A method and system for vertebra bone cement leakage early warning based on magnetic resonance imaging

Abstract

This invention proposes a method and system for early warning of vertebral bone cement leakage based on magnetic resonance imaging (MRI). The method first acquires MRI image data of the vertebral body, preprocesses the MRI image data to obtain standard image data, then segments the standard image data using Attention U-Net to extract the vertebral bone and soft tissue components, generating a three-dimensional geometric model. Subsequently, based on bone cement injection parameters and the three-dimensional geometric model, a fluid-structure coupled finite element model is constructed for simulation calculation. Finally, the simulation results are input into a multi-branch fusion network model for processing, and the vertebral bone cement leakage result is output. This invention, based on a multi-branch fusion deep network, multi-time finite element simulation, and medical images, achieves a more comprehensive, accurate, and scalable predictive analysis of bone cement leakage, demonstrating significant beneficial effects in clinical surgical planning and risk prevention.

Description

[Technical Field]

[0001] This invention belongs to the field of image processing technology, and in particular relates to a method and system for early warning of vertebral bone cement leakage based on magnetic resonance imaging. [Background Technology]

[0002] Bone cement has become a widely used material in spinal surgery in recent years, primarily for fracture repair, vertebroplasty, and spinal correction, especially in the treatment of vertebral kyphosis and vertebral compression fractures. As a functional material, bone cement's mechanism of action is based on its adhesiveness, mechanical strength, and rapid hardening properties. In vertebroplasty, the injection of cement not only fills the fracture site but also enhances the rigidity of the fracture area, allowing the damaged bone structure to recover. Cement leakage is a common complication in vertebroplasty. When bone cement is injected into the vertebral body, if it fails to fully stabilize in the target area, it may leak into adjacent soft tissues, nerve roots, or even around the spinal cord, leading to clinical symptoms such as pain, nerve compression, or even nerve damage. Therefore, developing a method for early warning of vertebral bone cement leakage based on magnetic resonance imaging is a technical problem that urgently needs to be solved by those skilled in the art. [Summary of the Invention]

[0003] In view of this, embodiments of the present invention provide a method and system for early warning of vertebral bone cement leakage based on magnetic resonance imaging.

[0004] In a first aspect, embodiments of the present invention provide a method for early warning of vertebral bone cement leakage based on magnetic resonance imaging, the method comprising:

[0005] S1. Acquire MRI image data of the vertebral body, and obtain standard image data after preprocessing the MRI image data;

[0006] S2. Use Attention U-Net to segment standard image data to extract vertebral bones and soft tissues, and generate a three-dimensional geometric model;

[0007] S3. Based on bone cement injection parameters and a three-dimensional geometric model, a fluid-structure coupled finite element model is constructed for simulation calculation;

[0008] S4. After processing the simulation results by inputting them into the multi-branch fusion network model, the vertebral body cement leakage results are output.

[0009] As described above and in any possible implementation, a further implementation is provided, wherein S1 includes:

[0010] S11. After denoising and artifact suppression using MRI image data as the source image, normalize the resolution and grayscale.

[0011] S12. Construct the thin-plate spline mapping function:

[0012] Where (x,y) are the coordinates of the point to be transformed in the source image, (x...y...) i ,y i Let be the coordinates of the i-th feature point in the source image, and U(r) = r 2 log(r) is the thin plate spline kernel function, and ||·|| is the Euclidean distance;

[0013] S13. Calculate deformation energy using formulas. Where, d j Let x be the position vector of the corresponding feature point in the target image. j ,y j Let be the coordinates of the j-th feature point in the source image, λ be the regularization coefficient, and E be the coordinates of the j-th feature point. bend (f) represents the bending energy of the deformation function f;

[0014] S14. By minimizing the deformation energy, establish a system of linear equations to solve for a1, a2, a3, and ω. i The mapping function f(x,y) is obtained. All pixel points of the source image are transformed to the target image coordinate system through the thin plate spline mapping function and aligned to obtain standard image data.

[0015] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein S2 specifically includes:

[0016] S21. Extract multi-layer encoder features and decoder upsampling features of standard images using the U-Net architecture, and perform 1×1 convolution on each layer feature to map it to the same dimension.

[0017] S22. After fusing the mapped features, perform non-linear activation, and then calculate the attention map through the Sigmoid layer;

[0018] S23. Pixel-wise weighting is performed on the feature machine of the multilayer encoder based on the attention map, and the weighted encoder features are fused with the sampled features of the corresponding layer of the decoder;

[0019] S24. The fusion result is transmitted to the decoder to restore the spatial resolution of the feature map, and the restored result is mapped into three categories: background, vertebral body and soft tissue.

[0020] S25. The segmentation result is obtained by converting the output of each pixel into a probability distribution belonging to each category through the Softmax activation function;

[0021] S26. The segmentation results are processed by the Marching Cubes algorithm to obtain a mesh. The mesh is then smoothed and repaired to obtain a three-dimensional geometric model.

