A three-dimensional image iterative registration method

Abstract

The present invention discloses a three-dimensional image iterative registration method, comprising the following steps: (1) obtaining a reference image and a target image, calculating the coordinates of the centers of each vertebral segment and obtaining the initial transformation relationship between the two; (2) generating multiple reference cross-sectional images in the reference image, and obtaining the corresponding target cross-sectional images by cutting out the target image according to step (1); (3) obtaining the transformation amount of a certain target cross-sectional image and the corresponding reference cross-sectional image through the best similarity matching, and calculating the posture of the corresponding target cross-sectional coordinate system; (4) judging whether the posture increments of multiple target cross-sectional coordinate systems are all less than a set threshold value, if so, calculating the final transformation relationship between the reference image coordinate system and the target image coordinate system; otherwise, updating the transformation amount and the posture of the corresponding target cross-sectional coordinate system, and repeating this step. The present invention can quickly complete the precise registration between two vertebral segment images, with high efficiency and strong adaptability, and can also achieve a good registration effect for poor image data.

Description

Technical Field

[0001] The invention relates to the field of image processing, and in particular to a three-dimensional image iterative registration method. Background Art

[0002] With the development of biomedical engineering and computer technology, medical imaging provides multimodal medical images for clinical diagnosis. In today's clinical field, these medical images are playing an increasingly important role, not only in medical diagnosis, but also in the formulation of surgical plans, formulation of radiotherapy plans, tracking of pathological changes and evaluation of treatment effects.

[0003] Medical image registration technology is an important technical link in medical imaging technology. Its essence is to find one (or a series of) spatial transformation relationships so that the corresponding points of two images can reach complete consistency in spatial position or anatomical structure, so that all anatomical points on the two image data, or at least all points with diagnostic significance, can achieve the best matching state.

[0004] However, the registration methods in the prior art have technical problems such as large registration errors and low registration efficiency. Summary of the invention

[0005] Purpose of the invention: In view of the above-mentioned shortcomings, the present invention provides an iterative registration method for three-dimensional images based on cross-sectional image template matching, which can quickly complete the accurate registration between the three-dimensional images of two vertebrae, has high efficiency and strong adaptability, and can achieve good registration effects even for image data with poor accuracy.

[0006] Technical solution:

[0007] A three-dimensional image iterative registration method comprises the steps of:

[0008] (1) Obtaining a reference image and a target image, and calculating the standard center coordinates of each vertebral segment in the reference image and the target image, respectively, and obtaining the initial transformation relationship between the reference image and the target image based on the coordinates;

[0009] (2) generating a plurality of reference cross-sectional images passing through the centers of the respective vertebral segments in the reference image, aligning the standard centers of the respective vertebral segments in the reference image and the target image according to the initial transformation relationship obtained in step (1), and using the aforementioned plurality of reference cross-sectional images to perform interception on the target image to obtain a plurality of corresponding target cross-sectional images;

[0010] (3) obtaining a transformation amount of a target cross-sectional image by best similarity matching with a corresponding reference cross-sectional image, and calculating the position and posture of the target cross-sectional coordinate system corresponding to the target cross-sectional image in the reference image based on the transformation amount;

[0011] (4) Determine whether the posture increments of multiple target section coordinate systems are all less than a set threshold. If so, calculate the final transformation relationship between the reference image coordinate system and the target image coordinate system in combination with step (1) and step (3). Otherwise, update the transformation amounts of multiple target section images and the corresponding postures of the target section coordinate systems respectively, and repeat this step until the posture increments of multiple target section coordinate systems are all less than the set threshold, and thereby obtain the final transformation relationship between the reference image coordinate system and the target image coordinate system.

[0012] The process of obtaining the standard center coordinates of each vertebral segment in the corresponding image in step (1) is as follows:

[0013] (11) Calculate the initial center coordinates of each vertebral segment in the reference image and the target image:

[0014] (12) performing distance clustering on all points in the image with respect to the initial center coordinates of each vertebra calculated in step (11) to obtain corresponding point sets, and calculating the current center coordinates of each point set;

[0015] (13) Perform distance clustering on all points in the image with respect to the current center coordinates to obtain new corresponding point sets, and calculate the current center coordinates of each new point set;

[0016] (14) Repeat step (13) until the current center coordinates of each point set remain unchanged, and use the final current center coordinates as the standard center coordinates of the corresponding vertebral segments.

