] Step 3: The maximum global similarity between each out-of-plane DRR image and the X-ray image to be registered can be obtained by the above method. Then, the maximum global similarity is obtained by integrating the results of all out-of-plane DRR images, and the rotation angle θ of a pair of out-of-plane DRR images with the maximum global similarity with the X-ray image to be registered among all out-of-plane DRR images is determined relative to the imaging plane coordinate system. 3D , and the final registration result of the X-ray image to be registered is obtained by combining the two-dimensional transformation parameters when the maximum global similarity is achieved.

[0075] Through the above steps 2 and 3, the final registration results (dx A ,dy A ,θ A ,θ 3D_A ) and the final registration result (dx B ,dy B ,θ B ,θ 3D_B ).

[0076] Step 4: Based on the final registration result (dx A ,dy A ,θ A ,θ 3D_A ) and the final registration result (dx B ,dy B ,θ B ,θ 3D_B ) to obtain the final registration result in the three-dimensional space coordinate system, including: A ,dy A ,θ A ,θ 3D_A ) and the final registration result (dx B ,dy B ,θ B ,θ 3D_B) to perform 2D-3D back-projection geometric transformation to obtain the final registration result in the three-dimensional space coordinate system. The specific operation of 2D-3D back-projection geometric transformation can refer to the existing practice, and this application will not repeat it. The final registration result (dx, dy, dz, θ x ,θ y ,θ z ), where dx is the translation parameter along the X-axis of the three-dimensional space coordinate system OXYZ, dy is the translation parameter along the Y-axis of the three-dimensional space coordinate system OXYZ, and dz is the translation parameter along the Z-axis of the three-dimensional space coordinate system OXYZ. x is the rotation angle around the X-axis of the three-dimensional space coordinate system OXYZ, θ y is the rotation angle around the Y axis of the three-dimensional space coordinate system OXYZ, θ z It is the rotation angle around the Z axis of the three-dimensional space coordinate system OXYZ.

[0077] Furthermore, the final registration result (dx, dy, dz, θ x ,θ y ,θ z ) and then, according to the final registration result (dx, dy, dz, θ x ,θ y ,θ z) Perform image transformation on the reference CT image to obtain a DRR image of 0 degrees that is registered with the X-ray image to be registered as the target DRR image. Divide the target DRR image and the corresponding X-ray image to be registered to obtain multiple display blocks, and calculate the similarity value between the target DRR image and the corresponding X-ray image to be registered in each display block, and overlay and visualize the target DRR image according to the display format corresponding to the similarity value in each display block. Different similarity values ​​correspond to different display formats, and different display formats can be distinguished by at least one of color, texture, and brightness. For example, different colors can be used to distinguish different similarity values, and display blocks with similarity values ​​lower than the first threshold μ1 are displayed in red, display blocks with similarity values ​​reaching the first threshold μ1 but lower than the second threshold μ2 are displayed in yellow, and display blocks with similarity values ​​reaching the second threshold μ2 are displayed in green, so that the similarity of different display blocks can be intuitively displayed in a visual manner. Among them, the specific formula for the similarity value in each display block is the same as the formula for calculating the similarity value in each valid block area in the above step 2, and will not be repeated here. However, in order to better visualize the similarities between the target DRR image and the corresponding X-ray image to be registered, this embodiment divides the image into display blocks instead of just the region of interest. In the above step 2, in order to ensure the validity of the statistical results and the accuracy of the registration during the rigid registration process, the size of the divided block area is usually small, such as each block area is typically a 16*16 pixel block, while here, for better visual effects, the divided display blocks are usually larger in size, such as each display block is typically a 64*64 pixel block. When calculating the similarity value, the similarity value within each display block is also calculated, and information entropy filtering is no longer used.

[0078] The above is only a preferred embodiment of the present application, and the present application is not limited to the above embodiments. It is understood that other improvements and changes directly derived or associated by those skilled in the art without departing from the spirit and concept of the present application should be considered to be included in the protection scope of the present application.