[0022] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein S3 specifically includes:

[0023] S31. The three-dimensional geometric model of the vertebral body and soft tissue is processed by the Marching Cubes algorithm to obtain a triangular mesh, while retaining the information on bone porosity and cracks; input the bone cement injection parameters and physical parameters, the bone cement injection parameters including: injection pressure, injection rate, initial viscosity and hardening parameters, the physical parameters including: bone elastic modulus, Poisson's ratio, cement density and external force information;

[0024] S32. At the current time step, after solving the solid and fluid problems, the fluid-structure coupled finite element model performs coupled iterations until convergence, and uses the output simulation results as the initial values ​​for the next time step to iteratively simulate the subsequent bone cement injection process.

[0025] S33. Output simulation results of cement flow field and bone mechanical field with multiple time steps.

[0026] In addition to the aspects and any possible implementations described above, an implementation is further provided, wherein S4 specifically includes:

[0027] S41. Obtain the time steps {t1,t2,...,t} at multiple time steps. N The cement flow field and bone biomechanical field data were processed into time-series field data in the same coordinate system. 1 ,Φ 2 ,...,Φ N};

[0028] S42. Use the time-series field data as the first input branch of the multi-branch fusion network model, and extract time-series features using methods such as temporal convolution, temporal graph neural network, or temporal attention.

[0029] S43. The segmented MRI image data is used as the second input branch of the multi-branch fusion network model, and anatomical features are extracted through a convolutional neural network.

[0030] S44. The bone cement injection parameters are used as the third input branch of the multi-branch fusion network model, and the embedded vector features are extracted using a fully connected layer.

[0031] S45. In the intermediate layer of the multi-branch fusion network model, the extracted time series features, anatomical structure features and embedded vector features are fused to construct a spatiotemporal representation vector.

[0032] S46. After the output head of the multi-branch fusion network model performs regression prediction and classification on the spatiotemporal representation vector, the bone cement leakage result is obtained.

[0033] In addition to the aspects and any possible implementations described above, an implementation is further provided, wherein S21 specifically includes:

[0034] The U-Net architecture includes an encoder and decoder at layer L, where the encoder at layer l has the encoder feature F. l e The sampled features of the decoder at the corresponding layer are F l d Perform a 1×1 convolution mapping on the above features: Let φ(F) l e ) = W e F l e ,φ(F l d ) = W d F l d Among them, W e and W d All are learnable convolutional kernels;

[0035] S22 specifically includes:

[0036] For φ(F l e ) and φ(F l d After feature fusion, the intermediate fused feature F is obtained by ReLU activation. l e F l fuse =RELU(φ(F) l e )+φ(F l d (b) The attention map α is obtained by calculating the intermediate fused features through a Sigmoid layer. l =Sigmoid(W f *F l fuse +b f ), where b and b f All are biased;

[0037] S23 specifically includes:

[0038] Based on attention map α l For encoder features F l e Perform pixel-by-pixel weighting: Weighted encoder features Sampling features F corresponding to the layer of the decoder l d After addition and fusion, we get F. l fuse2 ,Right now

[0039] S24 specifically includes:

[0040] The fusion result F l fuse2 The input is fed into the upsampling module corresponding to the l-th layer of the decoder. After bilinear interpolation, it is restored to the same spatial size as the original image layer by layer. Upsampling and merging are performed in reverse from the bottom layer until the original resolution is restored. The output is a feature map that matches the resolution of the input image. The last 1×1 convolution layer is used to map the number of channels to the number of categories. There are three categories, which correspond to the background, vertebral body and soft tissue respectively. The output is the result vector Z(x,y) corresponding to the three categories.

[0041] Specifically, S25 includes:

[0042] Applying Softmax to Z(x,y) along the channel dimension yields the probability distribution for each class:

[0043] c=0,1,2, p0(x,y)+p1(x,y)+p2(x,y)=1; find

[0044] Find the highest probability channel for each pixel and construct a mapping label as the segmentation result;

[0045] S26. Stack the segmentation results to form three-dimensional volume data, and select the target structure corresponding to the category mask; use the Marching Cubes algorithm to traverse the three-dimensional voxels and determine the vertex state of the voxel cubes according to the threshold, perform interpolation and topological lookup on the detected isosurfaces, and combine them to generate a triangular mesh; perform bilateral filtering to update the vertices of the triangular mesh; remove fragments in the mesh based on connected component analysis, fill holes or openings and verify the integrity of the mesh topology; output a smooth and closed three-dimensional geometric model.

[0046] As described above and in any possible implementation, a further implementation is provided, wherein S32 specifically includes:

[0047] S321. Initialize interface pressure;

[0048] S322. After receiving the input data, the fluid-structure coupled finite element model uses the interface pressure obtained in the previous time step as the normal load on the solid surface to assemble the solid finite element equations:

[0049] After solving the solid finite element equations using the nonlinear discretization method, new solid displacements are output, and the stress field within the bone is updated. The solid interface displacements are then transferred to the fluid domain to update the boundary geometry of the mesh.

