Personalized human body spine model generation method
Through high-resolution CT scan and radial basis function interpolation combined with principal component analysis, a high-precision personalized spine model was generated, which solved the problems of cumbersome operation and insufficient accuracy in traditional methods, and achieved high adaptability and detail restoration of the personalized spine model.
Patent Information
- Application Number
- CN202510342612.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-21
- Publication Date
- 2025-07-08
AI Technical Summary
The existing technology is difficult to generate high-precision and personalized spine models. The traditional methods are cumbersome and lack real-time interactiveness. Deep learning and statistical morphological methods have shortcomings in non-rigid matching and local details refinement processing.
Image data is obtained through high-resolution CT scan, standard templates are constructed and local flexible deformation is performed, and a personalized spine model is generated using radial basis function interpolation and principal component analysis. Combining regularized parameters and iterative optimization techniques, we ensure accurate reconstruction and real-time adjustment of the model.
It realizes the generation of high-precision and personalized spinal model, improves the adaptability and detailed restoration effect of the model, and meets the needs of clinical diagnosis and surgical planning.
Smart Images
Figure CN120280097A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of biomedical engineering, and more particularly, to a method for generating a personalized human spine model. Background Art
[0002] The human spine model is an important part of the human body model. Currently, with the wide application of medical imaging technologies such as CT and MRI, three-dimensional reconstruction of various parts of the human body using image data has become an important means in the fields of clinical diagnosis, surgical planning, rehabilitation training, and customized implant design. Traditional techniques usually rely on the processing of continuous tomographic images, and generate three-dimensional spine models through image segmentation, overlay, and reconstruction methods. However, due to factors such as the resolution of scanning equipment, imaging angle, noise interference, and the large anatomical differences of the human spine among different individuals, the obtained three-dimensional models often have insufficient accuracy, missing details, and inevitable deformation errors.
[0003] In addition, in the prior art, a method based on a fixed template and manual adjustment is often used to generate a spine model. This method is not only cumbersome to operate, but also lacks real-time interactivity and personalized regulation ability, and it is difficult to meet the clinical requirements for high-precision and individualized spine models. Although some studies have attempted to introduce deep learning and statistical morphology methods to automatically extract spine structure features in recent years, most of these methods still rely on fixed templates, and there are deficiencies in non-rigid matching and local detail refinement, and the diversity and complexity of inter-individual morphology have not been fully considered. Summary of the Invention
[0004] The purpose of the present invention is to provide a method for accurately reconstructing and dynamically adjusting a personalized human spine three-dimensional model in real time to meet the needs of clinical diagnosis and surgical planning.
[0005] The technical solution of the present invention is: A method for generating a personalized human spine model is provided, and the method includes:
[0006] S1. Perform high-resolution CT scanning on a human body to be measured to obtain medical image data of the spine region; use professional image processing software to convert the image data into a preliminary three-dimensional spine model;
[0007] S2. Collect three-dimensional models formed by spine scan data of multiple healthy human bodies; mark key points including vertebral body centers, vertebral body edges, and intervertebral space connection points for each spine model, and perform statistical analysis on the positions of the key points, calculate the average position, and construct a standard template representing the typical spine structure; perform translation, rotation, and scaling transformations on the standard template to make it roughly match the spine model of the human body to be measured in terms of position and size;
[0008] S3. Use the radial basis function interpolation method to perform local flexible deformation on the standard template described in step S2;
[0009] S4. Arrange the key point data of multiple healthy human spine three-dimensional models into a high-dimensional data matrix, perform centering processing and unified alignment to eliminate rotation, translation, and scale differences; then, calculate the covariance matrix using the centered data, obtain each principal component through eigenvalue decomposition, select some principal components after sorting by variance contribution rate, project the original data to obtain the principal component scores of each model, and calculate the variance contribution rate of each principal component as the contribution weight of morphological changes; establish a mapping relationship between the clinical parameters and the principal component scores to determine the regression coefficients of each principal component; finally, based on the average shape of all aligned spine models, weighted stack each principal component to construct a statistical shape model of the personalized spine;
[0010] S5. Output the finally generated personalized spine three-dimensional model as a standard data file; before output, perform necessary file format conversion and data optimization on the model to ensure that the output model has both high precision and is applicable to practical applications in different fields;
[0011] S6. When performing personalized adjustment on the saved model, the user inputs clinical parameters through the interactive interface, and the system calculates the principal component weights in real time according to the user input parameters, generates an updated spine model, and displays it to the user in real time.