[0017] The step (11) is specifically as follows:

[0018] The coordinate range of the known image in the image coordinate system is X min , X max , Y min , Y max , Z min , Z max , where the number of vertebrae is N;

[0019] Then the initial center coordinates of each vertebra are obtained:

[0020]

[0021] Among them, P i0 represents the coordinates of the initial center point of the i-th vertebra, and i represents the i-th vertebra in the image.

[0022] The distance clustering of step (12) is specifically as follows: the distance from each point in the image to the initial center coordinates of each vertebral segment is calculated respectively, and the point is classified into the vertebral segment corresponding to the initial center coordinate closest to it, and all points in the image are traversed to obtain N vertebral node sets.

[0023] The step (2) aligns the reference image and the target image by using the standard center coordinates of the reference image and the standard center coordinates of the target image obtained in step (1) as reference marking points, performing least squares rigid body alignment to obtain the transformation relationship between the reference image and the target image.

[0024] The best similarity matching in step (3) is specifically as follows:

[0025] (31) setting a reference ROI image on the reference cross-sectional image, traversing the corresponding target cross-sectional image of the target image to obtain the corresponding target ROI image and sequentially performing similarity matching evaluation on the reference ROI image to obtain the target ROI image corresponding to the first matching metric value that satisfies the set conditions, and based on this, obtaining the translation transformation relationship between the reference cross-sectional image and the corresponding target cross-sectional image;

[0026] (32) rotating the target cross-sectional coordinate system corresponding to the target cross-sectional image around an axis perpendicular to the cross-sectional image coordinate system and passing through the origin of the coordinate system at a set rotation angle and a set step length, calculating the matching metric value between the target cross-sectional image and the reference cross-sectional image after each rotation, and taking the matching metric value calculated last time as the second matching metric value when the matching metric value is less than the first matching metric value;

[0027] (33) Change the rotation direction and rotate the target section image coordinate system by setting a new step size with a step size reduced by 0.1°. After each rotation, update and calculate the matching metric value between the corresponding target section image and the corresponding reference section image and compare it with the second matching metric value until it is greater than the second matching metric value. At this time, the optimal rotation angle of the target section image is obtained.

[0028] The first matching metric value is the maximum matching metric value calculated in step (31).

[0029] The first matching metric value is the first matching metric value calculated in step (31) that is greater than 0.9.

[0030] In the step (31), the similarity matching of the two cross-sectional images is expressed as follows:

[0031] Assume that the reference roi image and the target roi image are I1 and I2 respectively, wherein the reference roi image is defined as the middle area image of the reference cross-sectional image, and the target roi image is obtained by setting the target roi image frame in the target cross-sectional image and traversing and cutting from the target cross-sectional image; and Represents the point p on the I1 and I2 images respectively ij The gray value of and Respectively represent the average gray value of all points on the I1 and I2 images, and the matching metric value is calculated as follows:

[0032]

[0033] Among them, η is the matching metric value.

[0034] The step length of the target roi image frame traversing the target cross-sectional image is set to 1 pixel value.

[0035] The step (3) is specifically:

[0036] For a certain vertebral segment, assuming that the optimal rotation angle is θ in the current target section coordinate system position, the rotation transformation is T rot ; The displacement transformation amounts are offX, offY, and offZ respectively, then the translation transformation is T trans , we get:

[0037]

[0038]

[0039] The step (4) is specifically as follows:

[0040] The multiple target cross-sectional images are respectively a sagittal cross-sectional image, an axial cross-sectional image and a coronal cross-sectional image. Assume that the transformation from the target image coordinate system to the reference image coordinate system is T transform , whose initial value is the initial transformation relationship obtained in step (1); the initial positions of the sagittal section coordinate system, axial section coordinate system and coronal section coordinate system corresponding to the multiple target section images are SA tag0 SS tag0 , SC tag0 , then the current The adjustment relationship and the position and posture of each target section coordinate system are adjusted as follows:

[0041]

[0042] In the formula, Respectively represent the rotation and translation transformation of the current target cross-sectional image; SP tag Take Substitute into the above formulas for iterative calculations, and enter the initial pose SA in the first calculation tag0 SS tag0 , SC tag0 Perform calculations; It represents the transformation relationship from the target image to the reference image after the last iteration. During the initial calculation, the initial transformation value T0 is substituted for calculation.

