Pedicle double-screw automatic planning method

Abstract

The invention discloses a vertebral pedicle double-screw automatic planning method. The method comprises the following steps: step (1), segmenting an image of a vertebral section to be subjected to screw placement through a pre-trained vertebral section segmentation model; (2) according to the segmentation result of the step (1) and clinical prior experience, initial screw placement channels of vertebral pedicles on the two sides of the vertebral section to be subjected to screw placement are obtained; (3) obtaining a screw mutual collision loss function and a screw touch boundary loss function, and constructing an optimization objective function through weighting; and (4) on the basis of the initial screw placement channel obtained in the step (2), carrying out iterative optimization on the optimization objective function constructed in the step (3) until the number of iterations reaches the maximum number of iterations or an optimization objective is achieved, and obtaining a final screw placement channel. According to the method, the automatic planning of the double-screw implanting track can be realized, a pair of optimal screw channels which are firm in mechanics, safe in space and free of mutual interference can be automatically determined, and the preoperative preparation time is greatly shortened.

Description

Technical Field

[0001] This invention relates to the field of pedicle screw technology, and more particularly to an automatic planning method for dual pedicle screws. Background Technology

[0002] In spinal fixation surgery, pedicle screw fixation has become a key method for achieving spinal stability reconstruction and correction. While traditional single-screw fixation provides sufficient stability in most routine cases, its mechanical properties remain insufficient in special situations such as severe osteoporosis, complex spinal fractures, severe deformity correction, or revision surgery. To overcome these limitations, double-screw fixation techniques are increasingly being used clinically, such as... Figure 1 As shown, two screws are implanted in the same pedicle. This technique significantly improves the overall anchoring force and stability of the screw system by increasing the bone-screw contact area.

[0003] However, double-screw implantation demands extremely high surgical precision. The pedicle anatomy is complex, adjacent to important neurovascular tissues, requiring careful surgical avoidance of penetrating the pedicle cortex or invading the spinal canal. Currently, the planning and execution of double-screw implantation rely entirely on the surgeon's experience, requiring multiple intraoperative fluoroscopic confirmations. This process is cumbersome and somewhat subjective, especially in complex cases. The challenge remains in clinical practice to rationally plan the implantation trajectory of the two screws, including the entry point, angle, depth, and relative spatial relationship, ensuring they do not interfere with each other within a limited anatomical space while achieving maximum support and optimal biomechanical performance, while simultaneously avoiding mutual interference. Summary of the Invention

[0004] Purpose of the invention: To address the above-mentioned shortcomings, this invention proposes an automatic planning method for pedicle screws, which can determine the optimal screw channels that are mechanically robust, spatially safe, and do not interfere with each other, thereby realizing automatic planning of dual screw channels based on images.

[0005] Technical solution: This invention provides an automatic planning method for pedicle screw fixation, comprising: Step (1): Segment the image of the vertebral segment to be nailed using a pre-trained segmentation model; Step (2): Based on the segmentation results of step (1) and prior clinical experience, obtain the initial nail placement channels of the pedicles on both sides of the vertebral segment to be nailed; Step (3): Obtain the screw collision loss function and the screw contact boundary loss function, and then construct the optimization objective function by weighting them. Step (4): Based on the initial pinning channel obtained in step (2), iteratively optimize the optimization objective function constructed in step (3) until the number of iterations reaches the maximum number of iterations or the optimization objective is achieved, and obtain the final pinning channel.

[0006] Specifically, step (2) is as follows: Based on the segmentation results of step (1), obtain the channel direction of the pedicle on both sides of the vertebral segment to be nailed; Based on prior clinical experience, random sampling is performed on the point cloud of the superior articular process and the transverse process on each side of the vertebral segment to be screwed, to obtain two initial channel outer points. Combined with the channel direction, the corresponding channel inner points are obtained, thereby obtaining the initial screw placement channels of the pedicles on both sides of the vertebral segment to be screwed.