[0012] In any of the above technical solutions, further, the method for performing local flexible deformation in step S3 includes:
[0013] Use an RBF interpolator to perform local deformation on the standard template. The basic form of RBF interpolation is:
[0014]
[0015] where f(x) is the position of the deformed key point, x is the coordinate of any key point on the standard template, x i is the coordinate of the key point on the three-dimensional model obtained in step S1, N is the total number of key points, is the radial basis function, ω i is the weighting coefficient, and P(x) is the polynomial term to ensure the smoothness and continuity of the overall deformation;
[0016] The RBF interpolator needs to set constraint conditions for the position of each key point:
[0017] f(x i ) = x i ;
[0018] That is, the target position of each template key point should be equal to the position of the corresponding key point in the patient model;
[0019] The system calculates the weighted coefficient ω of each radial basis function through the above two equations i , thereby completing the training of the RBF interpolator.
[0020] In any of the above technical solutions, further, step S3 further includes: the system introduces a regularization parameter in the RBF interpolator, and the regularization parameter controls the rigidity degree of the model. When the value is small, the model can show a more flexible deformation; when the value is large, the model will be closer to the original template in local features;
[0021] The user can balance the local fitting accuracy of the model and the smoothness of the overall shape by setting the regularization parameter, and avoid unnatural deformation caused by excessive displacement of individual points of the model.
[0022] In any of the above technical solutions, further, the deformation operation of step S3 is to obtain a higher matching accuracy, and the system will perform multiple iterative optimizations on the result of the RBF interpolator: after each deformation, the system calculates the displacement error between the template model and the patient's spine model at key points and non-key points; for the area with a large error, the system will add new key points in this area and recalculate the RBF interpolator to further refine the deformation of this area; when the overall matching error reaches the preset threshold, the iterative optimization ends.
[0023] In any of the above technical solutions, further, after obtaining the image data in step S1, preprocessing methods such as noise reduction, gray normalization, and contrast enhancement are adopted for the image data to segment the spine area.
[0024] In any of the above technical solutions, further, after converting the image data into a preliminary three-dimensional spine model in step S1, the Laplace smoothing algorithm is used to optimize the preliminary model to eliminate artifacts and reconstruction noise.
[0025] In any of the above technical solutions, further, step S4 specifically includes:
[0026] Convert the key point coordinates of the three-dimensional models formed by the spine scan data of multiple healthy human bodies collected in step S2 into vector form, and arrange them in sequence to form a high-dimensional data matrix X, where each component of each vector corresponds to the coordinate value of each key point in the unified coordinate system, and the high-dimensional data matrix X is a matrix with n rows and p columns;
[0027] Perform centering processing on the high-dimensional data matrix X to obtain the centered data matrix Xx:
[0028] X c = X - μ;
[0029] where μ is the mean vector of each variable;
[0030] Using the centralized data matrix X c Calculate the covariance matrix C to reflect the linear correlation between variables. The calculation formula is:
[0031]
[0032] Perform eigenvalue decomposition on the covariance matrix C:
[0033] C = VΛV T ;
[0034] where Λ is a diagonal matrix containing the eigenvalues λ1 ≥ λ2 ≥ … ≥ λ p ; V is the eigenvector matrix corresponding to Λ, where each eigenvector represents a shape change pattern and each column represents a principal component direction;
[0035] Select k principal components, project the original data onto these principal components, and obtain the scores Z of the model on each principal component:
[0036] Z = XV c V k ;
[0037] where V k is the matrix composed of the first k eigenvectors;
[0038] Calculate the variance contribution rate E of each principal component i :
[0039]
[0040] Through the variance contribution rate E i Obtain the functional mapping relationship between the coefficients of each principal component and the clinical parameters, that is, the regression coefficient F i ;
[0041] Finally, the constructed statistical shape model M can be expressed as: where represents the average shape, which is the average result of all aligned spinal models; realize the generation of personalized spinal models.
