A musculoskeletal personalized finite element rapid modeling method based on medical images

Abstract

A kind of medical image-based personalized finite element rapid modeling method of muscle and bone belongs to biomedical engineering, medical image processing, digital twin and finite element modeling technical field.The method of the present application comprises: collecting target muscle and bone tissue medical image of subject, automatically segmented to obtain subject semantic mask and is divided into anatomical component;Template semantic mask and subject semantic mask affine registration obtain affine transformation matrix;Based on differential homeomorphism deformation network, elasticity registration is carried out to obtain nonlinear deformation field;Human muscle and bone tissue template grid deformation obtains the specific deformed grid of subject;Ligament or tendon attachment point correction and the specific deformed grid of subject are assembled to generate specific subject finite element model.The whole process only needs to input target muscle and bone tissue medical image of subject to generate specific subject finite element model containing multi-level joint structure, to meet the demand of muscle and bone system digital twin rapid modeling.

Description

Technical Field

[0001] This invention belongs to the fields of biomedical engineering, medical image processing, digital twin and finite element modeling technology, and in particular relates to a rapid finite element modeling method for personalized musculoskeletal systems based on medical images. Background Technology

[0002] like Figure 1 As shown, the multi-scale human musculoskeletal system is composed of various structures, including bone tissue, articular cartilage tissue, intra-articular fibrocartilage tissue, ligament tissue, and tendon tissue. It is a crucial biomechanical system for maintaining human movement, load-bearing capacity, and postural stability. With the development of precision medicine, sports injury assessment, intervention for musculoskeletal diseases, rehabilitation training planning, and implant design, individualized digital twin models of the musculoskeletal system are gradually becoming an important foundation for biomechanical analysis and decision support. These digital twin models require mapping the subject's actual anatomical morphology into a computable, personalized finite element model to quantify the contact, tension, constraint, and load transfer relationships between different musculoskeletal structures.

[0003] Currently, digital twin modeling of the musculoskeletal system typically relies on multiple steps, including medical image segmentation, geometric reconstruction, mesh generation, material property configuration, boundary condition setting, and finite element method (FEM) solution. Medical images can provide individualized anatomical information of the target musculoskeletal tissue in a subject, but it is difficult to directly obtain the mechanical responses such as internal stress, strain, contact pressure, and attachment point forces from images alone. Finite element analysis can transform structural information from images into a computable model, making it a crucial technical approach for constructing digital twins of the musculoskeletal system.

[0004] However, existing personalized finite element modeling workflows for digital twins of the musculoskeletal system still heavily rely on manual processing. Traditional workflows typically include manual segmentation of medical images, tissue boundary repair, surface reconstruction, volume mesh generation, assembly between different anatomical tissues, correction of ligament or tendon attachment points, and collision detection. These processes are time-consuming, have poor repeatability, and struggle to meet the practical needs of multi-site, multi-tissue, and rapid modeling. The current technological bottlenecks are mainly reflected in the following three aspects: (1) High-precision and high-fidelity modeling requirements for digital twins of the musculoskeletal system. The bone tissue, cartilage tissue, intra-articular fibrocartilage tissue, ligament tissue and tendon tissue of different subjects and different target sites vary significantly in morphology, scale and relative position. Traditional manual modeling methods are prone to anatomical distortion due to segmentation errors, excessive smoothing of curved surfaces or inaccurate assembly between tissues, which in turn affects the simulation accuracy of personalized finite element models.

[0005] (2) The need for automated and rapid modeling of digital twins of the musculoskeletal system. Traditional manual reverse modeling is usually time-consuming, and the results vary between different modelers, making it difficult to form a stable and repeatable personalized modeling process. Therefore, there is an urgent need for an automated technology that can rapidly generate finite element models from medical images of the target musculoskeletal tissue of a subject.