[0043] In step (4), the increment threshold is set to 0.001, and the pose increment λ of a single target section coordinate system is defined as the pose increment matrix T diff The determinant value of is described as follows:

[0044]

[0045] Among them, θ is the optimal rotation angle; offX, offY, and offZ are the displacement transformation amounts of the target roi image matching.

[0046] Beneficial effects: The present invention can quickly complete the accurate registration between two CT vertebral images, has high efficiency and strong adaptability, and can achieve better registration effect even for poor image data. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 is a flow chart of the present invention;

[0048] Figure 2 This is a schematic diagram of the vertebral segment center obtained by the present invention;

[0049] Figure 3 is a cross-sectional view of the reference image in the present invention, wherein: Figure 3 (a) is a schematic diagram of the sagittal section of the reference image. Figure 3 (b) is a schematic diagram of the axial section of the reference image. Figure 3 (c) is a schematic diagram of the coronal cross section of the reference image;

[0050] Figure 4 is a schematic cross-sectional view of a reference image in the present invention;

[0051] Figure 5 is a cross-sectional view of the target image in the present invention, wherein: Figure 5 (a) is a schematic diagram of the sagittal section of the target image. Figure 5 (b) is a schematic diagram of the axial section of the target image. Figure 5 (c) is a schematic diagram of the coronal cross section of the target image;

[0052] Figure 6 It is a flowchart of iterative registration of the present invention;

[0053] Figure 7 is a schematic diagram of a reference cross-sectional image in the present invention;

[0054] Figure 8 It is a schematic diagram of the process of similarity matching between the reference cross-sectional image and the target cross-sectional image in the present invention. DETAILED DESCRIPTION

[0055] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments.

[0056] Figure 1 It is a flow chart of the present invention, as Figure 1 As shown, the present invention mainly solves the registration problem of two vertebral CT images, where the two vertebral CT images are a reference image and a target image respectively; the method of the present invention comprises the following steps:

[0057] (1) Obtain the reference image and target image, and establish the corresponding image coordinate systems respectively;

[0058] (2) Preliminary positioning of the vertebrae in the image is performed to obtain the standard center coordinates of each vertebrae in the reference image and the target image, respectively. i and P j , the specific process is:

[0059] (21) Calculate the initial center coordinates of each vertebral segment in the reference image and the target image as P i0 , P j0 :Let the coordinate range of the reference image in the reference image coordinate system be X min , X max , Y min , Y max , Z min , Z max , the coordinate value range varies the most in the length direction of the spine; each image contains N vertebrae, and the present invention takes each image containing three vertebrae as an example to explain the specific steps in detail. In other embodiments, the number of vertebrae in each image is determined according to the actual situation; the initial center coordinate P of each vertebrae in the reference image i0 The following formula is used for calculation:

[0060]

[0061] Among them, P i0 represents the coordinates of the initial center point of the i-th vertebra, i represents the i-th vertebra in the reference image;

[0062] Using the same method, calculate the initial center coordinates P of each vertebra in the target image. j0 , where j represents the jth vertebral segment in the target image;

[0063] (22) Calculate the standard center coordinates P of each vertebral segment in the reference image and the target image i , P j ;

[0064] (a) Align all pixels in the reference image to the initial center coordinates P of the corresponding vertebra calculated in step (21) i0 Perform clustering classification, that is, calculate the initial center coordinates P from each pixel point in the reference image to each vertebral segment i0 The distance D ik =|Pik P i0 |,P ik Represents the kth pixel in the reference image; the pixel P ik Classified into the vertebral segment corresponding to the initial center coordinate closest to it, all pixel points in the reference image can be segmented into one of the three corresponding vertebral segments, and three vertebral node sets Q are obtained. i =P i1 ,…,P ik ,…,P iK ; Where K represents the number of pixels in the reference image;

[0065] (b) Calculate the current center coordinates of each vertebral segment through the three vertebral node sets m is an integer greater than or equal to 0;

