Automatic Generation Method and System for 3D Printing Registration Guide

Abstract

The present invention provides a method and system for automatically generating a 3D printing registration guide plate, which relates to the technical field of autonomous path planning of open-circuit robots. The method includes accurately segmenting vertebral bodies in a patient's CT image by using a two-stage full-automatic spine segmentation method of 3D Swin Transformer to reconstruct a three-dimensional spine model; generating an open-circuit path by combining the patient's CT images and the reconstructed three-dimensional spine model; selecting 6 anatomical key points on the spine segment as target key points, and accurately positioning the 6 anatomical key points based on the SPU-Net three-dimensional space key point positioning algorithm, and respectively matching multiple preset guide plate models with the three-dimensional spine model; setting a bias matrix according to expert experience, and calculating the target position matrix of the best-matched guide plate model in the CT coordinate system; using Boolean subtraction operation to subtract the three-dimensional spine model from the best-matched guide plate model to obtain a patient-customized 3D printing registration guide plate with a fitting surface.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous path planning for open-loop robots, and particularly to a method and system for automatically generating a 3D printing registration guide plate. Background Art

[0002] With the continuous development of 3D printing technology, personalized guide plate manufacturing software provides new options for the large-scale clinical application of guide plates. It can greatly reduce the complexity of guide plate manufacturing, as well as software and usage costs. However, there are also problems such as low automation, a long learning curve for doctors, and weak learning interest, resulting in an inability to guarantee the planning effect and being unfavorable for clinical promotion. Summary of the Invention

[0003] Embodiments of the present invention provide a method and system for automatically generating a 3D printing registration guide plate, which can solve the problems in the prior art.

[0004] In the first aspect of the embodiments of the present invention,

[0005] A method for automatically generating a 3D printing registration guide plate is provided, including:

[0006] Precisely segment the vertebral bodies in the patient's CT image by using a two-stage full-automatic spine segmentation method of 3D Swin Transformer to reconstruct a three-dimensional spine model;

[0007] Combine the patient's CT images and the reconstructed three-dimensional spine model to generate an open-loop path;

[0008] Select 6 anatomical key points on the spinal segment as target key points, and precisely locate the 6 anatomical key points based on the SPU-Net three-dimensional space key point localization algorithm. The 6 anatomical key points are respectively the highest point of the left superior articular process, the highest point of the right superior articular process, the center of the left lamina, the center of the right lamina, the highest point of the left inferior articular process, and the highest point of the right inferior articular process;

[0009] Match multiple preset guide plate models with the reconstructed three-dimensional spine model respectively, where each guide plate model has 6 preset feature points;

[0010] Solve the rotation matrix, translation vector, and offset matrix of the best-matching guide plate model, and calculate the target position matrix of the best-matching guide plate model in the CT coordinate system;

[0011] Use Boolean subtraction operation to subtract the reconstructed three-dimensional spine model from the best-matching guide plate model to obtain a patient-customized 3D printing registration guide plate with a fitted surface.

[0012] In an alternative embodiment,

[0013] The step of respectively matching multiple preset guide plate models with the reconstructed three-dimensional spinal model includes:

[0014] Using the singular value decomposition (SVD) method to solve the rotation matrix and translation vector from the 6 anatomical key points to the feature points of each guide plate model;

[0015] Based on the rotation matrix and translation vector, calculate the matching error of each guide plate model, and determine the guide plate model with the minimum matching error as the best matching guide plate model.

[0016] In an alternative embodiment,

[0017] Calculating the matching error of each guide plate model through the following formula includes:

[0018]

[0019] where p j represents the spinal feature point, R i represents the rotation matrix, q j represents the guide plate feature point, and t i represents the translation vector.

[0020] In an alternative embodiment,

[0021] Using the singular value decomposition (SVD) method to solve the rotation matrix and translation vector from the 6 anatomical key points to the feature points of each guide plate model includes:

[0022] Extract several key points on the spinal model as the source point set, and extract the corresponding number of feature points on the guide plate model as the target point set;

[0023] Construct the correspondence matrix W from the source point set to the target point set, where Wij represents the correspondence weight between the i-th point in the source point set and the j-th point in the target point set;

[0024] Perform singular value decomposition on the matrix W to obtain three matrices U, S, and VT, where U is an m×m orthogonal matrix, S is an m×n non-negative real diagonal matrix, and VT is an n×n orthogonal matrix, satisfying W = U×S×VT;

[0025] Calculate the rotation matrix R = U×VT from the matrices U and VT, and calculate the scale factor s from the diagonal elements of the matrix S, where the scale factor s is the average value of the diagonal elements of S;

[0026] Denote the centroid of the source point set as Cs and the centroid of the target point set as Ct, then the translation vector t = Ct - s×R×Cs;

[0027] Using the rotation matrix R, the scale factor s, and the translation vector t, a rigid body transformation matrix is constructed to register the spinal model to the coordinate system of the guide plate model.