[0006] (3) High stability and end-to-end modeling requirements for digital twins of the musculoskeletal system. The coupling relationship between soft and hard tissues in musculoskeletal tissue is complex, and the location of attachment points of slender structures such as ligaments and tendons has a significant impact on model assembly and mechanical calculations. Existing automatic mesh generation and automatic assembly methods are prone to problems such as mesh distortion, inter-tissue penetration, or attachment point offset, making it difficult for the model to be used stably for subsequent finite element solutions. Therefore, there is an urgent need for an end-to-end rapid modeling scheme from medical image input to subject-specific finite element model output.

[0007] To address the aforementioned problems, this invention provides a rapid, personalized finite element modeling method for musculoskeletal systems based on medical images. This method uses medical images of the target musculoskeletal tissue of a subject as input. Through semantic segmentation, anatomical component partitioning, affine registration, differential homeomorphic elastic registration, template mesh coordinate mapping, ligament or tendon attachment point correction, and collision detection and assembly, it rapidly generates a specific subject finite element model containing multi-level joint structures. This provides a technical foundation for the construction of digital twin models of the musculoskeletal system and personalized biomechanical analysis. Summary of the Invention

[0008] This invention aims to address the problems of long modeling cycles, low automation, insufficient geometric fidelity, and poor tissue assembly stability in personalized finite element modeling of digital twins of the musculoskeletal system. It provides a rapid personalized finite element modeling method for musculoskeletal systems based on medical images, enabling end-to-end rapid construction from medical images of the target musculoskeletal tissue of the subject to a simulateable subject-specific finite element model. A schematic diagram of the overall method is shown below. Figure 2 As shown.

[0009] A rapid finite element method for personalized musculoskeletal modeling based on medical images includes the following steps: Step 1: Acquire medical images of the subject's target musculoskeletal tissue. Segment the images using an image segmentation network, such as nnU-Net, to obtain a subject semantic mask containing multiple anatomical structures within the target musculoskeletal tissue. Then, classify all subject semantic masks according to anatomical category and motor function. A separate anatomical component, in which The integer is positive; simultaneously, the human musculoskeletal tissue template mesh and the corresponding template semantic mask, which are pre-constructed using manual image segmentation and entity modeling methods, are obtained, and the human musculoskeletal tissue template mesh and its corresponding template semantic mask are also divided into... A separate anatomical component; Step 2, based on each individual anatomical component obtained in Step 1, for index Anatomical components, among which , will belong to the Affine registration is performed between the template semantic mask of the first anatomical component and the subject semantic mask to calculate the corresponding semantic mask of the first anatomical component. Affine transformation matrix of each anatomical component , so as to belong to the first The template semantic mask of each anatomical component is initially spatially aligned with the corresponding subject semantic mask in terms of pose and scale to obtain the initially spatially aligned template semantic mask. Step 3, based on the result obtained in Step 1, belonging to the first... The subject semantic mask for each anatomical component, and the template semantic mask obtained in step 2 after initial spatial alignment, are used to perform elastic registration calculations using a differential homeomorphic deformable network, such as VoxelMorph, to output the corresponding semantic mask for the first anatomical component. Nonlinear deformation field of each anatomical component ; Step 4: Based on the affine transformation matrix calculated in Step 2 and the nonlinear deformation field calculated in step 3 For the data obtained in step 1 belonging to the first... The nodes in the human musculoskeletal tissue template mesh of each anatomical component are sequentially subjected to coordinate mapping deformation to obtain a subject-specific deformed mesh that matches the subject's anatomical structure. Step 5: Based on the subject-specific deformed mesh obtained in Step 4, a hybrid strategy based on a distance threshold is used to correct the position of ligament or tendon attachment points in the subject-specific deformed mesh. All corrected subject-specific deformed meshes are then subjected to collision detection and assembly to generate a subject-specific finite element model containing multi-level joint structures.

[0010] Further, in step 1, the target musculoskeletal tissue of the subject includes multiple anatomical structures such as bone tissue, articular cartilage tissue, intra-articular fibrocartilage tissue, ligament tissue, and tendon tissue; all subject semantic masks are divided into Each independent anatomical component contains parameters It is a positive integer greater than or equal to 2; the independent anatomical components specifically include: bone and cartilage complex, intra-articular fibrocartilage and cartilage complex, ligament complex and tendon complex.