[0042] The beneficial effects of the present invention are:
[0043] The technical solution in the present invention uses high-resolution CT scanning combined with preprocessing technologies such as image noise reduction, gray normalization, and contrast enhancement, which can effectively eliminate noise and artifacts in the image, ensure that the initially reconstructed three-dimensional spinal model has high precision and clear details, and lay a solid data foundation for subsequent processing.
[0044] The present invention collects spinal data of multiple healthy human bodies, marks key anatomical points including vertebral body centers, vertebral body edges, and intervertebral space connection points, constructs a standard template representing a typical spinal structure, and uses geometric transformations such as translation, rotation, and scaling to preliminarily match the template with the spinal column of the human body to be measured. This technology effectively takes into account the differences in anatomical structures among individuals and improves the adaptability of personalized models. A radial basis function (RBF) interpolator is used to perform local flexible deformation on the standard template, and strict position constraints are set at key points. By introducing a regularization parameter to control the deformation rigidity, it not only ensures the fine fitting of local features of the model but also avoids over-stretching or distortion phenomena, thereby greatly improving the authenticity and detail restoration effect of the spinal model. Based on the RBF interpolation results, iterative optimization is carried out. New key points are added to areas with large errors to further adjust local deformation; at the same time, the main modes of morphological changes are extracted through principal component analysis, and a multiple linear regression model is established to map and correlate clinical parameters with principal component coefficients, so that the generated statistical shape model can more accurately reflect individual anatomical differences. Brief Description of the Drawings
[0045] The above and additional advantages of the present invention will become apparent and be easily understood in conjunction with the description of the embodiments with reference to the following drawings, where:
[0046] Figure 1 is a schematic diagram of a straight spine of a method for generating a personalized human spinal model according to an embodiment of the present invention;
[0047] Figure 2 is a schematic diagram of a curved spine of a method for generating a personalized human spinal model according to an embodiment of the present invention. Detailed Description of the Embodiments
[0048] In order to more clearly understand the above objects, features, and advantages of the present invention, the present invention will be further described in detail below in conjunction with the drawings and specific embodiments. It should be noted that, without conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.
[0049] In the following description, many specific details are set forth in order to fully understand the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.
[0050] As Figure 1 and 2 shown, this embodiment provides a method for generating a personalized human spinal model, including:
[0051] S1. Perform high-resolution CT scans on the human body to be measured to obtain medical image data of the spinal region. Then, use image processing software to perform noise reduction, gray-scale normalization, and contrast enhancement operations on several medical images, and then segment the spinal region. Convert the medical images of several spinal regions into a preliminary three-dimensional model through professional software, and then optimize the three-dimensional model using the Laplace smoothing algorithm to eliminate artifacts and reconstruction noise.
[0052] Next, based on the three-dimensional model obtained in step S1, construct a statistical shape model. The steps for constructing the statistical shape model include:
[0053] S2. Collect three-dimensional models formed by spinal scan data of multiple healthy human bodies, mark the key points (landmarks) of each spinal model. The key points include the centers of each vertebral body, the edges of the vertebral bodies, and the connection points of the intervertebral spaces. Perform statistical analysis on the positions of these key points, calculate their average positions, and construct a standard template representing the typical spinal structure. Compare this standard template with the three-dimensional model obtained in step S1, and perform translation, rotation, and scaling adjustments on the standard template to make it roughly coincide with the patient's spine in terms of position and size.
[0054] S3. Through a radial basis function (RBF) interpolator, perform local flexible deformation on the standard template so that each key point on the template can gradually approach the corresponding position on the patient's actual spine, enhancing shape matching.
[0055] The RBF interpolator is a mathematical tool that describes the displacement of spatial points in a specific function form, especially suitable for flexible deformation. This method introduces a radial basis function at each key point, and the value of this function gradually decays as the distance from the point to the key point increases. The basic form of RBF interpolation is:
[0056]
[0057] where f(x) is the position of the deformed key point, x is the coordinate of any key point on the standard template, x i is the coordinate of the key point on the three-dimensional model obtained in step S1, N is the total number of key points, is the radial basis function, ω i is the weighting coefficient, and P(x) is the polynomial term to ensure the smoothness and continuity of the overall deformation.
[0058] To ensure that the template shape can fit the patient's real spine after deformation, the RBF interpolator needs to set constraint conditions for the position of each key point:
[0059] f(x i ) = x i ;
[0060] That is, the target position of each template key point should be equal to the position of the corresponding key point in the patient model.