[0028] In an alternative embodiment,

[0029] Solve for the rotation matrix, translation vector, and bias matrix of the best-matching guide plate model, and calculate the target position matrix of the best-matching guide plate model in the CT coordinate system, including:

[0030] Based on the coordinates Pi of six feature points of the best-matching guide plate model in the model coordinate system, where i = 1, 2,..., 6, and the coordinates Qi of the corresponding six anatomical key points on the reconstructed three-dimensional spinal model, where i = 1, 2,..., 6, the following least-squares optimization problem is solved by the singular value decomposition (SVD) method to obtain the rotation matrix R and translation vector t from Pi to Qi:

[0031] min ∑||R * Pi + t - Qi||^2;

[0032] Set the bias matrix M to adjust the position of the best-matching guide plate model based on R and t to obtain a better initial position; wherein, the bias matrix M is determined by the prior knowledge of the relative position of the guide plate to the vertebral body and is used to improve the registration accuracy;

[0033] Multiply the rotation matrix R, the translation vector t, and the bias matrix M to obtain the target position matrix TCT of the best-matching guide plate model in the CT coordinate system:

[0034] TCT = M * [R, t];

[0035] Where [R, t] represents the transformation matrix formed by splicing R and t.

[0036] In an alternative embodiment,

[0037] The Boolean subtraction operation includes:

[0038] According to the calculated target position matrix, map the three-dimensional mesh of the best-matching guide plate model to the CT coordinate system;

[0039] Perform trilinear interpolation on the three-dimensional spinal model in the CT coordinate system and binarize the interpolated points;

[0040] Voxelize the mapped three-dimensional mesh model of the guide plate, assign a value of 1 to the voxels belonging to the guide plate model, and assign 0 to the rest;

[0041] Perform a Boolean subtraction operation on the assigned voxels, subtracting the voxels of the spinal model from the voxels of the guide plate model;

[0042] Perform surface rendering on the guide plate model after Boolean subtraction to reconstruct a three-dimensional mesh model, and generate a final patient-customized 3D printed registration guide plate.

[0043] In an alternative embodiment,

[0044] Select six anatomical key points on the spinal segment as target key points, and perform precise positioning based on the SPU-Net three-dimensional space key point positioning algorithm.

[0045] In the second aspect of the embodiments of the present invention,

[0046] Provide a 3D printed registration guide plate automatic generation system, including:

[0047] A first unit for precisely segmenting the vertebral bodies in the patient's CT image by using a two-stage full-automatic spinal segmentation method of 3D Swin Transformer to reconstruct a three-dimensional spinal model;

[0048] A second unit for generating an open path by combining the patient's CT images and the reconstructed three-dimensional spinal model;

[0049] A third unit for selecting six anatomical key points on the spinal segment as target key points, and precisely positioning the six anatomical key points based on the SPU-Net three-dimensional space key point positioning algorithm, where the six anatomical key points are respectively the highest point of the left superior articular process, the highest point of the right superior articular process, the center of the left lamina, the center of the right lamina, the highest point of the left inferior articular process, and the highest point of the right inferior articular process;

[0050] A fourth unit for respectively matching multiple preset guide plate models with the reconstructed three-dimensional spinal model, where each guide plate model has six preset feature points;

[0051] A fifth unit for solving the rotation matrix, translation vector, and offset matrix of the best-matching guide plate model, and calculating the target position matrix of the best-matching guide plate model in the CT coordinate system;

[0052] A sixth unit for using Boolean subtraction to subtract the reconstructed three-dimensional spinal model from the best-matching guide plate model to obtain a patient-customized 3D printed registration guide plate with a fitting surface.