[0011] Furthermore, in step 3, a differential homeomorphic deformation network is used for elastic registration calculation, outputting the corresponding... Nonlinear deformation field of each anatomical component The calculation process is as follows: The differential homeomorphic deformation network first predicts and generates a steady-state velocity field. ; Then, the steady-state velocity field is scaled and squared. In time Integrating within the inner region generates a nonlinear deformation field that ensures topological continuity and invertibility. Their relationship satisfies the formula:

[0012] in, This indicates the exponentiation mapping operator.

[0013] Furthermore, in step 3, the training process of the differential homeomorphic network is based on parameter optimization using a composite loss function, specifically targeting the first... The formula for calculating the composite loss function of each anatomical component is as follows:

[0014] in, Indicates the first Total loss term for each anatomical component; Indicates the first Subject semantic mask for each anatomical component; Indicates the first Template semantic mask for each anatomical component; Compound operators representing spatial transformations; Indicates passing through a nonlinear deformation field The template semantic mask after deformation; This represents the similarity loss function term used to measure overlap. This represents the regularization hyperparameter used to control the smoothing weights; This represents the smoothness loss function term used to penalize spatial gradients; Similarity loss function term The Dice loss is used, and the calculation formula is as follows:

[0015] in, Operators that count the number of voxel pixels within a set; The intersection operator for mask sets; Smoothness loss function term The calculation formula is:

[0016] in, The summation operator for all voxel positions in space; Represents the spatial gradient operator. express Steady-state velocity field of a voxel; The operator represents the operator for calculating the square of the second norm of a vector; Indicates the position of the voxel. Represents the image space.

[0017] Furthermore, in step 4, the calculation formula for sequentially performing coordinate mapping deformation on the nodes in the human musculoskeletal tissue template mesh is as follows:

[0018] in, This indicates that the result obtained after deformation belongs to the first... The target coordinates of the nodes of the deformed mesh for each anatomical component; Indicates that before the transformation it belongs to the first Initial coordinates of nodes in a human musculoskeletal tissue template mesh for an anatomical component; Represents nonlinear deformation fields The corresponding nonlinear displacement fields satisfy the following conditions: , where y is the spatial coordinate.

[0019] Furthermore, step 5, which employs a distance threshold-based hybrid strategy to correct the position of ligament or tendon attachment points in the deformed mesh specific to the subject, specifically includes the following sub-steps: Step 5.1: For each ligament or tendon endpoint in the deformed mesh specific to the subject, calculate the Euclidean distance from that ligament or tendon endpoint to the nearest bone surface, and preset a distance threshold. The unit is mm; Step 5.2, based on the Euclidean distance calculated in Step 5.1, make a judgment: when the Euclidean distance is less than or equal to... When the principal component analysis-guided projection is used to determine the location of the ligament or tendon attachment point, the first principal component direction of the ligament or tendon is extracted as the longitudinal axis direction, the endpoint of the ligament or tendon is projected onto the nearest bone surface along the longitudinal axis direction, and the mesh node closest to the projection location is set as the corrected ligament or tendon attachment point. Step 5.3, when the Euclidean distance is greater than At that time, the center of the insertion area of ​​the ligament or tendon is automatically determined as the corrected ligament or tendon attachment point based on the preset anatomical reference benchmark.

[0020] Furthermore, in step 5, collision detection and assembly of all the modified subject-specific deformed meshes includes: detecting the penetration area between cartilage, bone tissue and adjacent soft tissue based on triangular facets, and performing minimum displacement correction on the relevant nodes along the local surface normal so that the penetration depth between tissues does not exceed the preset contact tolerance.

