Fast repairing method for scoliosis model based on semantic feature optimization

Abstract

The application discloses a side bending spine model fast repairing method based on semantic feature optimization, relates to the technical field of medical image analysis and biomedical engineering, and comprises the following steps: performing segmentation and identification operation on the pre-collected medical image by using morphological features and geometric analysis technology, and acquiring three-dimensional semantic information of a spine structure based on the segmentation and identification result; extracting morphological parameters based on the three-dimensional semantic information of the spine structure, and exploring the correlation between the morphological parameters by using correlation analysis technology; generating a spine semantic parameter combination based on the correlation and the morphological parameters, and repairing the side bending spine model by using an optimization algorithm to find the best spine semantic parameter combination. The application adopts the global optimization strategy of the optimization algorithm, designs a reasonable evaluation function to quantize the performance of each parameter combination in the side bending model repairing task, automatically finds the best semantic parameter combination weight, and thus guides the effective adjustment of the side bending spine model parameters.

Description

Technical Field

[0001] This invention relates to the fields of medical image analysis and biomedical engineering technology, and more specifically, to a rapid repair method for scoliosis spine models based on semantic feature optimization. Background Technology

[0002] With the advancement of digital technology, three-dimensional model repair of scoliosis has been promoted. Detailed three-dimensional models can be constructed by using the bone cross-sectional contour lines generated by medical images, and geometric transformations such as scaling and shearing can be performed to simulate different correction strategies.

[0003] However, traditional shape-based adjustment methods have limitations when dealing with complex lesions, especially for scoliosis, which requires consideration of anatomical semantic information. It is difficult to accurately express these lesions through simple geometric deformation. Existing methods can effectively avoid the problem of poor local detail when simply using mesh deformation. However, the semantic parameter set referenced by the deformation in the existing technology is not comprehensive enough, and there is a lack of effective optimization algorithms to determine the optimal parameter combination during the adjustment process. Different parameter combinations can only be repeatedly tested during repair, which undoubtedly increases the complexity and uncertainty of the repair plan and has become a bottleneck currently faced in scoliosis repair.

[0004] Constructing a comprehensive set of semantic parameters for the spine with clear anatomical meaning, and exploring methods for intelligently optimizing these parameters, are the core issues in the current three-dimensional model repair of scoliosis. Furthermore, how to find the optimal or near-optimal solution in a high-dimensional parameter space with semantic relevance and constraints remains to be studied.

[0005] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0006] To address the problems in related technologies, this invention proposes a rapid repair method for scoliosis spine models based on semantic feature optimization, in order to overcome the aforementioned technical problems existing in existing related technologies.

[0007] Therefore, the specific technical solution adopted by the present invention is as follows:

[0008] A rapid repair method for scoliosis spine models based on semantic feature optimization is described below.

[0009] Morphological features and geometric analysis techniques are used to segment and recognize pre-collected medical images, and three-dimensional semantic information of the spinal structure is obtained based on the segmentation and recognition results.

[0010] Based on the three-dimensional semantic information of the spinal structure, morphological parameters were extracted, and correlation analysis was used to explore the relationship between the morphological parameters.

[0011] Based on the correlation and morphological parameters, a combination of spinal semantic parameters is generated. An optimization algorithm is used to find the optimal combination of spinal semantic parameters, and the scoliosis spinal model is repaired based on the optimal combination of spinal semantic parameters.

[0012] Preferably, the process involves segmenting and recognizing pre-collected medical images using morphological features and geometric analysis techniques, and obtaining three-dimensional semantic information of the spinal structure based on the segmentation and recognition results, including:

[0013] Tomographic and magnetic resonance images were collected as medical images, and affine transformation techniques were applied to register the medical images. The registration results were verified based on the analysis of the matching degree between the medical images.

[0014] Based on the verification results, the parameters were adjusted to complete the registration process. Then, filtering and equalization were used to denoise and enhance the contrast of the registered medical images to obtain optimized medical images.

[0015] Based on morphological manipulation and probability distribution techniques, information on the spinal region and ligaments in medical images is obtained, and the spinal structure is obtained by identifying growth feature lines in medical images using the watershed algorithm.

[0016] The spinal structure, spinal regions, and ligament information are used as the three-dimensional semantic information of the spinal structure.

[0017] Preferably, spinal region and ligament information in medical images are obtained based on morphological manipulation and probability distribution techniques, and the spinal structure is obtained by identifying growth feature lines in the medical images using a watershed algorithm, including:

[0018] Morphological operations are used to identify contour feature lines in medical images, and slicing and projection operations are performed on the contour feature lines to obtain point sets in the medical images. Based on the point sets, regional information of the spine is obtained.

[0019] A graph model is generated based on an adaptive sampling strategy and point set, and graph theory algorithms are used to analyze the graph model to complete hole identification and repair operations. Information about the ligament is extracted based on the repair results.

[0020] The watershed algorithm is applied to identify the growth feature lines of the point set. After optimizing the growth feature lines, a multidimensional tree is constructed. The detailed information of the spinal structure is analyzed based on the multidimensional tree.

[0021] Preferably, morphological operations are used to identify contour feature lines in medical images, and slicing and projection operations are performed on the contour feature lines to obtain a point set of the medical images. Based on the point set, the regional information of the spine is obtained, including:

[0022] The gradient magnitude and direction of each pixel in the medical image are calculated, and the contour of the medical image is identified by defining a threshold index based on historical experience. The pixels on the contour boundary are morphologically eroded.

[0023] After filling the gaps between objects on the contour boundary using dilation technology, adjacent objects are connected to obtain contour feature lines. The position and direction of the slicing plane are determined based on the position and direction of the contour feature lines.

[0024] The set of points intersecting the slice plane is calculated based on the contour feature lines, and the set of points is projected onto the slice plane. Based on the projection results, the regional information of the spine is obtained.

[0025] Preferably, a graph model is generated based on an adaptive sampling strategy and point set, and graph theory algorithms are used to analyze the graph model to complete the hole identification and repair operation. Based on the repair results, ligament information is extracted, including:

[0026] The point set is used as the original spine dataset, and a set of data points is randomly selected from the original spine dataset as the first sampling points and added to the sampling point set.

[0027] Based on a preset number of sampling points, sampling points are iteratively selected from the original spine dataset, and the sum of the distances between the sampling points and all data points in the original spine dataset is maximized.

[0028] When the number of iterations reaches the preset number of sampling points, the sampling points are added to the sampling point set to form a spine sampling dataset. The spine sampling dataset is then converted into a graph model, and nodes are created for the data points.

[0029] Determine the distance between any two nodes. If the distance is less than a set threshold, connect the two nodes to create an edge. Apply graph theory algorithms to detect edge creation status and analyze the hole information in the graph model.

[0030] The boundaries of the holes are determined and connectivity analysis is performed using the minimum spanning tree to fill the holes. The ligament information is then extracted from the filled graph model.