[0007] More specifically, step (2) includes: Step (21): Fit a plane to the voxel points of the superior endplate in the segmented image of the vertebral segment to be nailed; Step (22): Project the voxel points of the pedicles on both sides of the vertebral segment to be fitted onto the plane obtained in step (21) to obtain two sets of projection points. Fit straight lines to them respectively to obtain the channel direction of the pedicles on both sides of the vertebral segment to be fitted. Step (23): Randomly sample the point cloud of the superior articular process and the transverse process on both sides of the vertebral segment to be screwed, and obtain the initial channel outer points on both sides. Combined with the channel direction obtained in step (22), obtain the corresponding channel inner points, and then obtain the initial screw placement channel of the double screw of the pedicle on both sides of the vertebral segment to be screwed.

[0008] Furthermore, in step (22), the direction vector of the line connecting the centroid of the pedicle and the centroid of the vertebral body is obtained, and the dot product of the two direction vectors of the fitted line with the line is calculated respectively. The direction vector with a dot product greater than 0 is selected as the corresponding channel direction vector of the pedicle.

[0009] Furthermore, in step (23), after randomly sampling and obtaining the initial channel outer points of the pedicles on both sides of the vertebral segment to be screwed, it is determined whether the distance between the two channel outer points on the same side is greater than the sum of the radii of the two screw heads; If it is greater than, then according to the screw length and the channel direction obtained in step (22), the corresponding channel inner point is obtained, thereby obtaining the initial screw placement channel of the double screws on both sides of the pedicle of the vertebral segment to be placed. Otherwise, obtain the projection points of the two initial channel outer points on the same side on the plane fitted in step (21), and move any one of the channel outer points along the direction of the line connecting the two projection points with a set distance as the step size, so that the two are separated until the distance between them is greater than the sum of the radii of the two screw heads. The two points obtained at the end are used as the two initial channel outer points on the same side.

[0010] Specifically, in step (3), the process of obtaining the screw collision loss function is as follows: Obtain the loss function F for two screws i and j on the same side.ij : ; Where θ1 and θ2 are the weight ratios of non-penetration distance loss and penetration distance loss, respectively; , These are the non-penetrating distance loss and the penetrating distance loss, respectively, which can be obtained by acquiring the distance between the surfaces of the two screws on the same side. Then the collision loss function F of screws on the same side is obtained. collision ,as follows: .

[0011] Specifically, in step (3), the loss function for obtaining the screw contact boundary is as follows: Calculate the distances between each voxel in the pedicle point cloud and the two screws on the same side, and obtain the screw contact boundary loss function F on that side. border ,as follows: ; Where, d k The distance between any voxel point in the pedicle point cloud and the axes of any two screws on the same side is represented.

[0012] Specifically, in step (4), the gradient descent method is used to iteratively optimize the objective function constructed in step (3).

[0013] Specifically, in step (4), if the objective function value does not reach the optimization target when the number of iterations reaches the maximum number of iterations, the pinning channel corresponding to the minimum value is taken as the final optimization result.

[0014] Specifically, in step (1), the image of the vertebral segment to be nailed is segmented by a pre-trained segmentation model to obtain the probability that each voxel of the vertebral segment to be nailed belongs to each anatomical structure. The argmax function is used to calculate and extract the result of the highest probability that each voxel belongs to a certain anatomical structure. Each voxel is then classified into its corresponding anatomical structure, thus obtaining the three-dimensional segmentation result of the vertebral segment to be placed.

[0015] More specifically, in step (1), the pre-trained vertebral segmentation model is obtained by acquiring a dataset of vertebral images with pre-annotated anatomical structures, dividing it into a training set and a validation set, and inputting it into the Vnet network for training; During training, the Vnet network is optimized using a hybrid loss function of Dice and cross-entropy.

[0016] Furthermore, by collecting image data of any vertebral segment from different patients, each voxel point is labeled with its corresponding anatomical structure. All voxel points labeled with the same anatomical structure constitute the three-dimensional segmentation mask of that anatomical structure. After the labeling is completed, the three-dimensional segmentation mask of all anatomical structures is obtained, thus obtaining the vertebral segment image dataset of each anatomical structure.

[0017] Beneficial effects: This invention can realize the automatic planning of dual screw implantation trajectory, and can automatically determine the optimal screw channel that is mechanically strong, spatially safe and does not interfere with each other, which can significantly shorten the preoperative preparation time and reduce the number of intraoperative fluoroscopy sessions, and has a wide range of applications in high-difficulty spinal surgery. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of this invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 Example diagram of double screw implantation.

