A multi-perspective coupled method for table object positioning

Abstract

The present invention discloses a method for positioning a table object coupled with multiple perspective images, comprising the following steps: inputting precise position information of an object in a three-dimensional computer tomography image and a positioning scheme, placing the object on the table, and performing initial positioning of the object according to the positioning scheme; photographing the placed object from the side to form a number of two-dimensional X-ray perspective images at different imaging angles; preprocessing the two-dimensional X-ray perspective image so that the pixel value therein is proportional to the density value of the object through which the X-ray passes; performing rigid geometric registration from the two-dimensional X-ray perspective image to the three-dimensional computer tomography image, obtaining translation and rotation parameters, and moving the table according to the parameters to make the object reach a predetermined treatment position. The present invention can be applied to processing imaging hardware systems not limited to the above two types of hardware, and can be flexibly used for precise positioning of objects in different scenarios; can track the internal structure of the object; can be made without contacting the object and without damaging the surface of the object.

Description

Technical Field

[0001] The invention relates to an object positioning method, in particular to a multi-perspective image coupling table object positioning method. Background Art

[0002] Precise machining usually requires precise three-dimensional positioning of workpieces. Positioning through dual orthogonal X-ray registration is a new method for positioning workpieces. Also known as image-guided positioning technology, this technology collects 1 to N two-dimensional perspective images of the object before machining, compares them with the three-dimensional computed tomography image of the object, obtains the spatial transformation between the object's machining positioning position and the current position, and uses the control surface to move the object to the preset machining position.

[0003] The two-dimensional image can be acquired by a variety of hardware, the common ones are: 1. An X-ray imaging system independent of the processing tool with two sets of mutually orthogonal axes, each set of X-ray imaging system consists of an X-ray tube source and an X-ray detection plate; 2. A KV-level cone beam CT (Cone Beam, CBCT) system installed on a rotating device. Among them, the first type of hardware can quickly take two X-ray fluoroscopic images at orthogonal angles, and the second type of hardware can dynamically take multiple X-ray fluoroscopic images at different angles as the C-arm rotates. The first type has strict requirements on the installation position of the imaging system, which is difficult to adapt to different applications, resulting in a limited scope of application; the second type requires the device to be rotated at a certain angle during positioning, resulting in a longer positioning time, and because the object may be deformed during the rotation process, and the rotating device may be unstable during the rotation process, the positioning accuracy may be reduced. Summary of the invention

[0004] Purpose of the invention: The present invention aims to provide a multi-perspective coupled table object positioning method that can expand the scope of application and improve accuracy.

[0005] Technical solution: The multi-perspective coupled table object positioning method of the present invention comprises the following steps:

[0006] Step 1: Input the 3D CT image and the precise position information of the object in the positioning plan, place the object on the table, and perform the initial positioning of the object according to the positioning plan;

[0007] Step 2: Take a side shot of the placed object to form several two-dimensional X-ray perspective images at different imaging angles;

[0008] Step 3: Preprocess the two-dimensional X-ray image so that the pixel value is proportional to the density value of the object through which the X-ray passes;

[0009] Step 4: Perform rigid geometric registration from the two-dimensional X-ray image to the three-dimensional computed tomography image, obtain translation and rotation parameters, and move the table according to the parameters to make the object reach the predetermined treatment position.

[0010] Wherein, in step 4, the registration method is based on one or more two-dimensional X-ray images and three-dimensional computer tomography images to obtain a rigid registration transformation; in step 3, the preprocessing of the two-dimensional X-ray images includes the following steps:

[0011] Step 1: X-ray scatter correction, removing the low-frequency scattered components on the X-ray image;

[0012] Step 2: X-ray hardening correction, so that the pixel value of the X-ray perspective image is linearly related to the density of the object passed through.