[0066] (c) Repeat the process of aligning all pixels in the reference image to the latest current center coordinate P. im Perform distance clustering and calculate the new current center coordinates P im , until the current center coordinate of the vertebral segment does not change, and the final current center coordinate P im As the standard center coordinates P of each vertebra i ,like Figure 2 As shown;

[0067] (d) Using the methods of steps 22(a), 22(b), and 22(c), calculate the standard center coordinates P of each vertebral segment in the target image. j ;

[0068] (3) obtaining cross-sectional images of the reference image and the target image;

[0069] (31) The reference image coordinate system is used as the reference coordinate system, and the standard center coordinates of each vertebral segment in the reference image are used as the origin to establish the two-dimensional cross-sectional coordinate system S corresponding to the cross-sectional image of each vertebral segment. Ai , S Si , S Ci , respectively referred to as the first sagittal section coordinate system, the first axial section coordinate system and the first coronal section coordinate system, and a section image with a size of 521mmX512mm is intercepted on each section coordinate system to form a reference section image, namely the first sagittal section image, the first axial section image and the first coronal section image, such as Figure 3 (a), 3(b), 3(c); where the first axial section coordinate system is parallel to the XY plane of the reference image coordinate system, the first sagittal section is parallel to the YZ plane of the reference image coordinate system, and the first coronal section is parallel to the XZ plane of the reference image coordinate system. The cross-sectional image of the reference image is shown in Figure 4 As shown;

[0070] (32) The standard center coordinates P of the reference image calculated in step (2) i and the standard center coordinates P of the target image j The reference mark point is used to perform the least squares rigid body registration, and the overall preliminary registration of the reference image and the target image is completed to obtain the initial transformation relationship T0 between the two. This transformation relationship is applied to the target image, and the first sagittal section image, the first axial section image and the first coronal section image formed in step (31) are used to respectively intercept the target image to form corresponding target section images, which are the second sagittal section image, the second axial section image and the second coronal section image, respectively, as shown in FIG. Figure 5 (a), 5(b), and 5(c), the three two-dimensional coordinate systems where the second sagittal section image, the second axial section image, and the second coronal section image are located are S Aj , S Sj , S Cj ,;

[0071] (4) Figure 6 As shown, the cross-sectional image of the target image is matched with the cross-sectional image of the corresponding reference image to obtain the transformation relationship between the two, and the current target cross-sectional coordinate system position is updated;

[0072] Specifically:

[0073] (41) performing similarity matching on the target cross-sectional image and the corresponding reference cross-sectional image to obtain an optimal translation transformation relationship;

[0074] The similarity matching process between the reference cross-sectional image and the target cross-sectional image is as follows: Figure 8 As shown, the details are as follows:

[0075] (a) setting a reference ROI image on the reference cross-sectional image, the reference ROI image is defined as the middle area image of the reference cross-sectional image and is square, and the width of the reference ROI image is 1 / 2 of the width of the reference cross-sectional image; setting a target ROI image frame in the target cross-sectional image, the target ROI image frame traverses from the upper left corner of the target cross-sectional image, and sequentially intercepts the target ROI image frame to obtain multiple target ROI images until the target ROI image frame traverses to the lower right corner of the target cross-sectional image; wherein the traversal step is set to a pixel value;

[0076] (b) performing similarity matching evaluation calculations on the reference ROI image and the target ROI image obtained in step (a) of step 41 in sequence, and obtaining a matching metric value η and a position offset of the target ROI image;

[0077] Among them, the similarity matching of two roi images can be evaluated by the classic correlation coefficient evaluation value. Suppose the reference roi image and the target roi image are I1 and I2 respectively. and Represents the point p on the I1 and I2 images respectively ij The gray value of and Respectively represent the average gray value of all points on the I1 and I2 images, then the matching metric value can be expressed as follows:

[0078]

[0079] Among them, η is the matching metric value. The closer the η value is to 1, the more similar the two cross-sectional images are, and the closer it is to -1, the less similar the two cross-sectional images are. Figure 7 ;