[0031] Preferably, the watershed algorithm is applied to identify the point set to obtain growth feature lines, and after optimizing the growth feature lines, a multidimensional tree is constructed. The detailed information of the spinal structure is analyzed based on the multidimensional tree, including:

[0032] After preprocessing the data points in the point set, the corresponding gradient values ​​are calculated, and points whose gradient values ​​exceed the threshold are selected as the starting points for growing feature lines.

[0033] Gradient images are constructed using gradient values, and the data point with the largest gradient value is selected as the seed point. The seed point is then set as the marker point for the watershed transform.

[0034] Based on the marker points, the gradient image is segmented into several groups of regions using watershed changes, and growth feature lines are identified after determining the boundary information between regions and the gradient changes within the regions.

[0035] After performing noise removal and discontinuity removal operations on the growth feature line, the point set corresponding to the growth feature line is divided into several subsets, and a multidimensional tree is constructed on the subsets.

[0036] The growth region is constructed by finding similar points on the growth feature line based on multidimensional tree and K-nearest neighbor search technology, and detailed information of the spinal structure is extracted after merging and segmenting the growth region.

[0037] Preferably, morphological parameters are extracted based on the three-dimensional semantic information of the spinal structure, and correlation analysis is used to explore the relationships between the various morphological parameters, including:

[0038] Morphological parameters are extracted from the three-dimensional semantic information of the spinal structure, and the morphological parameters are normalized. Correlation techniques are used to analyze the correlation coefficients between pairs of morphological parameters.

[0039] The linear relationship between various morphological parameters is explained based on the magnitude of the correlation coefficient, and the correlation parameters are divided into different groups according to a preset classification standard;

[0040] Perform ANOVA on each group to determine the ratio of between-group variance to within-group variance and the probability of the ratio occurring. Use the ratio and probability to explain the differences in morphological parameters between different groups.

[0041] Clustering operations are performed on the morphological parameters, and the correlation and differences between the morphological parameters are interpreted based on the clustering results to obtain the parameter correlation relationship. Based on the parameter correlation relationship, the spinal morphological features represented by each morphological parameter are shown.

[0042] Preferably, the process involves generating a combination of spinal semantic parameters based on correlation and morphological parameters, using an optimization algorithm to find the optimal combination of spinal semantic parameters, and repairing the scoliosis spinal model based on the optimal combination of spinal semantic parameters, including:

[0043] Based on the morphological features, correlations and morphological parameters of the spine, the combination of semantic parameters of the spine is determined, and a parameter space is generated based on the combination of semantic parameters of the spine.

[0044] Within the parameter space, parameter combinations are encoded into individuals, and optimization objectives are defined based on the application of spinal semantic parameters in spinal model repair. Evaluation functions are defined based on the optimization objectives to calculate the evaluation values ​​of each individual.

[0045] Genetic operations are performed based on evaluation values ​​and optimization algorithms to simulate natural selection and genetic processes. The evaluation degree of individuals is re-evaluated according to the genetic process, and the individual with the highest evaluation degree is selected as the optimal solution to obtain the best combination of spinal semantic parameters.

[0046] A preliminary scoliosis spine model is generated based on spinal data, and the optimal combination of spinal semantic parameters is applied to the preliminary scoliosis spine model to repair the spinal condition.

[0047] Preferably, the formula for calculating the evaluation function is:

[0048] F(x) = α.P accuracy +β.P correlation -γ.P complexity +δ.P robustness ;

[0049] In the formula, F(x) represents the evaluation value of the x-th individual, and P... accuracy P represents the segmentation precision. correlation P represents parameter correlation. complexity P represents the computational complexity. robustness δ represents robustness, α represents the weighting coefficient of segmentation accuracy, β represents the weighting coefficient of parameter correlation, γ represents the weighting coefficient of computational complexity, and δ represents the weighting coefficient of robustness.

[0050] Preferably, generating a preliminary scoliosis spine model based on spinal data and applying the optimal combination of spinal semantic parameters to the preliminary scoliosis spine model to repair the spinal condition includes:

[0051] After preprocessing medical imaging data of a healthy spine, spinal contour data is extracted, and a preliminary scoliosis spine model is constructed using assisted design techniques.

[0052] Based on the preset repair requirements, the boundary conditions, initial displacement and velocity are defined. The deformation algorithm of the physical model guides the deformation of the prototype scoliosis spine model. The new position of each data point is obtained to obtain the scoliosis spine deformation model.

[0053] The optimal combination of spinal semantic parameters is applied to the scoliosis spinal deformation model to simulate the morphological changes of the spine, and the scoliosis spinal deformation model is repaired based on the morphological simulation results.

[0054] The beneficial effects of this invention are as follows:

[0055] 1. This invention is based on a global optimization strategy using optimization algorithms. It designs a reasonable evaluation function to quantify the performance of each parameter combination in the scoliosis model repair task, and automatically finds the optimal semantic parameter combination weights. This guides the effective adjustment of scoliosis spine model parameters, which can not only improve the quality of model repair, but also improve the efficiency of parameter optimization and avoid the tedious work caused by blindly modifying parameters manually.

[0056] 2. This invention utilizes well-defined and obtainable semantic parameters with clear anatomical significance to characterize in detail the degree, morphological features and potential correction needs of scoliosis. At the same time, it employs optimization algorithms to find the most suitable combination of model parameters to accurately simulate the real state of the scoliotic spine and the possible best correction results.

[0057] 3. This invention achieves efficient and accurate segmentation of spinal images, which not only improves the quality of spinal restoration but also ensures the integrity and continuity of the spinal structure. At the same time, it guides the initial deformation of the spinal model based on physical models and finite element analysis, providing a foundation for subsequent optimization based on genetic algorithms, and ensuring the rationality and accuracy of the spinal model in terms of morphology.

[0058] 4. The rapid repair method for lateral spine models implemented in this invention significantly improves the accuracy of spinal repair and the realism of three-dimensional models, effectively solving the problems of structural continuity and detail loss in traditional methods. It greatly promotes the accuracy and efficiency of medical image analysis, surgical planning and treatment evaluation, and is of great significance to promoting the development of personalized medicine and precision surgery. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0060] Figure 1 This is a flowchart of a rapid repair method for a scoliosis spine model based on semantic feature optimization according to an embodiment of the present invention;

[0061] Figure 2 This is a schematic diagram of a rapid repair method for scoliosis spine models based on semantic feature optimization according to an embodiment of the present invention.

[0062] Figure 3 These are lateral and anteroposterior images of a scoliosis patient in a rapid repair method for scoliosis spine models based on semantic feature optimization according to an embodiment of the present invention.

[0063] Figure 4 yes Figure 3 A schematic diagram of the spinal centerline in a patient with scoliosis. Detailed Implementation

[0064] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention.