[0061] The system calculates the weighting coefficient ω of each radial basis function through the above two equations i , thus completing the training of the RBF interpolator.
[0062] After obtaining the parameters of the RBF interpolator, the RBF interpolator can perform global smooth deformation on all key points on the template model. The position of each template point is calculated through the RBF function, and combined with the position constraints of the key points and the overall shape of the surrounding points, smooth and continuous deformation of each point is achieved.
[0063] To prevent the model from being overstretched, collapsed or unreasonably distorted during the deformation process, the system introduces a regularization parameter in the RBF interpolator. The regularization parameter controls the rigidity of the model. When the value is small, the model can exhibit more flexible deformation; when the value is large, the model will be closer to the original template in terms of local features.
[0064] Users can balance the local fitting accuracy of the model and the smoothness of the overall shape by setting the regularization parameter, avoiding unnatural deformation of the model caused by excessive displacement of individual points.
[0065] Since there may still be slight errors in the RBF interpolator after the initial deformation, to further improve the matching accuracy of the model, the system will perform multiple iterative optimizations on the results of the RBF interpolator: after each deformation, the system calculates the displacement errors between the template model and the patient's spine model at key points and non-key points; for areas with large errors, the system will add new key points in these areas and recalculate the RBF interpolator to further refine the deformation of these areas; when the overall matching error reaches the preset threshold, the iterative optimization ends.
[0066] S4. Perform statistical shape analysis on the standard template after the above matching and iterative adjustment, use the principal component analysis (PCA) method to extract the main patterns of spinal shape changes, and establish a mathematical relationship between these patterns and clinical parameters (clinical parameters include height, spinal length, spinal curvature), and construct a statistical shape model that can be used to generate personalized spinal models; the specific process is as follows:
[0067] Convert the key point coordinates of the three-dimensional models formed by the spinal scan data of multiple healthy human bodies collected in step S2 into vector form and arrange them in sequence to form a high-dimensional data matrix X, where each component of each vector corresponds to the coordinate value of each key point in the unified coordinate system. The high-dimensional data matrix X is a matrix with n rows and p columns.
[0068] To ensure data comparability, perform centering processing on the high-dimensional data matrix X to obtain the centered data matrix X c :
[0069] X c = X - μ;
[0070] where μ is the mean vector of each variable.
[0071] Using the centered data matrix X c calculate the covariance matrix C to reflect the linear correlation between variables. The calculation formula is:
[0072]
[0073] Perform eigenvalue decomposition on the covariance matrix C:
[0074] C = VΛV T ;
[0075] where Λ is a diagonal matrix containing the eigenvalues λ1 ≥ λ2 ≥ … ≥ λ p ; V is the eigenvector matrix corresponding to Λ, where each eigenvector represents a shape change pattern and each column represents a principal component direction.
[0076] Select k principal components and project the original data onto these principal components to obtain the scores Z of the model on each principal component:
[0077] Z = X c V k ;
[0078] where V k is the matrix composed of the first k eigenvectors. The principal component scores describe the deviation degree of each individual's spinal shape relative to the average shape.
[0079] At the same time, calculate the variance contribution rate E i of each principal component to quantify the contribution weight of each principal component to the parameters and ensure that the selected principal components can fully describe the main changes in the spinal shape. The variance contribution rate E i is calculated by the formula:
[0080]
[0081] Through the variance contribution rate E i obtain the functional mapping relationship between the coefficients of each principal component and the clinical parameters, that is, the regression coefficient F i .
[0082] Finally, the constructed statistical shape model M can be expressed as: where represents the average shape, which is the average result of all aligned spinal models; realizing the generation of personalized spinal models.
[0083] The experimental results show that the first three principal components can usually explain more than 90% of the morphological variations, and there is a high correlation between the coefficients of each principal component and the individual physiological parameters (such as height, weight, age), thus providing a reliable basis for subsequent parameter calibration.
[0084] S5. Output the finally generated personalized three-dimensional spinal model as a standard data file; before output, perform necessary file format conversion and data optimization on the model to ensure that the output model has both high precision and is applicable to practical applications in different fields.
[0085] S6. When performing personalized adjustment on the saved model, the user inputs clinical parameters through the interactive interface, and the system calculates the principal component weights in real time according to the input parameters of the user, generates an updated spinal model, and displays it to the user in real time.