[0080] (c) Selecting the reference ROI image and the target ROI image corresponding to the maximum matching metric value in step (b) of step 41 as the best similarity matching result between the target section image and the reference section image, and obtaining the corresponding translation matching metric value η1 and the corresponding offset of the target section image; in other embodiments, in order to improve the registration efficiency, the similarity matching evaluation calculation can be stopped when η is greater than or equal to 0.9, and the current reference ROI image and the target ROI image are used as the best similarity matching results between the target section image and the reference section image;

[0081] (d) According to step (c) in step 41, the translation matching metric value η1 and the corresponding offset of the target cross-sectional image and the corresponding reference cross-sectional image are obtained, that is, the offsets offX, offY, offZ of the target cross-sectional coordinate system to the reference cross-sectional coordinate system are obtained, thereby obtaining the translation transformation T between the two. trans ;

[0082] (42) performing similarity matching iterations on the target cross-sectional image and the corresponding reference cross-sectional image to obtain an optimal rotation transformation relationship;

[0083] (a) The direction in which the target cross-sectional coordinate system corresponding to the target cross-sectional image rotates clockwise or counterclockwise around a certain axis is specified as the angle increasing direction. In the present invention, the axis is perpendicular to the cross-sectional image coordinate system and passes through the origin of the coordinate system. The target cross-sectional coordinate system is rotated around the axis with the set step length using the set rotation angle as the set step length, and the target cross-sectional image is updated;

[0084] (b) After each rotation, the formula of step (b) in step (41) is used to update and calculate the first rotation matching metric value η2 between the target cross-sectional image and the corresponding reference cross-sectional image; the current first rotation matching metric value η2 is compared with the previous translation matching metric value η1. If η2>η1, the direction is increased according to the current rotation angle to set the step size and continue to rotate around the axis until η2<η1. The rotation matching metric value calculated last time is taken as the final first rotation matching metric value η2, and the process goes to step (c) in step 42;

[0085] (c) Changing the rotation direction, setting the step size to be reduced by 0.1° as a new step size, rotating the target section coordinate system around the axis with the new step size, and updating the target section image after each rotation, using the formula of step (41)(b) to update and calculate the second rotation matching metric value η3 between the target section image and the corresponding reference section image and compare it with the first rotation matching metric value η2, until η3>η2, at which time the best rotation matching metric value η3 between the target section image and the corresponding reference section image is obtained, thereby obtaining the best rotation angle;

[0086] (43) The update of the coordinate system of a certain current target section is performed as follows:

[0087] For a certain vertebral segment, assume that the optimal rotation angle of the current target section image under a certain current target section coordinate system is θ, that is, the rotation transformation is T rot ; The displacement transformation amounts of the target roi image matching are offX, offY, and offZ, that is, the translation transformation is T trans , we get:

[0088]

[0089]

[0090] According to the above formula, the target section coordinate system is transformed to obtain the current position and posture of the target section coordinate system;

[0091] (5) Calculate the transformation relationship from the target image to the reference image and the current position of the first sagittal section coordinate system, the first axial section coordinate system, and the first coronal section coordinate system:

[0092] Assume that the transformation relationship from the target image to the reference image is T transform , which is the transformation relationship between the target image coordinate system and the reference image coordinate system, and its initial transformation value is T0; the initial positions of the second sagittal section coordinate system, the second axial section coordinate system, and the second coronal section coordinate system are SA tag0 SS tag0 , SC tag0, obtained from step (31); then under the current adjustment, the transformation relationship from the current target image to the reference image is The coordinate system position adjustment of each target section is calculated using the following formula:

[0093]

[0094] In the formula, Respectively represent the rotation and translation transformation of the current target cross-sectional image; SP tag Take Substitute into the above formulas for iterative calculations, and enter the initial pose SA in the first calculation tag0 SS tag0 , SC tag0 Perform calculations; It represents the transformation relationship from the target image to the reference image after the last iteration. During the initial calculation, the initial transformation value T0 is substituted for calculation.