[0053] In the third aspect of the embodiments of the present invention,

[0054] Provide an electronic device, including:

[0055] A processor;

[0056] A memory for storing processor-executable instructions;

[0057] Wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0058] In the fourth aspect of the embodiments of the present invention,

[0059] A computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0060] In this embodiment, algorithms such as automatically segmenting vertebral bodies using 3D Swin Transformer and accurately positioning key points using SPU-Net significantly reduce manual intervention, improve the automation level of the guide plate design process, and reduce the labor intensity. By matching the preset guide plate model with the patient's actual spinal model and applying Boolean subtraction operations for optimization, a personalized customized guide plate model that highly fits each patient can be generated, avoiding the one-size-fits-all general design defect. Based on high-precision medical image segmentation and registration technology, the guide plate can accurately fit the patient's spinal anatomical structure, reducing surgical errors. At the same time, the Boolean subtraction operation removes the parts that may cause conflicts, further improving the adaptability. The precise and personalized guide plate design can effectively guide surgical operations, shorten the surgical time, reduce the surgical risk, improve the surgical quality, and has important clinical practical value. Using preoperative CT image data, a customized guide plate model can be pre-generated to support the planning of the preoperative surgical path and plan, improving the preparation quality of the surgery. It can be widely applied to various surgical scenarios that require guide plate guidance, so it has very strong versatility and expansion potential. The guide plate with a personalized fitting design can minimize the trauma to the patient's body, enhance the overall surgical experience, and contribute to accelerating postoperative recovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a schematic flowchart of the method for automatically generating a 3D printing registration guide plate according to an embodiment of the present invention;

[0062] Figure 2 It is a schematic structural diagram of a system for automatically generating a 3D printing registration guide plate according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0064] The technical solution of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0065] Figure 1 It is a schematic flowchart of the method for automatically generating a 3D printing registration guide in an embodiment of the present invention. As Figure 1 shown, the method includes:

[0066] S101. Use a two-stage full-automatic segmentation method of 3D Swin Transformer to accurately segment the vertebral bodies in the patient's CT image to reconstruct a three-dimensional spinal model;

[0067] Exemplarily, read the CT image sequence of the patient, perform resampling, unify the CT images to the same spatial resolution, perform data normalization, and limit the CT value within a fixed range;

[0068] Use a pre-trained 3D Swin Transformer rough segmentation model to perform rough segmentation of the foreground (vertebral body area) and background of the CT image, and use a threshold method or other post-processing strategies to extract the roughly segmented vertebral body area; use the pre-trained 3D Swin Transformer fine segmentation model to perform fine segmentation on each vertebral body with the rough segmentation result as the input, perform three-dimensional connected component analysis on the segmentation result, remove small connected components, and retain the largest connected component as the segmentation result of a single vertebral body;

[0069] Perform morphological opening operation on the segmentation result of a single vertebral body to smooth the segmentation boundary, further optimize the segmentation result according to the prior anatomical knowledge of the vertebral body to ensure the segmentation accuracy, detect the serial number of the segmented single vertebral body, mark the type and number of each vertebral body, reconstruct the segmented single vertebral body into a three-dimensional mesh model, and correctly splice each mesh model according to the vertebral body serial number to obtain a complete three-dimensional spinal model.

[0070] Perform mesh refinement on the spliced three-dimensional spinal model to improve the accuracy, and perform operations such as filtering and smoothing on the model to optimize the model quality.

[0071] In this embodiment, the 3D Swin Transformer is an advanced segmentation algorithm based on the Transformer self-attention mechanism, which can fully exploit the three-dimensional context information in CT images, accurately segment each vertebra, and reconstruct a high-fidelity three-dimensional spine model. Compared with traditional segmentation and reconstruction methods, the segmentation accuracy and the fidelity of model details of this algorithm have been greatly improved. The two-stage segmentation framework significantly reduces the need for manual intervention through a strategy of first rough and then fine, realizes the automation of spine segmentation and reconstruction, and reduces the workload of doctors. The high-quality three-dimensional spine model can truly reflect the patient's anatomical structure, provide accurate anatomical data support for subsequent surgical path planning, guide plate registration and other processes, and fundamentally improve the reliability of surgical planning.