[0065] According to embodiments of the present invention, a rapid repair method for scoliosis spine models based on semantic feature optimization is provided.

[0066] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figures 1 to 4 As shown, according to an embodiment of the present invention, a rapid repair method for a scoliosis spine model based on semantic feature optimization includes:

[0067] Step S1: Feature and geometric analysis techniques are used to segment and recognize the pre-collected medical images, and three-dimensional semantic information of the spinal structure is obtained based on the segmentation and recognition results.

[0068] Step S2: Based on the three-dimensional semantic information of the spinal structure, morphological parameters are extracted, and correlation analysis technology is used to explore the relationship between the morphological parameters.

[0069] Step S3: Generate a combination of spinal semantic parameters based on the correlation and morphological parameters, use an optimization algorithm to find the optimal combination of spinal semantic parameters, and repair the scoliosis spinal model based on the optimal combination of spinal semantic parameters.

[0070] In this embodiment, when performing segmentation and recognition operations on pre-collected medical images using morphological features and geometric analysis techniques, and obtaining three-dimensional semantic information of the spinal structure based on the segmentation and recognition results, tomographic scan images and magnetic resonance images can be collected as medical images. Affine transformation techniques are applied to register the medical images, and the registration results are verified based on the degree of matching between the medical images. The parameters are adjusted according to the verification results to complete the registration process, and the registered medical images are denoised and contrast-enhanced using filtering and equalization methods to obtain optimized medical images. The spinal region and ligament information in the medical images are obtained based on morphological operations and probability distribution techniques, and the growth feature lines in the medical images are identified using the watershed algorithm to obtain the spinal structure. The spinal structure, spinal region, and ligament information are used as the three-dimensional semantic information of the spinal structure.

[0071] Specifically, medical images refer to medical imaging data used for 3D reconstruction of the spine, such as CT (computed tomography) or MRI (magnetic resonance) images. Simultaneously, image registration is performed to ensure correct alignment between different slices, facilitating subsequent structural analysis. Image registration is the process of aligning spinal images acquired at different time points or using different imaging devices to ensure the consistent position of the same anatomical structure in different images. The image registration steps are as follows:

[0072] Key feature points, including corners and edges, are extracted from medical images. Registration algorithms, such as rigid transformation, affine transformation, and nonlinear transformation, are selected. A similarity metric is defined based on the registration algorithm to evaluate the degree of matching between images. Commonly used metrics include mutual information (MI) and mean squared error (MSE).

[0073] The formula for calculating mutual information is as follows:

[0074]

[0075] In the formula, MI(X,Y) represents the mutual information between the image X before registration and the image Y after registration, p(x,y) represents the joint probability density function, p(x) and p(y) represent the marginal probability density functions of X and Y respectively, and x and y represent the feature points before registration and the feature points after registration respectively.

[0076] The formula for calculating the mean square error is:

[0077]

[0078] In the formula, I1 and I2 represent the corresponding pixel values ​​of the images before and after registration, respectively, n is the total number of pixels, and i represents the number of observation data.

[0079] Meanwhile, optimization algorithms such as gradient descent, Newton's method, or genetic algorithms can be used to adjust transformation parameters and maximize the similarity measure.

[0080] Specifically, when performing image preprocessing operations such as denoising, contrast enhancement, and normalization to improve image quality, image preprocessing includes denoising, contrast enhancement, and normalization, as detailed below:

[0081] Image denoising: Specifically, mean filtering can be used to reduce random noise in the image.

[0082]

[0083] In the formula, f(x, y) represents the original image, g(x, y) represents the denoised image, m and n represent the number of pixels in the filtering window, s represents the index of the filtering window in the row direction (vertical direction), t represents the index of the filtering window in the column direction (horizontal direction), and a and b are respectively... and

[0084] Contrast enhancement aims to improve the visual quality of spine images. It employs commonly used histogram equalization, which enhances contrast by adjusting the image's grayscale distribution. The transform function T is defined as:

[0085]

[0086] In the formula, r represents the gray level of the input image, s represents the gray level of the output image, pr(j) represents the probability that the gray level of the input image is j, and j represents the gray level of the input image.

[0087] Meanwhile, the purpose of standardization is to adjust the grayscale range or distribution of an image to meet specific statistical characteristics. The linear standardization formula is:

[0088] g = (f - μ) / σ;

[0089] In the formula, f represents the pixel value of the original image, g represents the standardized pixel value, μ represents the mean of the pixel value, and σ represents the standard deviation.

[0090] Image denoising, contrast enhancement, and standardization can effectively improve the quality of spinal medical imaging data, laying a good foundation for subsequent three-dimensional reconstruction of the spine.

[0091] In this embodiment, when acquiring spinal region and ligament information from medical images based on morphological operations and probability distribution techniques, and using the watershed algorithm to identify growth feature lines in medical images to obtain the spinal structure, morphological operations can be used to identify contour feature lines in medical images. Slicing and projection operations are then performed on these contour feature lines to obtain a point set from the medical images. Based on this point set, regional information of the spine is obtained. A graph model is generated based on an adaptive sampling strategy and the point set. Graph theory algorithms are then used to analyze the graph model to complete hole identification and repair operations. Ligament information is extracted based on the repair results. The watershed algorithm is applied to identify the point set to obtain growth feature lines. After optimizing the growth feature lines, a multidimensional tree is constructed. The detailed information of the spinal structure is then analyzed based on the multidimensional tree.

[0092] In this embodiment, when using morphological operations to identify contour feature lines in medical images, and performing slicing and projection operations on the contour feature lines to obtain a point set of the medical images, and obtaining the regional information of the spine based on the point set, the gradient magnitude and direction of each pixel in the medical image can be calculated, and a threshold index can be defined based on historical experience to identify the contour of the medical image. Pixels on the contour boundary are morphologically eroded; after filling the gaps between objects on the contour boundary using dilation technology, adjacent objects are connected to obtain contour feature lines; the position and direction of the slicing plane are determined based on the position and direction of the contour feature lines; a point set intersecting the slicing plane is calculated based on the contour feature lines, and the point set is projected onto the slicing plane; the regional information of the spine is obtained based on the projection result.

[0093] It should be explained that when performing contour feature line recognition and slice projection, morphological features can be used to identify different regions of the spine. The human spine is composed of several vertebrae. Adults have 26 vertebrae, including 7 cervical vertebrae, 12 thoracic vertebrae, 5 lumbar vertebrae, 1 sacrum, and 1 coccyx.

[0094] The specific implementation steps are as follows:

[0095] Step 1: Contour feature line extraction, using morphological operations (such as edge detection) to identify contour feature lines;

[0096] The Canny edge detector is used to identify contours in the image. Gaussian filtering is used to reduce image noise. The gradient magnitude and direction of each pixel in the image are calculated. At the same time, the edges are thinned and non-edge points are suppressed to 0. Two thresholds, a high threshold and a low threshold, are selected to determine which edges are "determined edges" and "possible edges". Finally, the edges are determined by hysteresis thresholding.