[0050] S323. Read the new coordinates after the boundary geometry update, reconstruct the fluid mesh, and assemble the fluid equations: A f (v,p)x=b f (u) solves the fluid equation to obtain new fluid velocity and interface pressure, and then passes the interface pressure to the solid equation for the next iteration;

[0051] S324, Calculate the interface residual If ||Δσ|| is lower than the set tolerance, the coupled solution in this time step converges, and the process moves to the next time step, proceeding to S325; if ||Δσ|| is not lower than the set tolerance, the coupled solution in this time step does not converge, and the process returns to S322.

[0052] S325. Use the new fluid velocity, interface pressure and solid displacement obtained from the above steps as the initial values ​​for the next time step, and iteratively simulate the subsequent bone cement injection process.

[0053] Among them, M s Let C be the mass matrix. s Let K be the damping matrix. s Here, u represents the global stiffness matrix, and u represents the solid displacement. and These are the first and second time derivatives of the solid displacement, respectively, F ext Let F be the external force vector. fluid (p) represents the load vector transferred from the fluid pressure to the solid interface, A f (v,p)x represents a discrete fluid system, A f This indicates that the system coefficient matrix depends on the current solution (v, p), where x represents the fluid unknowns, and b... f (u) represents the right-hand vector of the fluid system.

[0054] As described above and in any possible implementation, a further implementation is provided, wherein S45 specifically includes: processing the time series features z time Anatomical structural features img and embedded vector features z param Stacking along the vector dimension forms a comprehensive feature z. fuse =Concat(z time ,z img ,z param ), will z fuse The input fuse layer obtains the spatiotemporal representation vector z through MLP mapping. final =MLPfuse (z fuse ).

[0055] In addition to the aspects and any possible implementations described above, an implementation is further provided, wherein S46 specifically includes:

[0056] z final The input is processed by the FC and ReLU layers of a multi-branch fusion network and then mapped to the target dimension to obtain z. out Using the output head of a multi-branch fusion network to pair z out The leakage type is obtained after performing Softma processing;

[0057] The multi-branch fusion network is trained in the following manner:

[0058] The training samples are input into the model to be trained in the form of three branches: time series branch, image branch, and parameter branch. After processing through LSTM layer, CNN layer, and MLP layer respectively, the time series features z are obtained. time Anatomical structural features img and embedded vector features z param Then, after fusion in the middle layer, z is generated. fuse , z fuse The input fuse layer obtains the spatiotemporal representation vector z through MLP mapping. final , will z final The input is processed by FC and ReLU layers before being mapped to the target dimension to obtain z. out The output head corresponds to z out The leakage type is obtained after performing Softmax processing;

[0059] The loss function is: B represents the total number of training samples, and C represents the total number of categories; y i,c The true label indicates whether sample i belongs to category c; Let L be the probability of the model's predicted output, representing the model's predicted probability that sample i belongs to class c. This is continuously trained until L... CE If the value is below the preset threshold, the final multi-branch fusion network model is obtained.

[0060] In a second aspect, embodiments of the present invention provide a vertebral bone cement leakage early warning system based on magnetic resonance imaging using the method of any one of claims 1-9, the system comprising:

[0061] The preprocessing module is used to acquire MRI image data of the vertebral body and preprocess the MRI image data to obtain standard image data.

[0062] The segmentation module is used to segment standard image data using Attention U-Net to extract vertebral bone and soft tissue parts and generate a three-dimensional geometric model.

[0063] The calculation module is used to construct a fluid-structure coupled finite element model for simulation calculation based on bone cement injection parameters and a three-dimensional geometric model;

[0064] The processing module is used to input the simulation results into the multi-branch fusion network model for processing and output the vertebral body bone cement leakage results.

[0065] One of the above technical solutions has the following beneficial effects:

[0066] This invention proposes a method and system for early warning of vertebral bone cement leakage based on magnetic resonance imaging (MRI). The method first acquires MRI image data of the vertebral body, preprocesses the MRI image data to obtain standard image data, then segments the standard image data using Attention U-Net to extract the vertebral bone and soft tissue components, generating a three-dimensional geometric model. Subsequently, based on bone cement injection parameters and the three-dimensional geometric model, a fluid-structure coupled finite element model is constructed for simulation calculation. Finally, the simulation results are input into a multi-branch fusion network model for processing, outputting the vertebral bone cement leakage result. This invention's method, based on multi-time cement flow field and bone biomechanical field data, models the diffusion of bone cement in the vertebral body and the evolution of bone stress in the time dimension, capturing the dynamic changes of fluid-structure interaction over time. Compared to single-time or static analysis, it can more accurately identify the time periods and characteristics of high-risk leakage during injection. The multi-branch model simultaneously integrates the fluid-structure finite element temporal field, MRI images, and injection parameters, providing richer and more complementary inputs for leakage prediction. In summary, the technical solution of this invention, based on multi-branch fusion deep networks, multi-time finite element simulation, and medical images, achieves a more comprehensive, accurate, and scalable predictive analysis of bone cement leakage, and has significant beneficial effects in clinical surgical planning and risk prevention. [Attached Image Description]

[0067] To more clearly illustrate the technical solutions of the embodiments of the present invention, 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.