[0072] S102. Combine the patient's CT images and the reconstructed three-dimensional spine model to generate an open path;

[0073] Exemplarily, first read the patient's CT image and the reconstructed three-dimensional spine model, manually or automatically mark the surgical target region ROI in the CT image, map the ROI to the three-dimensional spine model, extract the corresponding mesh model as the ROI three-dimensional model, and based on the ROI three-dimensional model, use the fast marching method (FMM) or other algorithms to generate a vector field, which encodes the shortest distance and direction information from any point to the ROI surface;

[0074] According to the surgical plan, determine one or more access points on the patient's skin surface, map the positions of the access points into the vector field, start from the access points, trace back along the vector field in the reverse direction, extract the shortest path to reach the ROI surface, and the optimal path among multiple shortest paths can be selected as the open path. Smooth the extracted open path to reduce the change of path curvature, introduce anatomical structure constraints, and avoid important organs and tissue structures;

[0075] Present the optimized open path in a three-dimensional view, combine the spine model and the three-dimensional models of other anatomical structures, intuitively display the relationship between the path and the surrounding environment, and parameterize and encode the open path. For example, use parametric curves such as B-Spline to represent it. The encoded path parameters can be used for subsequent applications such as robotic surgical planning.

[0076] In this embodiment, based on the patient's own anatomical model and surgical goals, the solution can generate an optimal open-circuit path that fits the patient's individual situation and meets the surgical needs, avoiding the limitations of path planning based on standard templates. The optimized open-circuit path not only has the shortest path length, but also avoids important organs and tissue structures, reducing surgical trauma. At the same time, the path accurately matches the patient's anatomical structure, can guide surgical instruments to be precisely in place, and improve surgical accuracy. After the open-circuit path is encoded, it can provide a precise motion planning path for the robotic surgical system, realize precise surgery with human-machine collaboration, and enhance the automation and complex operation capabilities of the surgery.

[0077] S103. Select 6 anatomical key points on the spinal segment as target key points, and accurately locate the 6 anatomical key points based on the SPU-Net three-dimensional space key point positioning algorithm, wherein the 6 anatomical key points are respectively the highest point of the left superior articular process, the highest point of the right superior articular process, the center of the left vertebral plate, the center of the right vertebral plate, the highest point of the left inferior articular process, and the highest point of the right inferior articular process;

[0078] Exemplarily, the complete spinal 3D mesh model reconstructed in S101 is read, the spinal segment of interest is determined according to the surgical requirements, the spinal model is rotationally corrected to be in a standard viewing angle, the model is rescaled to a fixed size required for network input, the circumscribed sphere of the model is calculated and moved to the center of the sphere, and the model is ensured to be located at the center of the field of view;

[0079] Using the pre-trained SPU-Net 3D key point detection model, the pre-processed spine model is input into SPU-Net, and the network outputs the coordinates of the six key points of each vertebra on the segment;

[0080] The coordinates of the key points output by the network are non-maximum suppressed to remove outliers, and only the key points with high confidence are retained. The key point positions are further optimized and corrected based on anatomical prior knowledge, and the optimized key point coordinates are converted from the normalized space output by the network back to the original model space. The six key points are marked as the left and right superior articular processes according to anatomical semantics, and the three-dimensional coordinates of the key points are exported and marked to the subsequent registration module.

[0081] S104. Matching a plurality of preset guide plate models with the reconstructed three-dimensional spine model respectively, wherein each guide plate model has six preset feature points;

[0082] Import the built guide plate model. The feature points on the guide plate model have been defined during modeling. For guide plate i, define Q = (q1, q2, q3, q4, q5, q6) i Six feature points are defined for each guide plate model.

[0083] In an optional embodiment,

[0084] The step of respectively matching a plurality of preset guide plate models with the reconstructed three-dimensional spine model includes:

[0085] Using the singular value decomposition (SVD) method to solve the rotation matrix and translation vector from the 6 anatomical key points to the feature points of each guide plate model;

[0086] Based on the rotation matrix and translation vector, calculate the matching error of each guide plate model, and determine the guide plate model with the minimum matching error as the best matching guide plate model.

[0087] Exemplarily, the SVD decomposition method is used to solve the relative relationship between the spine feature points and the guide plate feature points, and the guide plate with the minimum matching error is used for the operation. Let the spine feature points P = (p1, p2, p3, p4, p5, p6) and the guide plate feature points Q = (q1, q2, q3, q4, q5, q6) i The transformation matrix T:

[0088]

[0089] where R is a 3×3 rotation matrix and t is a 3D translation vector. The problem is transformed into finding suitable R and t such that for all i ∈ (1, n) it satisfies:

[0090] p i = Rq i + t;

[0091] In practical applications, due to the interference of errors and noises, the problem is transformed into a minimization problem of the following formula:

[0092]

[0093] In an alternative embodiment,

[0094] Calculating the matching error of each guide plate model through the following formula includes:

[0095]

[0096] where p j represents the spine feature point, R i represents the rotation matrix, q j represents the guide plate feature point, and t i represents the translation vector.