[0021] The beneficial effects of this invention are: (1) High modeling efficiency and automation: This invention reduces the tedious manual segmentation, geometric repair, mesh generation and model assembly steps in traditional modeling by using a template-based multi-stage deformation modeling strategy. Based on the preset human musculoskeletal tissue template mesh, template semantic mask and trained network model, the application stage only requires inputting the medical image of the subject's target musculoskeletal tissue to generate a subject-specific finite element model containing multi-level joint structures, meeting the needs of rapid modeling of digital twins of the musculoskeletal system.

[0022] (2) High mesh quality and computational stability: The present invention introduces differential homeomorphic deformation constraint in the elastic registration stage, which ensures the smoothness and reversibility of the deformation field, so that the generated subject-specific finite element model can inherit the topological characteristics of the human muscle and bone tissue template mesh, and reduce the risk of mesh distortion caused by automatic mesh generation.

[0023] (3) Accurate modeling and strong generalization ability: Through the division of anatomical components and the two-stage registration strategy of "affine + elastic", this method can compensate for the differences in target muscle and bone tissue size, posture and local morphology between the subject and the template. While capturing complex morphological details such as bone tissue, articular cartilage tissue, intra-articular fibrocartilage tissue, ligament tissue and tendon tissue, it can improve the applicability of the model under different target sites and different anatomical specificities.

[0024] (4) End-to-end applicability of digital twin modeling: This invention realizes a closed-loop operation from medical images of the subject's target musculoskeletal tissue to a subject-specific finite element model. Ligament or tendon attachment point correction and collision detection and assembly technology are beneficial for generating personalized models that can be used for finite element solution and digital twin analysis of the musculoskeletal system. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the human musculoskeletal system at multiple scales; Figure 2 This is a structural block diagram of a rapid finite element modeling method for personalized musculoskeletal systems based on medical images. Figure 3 This is a flowchart of a rapid finite element modeling method for personalized musculoskeletal systems based on medical images; Figure 4 This is a schematic diagram of the deformation of the coordinate mapping of the template mesh nodes of the human knee joint. Detailed Implementation

[0026] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0027] The following uses the knee joint as a specific example of the target musculoskeletal tissue to illustrate the complete working process of the present invention's personalized finite element rapid modeling method for musculoskeletal tissue based on medical imaging. A schematic diagram is shown below. Figure 3 As shown. It should be understood that this specific embodiment is only used to illustrate the implementation of the present invention and does not limit the present invention to the knee joint.

[0028] A rapid finite element method for personalized musculoskeletal modeling based on medical images includes the following steps: Step 1: When using the knee joint as the target musculoskeletal tissue, acquire medical images of the knee joints of multiple subjects using medical imaging equipment. In one embodiment, the knee joint medical images are knee joint magnetic resonance imaging (MRI) images, or other medical images capable of characterizing the three-dimensional anatomical structure of the knee joint. Manually annotate the femur, tibia, patella, fibula, femoral cartilage, medial tibial cartilage, lateral tibial cartilage, patellar cartilage, medial meniscus, lateral meniscus, anterior cruciate ligament, posterior cruciate ligament, medial collateral ligament, lateral collateral ligament, patellar ligament, and quadriceps tendon in some of the knee joint medical images to obtain corresponding semantic mask images. Based on the knee joint medical images and their semantic mask images, construct an image segmentation dataset and train an image segmentation network (nnU-Net).