[0095] (6) Update the pose of the target image coordinate system until the pose increment of the three target section coordinate systems of the target image is less than the set threshold:

[0096] After the poses of the three target section coordinate systems of a single vertebral segment are updated using step (4), the pose increments of the three target section coordinate systems (sagittal section coordinate system, axial section coordinate system, and coronal section coordinate system) in the target image need to be calculated in sequence;

[0097] Compare the position increments of the three target section coordinate systems with the set threshold. If the position increments of the three target section coordinate systems are all less than the threshold, the process ends and the current transformation relationship is used. As the final transformation relationship; if any of the position increments of the three target section coordinate systems is greater than the set threshold, the latest position of the three target section coordinate systems is calculated using the formula in step (5) and the position of the three target section coordinate systems is updated in the target image. At the same time, the translation transformation T of the updated section coordinate system is calculated using the formulas in steps (4) and (5). trans and the rotation transformation T rot And transformation relationship Based on this, the latest pose increments of the three target section coordinate systems are calculated and compared with the set threshold; this step is repeated until the pose increments of the three target section coordinate systems are all less than the set threshold.

[0098] In the present invention, the increment threshold is set to 0.001, and the pose increment λ of a single cross-sectional coordinate system is defined as the pose increment matrix T diff The determinant value of is described as follows:

[0099]

[0100] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the specific details in the above embodiments. Within the technical concept of the present invention, various equivalent transformations (such as the number of vertebrae, etc.) can be made to the technical scheme of the present invention, and these equivalent transformations all belong to the protection scope of the present invention.

Claims

1. A three-dimensional image iterative registration method, characterized in that: Includes steps: (1) Obtaining a reference image and a target image, and calculating the standard center coordinates of each vertebral segment in the reference image and the target image, respectively, and obtaining the initial transformation relationship between the reference image and the target image based on the coordinates; (2) generating a plurality of reference cross-sectional images passing through the centers of the respective vertebral segments in the reference image, aligning the standard centers of the respective vertebral segments in the reference image and the target image according to the initial transformation relationship obtained in step (1), and using the aforementioned plurality of reference cross-sectional images to perform interception on the target image to obtain a plurality of corresponding target cross-sectional images; (3) obtaining a transformation amount of a target cross-sectional image by best similarity matching with a corresponding reference cross-sectional image, and calculating the position and posture of the target cross-sectional coordinate system corresponding to the target cross-sectional image in the reference image based on the transformation amount; (4) Determine whether the posture increments of multiple target section coordinate systems are all less than a set threshold. If so, calculate the final transformation relationship between the reference image coordinate system and the target image coordinate system in combination with step (1) and step (3). Otherwise, update the transformation amounts of multiple target section images and the corresponding postures of the target section coordinate systems respectively, and repeat this step until the posture increments of multiple target section coordinate systems are all less than the set threshold, and thereby obtain the final transformation relationship between the reference image coordinate system and the target image coordinate system.

2. The three-dimensional image iterative registration method according to claim 1, characterized in that: The process of obtaining the standard center coordinates of each vertebral segment in the corresponding image in step (1) is as follows: (11) Calculate the initial center coordinates of each vertebral segment in the reference image and the target image: (12) performing distance clustering on all points in the image with respect to the initial center coordinates of each vertebra calculated in step (11) to obtain corresponding point sets, and calculating the current center coordinates of each point set; (13) Perform distance clustering on all points in the image with respect to the current center coordinates to obtain new corresponding point sets, and calculate the current center coordinates of each new point set; (14) Repeat step (13) until the current center coordinates of each point set remain unchanged, and use the final current center coordinates as the standard center coordinates of the corresponding vertebral segments.

3. The three-dimensional image iterative registration method according to claim 2, characterized in that: The step (11) is specifically as follows: The coordinate range of the known image in the image coordinate system is X min , X max , Y min , Y max , Z min , Z max , where the number of vertebrae is N; Then the initial center coordinates of each vertebra are obtained: Among them, P i0 represents the coordinates of the initial center point of the i-th vertebra, and i represents the i-th vertebra in the image.

4. The three-dimensional image iterative registration method according to claim 3, characterized in that: The distance clustering of step (12) is specifically as follows: the distance from each point in the image to the initial center coordinates of each vertebral segment is calculated respectively, and the point is classified into the vertebral segment corresponding to the initial center coordinate closest to it, and all points in the image are traversed to obtain N vertebral node sets.

5. The three-dimensional image iterative registration method according to claim 1, characterized in that: The step (2) aligns the reference image and the target image by using the standard center coordinates of the reference image and the standard center coordinates of the target image obtained in step (1) as reference marking points, performing least squares rigid body alignment to obtain the transformation relationship between the reference image and the target image.