[0068] Figure 1 This is a flowchart illustrating a method for early warning of vertebral bone cement leakage based on magnetic resonance imaging, provided in an embodiment of the present invention.

[0069] Figure 2This is a schematic block diagram of a vertebral bone cement leakage early warning system based on magnetic resonance imaging, provided in an embodiment of the present invention.

[0070] Figure 3 This is a schematic diagram of the hardware structure of a vertebral bone cement leakage early warning system based on magnetic resonance imaging, provided in an embodiment of the present invention.

Detailed Implementation Methods

[0071] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0072] Please refer to Figure 1 The figure shows a flowchart of a method for early warning of vertebral bone cement leakage based on magnetic resonance imaging, provided by an embodiment of the present invention. As shown in the figure, the method includes the following steps:

[0073] S1. Acquire MRI image data of the vertebral body, and obtain standard image data after preprocessing the MRI image data;

[0074] S2. Use Attention U-Net to segment standard image data to extract vertebral bones and soft tissues, and generate a three-dimensional geometric model;

[0075] S3. Based on bone cement injection parameters and a three-dimensional geometric model, a fluid-structure coupled finite element model is constructed for simulation calculation;

[0076] S4. After processing the simulation results by inputting them into the multi-branch fusion network model, the vertebral body cement leakage results are output.

[0077] Specifically, S1 includes:

[0078] S11. After denoising and artifact suppression using MRI image data as the source image, normalize the resolution and grayscale.

[0079] S12. Construct the thin-plate spline mapping function:

[0080] Where (x,y) are the coordinates of the point to be transformed in the source image, (x...y...) i ,y i Let be the coordinates of the i-th feature point in the source image, and U(r) = r 2 log(r) is the thin plate spline kernel function, and ||·|| is the Euclidean distance;

[0081] S13. Calculate deformation energy using formulas. Where, d j Let x be the position vector of the corresponding feature point in the target image. j ,y j Let be the coordinates of the j-th feature point in the source image, λ be the regularization coefficient, and E be the coordinates of the j-th feature point. bend (f) represents the bending energy of the deformation function f;

[0082] S14. By minimizing the deformation energy, establish a system of linear equations to solve for a1, a2, a3, and ω. i The mapping function f(x,y) is obtained. All pixel points of the source image are transformed to the target image coordinate system through the thin plate spline mapping function and aligned to obtain standard image data.

[0083] Specifically, S2 includes:

[0084] S21. Extract multi-layer encoder features and decoder upsampling features of standard images using the U-Net architecture, and perform 1×1 convolution on each layer feature to map it to the same dimension.

[0085] S22. After fusing the mapped features, perform non-linear activation, and then calculate the attention map through the Sigmoid layer;

[0086] S23. Pixel-wise weighting is performed on the feature machine of the multilayer encoder based on the attention map, and the weighted encoder features are fused with the sampled features of the corresponding layer of the decoder;

[0087] S24. The fusion result is transmitted to the decoder to restore the spatial resolution of the feature map, and the restored result is mapped into three categories: background, vertebral body and soft tissue.

[0088] S25. The segmentation result is obtained by converting the output of each pixel into a probability distribution belonging to each category through the Softmax activation function;

[0089] S26. The segmentation results are processed by the Marching Cubes algorithm to obtain a mesh. The mesh is then smoothed and repaired to obtain a three-dimensional geometric model.

[0090] Specifically, S3 includes:

[0091] S31. The three-dimensional geometric model of the vertebral body and soft tissue is processed by the Marching Cubes algorithm to obtain a triangular mesh, while retaining the information on bone porosity and cracks; input the bone cement injection parameters and physical parameters, the bone cement injection parameters including: injection pressure, injection rate, initial viscosity and hardening parameters, the physical parameters including: bone elastic modulus, Poisson's ratio, cement density and external force information;

[0092] S32. At the current time step, after solving the solid and fluid problems, the fluid-structure coupled finite element model performs coupled iterations until convergence, and uses the output simulation results as the initial values ​​for the next time step to iteratively simulate the subsequent bone cement injection process.

[0093] S33. Output simulation results of cement flow field and bone mechanical field with multiple time steps.

[0094] Specifically, S4 includes:

[0095] S41. Obtain the time steps {t1,t2,...,t} at multiple time steps. N The cement flow field and bone biomechanical field data were processed into time-series field data in the same coordinate system. 1 ,Φ 2 ,...,Φ N};

[0096] S42. Use the time-series field data as the first input branch of the multi-branch fusion network model, and extract time-series features using methods such as temporal convolution, temporal graph neural network, or temporal attention.