[0086] In summary, the present invention proposes a method for generating a personalized human spinal model, including:
[0087] S1. Perform high-resolution CT scanning on the human body to be measured to obtain medical image data of the spinal region; use preprocessing methods such as noise reduction, gray normalization, and contrast enhancement on the image data, and segment the spinal region; use professional image processing software to convert the segmented image data into a preliminary three-dimensional spinal model, and use the Laplace smoothing algorithm to optimize the preliminary model to eliminate artifacts and reconstruct noise.
[0088] S2. Collect three-dimensional models formed by scanning the spinal columns of multiple healthy human bodies; mark key points including vertebral body centers, vertebral body edges, and intervertebral space connection points for each spinal model, and perform statistical analysis on the positions of the key points, calculate the average positions, and construct a standard template representing the typical spinal structure; perform translation, rotation, and scaling transformations on the standard template so that it roughly matches the spinal model of the human body to be measured in terms of position and size.
[0089] S3. Use the radial basis function interpolation method to perform local flexible deformation on the standard template in step S2.
[0090] S4. Arrange the key point data of the three-dimensional spinal models of multiple healthy human bodies into a high-dimensional data matrix, and perform centering processing and unified alignment to eliminate rotation, translation, and scale differences; then, calculate the covariance matrix using the centered data, obtain each principal component through eigenvalue decomposition, select some principal components after sorting according to the variance contribution rate, project the original data to obtain the principal component scores of each model, and calculate the variance contribution rate of each principal component as the contribution weight of morphological changes; establish a mapping relationship between the clinical parameters and the principal component scores, and determine the regression coefficients of each principal component; finally, based on the average shape of all aligned spinal models, weighted superimpose each principal component to construct a statistical shape model of the personalized spine.
[0091] S5. Output the finally generated personalized three-dimensional spinal model as a standard data file; before output, perform necessary file format conversion and data optimization on the model to ensure that the output model has both high precision and is applicable to practical applications in different fields.
[0092] S6. When performing personalized adjustment on the saved model, the user inputs clinical parameters through the interaction interface, and the system calculates the principal component weights in real time according to the user input parameters, generates an updated spinal model, and displays it to the user in real time.
[0093] In the present invention, terms such as "install", "connect", "join", "fix", etc. should be understood in a broad sense. For example, "connect" can be a fixed connection, a detachable connection, or an integral connection; "join" can be a direct connection or an indirect connection through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0094] The shapes of the various components in the drawings are schematic, and there is no exclusion of a certain difference from their actual shapes. The drawings are only used to illustrate the principle of the present invention and are not intended to limit the present invention.
[0095] Although the present invention has been disclosed in detail with reference to the drawings, it should be understood that these descriptions are merely exemplary and are not used to limit the application of the present invention. The protection scope of the present invention is defined by the appended claims and may include various modifications, improvements, and equivalent solutions made to the invention without departing from the protection scope and spirit of the present invention.
Claims
1. A method for generating a personalized human spine model, characterized in that, The method includes: S1. Perform high-resolution CT scanning on the human body to be measured to obtain medical image data of the spinal region; use professional image processing software to convert the image data into a preliminary three-dimensional spinal model; S2. Collect three-dimensional models formed by spinal scan data of multiple healthy human bodies; mark key points including vertebral body centers, vertebral body edges, and intervertebral space connection points on each spinal model, and perform statistical analysis on the positions of the key points to obtain the average position, and construct a standard template representing the typical spinal structure; perform translation, rotation, and scaling transformations on the standard template so that it roughly matches the spinal model of the human body to be measured in terms of position and size; S3. Use the radial basis function interpolation method to perform local flexible deformation on the standard template in step S2; S4. Arrange the key point data of the three-dimensional spinal models of multiple healthy human bodies into a high-dimensional data matrix, and perform centering processing and unified alignment to eliminate rotation, translation, and scale differences; then, calculate the covariance matrix using the centered data, and obtain each principal component through eigenvalue decomposition, select some principal components after sorting by the variance contribution rate, project the original data to obtain the principal component scores of each model, and calculate the variance contribution rate of each principal component as the contribution weight of morphological changes; establish a mapping relationship between the clinical parameters and the principal component scores to determine the regression coefficients of each principal component; finally, based on the average shape of all aligned spinal models, weighted superimpose each principal component to construct a statistical shape model of the personalized spine; S5. Output the finally generated three-dimensional model of the personalized spine as a standard data file; perform necessary file format conversion and data optimization on the model before output to ensure that the output model has both high precision and is applicable to practical applications in different fields; S6. When performing personalized adjustment on the saved model, the user inputs clinical parameters through the interactive interface, and the system calculates the principal component weights in real time according to the user input parameters, generates an updated spinal model, and displays it to the user in real time.