[0097] In an alternative embodiment,

[0098] Using the singular value decomposition (SVD) method to solve the rotation matrix and translation vector from the 6 anatomical key points to the feature points of each guide plate model includes:

[0099] Extract a number of key points on the spine model as the source point set, and extract the corresponding number of feature points on the guide plate model as the target point set;

[0100] Construct the correspondence matrix W from the source point set to the target point set, where Wij represents the correspondence weight between the i-th point in the source point set and the j-th point in the target point set;

[0101] Perform singular value decomposition on the matrix W to obtain three matrices U, S, and VT. Among them, U is an m×m orthogonal matrix, S is an m×n non-negative real diagonal matrix, and VT is an n×n orthogonal matrix, satisfying W = U×S×VT;

[0102] Calculate the rotation matrix R = U×VT from the matrices U and VT, and calculate the scale factor s from the diagonal elements of the matrix S, where the scale factor s is the average value of the diagonal elements of S;

[0103] Denote the centroid of the source point set as Cs and the centroid of the target point set as Ct, then the translation vector t = Ct - s×R×Cs;

[0104] Use the rotation matrix R, the scale factor s, and the translation vector t to construct a rigid body transformation matrix to register the spine model to the coordinate system of the guide plate model.

[0105] Exemplarily, calculate the average values of the point sets P and Q and where,

[0106]

[0107] For perform SVD decomposition:

[0108] Define Then use Det(UV T ) to represent the determinant of UV T and define:

[0109]

[0110] Thus, R = UV T ,

[0111] For each guide plate Ti, a set of rotation matrices R i and translation t i can be calculated. Calculate the matching error of each guide plate, and adopt the winner-takes-all strategy to select the guide plate model with the minimum error as the finally adopted guide plate.

[0112] In this embodiment, by registering the actual spinal model of the patient with the ideal guide model, the optimal implantation position and angle of the guide on the patient's spine can be accurately calculated, thereby guiding the surgical process, reducing surgical errors, and improving implantation accuracy. The accurate implantation position of the guide can avoid unnecessary damage to surrounding tissues, reduce surgical risks, and ensure surgical safety. At the same time, the probability of postoperative complications is also reduced. Compared with the traditional method of manually measuring and adjusting the position of the guide, this method can quickly and automatically complete the registration of the guide and the spine, simplify the surgical procedure, and shorten the surgical time. It reduces the influence of the subjective experience of the surgeon, can provide a consistent reference for guide implantation among different surgeons, and improve the consistency of surgical quality. Through the spinal model and guide model constructed from preoperative imaging data such as CT, the registration calculation can be completed before the operation, the best surgical plan can be formulated, and the accuracy of surgical planning can be improved.

[0113] S105. Solve the rotation matrix, translation vector, and offset matrix of the best-matching guide model, and calculate the target position matrix of the best-matching guide model in the CT coordinate system;

[0114] Since the feature points and the feature points selected by the guide are all on the surface, directly using the initial pose matrix of the guide placed under CT may result in too small a registration surface and insufficient registration. Therefore, an offset matrix needs to be set additionally based on expert experience.

[0115] In an alternative embodiment,

[0116] Solving the rotation matrix, translation vector, and offset matrix of the best-matching guide model, and calculating the target position matrix of the best-matching guide model in the CT coordinate system, includes:

[0117] According to the coordinates P i of the 6 feature points of the best-matching guide model in the model coordinate system, where i = 1, 2,..., 6, and the coordinates Q i of the corresponding 6 anatomical key points on the reconstructed three-dimensional spinal model, where i = 1, 2,..., 6, by using the singular value decomposition (SVD) method, solve the following least squares optimization problem to obtain the rotation matrix R and translation vector t from P i to Q i :

[0118] min∑||R*P i + t - Q i ||^2;

[0119] Set the offset matrix M, which is used to adjust the position of the best-matching guide model based on R and t to obtain a better initial position; wherein, the offset matrix M is determined by the prior knowledge of the relative position of the guide to the vertebral body and is used to improve the registration accuracy;

[0120] Multiply the rotation matrix R, the translation vector t, and the bias matrix M to obtain the target position matrix TCT of the best-matching guide plate model in the CT coordinate system:

[0121] TCT = M * [R, t];

[0122] Where [R, t] represents the transformation matrix formed by splicing R and t.