[0097] Use morphological operations, including erosion and dilation, to further refine the edges.

[0098] Erosion: Used to eliminate small objects and reduce pixels at object boundaries.

[0099] Expansion: Used to fill gaps between objects and connect adjacent objects.

[0100] Use the findContours function (e.g., in OpenCV) to extract the contours after edge detection.

[0101] Step 2: Slice the extracted contour feature lines and project the 3D data onto a 2D plane to enhance spatial positioning accuracy;

[0102] Based on the position and orientation of the contour feature lines, the position and orientation of the slicing plane are determined. For each contour feature line Ci, the set of points P intersecting with the slicing plane is calculated, where the slicing operation formula can be expressed as:

[0103]

[0104] Projecting a 3D point set P onto a 2D plane, for example, projecting (x, y, z) onto the (x, y) plane, the projection formula can be expressed as:

[0105] P'(P)={(x,y)(x,y,z)∈P}

[0106] In the formula, P represents the 2D point set, represents the part of Ci that passes through the original 3D dataset D, and x, y, z represent the coordinates of the contour feature lines.

[0107] In this embodiment, when generating a graph model based on an adaptive sampling strategy and a point set, and combining graph theory algorithms to analyze the graph model to complete the hole identification and repair operation, and extracting ligament information based on the repair results, the point set can be used as the original spinal dataset, and a set of data points is randomly selected from the original spinal dataset as the first sampling points and added to the sampling point set; sampling points are iteratively selected from the original spinal dataset based on a preset number of sampling points, and the sum of the distances between the sampling points and all data points in the original spinal dataset is maximized; when the number of iterations reaches the preset number of sampling points, the sampling points are added to the sampling point set to form a spinal sampling dataset, the spinal sampling dataset is converted into a graph model, and nodes are created for the data points; the distance between any two nodes is judged, and when the distance value is less than a set threshold, the two nodes are connected to create an edge, and the graph theory algorithm is applied to detect the edge creation status and analyze the hole information of the graph model; the boundary of the hole is determined and connectivity analysis is performed in combination with the minimum spanning tree to fill the hole, and ligament information is extracted based on the filled graph model.

[0108] It should be explained that, in order to improve the representativeness and efficiency of sampling, an adaptive sampling strategy is adopted in this embodiment. Adaptive sampling not only considers randomness, but also determines the next sampling point based on the currently selected sampling point, thereby better covering the spatial distribution of the entire 3D point cloud dataset, enabling the extraction of key information of the ligament based on probability distribution and graph theory methods.

[0109] Through an adaptive mechanism, each newly selected point is chosen to have the maximum information gain relative to existing sample points. Assume D is the original 3D spine dataset, and S... i Let k be the set of points obtained from the i-th sampling (sampling point set), and k represents the number of sampling points required.

[0110] Then, a point is randomly selected from D as the first sampling point and added to S1. For each iteration i, from D\S i Select the next point p i+1 , making p i+1 With S i The goal is to maximize the sum of distances between all points, which can be achieved by calculating the following objective function:

[0111]

[0112] In the formula, d(p, q) represents the Euclidean distance between points p and q. i+1 Add to S i S formed in i+1 When |S i Sampling stops when | = k.

[0113] Specifically, connectivity analysis in graph theory is used to identify and repair holes to improve the integrity of ligament information. First, the 3D point cloud data is converted into a graph model, where each point in the point cloud corresponds to a node in the graph, and the connections between points (usually pairs of points with a Euclidean distance less than a certain threshold) correspond to edges in the graph. A node is created for each point in the point cloud, and if the distance between two nodes is less than the set threshold, an edge is created between the nodes.

[0114] In a graph model, a hole can be viewed as a disconnected part of the graph. Graph theory algorithms (such as Depth-First Search (DFS) or Breadth-First Search (BFS)) can be used to detect connected components in the graph. If a connected component is completely surrounded by other connected components, a hole exists. Repairing a hole usually involves determining the hole boundary and filling the hole region. This involves identifying the set of nodes that constitute the hole boundary and using graph theory algorithms (such as Minimum Spanning Tree (MST) or Shortest Path algorithm) to determine how to fill the hole.

[0115] In this embodiment, after applying the watershed algorithm to identify the growth feature lines of the point set and optimizing the growth feature lines to construct a multidimensional tree, and analyzing the detailed information of the spinal structure based on the multidimensional tree, the data points in the point set can be preprocessed to calculate the corresponding gradient values, and the points with gradient values ​​exceeding the threshold can be selected as the starting points of the growth feature lines; a gradient image can be constructed using the gradient values, and the data point with the largest gradient value can be selected as the seed point, which is set as the marker point of the watershed transformation; based on the marker points, the gradient image can be divided into several groups of regions using the watershed transformation, and the growth feature lines can be identified after judging the boundary information between regions and the gradient changes within the regions; after performing noise removal and discontinuity point removal operations on the growth feature lines, the point set corresponding to the growth feature lines is divided into several subsets, and a multidimensional tree is constructed on the subsets; based on the multidimensional tree and K-nearest neighbor search technology, similar points on the growth feature lines are found to form growth regions, and the detailed information of the spinal structure is extracted after merging and segmenting the growth regions.

[0116] It needs to be explained that the growth feature lines and parallel K-nearest neighbor region growth are methods based on growth algorithms and graph theory to segment detailed structures such as intervertebral discs. Simultaneously, the watershed algorithm is used to identify growth feature lines in 3D data. To enhance the watershed algorithm's ability to identify feature lines in 3D data, machine learning models can be combined to improve gradient calculation, and more advanced segmentation algorithms can be used to improve the accuracy of feature line identification. The specific steps are as follows:

[0117] Step 1: Gradient Calculation: Gradient calculation is the first step in identifying feature lines. Traditionally, the Laplacian operator is used to estimate the gradient of each voxel. This embodiment utilizes a machine learning model to further improve the accuracy of gradient estimation. First, the 3D data is preprocessed to ensure data quality and consistency. For each voxel v, its gradient is calculated using the Laplacian operator. Simultaneously, a machine learning model (e.g., a Convolutional Neural Network, CNN) is trained. This model accepts a local 3D voxel block as input and outputs the gradient vector of the center point of that voxel block. The specific formula is expressed as follows:

[0118]

[0119] Among them, V block It is a local voxel block centered at v, and ML stands for machine learning model.

[0120] At the same time, based on the gradient magnitude, possible feature line starting points can be determined, and points with gradient magnitudes exceeding a certain threshold can be selected as candidate starting points for feature lines.