2. The personalized human spine model generation method according to claim 1, wherein The method for performing local flexible deformation in step S3 includes: Use an RBF interpolator to perform local deformation on the standard template. The basic form of RBF interpolation is: Among them, f(x) is the position of the key point after deformation, x is the coordinate of the key point on any standard template, and x i is the coordinate of the key point on the three-dimensional model obtained in step S1, N is the total number of key points, is the radial basis function, and ω i is the weighting coefficient, and P(x) is the polynomial term to ensure the smoothness and continuity of the overall deformation; The RBF interpolator needs to set constraint conditions for the position of each key point: f(x i ) = x i ; That is, the target position of each template key point should be equal to the position of the corresponding key point in the patient model; The system calculates the weighted coefficient ω of each radial basis function through the above two equations i , thereby completing the training of the RBF interpolator.
3. The personalized human spine model generation method according to claim 2, wherein, Step S3 also includes: The system introduces a regularization parameter in the RBF interpolator. The regularization parameter controls the rigidity of the model. When the value is small, the model can show more flexible deformation; when the value is large, the model will be closer to the original template in terms of local features; The user can balance the local fitting accuracy and the smoothness of the overall shape of the model by setting the regularization parameter, and avoid unnatural deformation of the model caused by excessive displacement of individual points.
4. The personalized human spine model generation method according to claim 1, wherein The deformation operation of step S3 is to obtain a higher matching accuracy. The system will perform multiple iterative optimizations on the results of the RBF interpolator: after each deformation, the system calculates the displacement errors between the template model and the patient's spinal model at key points and non-key points; for areas with larger errors, the system will add new key points in these areas and recalculate the RBF interpolator to further refine the deformation of these areas; when the overall matching error reaches the preset threshold, the iterative optimization ends.
5. The personalized human spine model generation method according to claim 1, characterized in that, After obtaining the image data in step S1, preprocessing methods such as noise reduction, gray-scale normalization, and contrast enhancement are used on the image data to segment the spinal region.
6. The personalized human spine model generation method according to claim 1, wherein, After converting the image data into a preliminary three-dimensional spinal model in step S1, the Laplace smoothing algorithm is used to optimize the preliminary model to eliminate artifacts and reconstruction noise.
7. The personalized human spine model generation method according to claim 1, characterized in that, Step S4 specifically includes: Convert the key point coordinates of the three-dimensional models formed by the spinal scan data of multiple healthy human bodies collected in step S2 into vector forms, and arrange them in sequence to form a high-dimensional data matrix X, where each component of each vector corresponds to the coordinate value of each key point in a unified coordinate system. The high-dimensional data matrix X is a matrix with n rows and p columns; Center the high-dimensional data matrix X to obtain the centered data matrix X c : X c = X - μ; where μ is the mean vector of each variable; Using the centralized data matrix X c Calculate the covariance matrix C to reflect the linear correlation between variables. The calculation formula is as follows: Perform eigenvalue decomposition on the covariance matrix C: C = V Λ V T ; where, Λ is a diagonal matrix containing the eigenvalues λ1 ≥ λ2 ≥ … ≥ λ arranged in descending order p ; V is the eigenvector matrix corresponding to Λ, where each eigenvector represents a shape change mode and each column represents a principal component direction; Select k principal components, project the original data onto these principal components, and obtain the scores Z of the model on each principal component: Z = X c V k ; where, V k is the matrix composed of the first k eigenvectors; Calculate the variance contribution rate E of each principal component i : Through the variance contribution rate E i The functional mapping relationship between the coefficients of each principal component and the clinical parameters is obtained, that is, the regression coefficient F i ; Finally, the constructed statistical shape model M can be expressed as: where represents the average shape, which is the average result of all aligned spine models; realizing the generation of personalized spine models.