[0013] In the present invention, the translation vector under the x, y, z coordinate axes and the rotation angle around the x, y, z axes are interactively adjusted until the digitally reconstructed X-ray perspective image dynamically generated according to the translation and rotation coincides with the perspective image obtained by shooting to realize rigid geometric registration; based on the two-dimensional registration results of each perspective image and the digitally reconstructed X-ray perspective image of the corresponding angle, iterative optimization is performed to form a three-dimensional transformation to realize rigid geometric registration; based on the iterative three-dimensional projection gradient descent method, 1 to 2 perspective images are randomly selected from multiple perspective images in each iteration, and the gradient of the spatial transformation is calculated to obtain the rigid registration transformation of multiple perspective images and the computer tomography image; the rigid geometric registration can choose to abandon any degree of freedom, and calculate and output the registration results of only translation or translation combined with rotation.

[0014] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: it can be applied to processing imaging hardware systems not limited to the above two types of hardware, supports three-dimensional registration between X-ray perspective images based on 1 to N angles and three-dimensional computer tomography images, and is flexibly used for precise positioning of objects in different scenes; it can track the internal structure of objects; it does not need to touch the objects and does not damage the surface of the objects. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a structural schematic diagram of the present invention;

[0016] Figure 2 It is a schematic diagram of an X-ray imaging device with two sets of axes orthogonal to each other according to the present invention;

[0017] Figure 3 Schematic diagram of the imaging system installed on the C-arm of the present invention. DETAILED DESCRIPTION

[0018] The following is combined with Figure 1-3 The technical solution of the present invention is further described.

[0019] As shown in the figure, a multi-perspective coupled table object positioning method includes the following steps:

[0020] Step 1: Input the 3D CT image and the precise position information of the object in the positioning scheme, place the object on the table, and perform the initial positioning of the object according to the positioning scheme; the 3D image value at position p in the object coordinate system under the initial positioning is determined by the input information, that is, V(p), where p represents the coordinates of any position within the CT range of the object coordinate system, p = (x, y, z) T .

[0021] Step 2: Take a side shot of the placed object to form several two-dimensional X-ray perspective images at different imaging angles; Img i i=1,…n,n≥1, where each image combines the source and imaging plane coordinate system information, and can be used to point the source to the image. From source to point All density information on the rays between them is linked;

[0022] Step 3: Preprocess the two-dimensional X-ray image so that the pixel value is proportional to the density value of the object through which the X-ray passes;

[0023] Step 4: Perform rigid geometric registration from the two-dimensional X-ray image to the three-dimensional computed tomography image, obtain translation and rotation parameters, and move the table according to the parameters to make the object reach the predetermined treatment position.

[0024] In step 4, the registration method is based on registering one or more two-dimensional X-ray images with the three-dimensional computed tomography image to obtain a rigid registration transformation, that is, the number of two-dimensional X-ray images in A can be one or any number;

[0025] Interactively adjust the translation vectors under the x, y, z coordinate axes and the rotation angles around the x, y, z axes until the digitally reconstructed X-perspective image dynamically generated according to the translation and rotation coincides with the photographed perspective image to achieve rigid geometric alignment;

[0026] Based on the two-dimensional registration results of each perspective image and the digitally reconstructed X-ray perspective image at the corresponding angle, iterative optimization forms a three-dimensional transformation to achieve rigid geometric registration;

[0027] Based on the iterative 3D projection gradient descent method, 1 to 2 perspective images are randomly selected from multiple perspective images in each iteration, and the gradient of the spatial transformation is calculated to obtain the rigid registration transformation between the multiple perspective images and the computed tomography image;

[0028] Rigid geometric registration can choose to give up any degree of freedom and calculate and output the registration results of only translation or translation combined with rotation.

[0029] In step 3, the preprocessing of the 2D X-ray image includes the following steps:

[0030] Step 1: X-ray scatter correction, remove the low-frequency scattered components on the X-ray image; when X-rays penetrate an object, the direction of the rays remains unchanged before scattering occurs, and the ray intensity decays according to Beer's law. If scattering occurs, the direction and energy of the rays change. Therefore, the pixel values ​​in the final image will correspond to a part of the scattered X-rays. For this purpose, scatter correction is required, such as using principal component modulation filtering, adding a filter grid at the source, etc.