[0097] S43. The segmented MRI image data is used as the second input branch of the multi-branch fusion network model, and anatomical features are extracted through a convolutional neural network.

[0098] S44. The bone cement injection parameters are used as the third input branch of the multi-branch fusion network model, and the embedded vector features are extracted using a fully connected layer.

[0099] S45. In the intermediate layer of the multi-branch fusion network model, the extracted time series features, anatomical structure features and embedded vector features are fused to construct a spatiotemporal representation vector.

[0100] S46. After the output head of the multi-branch fusion network model performs regression prediction and classification on the spatiotemporal representation vector, the bone cement leakage result is obtained.

[0101] Specifically, S21 includes:

[0102] The U-Net architecture includes an encoder and decoder at layer L, where the encoder at layer l has the encoder feature F. l e The sampled features of the decoder at the corresponding layer are F l d Perform a 1×1 convolution mapping on the above features: Let φ(F) l e ) = W e F le ,φ(F l d ) = W d F l d Among them, W e and W d All are learnable convolutional kernels;

[0103] S22 specifically includes:

[0104] For φ(F l e ) and φ(F l d After feature fusion, the intermediate fused feature F is obtained by ReLU activation. l e F l fuse =RELU(φ(F) l e )+φ(F l d (b) The attention map α is obtained by calculating the intermediate fused features through a Sigmoid layer. l =Sigmoid(W f *F l fuse +b f ), where b and b f All are biased;

[0105] S23 specifically includes:

[0106] Based on attention map α l For encoder features F l e Perform pixel-by-pixel weighting: Weighted encoder features Sampling features F corresponding to the layer of the decoder l d After addition and fusion, we get F. l fuse2 ,Right now

[0107] S24 specifically includes:

[0108] The fusion result F l fuse2The input is fed into the upsampling module corresponding to the l-th layer of the decoder. After bilinear interpolation, it is restored to the same spatial size as the original image layer by layer. Upsampling and merging are performed in reverse from the bottom layer until the original resolution is restored. The output is a feature map that matches the resolution of the input image. The last 1×1 convolution layer is used to map the number of channels to the number of categories. There are three categories, which correspond to the background, vertebral body and soft tissue respectively. The output is the result vector Z(x,y) corresponding to the three categories.

[0109] Specifically, S25 includes:

[0110] Applying Softmax to Z(x,y) along the channel dimension yields the probability distribution for each class:

[0111] c=0,1,2, p0(x,y)+p1(x,y)+p2(x,y)=1; find

[0112] Find the highest probability channel for each pixel and construct a mapping label as the segmentation result;

[0113] S26. Stack the segmentation results to form three-dimensional volume data, and select the target structure corresponding to the category mask; use the Marching Cubes algorithm to traverse the three-dimensional voxels and determine the vertex state of the voxel cubes according to the threshold, perform interpolation and topological lookup on the detected isosurfaces, and combine them to generate a triangular mesh; perform bilateral filtering to update the vertices of the triangular mesh; remove fragments in the mesh based on connected component analysis, fill holes or openings and verify the integrity of the mesh topology; output a smooth and closed three-dimensional geometric model.

[0114] Specifically, S32 includes:

[0115] S321. Initialize interface pressure;

[0116] S322. After receiving the input data, the fluid-structure coupled finite element model uses the interface pressure obtained in the previous time step as the normal load on the solid surface to assemble the solid finite element equations:

[0117] After solving the solid finite element equations using the nonlinear discretization method, new solid displacements are output, and the stress field within the bone is updated. The solid interface displacements are then transferred to the fluid domain to update the boundary geometry of the mesh.

[0118] S323. Read the new coordinates after the boundary geometry update, reconstruct the fluid mesh, and assemble the fluid equations: A f (v,p)x=b f (u) solves the fluid equation to obtain new fluid velocity and interface pressure, and then passes the interface pressure to the solid equation for the next iteration;

[0119] S324, Calculate the interface residual If ||Δσ|| is lower than the set tolerance, the coupled solution in this time step converges, and the process moves to the next time step, proceeding to S325; if ||Δσ|| is not lower than the set tolerance, the coupled solution in this time step does not converge, and the process returns to S322.

[0120] S325. Use the new fluid velocity, interface pressure and solid displacement obtained from the above steps as the initial values ​​for the next time step, and iteratively simulate the subsequent bone cement injection process.

[0121] Among them, M s Let C be the mass matrix. s Let K be the damping matrix. s Here, u represents the global stiffness matrix, and u represents the solid displacement. and These are the first and second time derivatives of the solid displacement, respectively, F ext Let F be the external force vector. fluid (p) represents the load vector transferred from the fluid pressure to the solid interface, A f (v,p)x represents a discrete fluid system, A f This indicates that the system coefficient matrix depends on the current solution (v, p), where x represents the fluid unknowns, and b... f (u) represents the right-hand vector of the fluid system.