[0029] In the application phase, medical images of the subject's knee joint are acquired, and these images are segmented using a trained image segmentation network to obtain a subject semantic mask containing multiple anatomical structures within the subject's knee joint. Based on anatomical category and motor function, all subject semantic masks are divided into... A separate anatomical component; in one implementation, parameters The independent anatomical components include the femoral complex, tibia-fibula complex, patellar complex, anterior cruciate ligament, posterior cruciate ligament, medial collateral ligament, lateral collateral ligament, medial and lateral meniscus complex, quadriceps tendon, and patellar ligament. Simultaneously, a human knee joint template mesh and its corresponding semantic mask, pre-constructed using manual image segmentation and solid modeling methods, are obtained. Bone tissue can be modeled using solid elements, articular cartilage tissue and meniscus can be modeled using solid elements, and ligament and tendon tissue can be modeled using one-dimensional tension connectors or equivalent continuum forms. The human knee joint template mesh and its corresponding semantic mask are also divided into 10 independent anatomical components. Step 2, based on each individual anatomical component obtained in Step 1, for index Anatomical components, among which , will belong to the Affine registration is performed between the template semantic mask of each anatomical component and the subject semantic mask. During affine registration, the template semantic mask is used as a moving image, and the subject semantic mask is used as a fixed image. The corresponding semantic mask for each anatomical component is then calculated. Affine transformation matrix of each anatomical component This matrix contains translation, rotation, scaling, and shearing transformations to transform elements belonging to the first... The template semantic mask of each anatomical component is initially spatially aligned with the corresponding subject semantic mask in terms of pose and scale to obtain the initially spatially aligned template semantic mask. Step 3, based on the result obtained in Step 1, belonging to the first... The subject semantic mask for each anatomical component, and the initial spatially aligned template semantic mask obtained in step 2, are used for elastic registration calculation using a differential homeomorphic deformation network (VoxelMorph). Before application, the differential homeomorphic deformation network is trained offline based on multiple knee joint semantic mask images in the training dataset. The training process optimizes parameters based on a composite loss function, specifically targeting the [missing information - likely a specific type of image or feature]. The formula for calculating the composite loss function of each anatomical component is as follows:

[0030] in, Indicates the first Total loss term for each anatomical component; Indicates the first Subject semantic mask for each anatomical component; Indicates the first Template semantic mask for each anatomical component; Compound operators representing spatial transformations; Indicates passing through a nonlinear deformation field The template semantic mask after deformation; This represents the similarity loss function term used to measure overlap. This represents the regularization hyperparameter used to control the smoothing weights; This represents the smoothness loss function term used to penalize spatial gradients; This represents the steady-state velocity field predicted by the differential homeomorphic network; Similarity loss function term The Dice loss is used, and the calculation formula is as follows:

[0031] in, Operators that count the number of voxel pixels within a set; The intersection operator for mask sets; Smoothness loss function term The calculation formula is:

[0032] in, The summation operator for all voxel positions in space; Represents the spatial gradient operator. express Steady-state velocity field of a voxel; The operator represents the operator for calculating the square of the second norm of a vector; Indicates the position of the voxel. Represents the image space.

[0033] In the application phase, the trained differential homeomorphism network is invoked to elastically register the initially spatially aligned template semantic mask with the subject semantic mask. The differential homeomorphism network first predicts and generates a steady-state velocity field. Then, the steady-state velocity field is scaled and squared. In time Integrating within the inner region generates a nonlinear deformation field that ensures topological continuity and invertibility. Their relationship satisfies the formula:

[0034] in, This indicates the exponentiation mapping operator.

[0035] Step 4, as follows Figure 4 As shown, the affine transformation matrix calculated in step 2... and the nonlinear deformation field calculated in step 3 For the data obtained in step 1 belonging to the first... The nodes in the human knee joint template mesh of each anatomical component are sequentially deformed by coordinate mapping. The calculation formula for the sequential coordinate mapping deformation of the nodes in the human knee joint template mesh is as follows:

[0036] in, This indicates that the result obtained after deformation belongs to the first... The target coordinates of the nodes of the deformed mesh for each anatomical component; Indicates that before the transformation it belongs to the first Initial coordinates of nodes in a human knee joint template mesh for an anatomical component; Represents nonlinear deformation fields The corresponding nonlinear displacement fields satisfy the following conditions: , where y is the spatial coordinate.

[0037] During the actual mapping process, due to the nonlinear displacement field This is a discrete vector field defined based on a voxel grid. This step uses a trilinear interpolation algorithm, based on the grid node positions after affine transformation. The displacement of the node is obtained by interpolation from the displacement vector of the adjacent voxels, thereby achieving precise deformation from the template mesh node to the subject mesh node, and obtaining a subject-specific deformed mesh that matches the subject's anatomical structure.