[0031] Step 2: X-ray hardening correction, so that the pixel value of the X-ray perspective image is linearly related to the density of the object passed through. The X-rays generated by the CT tube have a certain spectrum width. Since the low-energy rays of the continuous energy spectrum are easily absorbed after passing through the object, the high-energy rays are easier to pass through. During the propagation of the rays, the average energy will become higher and the rays will gradually harden. This is called the beam hardening effect. The beam hardening effect makes the pixel brightness in the X-ray perspective image not proportional to the thickness of the object passed by the rays. Through hardening correction, the X-ray perspective image can truly reflect the thickness and density information of the object, reducing the difficulty of image registration.

[0032] According to the actual situation, one of the following three registration methods can be used to perform three-dimensional registration from the perspective image to the three-dimensional computer tomography image to obtain the table displacement d = (x, y, z), rotation (Rotation) angle θ, roll (Roll) angle φ, tilt (Pitch) angle ψ, where the spatial rotation matrix corresponding to each rotation angle is as follows:

[0033]

[0034]

[0035]

[0036] Assume that the coordinate of a point on the bed is x, and the coordinate transformation formula is:

[0037]

[0038] This method supports the following common forms of registration, and during the registration process, θ, φ and ψ can be locked to 0 to ensure that the output spatial transformation parameters can be implemented on a table that only supports four-dimensional spatial transformation and a table that supports six-dimensional spatial transformation. To obtain these parameters, one of the following methods is selected according to the actual situation:

[0039] (1) Manual registration: x, y, z, θ, φ, ψ are manually adjusted. Each adjustment moves the 3D object corresponding to the 3D CT image in the computer virtual space according to the corresponding adjusted parameters, and simulates the X-ray transmission process of the object according to the shooting parameters of each X-ray image to generate a digital reconstructed image (DRR). Each DRR is compared with the X-ray image one by one to confirm whether it has been moved into place;

[0040] (2) Multiple perspectives Figure 2 3D registration and fitting of 3D space transformation: for each set of X-ray images and DRR, perform 2D registration on them, obtain the translation vector and rotation angle in the 2D plane, establish the functional relationship between the global 3D space transformation parameters and the translation and rotation parameters of each plane, and obtain the optimal global 3D space transformation parameters through the gradient descent method. Iterate this process until the global 3D space transformation parameters converge;

[0041] (3) Direct 3D registration based on perspective images: Establish the correspondence between CT points and pixels on the DRR of the perspective image: Where T(x i ) represents the coordinates of the ith pixel after spatial transformation, V(T(x i )) represents the simulated reconstructed pixel value of the point, Vol(z) represents the density of point z in space, z∈ray(T(x i ),c) refers to the distance from the X-ray source point c to T(x i ) Based on the objective functions such as the maximum mutual information between the X-ray image and the DRR or the maximum gradient of the contour, the optimal spatial transformation parameters are obtained by the gradient descent method. For X-ray images at multiple angles, the objective functions of each angle can be accumulated as the total objective function to participate in the optimization and complete the registration process.

[0042] (4) During the registration process, some spatial transformation parameters can be locked according to the displacement and rotational degrees of freedom supported by the target table. During the optimization process, the partial differentials of these parameter dimensions are ignored and partial optimization is performed to obtain the spatial transformation parameters that ultimately meet the displacement and rotation capabilities of the table.

[0043] The following describes the specific implementation methods of two types of positioning imaging systems: dual orthogonal and C-arm:

[0044] Reference numerals: 1 - imaging plate; 2 - object; 3 - table; 4 - X-ray source; 5 - rotation axis.

[0045] 1. If Figure 1 The dual orthogonal X-ray imaging system is shown installed on the radiation table 3, next to the treatment head. The application steps are as follows:

[0046] a) The object 2 is placed on the table 3, and the table 3 is moved to the initial position according to the coordinates of the isocenter point in the treatment plan;

[0047] b) simultaneously taking two images in which the imaging planes are orthogonal to each other and the X-ray source 4 is perpendicular to the isocenter of the accelerator;

[0048] c) When shooting, a light source modulator is inserted at the source end, and after imaging, the non-scattering component is stripped and obtained by combining the phase shift of the main component (non-scattering component) introduced by the modulator through a two-dimensional discrete Fourier transform to achieve scattering correction, and the image grayscale is corrected according to a pre-stored nonlinear beam hardening correction curve, so that the obtained X-ray perspective image is linearly related to the equivalent water thickness of the X-ray penetrating the object;