6. The three-dimensional image iterative registration method according to claim 1, characterized in that: The best similarity matching in step (3) is specifically as follows: (31) setting a reference ROI image on the reference cross-sectional image, traversing the corresponding target cross-sectional image of the target image to obtain the corresponding target ROI image and sequentially performing similarity matching evaluation on the reference ROI image to obtain the target ROI image corresponding to the first matching metric value that satisfies the set conditions, and based on this, obtaining the translation transformation relationship between the reference cross-sectional image and the corresponding target cross-sectional image; (32) rotating the target cross-sectional coordinate system corresponding to the target cross-sectional image around an axis perpendicular to the cross-sectional image coordinate system and passing through the origin of the coordinate system with a set rotation angle as a step size, calculating the matching metric value between the target cross-sectional image and the reference cross-sectional image after each rotation, and taking the matching metric value calculated last time as the second matching metric value when the matching metric value is less than the first matching metric value; (33) Change the rotation direction, and rotate the target section image coordinate system with the aforementioned set rotation angle reduced by 0.1° as the new step size, and update the matching metric value between the corresponding target section image and the corresponding reference section image after each rotation and compare it with the second matching metric value until it is greater than the second matching metric value, at which time the optimal rotation angle of the target section image is obtained.

7. The three-dimensional image iterative registration method according to claim 6, characterized in that: The first matching metric value is the maximum matching metric value calculated in step (31).

8. The three-dimensional image iterative registration method according to claim 6, characterized in that: The first matching metric value is the first matching metric value calculated in step (31) that is greater than 0.

9.

9. The three-dimensional image iterative registration method according to claim 6, characterized in that: In the step (31), the similarity matching of the two cross-sectional images is expressed as follows: Assume that the reference roi image and the target roi image are I1 and I2 respectively, wherein the reference roi image is defined as the middle area image of the reference cross-sectional image, and the target roi image is obtained by setting the target roi image frame in the target cross-sectional image and traversing and cutting from the target cross-sectional image; and Represents the point p on the I1 and I2 images respectively ij The gray value of and Respectively represent the average gray value of all points on the I1 and I2 images, and the matching metric value is calculated as follows: Among them, η is the matching metric value.

10. The three-dimensional image iterative registration method according to claim 9, characterized in that: The step length of the target roi image frame traversing the target cross-sectional image is set to 1 pixel value.

11. The three-dimensional image iterative registration method according to claim 6, characterized in that: The step (3) is specifically: For a certain vertebral segment, assuming that the optimal rotation angle is θ in the current target section coordinate system position, the rotation transformation is T rot ; The displacement transformation amounts are offX, offY, and offZ respectively, then the translation transformation is T trans , we get:

12. The three-dimensional image iterative registration method according to claim 11, characterized in that: The step (4) is specifically: The multiple target cross-sectional images are respectively a sagittal cross-sectional image, an axial cross-sectional image and a coronal cross-sectional image. Assume that the transformation from the target image coordinate system to the reference image coordinate system is T transform , whose initial value is the initial transformation relationship obtained in step (1); the initial positions of the sagittal section coordinate system, axial section coordinate system and coronal section coordinate system corresponding to the multiple target section images are SA tag0 SS tag0 , SC tag0 , then the current The adjustment relationship and the position and posture of each target section coordinate system are adjusted as follows: In the formula, Respectively represent the rotation and translation transformation of the current target cross-sectional image; SP tag Take Substitute into the above formulas for iterative calculations, and enter the initial pose SA in the first calculation tag0 SS tag0 , SC tag0 Perform calculations; It represents the transformation relationship from the target image to the reference image after the last iteration. During the initial calculation, the initial transformation value T0 is substituted for calculation.

13. The three-dimensional image iterative registration method according to claim 1 or 11, characterized in that: In step (4), the increment threshold is set to 0.001, and the pose increment λ of a single target section coordinate system is defined as the pose increment matrix T diff The determinant value of is described as follows: Among them, θ is the optimal rotation angle; offX, offY, and offZ are the displacement transformation amounts of the target roi image matching.

Images