[0038] Step 5: Based on the subject-specific deformed mesh obtained in Step 4, a hybrid strategy based on a distance threshold is used to correct the position of ligament or tendon attachment points in the subject-specific deformed mesh. In this embodiment, mm. For the endpoints of the anterior cruciate ligament, posterior cruciate ligament, medial collateral ligament, lateral collateral ligament, patellar ligament, or quadriceps tendon in the knee joint, the Euclidean distance from the endpoint to the nearest bone surface is calculated. When the Euclidean distance is less than or equal to 4 mm, the first principal component direction of the ligament or tendon is extracted as the longitudinal axis, and the endpoint of the ligament or tendon is projected onto the nearest bone surface along the longitudinal axis. The mesh node closest to this projection position is set as the corrected ligament or tendon attachment point. When the Euclidean distance is greater than 4 mm, the center of the insertion area of ​​the ligament or tendon is automatically determined as the corrected ligament or tendon attachment point based on a preset anatomical reference. Finally, all the corrected subject-specific deformed meshes are assembled, and the penetration areas between cartilage, meniscus, bone tissue, and adjacent soft tissues are identified through collision detection based on triangular facets. The relevant nodes are then corrected with minimum displacement along the local surface normal to ensure that the penetration depth between tissues does not exceed 0.1 mm, generating a finite element model of the subject's knee joint containing multi-level joint structures.

Claims

1. A rapid finite element method for personalized musculoskeletal modeling based on medical images, characterized in that, Includes the following steps: Step 1: Acquire medical images of the subject's target musculoskeletal tissue. Segment the images using an image segmentation network to obtain a subject semantic mask containing multiple anatomical structures within the target musculoskeletal tissue. Then, classify all subject semantic masks according to anatomical category and motor function. A separate anatomical component, in which The integer is positive; simultaneously, the human musculoskeletal tissue template mesh and the corresponding template semantic mask, which are pre-constructed using manual image segmentation and entity modeling methods, are obtained, and the human musculoskeletal tissue template mesh and its corresponding template semantic mask are also divided into... A separate anatomical component; Step 2, based on each individual anatomical component obtained in Step 1, for index ... Anatomical components, among which , will belong to the Affine registration is performed between the template semantic mask of the first anatomical component and the subject semantic mask to calculate the corresponding semantic mask of the first anatomical component. Affine transformation matrix of each anatomical component , so as to belong to the first The template semantic mask of each anatomical component is initially spatially aligned with the corresponding subject semantic mask in terms of pose and scale to obtain the initially spatially aligned template semantic mask. Step 3, based on the result obtained in Step 1, belonging to the first... The subject semantic mask of the first anatomical component, and the template semantic mask obtained in step 2 after initial spatial alignment, are used for elastic registration calculation using a differential homeomorphic deformation network to output the corresponding first anatomical component. Nonlinear deformation field of each anatomical component ; Step 4: Based on the affine transformation matrix calculated in Step 2 and the nonlinear deformation field calculated in step 3 For the data obtained in step 1 belonging to the first... The nodes in the human musculoskeletal tissue template mesh of each anatomical component are sequentially subjected to coordinate mapping deformation to obtain a subject-specific deformed mesh that matches the subject's anatomical structure. Step 5: Based on the subject-specific deformed mesh obtained in Step 4, a hybrid strategy based on a distance threshold is used to correct the position of ligament or tendon attachment points in the subject-specific deformed mesh. All corrected subject-specific deformed meshes are then subjected to collision detection and assembly to generate a subject-specific finite element model containing multi-level joint structures.

2. The rapid finite element modeling method for musculoskeletal personalized models based on medical images according to claim 1, characterized in that, In step 1, the target musculoskeletal tissue of the subject includes multiple anatomical structures such as bone tissue, articular cartilage tissue, intra-articular fibrocartilage tissue, ligament tissue, and tendon tissue; all subject semantic masks are divided into Each independent anatomical component contains parameters It is a positive integer greater than or equal to 2; the independent anatomical components specifically include: bone and cartilage complex, intra-articular fibrocartilage and cartilage complex, ligament complex and tendon complex.