[0020] Figure 2 This is a flowchart of the automatic planning method for pedicle screws of the present invention.

[0021] Figure 3 This is an example diagram showing the segmentation result of the vertebral segment to be screwed according to the present invention.

[0022] Figure 4 This is an example diagram of a voxel-point fitting plane for the superior endplate of a vertebral segment.

[0023] Figure 5 Example diagram to obtain the channel direction of the pedicles on both sides of the vertebral segment to be screwed.

[0024] Figure 6 Example image for obtaining the initial channel outside point.

[0025] Figure 7 This is a sagittal example diagram of the dual-screw channel planning of the present invention.

[0026] Figure 8 This is an example diagram of the coronal position of the dual-screw channel planning of the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions and advantages of the present invention clearer, the present application will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0028] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of the present invention should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0029] The automatic planning method for pedicle screws of the present invention is as follows: Figure 2 As shown, it includes: Step (1): Segment the image of the vertebral segment to be nailed using a pre-trained segmentation model.

[0030] In this invention, the pre-trained vertebral segmentation model is obtained by acquiring a dataset of vertebral images with pre-annotated anatomical structures, dividing it into a training set and a validation set, and inputting it into the Vnet network for training.

[0031] In this invention, image data of any vertebral segment from different patients are collected, and each voxel point is labeled with its corresponding anatomical structure. All voxel points labeled with the same anatomical structure constitute a three-dimensional segmentation mask for that anatomical structure. After the labeling is completed, the three-dimensional segmentation mask of all anatomical structures is obtained, thus obtaining the vertebral segment image dataset of each anatomical structure.

[0032] In this invention, during the training process of the vertebral segmentation model, the Dice + cross-entropy hybrid loss function is used to optimize the Vnet network parameters to improve the stability of training and segmentation accuracy. The weighted average of the Dice coefficients of the three-dimensional segmentation masks of each anatomical structure and the contour-based index ASD are used as evaluation indicators to monitor and ensure the accuracy of the vertebral segmentation model.

[0033] In this invention, the anatomical structures of the vertebral segments in the imaging may include the vertebral body, pedicle, superior articular process, transverse process, superior endplate, spinous process, and lamina.

[0034] In this invention, a pre-trained segmentation model is used to segment the image of the segment to be screwed, thus obtaining the probability that each voxel of the segment belongs to each anatomical structure. Then, the argmax function is used to calculate and extract the result of the maximum probability of each voxel belonging to a certain anatomical structure. This allows each voxel to be classified into its corresponding anatomical structure, thereby obtaining the three-dimensional segmentation result of the segment to be screwed. (See reference...) Figure 3 .

[0035] In this step, the pre-trained vertebral segmentation model can automatically segment the image of the vertebral segment to obtain the accurate three-dimensional spatial position and morphological features of the vertebral body, pedicle, superior articular process, transverse process and superior endplate, providing reliable data support for the preliminary and accurate positioning of the subsequent screw placement channel point.

[0036] Step (2): Based on the segmentation results of step (1) and prior clinical experience, obtain the initial nail placement channels of the pedicles on both sides of the vertebral segment to be nailed.

[0037] In this invention, based on the segmentation results of step (1), the channel direction of the pedicles on both sides of the vertebral segment to be nailed can be obtained; based on prior clinical experience, random sampling can be performed on the point cloud of the superior articular process and the point cloud of the transverse process on each side of the vertebral segment to be nailed to obtain two initial channel outer points. Combined with the aforementioned channel direction, the corresponding channel inner points can be obtained respectively, thereby obtaining the initial nail placement channel of the pedicles on both sides of the vertebral segment to be nailed.