Claims

1. A method for rigid registration of 2D-3D images of deformable images, characterized in that: The two-dimensional-three-dimensional image rigid registration method comprises: A first X-ray image of a three-dimensional imaged object to be registered is acquired through a first imaging component in an X-ray imaging device, and a second X-ray image of the three-dimensional imaged object to be registered is acquired through a second imaging component in the X-ray imaging device, wherein the two imaging components are arranged crosswise and each imaging component includes a ray generator and a two-dimensional imaging flat panel arranged opposite to each other on both sides of the three-dimensional imaged object; Acquire a reference CT image of the three-dimensional imaged body, and rotate the reference CT image by a plurality of angles around the x-axis and the y-axis of the imaging plane coordinate system of the two-dimensional imaging flat panel in the first imaging component with the positioning center point as the center to obtain N first out-of-plane DRR images, and rotate the reference CT image by a plurality of angles around the x-axis and the y-axis of the imaging plane coordinate system of the two-dimensional imaging flat panel in the second imaging component with the positioning center point as the center to obtain N second out-of-plane DRR images, where the integer parameter N≥2; The N first out-of-plane DRR images are rigidly registered with the first X-ray image to be registered, and the N second out-of-plane DRR images are rigidly registered with the second X-ray image to be registered; the rigid registration of the N out-of-plane DRR images with the corresponding X-ray images to be registered includes: For any i-th out-of-plane DRR image, based on the two-dimensional transformation parameters (dx i ,dy i ,θ i ) performs image transformation on the i-th out-of-plane DRR image to obtain the i-th transformed DRR image, divides the i-th transformed DRR image and the region of interest in the to-be-registered X-ray image to obtain a plurality of blocks, calculates the bone texture similarity value between the i-th transformed DRR image and the corresponding to-be-registered X-ray image in each block region, and performs statistics on the bone texture similarity values ​​in all block regions to obtain the global similarity S between the i-th out-of-plane DRR image and the to-be-registered X-ray image under the current two-dimensional transformation parameters i ; Adjust the two-dimensional transformation parameters until the maximum global similarity that the i-th out-of-plane DRR image can achieve with the X-ray image to be registered is determined; wherein the integer parameter 1≤i≤N, dx i is the translation parameter along the x-axis of the imaging plane coordinate system of the two-dimensional imaging plate, dy i is the translation parameter along the y-axis of the imaging plane coordinate system of the two-dimensional imaging plate, θ i is the in-plane rotation angle of the imaging plane coordinate system of the two-dimensional imaging flat panel around the origin of the coordinate system; determine the rotation angle θ of a pair of out-of-plane DRR images with the greatest global similarity to the X-ray image to be registered relative to the imaging plane coordinate system among all out-of-plane DRR images 3D , and combining the two-dimensional transformation parameters when the maximum global similarity is achieved to obtain the final registration result of the X-ray image to be registered; According to the final registration result (dx A ,dy A ,θ A ,θ 3D_A ) and the final registration result (dx B ,dy B ,θ B ,θ 3D_B ) to obtain the final registration result in the three-dimensional space coordinate system.

2. The two-dimensional-three-dimensional image rigid registration method according to claim 1, characterized in that: Adjusting the two-dimensional transformation parameters until determining the maximum global similarity that the i-th out-of-plane DRR image can achieve with the X-ray image to be registered includes: Initialize the two-dimensional transformation parameters (dx i ,dy i ,θ i ) is (0,0,0), and the global similarity S between the i-th out-of-plane DRR image and the X-ray image to be registered under the two-dimensional transformation parameters is i As the cost function, the gradient descent method is used to update the two-dimensional transformation parameters and obtain the global similarity S i The maximum value of and the two-dimensional transformation parameters when the maximum value is reached.

3. The method for rigid registration of two-dimensional and three-dimensional images according to claim 2, characterized in that: Use the gradient descent method to update the two-dimensional transformation parameters and obtain the global similarity S i The maximum values ​​include: Downsampling the i-th out-of-plane DRR image and the X-ray image to be registered multiple times to different degrees to obtain Q sampling image groups at different resolutions, each sampling image group at each resolution includes the i-th out-of-plane DRR image and the X-ray image to be registered at the resolution obtained by downsampling, and the integer parameter Q ≥ 2; Initialize the initial values ​​of the two-dimensional transformation parameters of the first iteration to (0,0,0), and initialize the number of iterations q = 1; Based on the initial values ​​of the two-dimensional transformation parameters of the qth iteration, the maximum global similarity between the i-th out-of-plane DRR image in the qth sampling image group and the X-ray image to be registered in the same sampling image group is calculated by using the gradient descent method; wherein the qth sampling image group is the sampling image group at the qth resolution from small to large; The two-dimensional transformation parameters that reach the maximum global similarity in the qth iteration are used as the initial values ​​of the two-dimensional transformation parameters in the q+1th iteration, and the q+1th iteration is entered until the number of iterations q=Q.