[0121] Step 2: Apply the watershed algorithm to identify feature lines: Construct a gradient image using gradient information. The intensity value of each voxel corresponds to the magnitude of its gradient, and the point with the largest gradient magnitude is selected as the seed point. The seed point will serve as the marker point for the watershed transform. The watershed algorithm is applied to segment the gradient image into multiple regions. The goal of the watershed transform is to find the "valleys" in the image, i.e., the places where the gradient changes most gently. These places are considered as the boundaries of different regions, specifically represented as follows:

[0122] R=Watershed(Gradient Image, Seeds);

[0123] Where R represents the segmented region, Watershed represents the watershed algorithm, Gradient Image represents the gradient image, and Seeds represent seed points.

[0124] By analyzing the boundaries between regions and the changes in gradient, feature lines are identified. Feature lines are usually the locations with the greatest gradient changes, which are the "ridges" in the watershed algorithm. The identified feature lines are then optimized by removing noise or discontinuous parts to make them smoother and more coherent.

[0125] Step 3: Parallel K-Nearest Neighbor Region Growing: To accelerate the K-nearest neighbor search process while ensuring accuracy, parallel processing techniques can be employed. By introducing parallel processing, the speed of the K-nearest neighbor search can be significantly increased without sacrificing accuracy, thereby improving the overall efficiency of the region growing algorithm. Details are as follows:

[0126] A KD-tree (multidimensional tree) is constructed on the set of points around the growth feature line to enable fast K-nearest neighbor search. To support parallelization, the point set is divided into multiple subsets, and a KD-tree is constructed independently on each subset: the point set P is divided into n subsets P1, P2, ..., Pn. n For each subset P i Construct KD-trees in parallel i The formula is expressed as:

[0127] T i =BuildKDTree(P i ),

[0128] All subtrees T i Merge them into a complete KD-tree T. During the merging process, it is necessary to ensure that the root nodes of all subtrees are correctly connected. For each point on the growth feature line, a parallelized K-nearest neighbor search is used to find the K nearest points, and a threshold is set to determine similarity points: for each point p on the growth feature line, parallel K-nearest neighbor search is performed on each subtree T. i Perform a K-nearest neighbor search on the top to find the K points in each subtree that are closest to p. The specific formula is as follows:

[0129] N i =KNearestNeighbors(T i ,p,K),

[0130] Return the set of nearest neighbors N for all subtrees i Then, select the K nearest points to p as the final nearest neighbor set N, expressed by the formula:

[0131]

[0132] A threshold ∈ is set. Only points whose distance is less than this threshold are considered similar points. The found similar points are merged into the growth region. To speed up this process, the points on each growth feature line are processed in parallel. A growth region R is initialized, and the points on the growth feature line are added to this region. For each newly added point q, a K-nearest neighbor search is performed in parallel, and the found similar points are added to the growth region R. The above process is repeated in parallel until no new points can be added to the growth region R.

[0133] To optimize the growth region and remove noise or discontinuous parts, parallelized region merging or segmentation techniques can be used to improve the quality of the region. Parallelized smoothing filters can be used to remove noise points within the region. If two regions are very close or overlap, region merging can be performed in parallel to reduce unnecessary segmentation.

[0134] Post-processing of the segmentation results, such as filling holes and removing small regions, can also be parallelized. Parallelized filling algorithms can be used to repair holes in the regions, and parallel detection and removal of regions that are too small can be performed.

[0135] In this embodiment, when extracting morphological parameters based on the three-dimensional semantic information of the spinal structure and exploring the correlation between these parameters using correlation analysis, morphological parameters can be extracted from the three-dimensional semantic information of the spinal structure and normalized. Correlation analysis is then used to analyze the correlation coefficients between pairs of morphological parameters. The linear relationship between each morphological parameter is explained based on the magnitude of the correlation coefficients, and the relevant parameters are divided into different groups according to a preset classification standard. Variance analysis is performed on each group to determine the ratio of between-group variance to within-group variance and the probability of the ratio occurring. The differences in morphological parameters between different groups are explained based on the ratio and probability values. Clustering is performed on the morphological parameters, and the correlation and differences between the morphological parameters are explained based on the clustering results to obtain parameter correlation relationships. Based on these parameter correlation relationships, the spinal morphological features represented by each morphological parameter are shown.

[0136] It should be explained that the calculation of spinal morphological parameters and the establishment of correlations are based on extracting key morphological parameters from the results of spinal semantic segmentation, such as: cervical lordosis angle, thoracic kyphosis angle, thoracolumbar kyphosis angle, lumbar lordosis angle, sacral curvature, scoliosis angle, rotation angle, transverse process angle of ribs, and vertebral body angle.

[0137] The extracted parameters are standardized using the following formula:

[0138]

[0139] Among them, Z i P represents the normalized parameters. i σ represents the original parameters; μ represents the mean of the parameters; σ represents the standard deviation of the parameters.

[0140] Correlation analysis is used to explore the linear relationship between different parameters. Specifically, all extracted spinal morphology parameters can be collected, and a correlation analysis method, such as Pearson correlation coefficient or Spearman's rank correlation coefficient, can be selected. The correlation coefficient is calculated for each pair of parameters to assess the strength of the linear relationship between them, and a significance test (such as a t-test) is performed on the correlation of each pair of parameters to determine whether the correlation is statistically significant.

[0141] Based on the magnitude of the correlation coefficient and the results of the significance test, the linear relationship between the parameters is explained, and analysis of variance is used to explore the differences between different parameters.

[0142] Specifically, depending on the research objective, the parameter data can be divided into different groups, such as by gender, age, etc., and one-way ANOVA or multi-way ANOVA can be selected, mainly depending on the number and type of groups. ANOVA is performed on each parameter, and F-values ​​and p-values ​​are obtained to assess whether there are significant differences between different groups.

[0143] If the analysis of variance shows significant differences, perform multiple comparison tests (such as the Tukey test or Bonferroni correction) to determine which specific groups have differences, and interpret the differences in parameters between different groups based on the F-value and p-value.

[0144] The normalized parameter data is standardized or normalized to ensure that the data are on the same scale. An appropriate clustering algorithm is selected, such as K-means clustering, hierarchical clustering, or DBSCAN. The number of clusters (k value) is determined according to the research purpose or the use of the Elbow method. The selected algorithm is used to perform clustering operations on the parameter data. The silhouette score or other evaluation indicators are used to evaluate the quality of the clustering. Based on the clustering results, the similarities and differences between parameters and the spinal morphological features they may represent are interpreted.

[0145] In this embodiment, when generating a combination of spinal semantic parameters based on correlation and morphological parameters, using an optimization algorithm to find the optimal combination of spinal semantic parameters, and repairing the scoliosis spinal model based on the optimal combination of spinal semantic parameters, the combination of spinal semantic parameters can be determined based on spinal morphological features, correlation, and morphological parameters, and a parameter space can be generated based on the combination of spinal semantic parameters. Within the parameter space, the parameter combination is encoded as an individual, and an optimization objective is defined based on the application of spinal semantic parameters in spinal model repair. An evaluation function is defined based on the optimization objective to calculate the evaluation value of each individual. Based on the evaluation value and the optimization algorithm, genetic operations are performed to simulate natural selection and genetic processes, and the evaluation degree of the individual is re-evaluated based on the genetic process. The individual with the highest evaluation degree is selected as the optimal solution to obtain the optimal combination of spinal semantic parameters. A preliminary scoliosis spinal model is generated based on spinal data, and the optimal combination of spinal semantic parameters is applied to the preliminary scoliosis spinal model to repair the spinal state.