[0049] d) In the registration stage, first generate the DRRs corresponding to the two X-ray images, and then use the following methods to perform the registration according to the situation:

[0050] i) Compare the DRR and X-ray images. If there is a big difference between the two or the automatic registration method cannot find the appropriate spatial transformation parameters, or the manual registration method is selected, perform manual registration or obtain preliminary spatial transformation parameters;

[0051] ii) Perform two-dimensional plane registration on the two DRRs and X-ray images respectively, obtain the two-dimensional translation T1, T2 and the rotation angles θ1 and θ2 relative to the imaging center in the corresponding plane coordinate system, combine the above two-dimensional translation and rotation to form the translation and rotation parameters in three-dimensional space according to the orthogonal relationship between the two registration planes, update the DRR and further align until convergence, and combine the three-dimensional space transformations of each iteration to obtain the three-dimensional space transformation after complete registration;

[0052] iii) The spatial transformation after registration obtained in step ii is generally a rotation matrix and translation vector d = (x, y, z) T ,according to:

[0053]

[0054] Available In this way, the x, y, z, θ, φ, ψ required for bed moving are obtained.

[0055] 2. If Figure 2 The C-arm imaging system is shown, and the application steps are as follows:

[0056] a) The object 2 is placed on the table 3, and the table 3 is moved to the initial position according to the processing plan position;

[0057] b) rotating the gantry to take N transmission images of the X-ray source 4 at different angles;

[0058] c) constructing a three-dimensional computed tomography image of the target area using the FDK algorithm based on the N X-ray perspective images, obtaining the scattering component of the corresponding X-ray perspective image by simulation method based on the image and removing the component, iterating the above steps until convergence or a certain time;

[0059] d) Select M images from the N X-ray perspective images (the value of M is determined according to real-time requirements), generate DRR images corresponding to the M X-ray perspective images, and obtain the target three-dimensional space transformation through the following steps:

[0060] i) Compare the DRR and X-ray images. If there is a big difference between the two or the automatic registration method cannot find the appropriate spatial transformation parameters, or the manual registration method is selected, perform manual registration or obtain preliminary spatial transformation parameters;

[0061] ii) Perform two-dimensional plane registration on the M DRR and X-ray images respectively to obtain the two-dimensional translations T1, T2, ..., T in the corresponding plane coordinate system M and the rotation angles θ1, θ2, ..., θ relative to the imaging center M. ;

[0062] iii) Direct 3D registration based on perspective images: Establish the correspondence between the pixels on the DRR of the kth perspective image of the CT point: Where T k (x i ) represents the coordinate of the i-th pixel in the image in space, V k (T(x i )) represents the pixel value of the point, Vol(z) represents the density of point z in space, z∈ray(T(x i ),c) refers to the distance from the X-ray source point c to T(x i ) points on this ray, based on the maximum total mutual information of each pair of X-ray perspective images and DRR as the objective function, the optimal spatial transformation parameters are obtained by the gradient descent method;

[0063] iv) The spatial transformation after registration obtained in the previous three steps is converted into a rotation matrix and translation vector d = (x, y, z) T ,according to:

[0064]

[0065] Available In this way, we obtain the x, y, z, θ, φ, ψ required for the table movement.