[0038] Specifically, including: Step (21): Based on the segmentation results of step (1), fit a plane to the voxel points of the upper end plate; in some embodiments, the Random Sample Consensus (RANSAC) algorithm can be used to fit a plane to the voxel points of the upper end plate, such as... Figure 4 As shown; Step (22): Based on the segmentation results of step (1), the voxel points of the pedicles on both sides of the vertebral segment to be nailed are projected onto the plane obtained by fitting in step (21) to obtain two sets of projection point sets. Straight lines are fitted to them respectively to obtain the channel direction of the pedicles on both sides of the vertebral segment to be nailed. In this embodiment, the Random Sample Consensus (RANSAC) algorithm can be used to fit a straight line to the aforementioned set of projection points, thereby obtaining the channel direction vectors V1 and V2 of the pedicles on both sides of the vertebral segment to be implanted. Furthermore, since the fitted straight line has two directions, to determine the channel directions V1 and V2 of the pedicles on both sides of the vertebral segment to be implanted, the direction vector V1 can be obtained from the line connecting the centroid of the pedicle and the centroid of the vertebral body. b Calculate the dot product of the two direction vectors of the fitted line with the line, and select the direction vectors whose dot product is greater than 0 as the corresponding channel direction vectors of the pedicle. Figure 5 As shown; Step (23): Based on the segmentation results of step (1), randomly sample the point clouds of the superior articular processes and transverse processes on both sides of the vertebra to be screwed, as follows: Figure 6 S1 and S2 are shown, the initial channel outer points on both sides are obtained, and combined with the channel direction obtained in step (22), the corresponding channel inner points are obtained respectively, and then the initial double screw placement channel of the pedicle on both sides of the vertebral segment to be placed is obtained.

[0039] In this embodiment, after randomly sampling and obtaining the initial channel outer points of the pedicles on both sides of the vertebral segment to be screwed, it can also be determined whether the distance between the two channel outer points on the same side is greater than the sum of the radii of the two screw heads; If it is greater than, the corresponding channel point can be obtained according to the screw length and the channel direction obtained in step (22), thereby obtaining the initial double screw placement channels on both sides of the vertebral segment to be placed. Otherwise, obtain the projection points of the two channel outer points on the plane fitted in step (21) respectively, and move any one of the channel outer points along the direction of the line connecting the two projection points with a set distance as the step size, so that the two are separated until the distance between them is greater than the sum of the radii of the two screw heads. The two points obtained at the end are taken as the outer points of the two channels on the same side.

[0040] In some embodiments, the distance can be set to 1 pixel.

[0041] Step (3): Obtain the screw collision loss function and the screw contact boundary loss function, and then construct the optimization objective function by weighting.

[0042] In this invention, the distance between the surfaces of two screws on the same side is calculated, thereby obtaining the screw collision loss function on the same side, specifically: The distance between the surfaces of two screws on the same side can be converted into the distance between the axes of the two screws on the same side and then the radius of the two screws. The distance between the axes of two screws on the same side can be divided into two cases: the axes of the two screws are parallel and they are not parallel, as detailed below: ; Where, d axis V is the distance between the axes of two screws on the same side. i V j Let E represent the direction vectors of the two screws i and j, respectively, where i is not equal to j. i E j Let E and E represent any point on the axis of the two screws; ε is a minimum value used to determine whether the direction vectors of the two screws are parallel; in this embodiment, E i E j The ends of the two screws can be taken separately, that is, the points inside the channel; Let the radii of the two screws be r1 and r2, then the distance between the surfaces of the two screws on the same side can be expressed as: d cyl =d axis -(r1+r2); When d cyl When the distance is greater than 0, the two screws on the same side do not contact each other, that is, they do not penetrate each other. Let the non-penetrating distance d be... p = d cyl ; When d cyl When the value is ≤0, the two screws on the same side intersect, meaning they penetrate each other. Let the penetration distance be d. n = d cyl ; Then the loss function F of the two screws on the same side ij It can be represented as: ; in, , θ1 and θ2 are the non-penetrating distance loss and the penetrating distance loss, respectively; θ1 and θ2 are the weight ratios of the non-penetrating distance loss and the penetrating distance loss, respectively, and both of them have a value range of [0,1]. In some embodiments, θ1 can be 0.7 and θ2 can be 0.3.

[0043] Then the collision loss function F of screws on the same side can be obtained. collision ,as follows: ; The closer the distance between two screws on the same side, the greater the value of the collision loss function of the screws.

[0044] The same applies to the other side.