[0146] It should be explained that a genetic algorithm can be used to find the optimal combination of spinal semantic parameters. The specific steps are as follows:

[0147] Step 1: Population initialization;

[0148] Parameter space definition: Determine all possible combinations of spine semantic parameters, including but not limited to segmentation thresholds, edge detection parameters (such as those in the Sobel operator and Canny edge detection algorithm), feature extraction parameters (such as those in HOG and Haar features), classifier parameters (such as the C value and gamma value in SVM), and clustering parameters (such as the number of clusters in K-means).

[0149] Multiple candidate solutions (populations) are created based on random generation of spinal semantic parameter combinations or using heuristic methods: a certain number of candidate solutions are randomly generated in the parameter space, and some promising parameter combinations are selected as the initial population based on prior knowledge or experience. Each parameter combination is encoded as a chromosome, which can be binary encoding, real number encoding, or Gray encoding, etc.

[0150] Step 2: Fitness Assessment;

[0151] The fitness function (evaluation function) is defined based on specific objectives (such as segmentation accuracy, parameter correlation, computational complexity, and robustness), as follows:

[0152] F(x) = α.P accuracy +β.P correlation -γ.P complexity +δ.P robustness ;

[0153] In the formula, F(x) represents the evaluation value of the x-th individual, and P... accuracy P represents the segmentation accuracy, specifically the ratio of the correctly segmented spinal region to the actual spinal region. correlation The correlation coefficient (P) represents the correlation between parameters, specifically the correlation within a combination of parameters. complexity P represents computational complexity, which measures the running time and resource consumption of an algorithm. robustness Robustness is a measure of an algorithm's ability to resist external noise or data changes. α represents the weighting coefficient of segmentation accuracy, β represents the weighting coefficient of parameter correlation, γ represents the weighting coefficient of computational complexity, and δ represents the weighting coefficient of robustness.

[0154] The fitness value is calculated for each individual by comparing the algorithm's segmentation results with the real labeled results. The correlation coefficient (such as the Pearson correlation coefficient) is used to measure the correlation between parameters. At the same time, the algorithm's running time and resource consumption are recorded. The stability of the algorithm is tested by introducing noise or changes into the dataset.

[0155] Step 3: Genetic manipulation;

[0156] Genetic algorithms simulate natural selection and the genetic process through genetic operations (selection, crossover, mutation): individuals are selected based on fitness values ​​to produce the next generation, and two individuals are randomly selected and some parts of them are exchanged to produce new individuals, and a part of an individual is randomly changed to produce new individuals. Through the iterative process, the population gradually converges to the optimal solution.

[0157] Genetic operations are used to generate a new generation of population. The fitness of each individual in the new population is re-evaluated. When the preset maximum number of iterations is reached or a certain termination condition is met, the iteration stops and the results are output. The individual with the highest fitness in the current population is selected as the optimal solution and the optimal solution, i.e., the best combination of spinal semantic parameters, is output.

[0158] In this embodiment, when generating a preliminary scoliosis spine model based on spinal data and applying the optimal combination of spinal semantic parameters to the preliminary scoliosis spine model to repair the spinal state, the medical imaging data of the healthy spine can be preprocessed to extract spinal contour data, and a preliminary scoliosis spine model can be constructed by combining auxiliary design technology; boundary conditions and initial displacement and velocity are defined based on preset repair requirements, and the deformation of the preliminary scoliosis spine model is guided by the physical model deformation algorithm to obtain the new position of each data point and obtain the scoliosis spine deformation model; the optimal combination of spinal semantic parameters is applied to the scoliosis spine deformation model to simulate the morphological changes of the spine, and the scoliosis spine deformation model is repaired based on the morphological simulation results.

[0159] It should be explained that the optimal semantic feature combination guiding spinal model deformation first guides the model deformation based on healthy spinal data to generate a preliminary spinal model. Then, the optimal combination of spinal semantic parameters obtained through genetic algorithm is applied to the model to simulate and adjust the spinal shape, explore possible correction strategies, and predict the ideal state after correction.

[0160] The process of generating a spinal model is as follows: Medical imaging data of a healthy spine, such as X-rays, CT scans, or MRI scans, are collected in advance. The collected image data is preprocessed, including denoising, standardization, and segmentation, to extract the contour of the spine. Based on the preprocessed data, a three-dimensional model of the spine is constructed using computer-aided design (CAD) software or 3D modeling tools. The accuracy of the model is verified by comparing it with the anatomical structure of an actual healthy spine. Using healthy spine data, the initial deformation of the spinal model is guided by deformation algorithms (such as physics-based deformation models or machine learning models). The process of guiding the initial deformation of the spinal model using physics-based deformation algorithms involves modeling and simulating the physical properties of the spine.

[0161] Modeling and simulating the physical properties of the spine requires establishing a model that reflects these properties and describes the spine's response under different conditions. For example, for an elastic body, the Navier-Cauchy equation can be used to describe its behavior. Where σ1 represents the stress tensor, b represents the volume force (such as gravity), ρ represents the material density, and u represents the second derivative of the displacement.

[0162] For linear elastic materials, the relationship between the stress tensor and the strain tensor can be expressed by Hooke's law: σ = E.ε, where E represents Young's modulus and ε represents the strain tensor.

[0163] Secondly, specify appropriate boundary conditions to simulate the real physical environment. Boundary conditions can be displacement boundary conditions (fixing the displacement of certain areas) or force boundary conditions (applying external loads). At the same time, set initial conditions, such as initial displacement and velocity.

[0164] Numerical methods (such as the finite element method) are used to solve the equations. The finite element method discretizes a continuous domain into a finite number of small elements, approximates the solution within each element, and finds the solution for the entire domain by solving a system of linear equations. In the finite element method, the solution can be expressed as:

[0165]

[0166] In the formula, N i Represents the shape function, u i The displacement at the nodes is represented by KU. Substituting this into the equations of elasticity, we can obtain a set of linear equations: KU = F, where K represents the stiffness matrix; U represents the nodal displacement vector; and F represents the external force vector. Once the linear equations are solved, the new position of each node can be obtained, thus yielding the deformed spine model.