[0123] In this embodiment, by solving the least squares optimization problem through singular value decomposition, the rotation matrix R and the translation vector t for optimally registering the guide plate model to the patient's actual spinal model can be obtained. Then, combined with the bias matrix M introduced by expert experience, the initial registration result is optimized and adjusted. This way of combining mathematical optimization with expert knowledge can greatly improve the registration accuracy between the guide plate model and the actual spine, thereby guiding the more accurate implantation of the guide plate into the ideal position during surgery. Due to the improvement of registration accuracy, the guide plate can be implanted more precisely into the planned position and angle, avoiding unnecessary damage to surrounding tissues, thus reducing the surgical risk and improving surgical safety. Traditionally, determining the implantation position of the guide plate requires a large amount of manual measurement and repeated adjustment by the surgeon, which is time-consuming and laborious. By automatically completing the optimal registration before surgery, the surgical process is simplified and the surgical time is shortened. And it reduces the influence of the surgeon's subjective experience, can provide a consistent reference for guide plate implantation for different surgeons, and improves the consistency of surgical quality.

[0124] S106. Using the Boolean subtraction operation, subtract the reconstructed three-dimensional spinal model from the best-matching guide plate model to obtain a patient-customized 3D printed registration guide plate with a fitted surface.

[0125] After completing the matching of the guide plate and the vertebral body position, a fitted surface between the guide plate and the vertebral body needs to be generated, that is, subtracting the model of the vertebral body from the model of the guide plate.

[0126] In an alternative embodiment,

[0127] The Boolean subtraction operation includes:

[0128] According to the calculated target position matrix, map the three-dimensional grid of the best-matching guide plate model to the CT coordinate system;

[0129] Perform trilinear interpolation on the three-dimensional spinal model in the CT coordinate system, and perform binarization processing on the interpolated points;

[0130] Voxelize the mapped three-dimensional grid model of the guide plate, assign a value of 1 to the voxels belonging to the guide plate model, and assign other values to 0;

[0131] Perform a Boolean subtraction operation on the assigned voxels, subtracting the voxels of the spinal model from the voxels of the guide plate model;

[0132] Perform surface rendering on the guide plate model after Boolean subtraction to reconstruct the 3D mesh model, and generate the final patient-customized 3D printed registration guide plate.

[0133] Exemplarily, define the original guide plate model as A and the vertebral body model as B. The generation method of the guide plate model C with a fitted surface is represented by the following formula. This operation is a Boolean subtraction in model operations:

[0134] C = A - A ∩ B;

[0135] Specifically, map the 3D mesh model of the guide plate to the CT coordinate system according to the calculated target position matrix, and perform subdivision interpolation on the vertebral body model in the CT coordinate system. Since the fineness of the CT data is insufficient, the CT slice thickness used in the research process is 0.6 mm. If no processing is performed, it will cause jagged edges after the guide plate model is voxelized. Therefore, a trilinear interpolation method is used to interpolate the CT space and the target vertebral body.

[0136] Among them, the larger the interpolation coefficient, the higher the resolution of the interpolated data. To balance accuracy and operation speed, the interpolation coefficient is taken as 3. After completing the three-dimensional linear interpolation, binarize the interpolated points, assign points greater than 0.5 as 1, and the rest as 0.

[0137] Voxelize the mapped stl model and map it to the expanded CT coordinate system. It uses a geometric intersection method to quickly voxelize the 3D mesh surface, and then uses a scan line seed filling algorithm to fill the internal and external voxels, assign voxels belonging to the model as 1, and the rest as 0.

[0138] Perform Boolean subtraction on the voxel points of the segmented vertebral body model. For a certain point, if the voxel value I of the guide plate model at this point A is 0, then the value I of this point C is 0; if the voxel value I of the guide plate model at this point A is 1 and the voxel value I of the vertebral body model is B 1, then the value I of this point C is 0, otherwise it is 1. That is:

[0139]

[0140] Perform surface rendering on the vertebral body model and the calculated guide plate respectively. Use the Marching-cube algorithm to perform surface rendering on the guide plate after Boolean operation, re-establish its 3D mesh model, finally generate the guide plate, and finally perform 3D printing and processing on the generated guide plate before surgery and install the visual corner points, that is, complete the generation and production of the guide plate.

[0141] In an alternative embodiment,

[0142] Six anatomical key points on the spinal segment are selected as target key points, and accurate positioning is carried out based on the SPU-Net three-dimensional space key point positioning algorithm.