[0045] Calculate the distance between each voxel in the pedicle point cloud and the axis of the two screws on the same side, and obtain the screw contact boundary loss function F on that side. border ,as follows: ; Where, d k This represents the distance between any voxel point in the pedicle point cloud and any screw axis on the same side.

[0046] The closer any screw is to a voxel point in the pedicle point cloud, the greater the value of the screw's boundary contact loss function.

[0047] Therefore, the constructed optimization objective function is as follows: F (x) =αF collision +βF border； Wherein, α and β are the corresponding weight parameters, and both of them take values ​​in the range of [0,1]. In some embodiments, α can be 0.65 and β can be 0.35.

[0048] Step (4): Based on the initial pinning channel obtained in step (2), iteratively optimize the objective function constructed in step (3) until the number of iterations reaches the maximum number of iterations or the optimization objective is achieved. This yields the final pinning channel, completing the planning process. Figure 7 , 8 As shown.

[0049] In this invention, the gradient descent method can be used to iteratively optimize the objective function constructed in step (3).

[0050] In this invention, the maximum number of iterations can be 50-200.

[0051] In this invention, the optimization target is a set value.

[0052] In this invention, if the objective function value fails to reach the optimization target when the maximum number of iterations is reached, the pinning channel corresponding to the minimum value of the objective function is taken as the final optimization result.

[0053] In this invention, if the number of iterations has not yet reached the maximum number of iterations, but the objective function value has reached the optimization target, then the pinned channel corresponding to the current objective function value is taken as the minimum optimization result.

[0054] This invention integrates medical imaging data, segments the vertebral body, and classifies key areas. Based on prior clinical information and the segmentation and classification results, it determines the initial screw placement channel, formulates constraint rules based on clinical experience, and iteratively optimizes the initial screw placement channel to plan the optimal screw placement channel. This enables automatic planning of the dual screw implantation trajectory, automatically determining a pair of mechanically robust, spatially safe, and non-interfering optimal screw channels. It significantly shortens preoperative preparation time and reduces the number of intraoperative fluoroscopy sessions, and has broad application potential in complex spinal surgeries.

[0055] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples; within the framework of the invention, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of the invention as described above, which are not provided in the details for the sake of brevity.

[0056] The embodiments of this invention are intended to cover all such substitutions, modifications, and variations falling within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this invention should be included within the protection scope of this invention.

Claims

1. An automatic planning method for pedicle screw fixation, characterized in that, include: Step (1): Segment the image of the vertebral segment to be nailed using a pre-trained segmentation model; Step (2): Based on the segmentation results of step (1) and prior clinical experience, obtain the initial nail placement channels of the pedicles on both sides of the vertebral segment to be nailed; Step (3): Obtain the screw collision loss function and the screw contact boundary loss function, and then construct the optimization objective function by weighting them. Step (4): Based on the initial pinning channel obtained in step (2), iteratively optimize the optimization objective function constructed in step (3) until the number of iterations reaches the maximum number of iterations or the optimization objective is achieved, and obtain the final pinning channel.

2. The automatic planning method for double pedicle screws according to claim 1, characterized in that, Step (2) specifically involves: Based on the segmentation results of step (1), obtain the channel direction of the pedicle on both sides of the vertebral segment to be nailed; Based on prior clinical experience, random sampling is performed on the point cloud of the superior articular process and the transverse process on each side of the vertebral segment to be screwed, to obtain two initial channel outer points. Combined with the channel direction, the corresponding channel inner points are obtained, thereby obtaining the initial screw placement channels of the pedicles on both sides of the vertebral segment to be screwed.

3. The automatic planning method for double pedicle screws according to claim 2, characterized in that, Step (2) includes: Step (21): Fit a plane to the voxel points of the superior endplate in the segmented image of the vertebral segment to be nailed; Step (22): Project the voxel points of the pedicles on both sides of the vertebral segment to be fitted onto the plane obtained in step (21) to obtain two sets of projection point sets. Fit straight lines to them respectively to obtain the channel direction of the pedicles on both sides of the vertebral segment to be fitted. Step (23): Randomly sample the point cloud on the surface of the superior articular process and the transverse process on both sides of the vertebral segment to be screwed, and obtain the initial channel outer points on both sides. Combined with the channel direction obtained in step (22), obtain the corresponding channel inner points, and then obtain the initial screw placement channel of the double screw of the pedicle on both sides of the vertebral segment to be screwed.