[0167] Specifically, the steps for adjusting spinal morphology based on the optimal parameter combination are as follows: First, identify the key semantic parameters affecting spinal morphology, such as vertebral body size, intervertebral disc thickness, and intervertebral angle. Then, use a genetic algorithm to optimize these parameters and find the best parameter combination. To improve the performance of the genetic algorithm, increase its diversity, and prevent premature convergence to local optima, the genetic algorithm is improved. The specific steps are as follows:

[0168] Set the initial population size N, where each individual in the population represents a set of parameter configurations. Randomly generate N sets of parameters as the initial population P0, specifically: P0 = {x1, x2, ..., x3}. N}; where x i This represents the parameter vector of the i-th individual.

[0169] The fitness value of each individual is calculated. The fitness function can be defined based on the difference between the spinal model and the ideal state. The fitness value is f(x). i The higher the value of x, the better for individual x. i The closer to the optimal solution, the more specific the formula is: f(x) i ) = evaluate(x i ).

[0170] This approach combines roulette wheel selection with elitism to ensure that the best individuals are preserved. Elitism guarantees that at least one best individual will be passed on to the next generation, as expressed by the formula: P elite ={x best};

[0171] P next =P elite ∪{roulette wheel selection(P current )};

[0172] In the formula, P next P represents the next generation of the population. elite P represents the set of elite individuals. current x represents the current generation population. best This represents the best individual in the current generation.

[0173] Multiple crossover strategies (such as single-point crossover, two-point crossover, uniform crossover, etc.) are employed to increase population diversity, with a crossover probability C. p The value should be moderate; too high a value will lead to premature convergence, while too low a value may result in slow convergence. The formula is as follows:

[0174]

[0175] Mutation operations can help escape local optima and increase population diversity. The mutation probability M p The value should be moderate; too high a value will destroy excellent individuals, while too low a value may not increase sufficient diversity. The specific formula is: if random() < M p then{x i ' = mutate(x) i );

[0176] A new generation population, P, is generated based on selection, crossover, and mutation operations. t+1 It should include elite individuals and new individuals generated through selection, crossover, and mutation operations. The specific formula is expressed as: if t≥Torf(x)converges thenstop.

[0177] In the formula, X1 and X2 represent two individuals in the current generation population, the parent individuals used for crossover, X′1 and X′2 represent the offspring individuals generated by the crossover operation, random() is a random number generation function, which usually generates a random number between 0 and 1, x i ' represents a new individual generated through mutation, x i Let represent the current individual, t represent the current generation, T represent the maximum number of iterations, and f(x) be the fitness function.

[0178] Furthermore, by adjusting the parameters in the model, the morphological changes of the spine can be simulated. Virtual reality (VR) technology can be introduced during the process of spinal morphology adjustment, providing doctors, researchers, and patients with a more intuitive and interactive learning and treatment environment. This not only increases the functionality of spinal morphology adjustment but also enhances its fun and educational value.

[0179] Using VR software development kits (SDKs), such as Unity or Unreal Engine, a virtual environment is created that can display a spinal model and allow users to interact with it through VR devices (such as head-mounted displays (HMDs) and controllers). Users wear VR devices to enter the virtual environment and import the spinal parameter model optimized based on genetic algorithms into the VR environment to ensure that the model can be displayed correctly in VR.

[0180] The user interface (UI) is designed to allow users to directly adjust parameters of the spinal model, such as vertebral body size, intervertebral disc thickness, and intervertebral angles, using VR controllers. A real-time feedback mechanism is implemented; when the user adjusts parameters, the spinal model updates immediately, allowing the user to observe the changes. After adjustment, the system saves the final spinal morphology and can export the data for practical medical applications or research. Simultaneously, data from user adjustments to the spinal model in the VR environment is collected for subsequent analysis and optimization of the genetic algorithm, and for evaluating the difference between the adjusted spinal model and the ideal state.

[0181] The steps involved in exploring correction strategies and predicting the ideal state after correction are as follows: Analyze the differences between the current spinal model and the ideal state to identify the parts that need correction; design possible correction strategies, such as physical therapy, surgical correction, or bracing; simulate the application of correction strategies on the model and observe changes in spinal morphology; predict the ideal spinal state after applying the correction strategies based on the simulation results; verify the accuracy of the prediction results through actual cases and adjust the model and correction strategies based on feedback.

[0182] In summary, by utilizing the above-mentioned technical solution of this invention, the present invention employs a global optimization strategy based on optimization algorithms to design a reasonable evaluation function to quantify the performance of each parameter combination in the scoliosis model repair task. It automatically finds the optimal semantic parameter combination weights, thereby guiding the effective adjustment of scoliosis spine model parameters. This not only improves the quality of model repair but also enhances parameter optimization efficiency, avoiding the tedious work of manually modifying parameters blindly. This invention utilizes well-defined and obtainable intact semantic parameters with clear anatomical significance to characterize the degree, morphological features, and potential correction needs of scoliosis in detail. Simultaneously, it employs optimization algorithms to find the most suitable model parameter combination to accurately simulate the real state of the scoliosis spine and the possible best correction results.

[0183] This invention achieves efficient and accurate segmentation of spinal images, improving the quality of spinal reconstruction and ensuring the integrity and continuity of the spinal structure. Furthermore, it guides the initial deformation of the spinal model based on physical models and finite element analysis, providing a foundation for subsequent optimization using genetic algorithms and ensuring the morphological rationality and accuracy of the spinal model. The rapid lateral spinal model reconstruction method implemented in this invention significantly improves the accuracy of spinal reconstruction and the realism of the 3D model, effectively solving the problems of structural continuity and detail loss in traditional methods. This greatly promotes the accuracy and efficiency of medical image analysis, surgical planning, and treatment evaluation, and is of great significance for advancing personalized medicine and precision surgery.

[0184] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for fast repair of a scoliosis model based on semantic feature optimization, characterized in that, The method comprises the following steps: Collecting tomography images and magnetic resonance images as medical images, and performing registration processing on the medical images by using an affine transformation technique, and verifying the registration result based on the matching degree between the medical images; Adjusting the change parameters according to the verification result to complete the registration processing, and performing denoising and contrast enhancement processing on the registered medical images by using filtering and equalization methods to obtain optimized medical images; Calculating the gradient amplitude and direction of each pixel point in the medical images, and identifying the contour of the medical images according to the threshold index defined based on historical experience, and eroding the pixels on the contour boundary based on morphological erosion; Using the inflation technology to fill the gap between the objects on the contour boundary to obtain the contour feature line, and determining the position and direction of the slice plane based on the position and direction of the contour feature line; Calculating the point set intersecting with the slice plane based on the contour feature line, and projecting the point set onto the slice plane to obtain the region information of the spine based on the projection result; Generating a graph model based on the adaptive sampling strategy and the point set, and analyzing the graph model based on the graph theory algorithm to complete the hole recognition and repair operation, and extracting the ligament information based on the repair result; Applying the watershed algorithm to identify the point set to obtain the growth feature line, and constructing a multi-dimensional tree after optimizing the growth feature line, and analyzing the detail information of the spine structure based on the multi-dimensional tree; Taking the spine structure, the spine region and the ligament information as the three-dimensional semantic information of the spine structure; Extracting the morphological parameters based on the three-dimensional semantic information of the spine structure, and exploring the correlation between the morphological parameters by using the correlation analysis technique; Generating a spine semantic parameter combination based on the correlation and the morphological parameters, finding the best spine semantic parameter combination by using an optimization algorithm, and repairing the scoliosis spine model according to the best spine semantic parameter combination.