[0122] Specifically, S45 includes: processing the time series features z time Anatomical structural features img and embedded vector features z param Stacking along the vector dimension forms a comprehensive feature z. fuse =Concat(z time ,z img ,z param ), will z fuse The input fuse layer obtains the spatiotemporal representation vector z through MLP mapping. final =MLP fuse (z fuse ).

[0123] Specifically, S46 includes:

[0124] z final The input is processed by the FC and ReLU layers of a multi-branch fusion network and then mapped to the target dimension to obtain z. out Using the output head of a multi-branch fusion network to pair z out The leakage type is obtained after performing Softma processing;

[0125] The multi-branch fusion network is trained in the following manner:

[0126] The training samples are input into the model to be trained in the form of three branches: time series branch, image branch, and parameter branch. After processing through LSTM layer, CNN layer, and MLP layer respectively, the time series features z are obtained. time Anatomical structural features img and embedded vector features z param Then, after fusion in the middle layer, z is generated. fuse , z fuse The input fuse layer obtains the spatiotemporal representation vector z through MLP mapping. final , will z final The input is processed by FC and ReLU layers before being mapped to the target dimension to obtain z. out The output head corresponds to z out The leakage type is obtained after performing Softmax processing;

[0127] The loss function is: B represents the total number of training samples, and C represents the total number of categories; y i,c The true label indicates whether sample i belongs to category c; Let L be the probability of the model's predicted output, representing the model's predicted probability that sample i belongs to class c. This is continuously trained until L... CE If the value is below the preset threshold, the final multi-branch fusion network model is obtained.

[0128] Through the above steps, the present invention achieves the following technical effects:

[0129] This invention proposes a method and system for early warning of vertebral bone cement leakage based on magnetic resonance imaging (MRI). The method first acquires MRI image data of the vertebral body, preprocesses the MRI image data to obtain standard image data, then segments the standard image data using Attention U-Net to extract the vertebral bone and soft tissue components, generating a three-dimensional geometric model. Subsequently, based on bone cement injection parameters and the three-dimensional geometric model, a fluid-structure coupled finite element model is constructed for simulation calculation. Finally, the simulation results are input into a multi-branch fusion network model for processing, outputting the vertebral bone cement leakage result. This invention's method, based on multi-time cement flow field and bone biomechanical field data, models the diffusion of bone cement in the vertebral body and the evolution of bone stress in the time dimension, capturing the dynamic changes of fluid-structure interaction over time. Compared to single-time or static analysis, it can more accurately identify the time periods and characteristics of high-risk leakage during injection. The multi-branch model simultaneously integrates the fluid-structure finite element temporal field, MRI images, and injection parameters, providing richer and more complementary inputs for leakage prediction. In summary, the technical solution of this invention, based on multi-branch fusion deep networks, multi-time finite element simulation, and medical images, achieves a more comprehensive, accurate, and scalable predictive analysis of bone cement leakage, and has significant beneficial effects in clinical surgical planning and risk prevention.

[0130] The present invention further provides an apparatus embodiment for implementing the steps and methods in the above method embodiments.

[0131] Please refer to Figure 2 , it is Figure 2 This is a schematic block diagram of a vertebral bone cement leakage early warning system based on magnetic resonance imaging, provided in an embodiment of the present invention. The system includes:

[0132] Preprocessing module 210 is used to acquire MRI image data of the vertebral body and preprocess the MRI image data to obtain standard image data;

[0133] The segmentation module 220 is used to segment standard image data using Attention U-Net to extract vertebral bone and soft tissue parts and generate a three-dimensional geometric model;

[0134] Calculation module 230 is used to construct a fluid-structure coupled finite element model for simulation calculation based on bone cement injection parameters and a three-dimensional geometric model;

[0135] The processing module 240 is used to input the simulation results into the multi-branch fusion network model for processing and output the vertebral body bone cement leakage results.

[0136] Because each unit module in this embodiment can execute Figure 1For the methods shown, parts not described in detail in this embodiment, please refer to the [examples / descriptions]. Figure 1 Related explanations.

[0137] Please refer to Figure 3 This is a schematic diagram of the hardware structure of a vertebral bone cement leakage early warning system based on magnetic resonance imaging provided in an embodiment of the present invention. The data prediction device includes at least one processor and a memory. The at least one processor is coupled to the memory and is used to read and execute instructions in the memory to execute the vertebral bone cement leakage early warning method based on magnetic resonance imaging provided in this embodiment of the present invention.

[0138] Thirdly, embodiments of the present invention provide a computer-readable medium. The computer-readable medium stores program code that, when executed on a computer, causes the computer to perform the vertebral bone cement leakage early warning method based on magnetic resonance imaging provided in the embodiments of the present invention.