3. The rapid finite element modeling method for musculoskeletal personalized models based on medical images according to claim 2, characterized in that, In step 3, a differential homeomorphic deformation network is used for elastic registration calculation, and the corresponding first step is output. Nonlinear deformation field of each anatomical component The calculation process is as follows: The differential homeomorphic deformation network first predicts and generates a steady-state velocity field. ; Then, the steady-state velocity field is scaled and squared. In time Integrating within the inner region generates a nonlinear deformation field that ensures topological continuity and invertibility. Their relationship satisfies the formula: in, This indicates the exponentiation mapping operator.

4. The rapid finite element modeling method for musculoskeletal personalized models based on medical images according to claim 3, characterized in that, In step 3, the training process of the differential homeomorphic network is based on parameter optimization using a composite loss function, specifically targeting the first... The formula for calculating the composite loss function of each anatomical component is as follows: in, Indicates the first Total loss term for each anatomical component; Indicates the first Subject semantic mask for each anatomical component; Indicates the first Template semantic mask for each anatomical component; Compound operators representing spatial transformations; Indicates passing through a nonlinear deformation field The template semantic mask after deformation; This represents the similarity loss function term used to measure overlap. This represents the regularization hyperparameter used to control the smoothing weights; This represents the smoothness loss function term used to penalize spatial gradients; Similarity loss function term The Dice loss is used, and the calculation formula is as follows: in, Operators that count the number of voxel pixels within a set; The intersection operator for mask sets; Smoothness loss function term The calculation formula is: in, The summation operator for all voxel positions in space; Represents the spatial gradient operator, express Steady-state velocity field of a voxel; The operator represents the operator for calculating the square of the second norm of a vector; Indicates the position of the voxel. Represents the image space.

5. The rapid finite element modeling method for musculoskeletal personalized models based on medical images according to claim 4, characterized in that, In step 4, the calculation formula for the coordinate mapping deformation of the nodes in the human musculoskeletal tissue template mesh is as follows: in, This indicates that the result obtained after deformation belongs to the first... The target coordinates of the nodes of the deformed mesh for each anatomical component; Indicates that before the transformation it belongs to the first Initial coordinates of nodes in a human musculoskeletal tissue template mesh for an anatomical component; Represents nonlinear deformation fields The corresponding nonlinear displacement fields satisfy the following conditions: , where y is the spatial coordinate.

6. The rapid finite element modeling method for musculoskeletal personalized models based on medical images according to claim 5, characterized in that, Step 5 involves using a distance threshold-based hybrid strategy to correct the position of ligament or tendon attachment points in the subject's specific deformed mesh. This process includes the following sub-steps: Step 5.1: For each ligament or tendon endpoint in the deformed mesh specific to the subject, calculate the Euclidean distance from that ligament or tendon endpoint to the nearest bone surface, and preset a distance threshold. ; Step 5.2, based on the Euclidean distance calculated in Step 5.1, make a judgment: when the Euclidean distance is less than or equal to... When the principal component analysis-guided projection is used to determine the location of the ligament or tendon attachment point, the first principal component direction of the ligament or tendon is extracted as the longitudinal axis direction, the endpoint of the ligament or tendon is projected onto the nearest bone surface along the longitudinal axis direction, and the mesh node closest to the projection location is set as the corrected ligament or tendon attachment point. Step 5.3, when the Euclidean distance is greater than At that time, the center of the insertion area of ​​the ligament or tendon is automatically determined as the corrected ligament or tendon attachment point based on the preset anatomical reference benchmark.

7. The rapid finite element modeling method for musculoskeletal personalized models based on medical images according to claim 6, characterized in that, In step 5, collision detection and assembly of all the modified subject-specific deformed meshes are performed, including: detecting the penetration area between cartilage, bone tissue and adjacent soft tissue based on triangular facets, and performing minimum displacement correction on the relevant nodes along the local surface normal so that the penetration depth between tissues does not exceed the preset contact tolerance.

Images