2. The method of claim 1, wherein, The method comprises the following steps: Taking the point set as an original spine data set, and randomly selecting a group of data points from the original spine data set as first sampling points and adding them to the sampling point set; Iteratively selecting sampling points from the original spine data set based on a preset number of sampling points, and the sum of distances between the sampling points and all data points in the original spine data set is maximum; When the number of iterations reaches the preset number of sampling points, the sampling points are added to the sampling point set to form a spine sampling data set, the spine sampling data set is converted into a graph model, and nodes are created for the data points; Determining the distance between any two nodes, connecting the two nodes to create an edge when the distance value is less than a set threshold, and detecting the edge creation condition by using a graph theory algorithm to analyze the hole information of the graph model; Determining the boundary of the hole and filling the hole by combining the minimum spanning tree, and extracting the ligament information based on the filled graph model.

3. The method of claim 2, wherein, The method comprises the following steps: Calculating the gradient amplitude and direction of each pixel point in the medical images, and identifying the contour of the medical images according to the threshold index defined based on historical experience, and eroding the pixels on the contour boundary based on morphological erosion; Using the inflation technology to fill the gap between the objects on the contour boundary to obtain the contour feature line, and determining the position and direction of the slice plane based on the position and direction of the contour feature line; Calculating the point set intersecting with the slice plane based on the contour feature line, and projecting the point set onto the slice plane to obtain the region information of the spine based on the projection result; Generating a graph model based on the adaptive sampling strategy and the point set, and analyzing the graph model based on the graph theory algorithm to complete the hole recognition and repair operation, and extracting the ligament information based on the repair result; Applying the watershed algorithm to identify the point set to obtain the growth feature line, and constructing a multi-dimensional tree after optimizing the growth feature line, and analyzing the detail information of the spine structure based on the multi-dimensional tree; Taking the spine structure, the spine region and the ligament information as the three-dimensional semantic information of the spine structure; Extracting the morphological parameters based on the three-dimensional semantic information of the spine structure, and exploring the correlation between the morphological parameters by using the correlation analysis technique; Generating a spine semantic parameter combination based on the correlation and the morphological parameters, finding the best spine semantic parameter combination by using an optimization algorithm, and repairing the scoliosis spine model according to the best spine semantic parameter combination. After the data points in the point set are preprocessed, the corresponding gradient values are calculated, and the points with gradient values exceeding a threshold value are selected as the starting points of the growing characteristic lines; A gradient image is constructed using the gradient values, and the data point with the maximum gradient value is selected as the seed point, which is set as the marker point of the watershed transform; Based on the marker point, the gradient image is segmented into several regions by the watershed transform, and after the boundary information between regions and the gradient change within the regions are determined, the growing characteristic lines are identified; After noise removal and discontinuous point removal operations are performed on the growing characteristic lines, the point set corresponding to the growing characteristic lines is divided into several subsets, and a multi-dimensional tree is constructed on the subsets; Based on the multi-dimensional tree and the K-nearest neighbor search technique, similar points on the growing characteristic lines are found to form growing regions, and after the growing regions are merged and segmented, the detailed information of the spinal structure is extracted.

4. The method of claim 1, wherein, The morphological parameters are extracted based on the three-dimensional semantic information of the spinal structure, and the correlation analysis technique is used to explore the correlation between the morphological parameters, including: The morphological parameters are extracted from the three-dimensional semantic information of the spinal structure, and standardized processing is performed on the morphological parameters. The correlation technique is used to analyze the correlation coefficients between each pair of morphological parameters. The linear relationships between the morphological parameters are explained according to the sizes of the correlation coefficients, and the correlation parameters are divided into different groups according to the pre-set division standard. The variance analysis is performed on each group to determine the ratio value of the inter-group variance to the intra-group variance and the probability value of the ratio value, and the differences between the morphological parameters in different groups are explained according to the ratio value and the probability value. The morphological parameters are clustered, and the correlation and differences between the morphological parameters are explained based on the clustering results to obtain the parameter correlation relationship. The morphological features of the spinal column represented by each morphological parameter are shown based on the parameter correlation relationship.

5. The method of claim 1, wherein, The spinal semantic parameter combination is generated based on the correlation relationship and the morphological parameters, and the optimal spinal semantic parameter combination is found using an optimization algorithm. The scoliosis spinal model is repaired based on the optimal spinal semantic parameter combination, including: The spinal semantic parameter combination is determined based on the morphological features of the spinal column, the correlation relationship, and the morphological parameters, and a parameter space is generated based on the spinal semantic parameter combination. In the parameter space, the parameter combination is encoded as an individual, and the optimization objective is defined according to the application of the spinal semantic parameters in the repair of the spinal model. The evaluation function is calculated based on the optimization objective to determine the evaluation value of each individual. Based on the evaluation value and the optimization algorithm, genetic operations are performed to simulate the natural selection and genetic process. The evaluation degree of the individual is re-evaluated based on the genetic process, and the individual with the highest evaluation degree is selected as the optimal solution to obtain the best spinal semantic parameter combination. The scoliosis spinal model is generated based on the spinal data, and the best spinal semantic parameter combination is applied to the scoliosis spinal model to repair the spinal state.

6. The method of claim 5, wherein the method further comprises: The calculation formula of the evaluation function is: ； In the formula, F(x) represents the evaluation value of the xth individual, P accuracy represents the segmentation accuracy, P correlation represents the parameter correlation, P complexity represents the calculation complexity, P robustness represents the robustness, and α represents the weight coefficient of the segmentation accuracy, β represents the weight coefficient of the parameter correlation, γ represents the weight coefficient of the calculation complexity, and δ represents the weight coefficient of the robustness.

7. The method of claim 6, wherein the method further comprises: The scoliosis spinal model is generated based on the spinal data, and the best spinal semantic parameter combination is applied to the scoliosis spinal model to repair the spinal state, including: After preprocessing the medical image data of a healthy spine, the spinal contour data is extracted, and an auxiliary design technique is used to construct a scoliosis spinal model. The boundary conditions, initial displacement and velocity are defined based on preset repair requirements, and a physical model deformation algorithm is combined to guide the deformation of the scoliosis model to obtain a new position of each data point and obtain a scoliosis deformation model; The best spinal semantic parameter combination is applied to the scoliosis deformation model to simulate the morphological changes of the spine, and the scoliosis deformation model is repaired based on the morphological simulation results.

Images