[0139] At the hardware level, the device may include a processor, and optionally an internal bus, network interface, and memory. The memory may include main memory, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk drive. Of course, the device may also include other hardware required for other business operations.

[0140] The processor, network interface, and memory can be interconnected via an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. This bus can be categorized into address bus, data bus, control bus, etc.

[0141] Memory is used to store programs. Specifically, programs may include program code, which includes computer operation instructions. Memory may include main memory and non-volatile memory, and provides instructions and data to the processor.

[0142] The steps of the method disclosed in the embodiments of this invention can be directly manifested as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0143] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0144] For ease of description, the above apparatus is described by dividing it into various units or modules according to their functions. Of course, in implementing this invention, the functions of each unit or module can be implemented in one or more software and / or hardware.

[0145] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0146] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0147] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0148] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0149] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0150] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0151] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0152] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0153] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0154] This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0155] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0156] The above description is merely an embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A method for early warning of vertebral bone cement leakage based on magnetic resonance imaging, characterized in that, The method includes: S1. Acquire MRI image data of the vertebral body, and obtain standard image data after preprocessing the MRI image data; S2. Use Attention U-Net to segment standard image data to extract vertebral bones and soft tissues, and generate a three-dimensional geometric model; S3. Based on bone cement injection parameters and a three-dimensional geometric model, a fluid-structure coupled finite element model is constructed for simulation calculation; S4. After processing the simulation results by inputting them into the multi-branch fusion network model, the vertebral body cement leakage results are output. S3 specifically includes: S31, processing the three-dimensional geometric model of the vertebral body and soft tissue using the Marching Cubes algorithm to obtain a triangular mesh, while retaining information on intraosseous porosity and cracks; inputting bone cement injection parameters and physical parameters; S32, at the current time step, after performing solid and fluid solutions, the fluid-structure coupled finite element model undergoes coupled iteration until convergence, using the output simulation results as the initial values ​​for the next time step, iteratively simulating the subsequent bone cement injection process; S33, outputting the simulation results of the cement flow field and bone mechanical field across multiple time steps; S4 specifically includes: S41, obtaining data at multiple time steps. The cement flow field and bone biomechanical field data were processed into time-series field data in the same coordinate system. S42. Using the temporal field data as the first input branch of the multi-branch fusion network model, extract time-series features using temporal convolution, temporal graph neural networks, or temporal attention. S43. Using the segmented MRI image data as the second input branch of the multi-branch fusion network model, extract anatomical structure features using a convolutional neural network. S44. Using bone cement injection parameters as the third input branch of the multi-branch fusion network model, extract embedding vector features using a fully connected layer. S45. In the intermediate layer of the multi-branch fusion network model, the extracted time-series features, anatomical structure features, and embedding vector features are fused to construct a spatiotemporal representation vector. S46. The output head of the multi-branch fusion network model performs regression prediction and classification on the spatiotemporal representation vector to obtain the bone cement leakage result. The solid finite element equation is as follows: The fluid equation is: ;in, For the quality matrix, Here is the damping matrix. The global stiffness matrix, For solid displacement, and These are the first and second time derivatives of the solid displacement, respectively. For external force vectors, Let be the load vector transferred from the fluid pressure to the solid interface. Represents a discrete fluid system. This indicates that the system coefficient matrix depends on the current solution. , Represents unknown quantities in the fluid. This represents the right-hand vector of the fluid system.

2. The method for early warning of vertebral bone cement leakage using magnetic resonance imaging according to claim 1, characterized in that, S1 includes: S11. After denoising and artifact suppression using MRI image data as the source image, normalize the resolution and grayscale. S12. Construct the thin-plate spline mapping function: ,in, The coordinates of the point to be transformed in the source image. For the source image, the first Coordinates of feature points For thin plate spline kernel functions, Euclidean distance; S13. Calculate deformation energy using formulas. ,in, This is the position vector of the corresponding feature point in the target image. For the source image, the first Coordinates of feature points The regularization coefficient is . Deformation function The bending energy; S14. Solve the linear equations by minimizing the deformation energy. , , and Solve for the mapping function The standard image data is obtained by transforming all the pixel mappings of the source image to the coordinate system of the target image through a thin-plate spline mapping function and aligning them.

3. The method for early warning of vertebral bone cement leakage based on magnetic resonance imaging according to claim 2, characterized in that, S2 specifically includes: S21. Extract multi-layer encoder features and decoder upsampling features of standard images using the U-Net architecture, and perform processing on each layer of features. Convolution maps to the same dimension; S22. After fusing the mapped features, perform non-linear activation, and then calculate the attention map through the Sigmoid layer; S23. Pixel-wise weighting is performed on the feature machine of the multilayer encoder based on the attention map, and the weighted encoder features are fused with the sampled features of the corresponding layer of the decoder; S24. The fusion result is transmitted to the decoder to restore the spatial resolution of the feature map, and the restored result is mapped into three categories: background, vertebral body and soft tissue. S25. The segmentation result is obtained by converting the output of each pixel into a probability distribution belonging to each category through the Softmax activation function; S26. The segmentation results are processed by the Marching Cubes algorithm to obtain a mesh. The mesh is then smoothed and repaired to obtain a three-dimensional geometric model.