4. The method for rigid registration of two-dimensional and three-dimensional images according to claim 1, characterized in that: Calculating the bone texture similarity value between the i-th transformed DRR image and the corresponding X-ray image to be registered in each block area includes: Calculating the information entropy of each block region according to the gradient value of the X-ray image to be registered in each block region, and selecting the block region containing the bone texture as the valid block region according to the information entropy of each block region; The similarity between the i-th transformed DRR image and the X-ray image to be registered in each valid block area is calculated as the bone texture similarity value in the valid block area.

5. The method for rigid registration of two-dimensional and three-dimensional images according to claim 4, characterized in that: The bone texture similarity value s of the i-th transformed DRR image and the X-ray image to be registered in each valid block area is: Among them, x i is the gradient value of the i-th pixel point in the effective block area of ​​the i-th transformed DRR image, y i is the gradient value of the i-th pixel point of the X-ray image to be registered in the effective block area, and the i-th pixel point of the i-th transformed DRR image in the effective block area corresponds to the i-th pixel point of the X-ray image to be registered in the effective block area; n is the number of pixels in the effective block area.

6. The method for rigid registration of two-dimensional and three-dimensional images according to claim 4, characterized in that: The global similarity S between the i-th out-of-plane DRR image and the X-ray image to be registered under the current two-dimensional transformation parameters is obtained by statistics. i include: The bone texture similarity values ​​in all valid block areas are normalized respectively to obtain the normalized bone texture similarity values ​​in each valid block area, and the number m of valid block areas whose normalized bone texture similarity values ​​are greater than the similarity threshold σ is determined to obtain the global similarity between the i-th out-of-plane DRR image and the X-ray image to be registered under the current two-dimensional transformation parameters. M is the total number of valid block areas, and the similarity threshold is 0.5<σ<1.

7. The method for rigid registration of two-dimensional and three-dimensional images according to claim 4, characterized in that: Calculate the information entropy of each block area and filter out the valid block areas including: Calculate the information entropy of each block area Among them, p k is the ratio of the number of pixels with the kth gradient value to the total number of pixels in the block area, and K is the number of gradient values ​​in the block area; The information entropy of all block regions is normalized to obtain the normalized information entropy of each block region, and the block region whose normalized information entropy is greater than the information entropy threshold is taken as a valid block region.

8. The method for rigid registration of two-dimensional and three-dimensional images according to claim 7, characterized in that: Determining the information entropy threshold includes: Calculate the average and minimum values ​​of the normalized information entropy of all block areas, and take the median of the average and minimum values ​​of the normalized information entropy as the information entropy threshold.

9. The method for rigid registration of two-dimensional and three-dimensional images according to claim 1, characterized in that: The two-dimensional-three-dimensional image rigid registration method further comprises: Performing image transformation on the reference CT image according to the final registration result in the three-dimensional space coordinate system to obtain a DRR image that is completely registered with the X-ray image to be registered as a target DRR image; The target DRR image and the corresponding X-ray image to be registered are divided into multiple display blocks, and the similarity value between the target DRR image and the corresponding X-ray image to be registered in each display block is calculated, and the target DRR image is superimposed and visually displayed according to the display format corresponding to the similarity value in each display block.

10. The method for rigid registration of two-dimensional and three-dimensional images according to claim 1, characterized in that: The final registration result in the three-dimensional space coordinate system is obtained according to the final registration result of the first X-ray image to be registered and the final registration result of the second X-ray image to be registered, including: The final registration result (dx A ,dy A ,θ A ,θ 3D_A ) and the final registration result (dx B ,dy B ,θ B ,θ 3D_B ) performs 2D-3D back-projection geometric transformation to obtain the registration result in the three-dimensional space coordinate system.

Images