4. The automatic planning method for double pedicle screws according to claim 3, characterized in that, In step (22), the direction vector of the line connecting the centroid of the pedicle and the centroid of the vertebral body is obtained, and the dot product of the two direction vectors of the fitted line with the line is calculated. The direction vector with a dot product greater than 0 is selected as the corresponding channel direction vector of the pedicle.

5. The automatic planning method for double pedicle screws according to claim 3, characterized in that, In step (23), after randomly sampling and obtaining the initial channel outer points of the pedicles on both sides of the vertebral segment to be placed, it is determined whether the distance between the two channel outer points on the same side is greater than the sum of the radii of the two screw heads. If it is greater than, then according to the screw length and the channel direction obtained in step (22), the corresponding channel inner point is obtained, thereby obtaining the initial screw placement channel of the double screws on both sides of the pedicle of the vertebral segment to be placed. Otherwise, obtain the projection points of the two initial channel outer points on the same side on the plane fitted in step (21), and move any one of the channel outer points along the direction of the line connecting the two projection points with a set distance as the step size, so that the two are separated until the distance between them is greater than the sum of the radii of the two screw heads. The two points obtained at the end are used as the two initial channel outer points on the same side.

6. The automatic planning method for double pedicle screws according to claim 1, characterized in that, In step (3), the specific steps for obtaining the screw collision loss function are as follows: Obtaining the loss function F of the two screws i, j on the same side ij : ； Where θ1 and θ2 are the weight ratios of non-penetration distance loss and penetration distance loss, respectively; , These are the non-penetrating distance loss and the penetrating distance loss, respectively, which can be obtained by acquiring the distance between the surfaces of the two screws on the same side. Then the collision loss function F of screws on the same side is obtained. collision ,as follows: 。 7. The automatic planning method for double pedicle screws according to claim 1, characterized in that, In step (3), the loss function for obtaining the screw contact boundary is specifically as follows: Calculate the distances between each voxel in the pedicle point cloud and the two screws on the same side, and obtain the screw contact boundary loss function F on that side. border ,as follows: ； Where, d k The distance between any voxel point in the pedicle point cloud and the axes of any two screws on the same side is represented.

8. The automatic planning method for double pedicle screws according to claim 1, characterized in that, In step (4), the gradient descent method is used to iteratively optimize the objective function constructed in step (3).

9. The automatic planning method for double pedicle screws according to claim 1, characterized in that, In step (4), if the objective function value does not reach the optimization target when the number of iterations reaches the maximum number of iterations, the pinning channel corresponding to the minimum value is taken as the final optimization result.

10. The automatic planning method for double pedicle screws according to claim 1, characterized in that, In step (1), the image of the vertebral segment to be nailed is segmented by a pre-trained segmentation model to obtain the probability that each voxel of the vertebral segment to be nailed belongs to each anatomical structure. The argmax function is used to calculate and extract the result of the highest probability that each voxel belongs to a certain anatomical structure. Each voxel is then classified into its corresponding anatomical structure, thus obtaining the three-dimensional segmentation result of the vertebral segment to be placed.

11. The automatic planning method for double pedicle screws according to claim 1 or 10, characterized in that, In step (1), the pre-trained vertebral segmentation model is obtained by acquiring a dataset of vertebral images with pre-annotated anatomical structures, dividing it into a training set and a validation set, and inputting it into the Vnet network for training. During training, the Vnet network is optimized using a hybrid loss function of Dice and cross-entropy.

12. The automatic planning method for double pedicle screws according to claim 11, characterized in that, By collecting image data of any vertebral segment from different patients, each voxel point is labeled with its corresponding anatomical structure. All voxel points labeled with the same anatomical structure constitute the three-dimensional segmentation mask of that anatomical structure. After the labeling is completed, the three-dimensional segmentation mask of all anatomical structures is obtained, thus obtaining the vertebral segment image dataset of each anatomical structure.

Images