[0143] In this embodiment, through Boolean subtraction operation, the part of the guide plate model that coincides with the patient's spinal anatomical structure can be effectively removed, ensuring that the finally printed guide plate can accurately fit the patient's spine and avoiding surgical conflicts. The guide plate model after Boolean subtraction contains the negative imprint information of the patient's spinal anatomical structure, and this shape feature can provide better visual and tactile feedback during the operation, helping the surgeon quickly locate the implantation position of the guide plate. The customized guide plate that fits precisely can minimize the damage to surrounding tissues and reduce the surgical risk. At the same time, the Boolean subtraction operation also eliminates the parts that may cause conflicts. After removing the part overlapping with the spine, the shape of the guide plate is more reasonable, and it can maximize its role in guiding the screw implantation. The use of a customized guide plate can avoid the traditional process of repeatedly measuring and adjusting the position of the guide plate, simplify the surgical procedure, and shorten the operation time. By generating a customized guide plate model for each patient, which perfectly fits the patient's own anatomical characteristics, it helps to achieve personalized and precise treatment.

[0144] Figure 2 It is a schematic structural diagram of the 3D printing registration guide plate automatic generation system according to the embodiment of the present invention, as Figure 2 shown, the system includes:

[0145] The first unit is used to accurately segment the vertebral bodies in the patient's CT image by using the two-stage full-automatic segmentation method of the 3D Swin Transformer to reconstruct the three-dimensional spinal model;

[0146] The second unit is used to generate an open path by combining the patient's CT pictures and the reconstructed three-dimensional spinal model;

[0147] The third unit is used to select six anatomical key points on the spinal segment as target key points, and accurately locate the six anatomical key points based on the SPU-Net three-dimensional space key point positioning algorithm. The six anatomical key points are respectively the highest point of the left superior articular process, the highest point of the right superior articular process, the center of the left lamina, the center of the right lamina, the highest point of the left inferior articular process, and the highest point of the right inferior articular process;

[0148] The fourth unit is used to match multiple preset guide plate models with the reconstructed three-dimensional spinal model respectively, and each guide plate model has six preset feature points;

[0149] The fifth unit is used to solve the rotation matrix, translation vector and offset matrix of the best-matching guide plate model, and calculate the target position matrix of the best-matching guide plate model in the CT coordinate system;

[0150] The sixth unit is configured to subtract the reconstructed three-dimensional spinal model from the best-matching guide model by using Boolean subtraction to obtain a patient-customized 3D printed registration guide with a fitted surface.

[0151] In a third aspect of the embodiments of the present invention,

[0152] a kind of electronic device is provided, including:

[0153] a processor;

[0154] a memory for storing instructions executable by the processor;

[0155] wherein, the processor is configured to call the instructions stored in the memory to execute the method described above.

[0156] In a fourth aspect of the embodiments of the present invention,

[0157] a computer-readable storage medium is provided, on which computer program instructions are stored, and when the computer program instructions are executed by a processor, the method described above is implemented.

[0158] The present invention may be a method, an apparatus, a system, and / or a computer program product. The computer program product may include a computer-readable storage medium having thereon computer-readable program instructions for performing various aspects of the present invention.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatically generating a 3D printing registration guide, characterized in that: include: The two-stage fully automatic spine segmentation method using 3D Swin Transformer is used to accurately segment the vertebrae in the patient's CT images to reconstruct the three-dimensional model of the spine; combining the patient's CT image and the reconstructed three-dimensional model of the spine to generate an open path; Six anatomical key points on the spinal segment are selected as target key points, and the six anatomical key points are accurately positioned based on the SPU-Net three-dimensional space key point positioning algorithm. The six anatomical key points are the highest point of the left superior articular process, the highest point of the right superior articular process, the center of the left vertebral plate, the center of the right vertebral plate, the highest point of the left inferior articular process, and the highest point of the right inferior articular process; Matching a plurality of preset guide plate models with the reconstructed three-dimensional spine model respectively, wherein each guide plate model has 6 preset feature points; Solve the rotation matrix, translation vector and offset matrix of the best matching guide model, and calculate the target position matrix of the best matching guide model in the CT coordinate system; Using a Boolean subtraction operation, subtracting the reconstructed three-dimensional spine model from the best matching guide model to obtain a patient-customized 3D printed registration guide with a fitting surface; Solve the rotation matrix, translation vector and offset matrix of the best matching guide model, and calculate the target position matrix of the best matching guide model in the CT coordinate system, including: According to the coordinates P of the six feature points of the best matching guide plate model in the model coordinate system i , and the coordinates Q of the six anatomical key points corresponding to the reconstructed three-dimensional model of the spine i , through the singular value decomposition SVD method, we can get i To Q i The rotation matrix R and translation vector t; Setting a bias matrix M for adjusting the position of the best matching guide plate model based on R and t; The rotation matrix R, the translation vector t and the offset matrix M are multiplied to obtain the target position matrix of the best matching guide model in the CT coordinate system.