Claims

1. A multi-perspective coupled table object positioning method, characterized in that: The following steps are involved: Step 1: Input the 3D CT image and the precise position information of the object in the positioning plan, place the object on the table, and perform the initial positioning of the object according to the positioning plan; Step 2: Take a side shot of the placed object to form several two-dimensional X-ray perspective images at different imaging angles; Step 3: Preprocess the two-dimensional X-ray image so that the pixel value is proportional to the density value of the object through which the X-ray passes; Step 4: Perform rigid geometric registration from the two-dimensional X-ray image to the three-dimensional computed tomography image, obtain translation and rotation parameters, and move the table according to the parameters to make the object reach the predetermined treatment position; perform registration based on one or more two-dimensional X-ray images and three-dimensional computed tomography images to obtain a rigid registration transformation; The registration includes obtaining the table displacement d=(x, y, z), the rotation angle θ, the roll angle φ, and the tilt angle ψ, wherein the spatial rotation matrix corresponding to each angle is as follows: Assume that the coordinate of a point on the bed is x, and the coordinate transformation formula is: The registration method includes: manual registration, multi-perspective two-dimensional registration fitting three-dimensional space transformation and three-dimensional registration directly based on perspective; The registration process supports locking θ, φ, and ψ to 0 to ensure that the output spatial transformation parameters are implemented on both the table that only supports four-dimensional spatial transformation and the table that supports six-dimensional spatial transformation. Specifically, Manual registration: manually adjusting x, y, z, θ, φ, ψ, each adjustment moves the three-dimensional object corresponding to the three-dimensional computer tomography image in the computer virtual space according to the corresponding adjusted parameters, and simulates the X-ray transmission process of the object according to the shooting parameters of each X-ray perspective image to generate a digital reconstructed image DRR, and each DRR is compared with the X-ray perspective image one by one to confirm whether it has been moved into place; The multi-perspective two-dimensional registration fitting three-dimensional space transformation: for each set of X-ray perspective images and DRR, two-dimensional registration is performed to obtain the translation vector and rotation angle in the two-dimensional plane where they are located, and a functional relationship between the global three-dimensional space transformation parameters and the translation and rotation parameters of each plane is established, and the optimal global three-dimensional space transformation parameters are obtained by the gradient descent method; this process is iterated until the global three-dimensional space transformation parameters converge; The three-dimensional registration directly based on the perspective image: establish the correspondence between the CT point and the pixel point on the DRR of the perspective image: V(T(x i ))=∑ z∈ray(T(xi),c) Vol(z), where T(x i ) represents the coordinates of the ith pixel after spatial transformation, V(T(x i )) represents the simulated reconstructed pixel value of the point, Vol(z) represents the density of point z in space, z∈ray(T(x i ),c) refers to the distance from the X-ray source point c to T(x i ) For the points on this ray, the optimal spatial transformation parameters are obtained by the gradient descent method based on the maximum mutual information between the X-ray perspective image and the DRR or the maximum contour gradient objective function; for X-ray perspective images at multiple angles, the objective functions of each angle are accumulated as the total objective function, and the registration process is completed by participating in the optimization together; During the alignment process, it is supported to lock some spatial transformation parameters according to the displacement and rotational degrees of freedom supported by the target table, and to ignore the partial differentials of the locked parameter dimensions during the optimization process to perform partial optimization, so as to obtain the spatial transformation parameters that ultimately meet the displacement and rotation capabilities of the table.

2. A multi-perspective coupled tabletop object positioning method as claimed in claim 1, characterized in that: In the step 3, the preprocessing of the two-dimensional X-ray image includes the following steps: Step 1: X-ray scatter correction, removing the low-frequency scattered components on the X-ray image; Step 2: X-ray hardening correction, so that the pixel value of the X-ray perspective image is linearly related to the density of the object passed through.

3. The multi-perspective coupled table object positioning method according to claim 1, characterized in that: In the step 4, the translation vectors under the x, y, and z coordinate axes and the rotation angles around the x, y, and z axes are interactively adjusted until the digitally reconstructed X-perspective image dynamically generated according to the translation and rotation coincides with the photographed perspective image to achieve rigid geometric alignment.

4. The multi-perspective coupled table object positioning method according to claim 1, characterized in that: In the step 4, based on the two-dimensional registration results of each perspective view and the digitally reconstructed X-ray perspective view at the corresponding angle, iterative optimization is performed to form a three-dimensional transformation to achieve rigid geometric registration.

5. The multi-perspective coupled table object positioning method according to claim 1, characterized in that: In the step 4, based on the iterative three-dimensional projection gradient descent method, 1 to 2 perspective images are randomly selected from multiple perspective images in each iteration, and the gradient of the spatial transformation is calculated to obtain the rigid registration transformation of the multiple perspective images and the computed tomography image.

6. The multi-perspective coupled table object positioning method according to claim 1, characterized in that: In the above-mentioned step 4, the rigid geometric registration can choose to abandon any degree of freedom, calculate and output the registration result of only translation or translation combined with rotation.

Images