2. The method according to claim 1, characterized in that The step of matching a plurality of preset guide plate models with the reconstructed three-dimensional spine model respectively comprises: Using the singular value decomposition (SVD) method, the rotation matrix and translation vector from the six anatomical key points to each guide model feature point are solved; Based on the rotation matrix and the translation vector, the matching error of each guide plate model is calculated, and the guide plate model with the smallest matching error is determined as the best matching guide plate model.

3. The method according to claim 2, characterized in that The matching error of each guide plate model is calculated by the following formula: ； in, p j represents the characteristic points of the spine, represents the rotation matrix, q j represents the feature points of the guide plate, Represents the translation vector.

4. The method according to claim 2, characterized in that: The singular value decomposition (SVD) method is used to solve the rotation matrix and translation vector from the six anatomical key points to each guide model feature point, including: A number of key points are extracted from the spine model as a source point set, and a corresponding number of feature points are extracted from the guide plate model as a target point set; Construct the correspondence matrix W from the source point set to the target point set, where W ij Represents the weight of the correspondence between the i-th point in the source point set and the j-th point in the target point set; Perform singular value decomposition on the matrix W to obtain three matrices U, S, and VT, where U is an m×m orthogonal matrix, S is an m×n non-negative real diagonal matrix, and VT is an n×n orthogonal matrix, satisfying W=U×S×VT; The rotation matrix R = U × VT is calculated from the matrices U and VT, and the scale factor s is calculated from the diagonal elements of the matrix S, where the scale factor s is the average value of the diagonal elements of S; The centroid of the source point set is recorded as Cs, and the centroid of the target point set is recorded as Ct, then the translation vector t=Ct-s×R×Cs; The rigid body transformation matrix is ​​constructed using the rotation matrix R, scale factor s and translation vector t to align the spine model to the guide model coordinate system.

5. The method according to claim 1, characterized in that Six anatomical key points on the spinal segment are selected as target key points, and precise positioning is performed based on the SPU-Net three-dimensional space key point positioning algorithm.

6. The method according to claim 1, characterized in that Boolean subtraction operations include: According to the calculated target position matrix, mapping the three-dimensional grid of the best matching guide plate model to the CT coordinate system; Perform trilinear interpolation on the three-dimensional model of the spine in the CT coordinate system, and perform binarization on the interpolated points; voxelize the mapped three-dimensional mesh model of the guide plate, assign the voxels belonging to the guide plate model a value of 1, and assign the rest to 0; Performing Boolean subtraction operation on the assigned voxels, subtracting the spine model voxels from the guide model voxels; The guide model after Boolean subtraction operation is surface rendered to reconstruct the three-dimensional mesh model and generate the final patient-customized 3D printed registration guide.

7. A 3D printing registration guide automatic generation system, used to implement the method described in any one of claims 1 to 6, characterized in that: include: The first unit is used to accurately segment the vertebrae in the patient's CT images using a two-stage fully automatic spine segmentation method using 3D Swin Transformer to reconstruct a three-dimensional spine model; The second unit is used to generate an open path by combining the CT image of the patient and the reconstructed three-dimensional model of the spine; The third unit is used to select 6 anatomical key points on the spinal segment as target key points, and accurately locate the 6 anatomical key points based on the SPU-Net three-dimensional space key point positioning algorithm, wherein the 6 anatomical key points are respectively the highest point of the left superior articular process, the highest point of the right superior articular process, the center of the left vertebral plate, the center of the right vertebral plate, the highest point of the left inferior articular process and the highest point of the right inferior articular process; A fourth unit is used to match a plurality of preset guide plate models with the reconstructed three-dimensional spine model respectively, wherein each guide plate model has 6 preset feature points; The fifth unit is used to solve the rotation matrix, translation vector and offset matrix of the best matching guide plate model, and calculate the target position matrix of the best matching guide plate model in the CT coordinate system; The sixth unit is used to subtract the reconstructed three-dimensional spine model from the best matching guide model by using Boolean subtraction operation to obtain a patient-customized 3D printed registration guide with a fitting surface.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images