4. The method for early warning of vertebral bone cement leakage based on magnetic resonance imaging according to claim 3, characterized in that, S21 specifically includes: The U-Net architecture includes The encoder and decoder of the layer, the encoder in the first layer The encoder features of the layer are The decoder's sampling features at the corresponding layer are ; to perform the above features Convolution mapping: Let , ,in, and All are learnable convolutional kernels; S22 specifically includes: right and After feature fusion, ReLU activation is used to obtain intermediate fused features. , The attention map is obtained by calculating the intermediate fusion features through a Sigmoid layer. ,in, and All are biased; S23 specifically includes: Based on attention map Encoder features Perform pixel-by-pixel weighting: The weighted encoder features Sampling features of the corresponding layer of the decoder After performing addition and fusion, we get ,Right now ; S24 specifically includes: fusion results Input to decoder The upsampling module corresponding to each layer performs bilinear interpolation, restoring the spatial size to the same level as the original image layer by layer. Upsampling and merging are then performed in reverse from the bottom layer until the original resolution is restored, outputting a feature map matching the input image resolution. The last layer is then used... Convolution maps the number of channels to the number of categories, with three categories corresponding to background, vertebral body, and soft tissue, respectively, and outputs a result vector corresponding to the three categories. ; Specifically, S25 includes: right Performing Softmax along the channel dimension yields the probability distribution for each class: , Find the channel with the highest probability for each pixel and construct a mapping label as the segmentation result; S26. Stack the segmentation results to form three-dimensional volume data, and select the target structure corresponding to the category mask; use the MarchingCubes algorithm to traverse the three-dimensional voxels and determine the vertex state of the voxel cubes according to the threshold, perform interpolation and topological lookup on the detected isosurfaces, and combine them to generate a triangular mesh; perform bilateral filtering to update the vertices of the triangular mesh; remove fragments in the mesh based on connected component analysis, fill holes or openings and verify the integrity of the mesh topology; output a smooth and closed three-dimensional geometric model.

5. The method for early warning of vertebral bone cement leakage based on magnetic resonance imaging according to claim 1, characterized in that, Specifically, S32 includes: S321. Initialize interface pressure; S322. After receiving the input data, the fluid-structure coupled finite element model takes the interface pressure obtained in the previous time step as the normal load on the solid surface, assembles the solid finite element equation, solves the solid finite element equation by nonlinear discretization method, outputs the new solid displacement, updates the stress field in the bone, and transfers the solid interface displacement to the fluid domain to update the boundary geometry of the mesh. S323. Read the new coordinates after the boundary geometry update, reconstruct the fluid mesh, assemble the fluid equations, solve the fluid equations to obtain the new fluid velocity and interface pressure, and pass the interface pressure to the solid equations for the next iteration. S324, Calculate the interface residual ,like If the value is below the set tolerance, the coupled solution for this time step converges, and the process moves to the next time step, proceeding to S325; if... If the set tolerance is not lower than the current time step coupled solution has not converged, return to S322; S325. Use the new fluid velocity, interface pressure, and solid displacement obtained from the above steps as the initial values ​​for the next time step, and iteratively simulate the subsequent bone cement injection process.

6. The method for early warning of vertebral bone cement leakage based on magnetic resonance imaging according to claim 1, characterized in that, S45 specifically includes: processing time series features Anatomical features and embedded vector features Stacking along the vector dimension forms a comprehensive feature. ,Will The input fuse layer obtains the spatiotemporal representation vector through MLP mapping. .

7. The method for early warning of vertebral bone cement leakage based on magnetic resonance imaging according to claim 6, characterized in that, Specifically, S46 includes: Will The input is processed by the FC and ReLU layers of the multi-branch fusion network and then mapped to the target dimension to obtain the result. The output head of the multi-branch fusion network is used to... The leakage type is obtained after performing Softma processing; The multi-branch fusion network is trained in the following manner: The training samples are input into the model to be trained in the form of three branches: time series branch, image branch, and parameter branch. The time series features are obtained after being processed by LSTM layer, CNN layer, and MLP layer, respectively. Anatomical features and embedded vector features Then, after fusion in the middle layer, it is generated. , The input fuse layer obtains the spatiotemporal representation vector through MLP mapping. ,Will The input is processed by FC and ReLU layers before being mapped to the target dimension. By output head The leakage type is obtained after performing Softmax processing; The loss function is: , The total number of training samples, Total number of categories; The true label represents the sample. Does it belong to a category? Instructions; The probability of the model's predicted output represents the model's prediction of the sample. Category The predicted probability is continuously trained until... If the value is below the preset threshold, the final multi-branch fusion network model is obtained.

8. A vertebral bone cement leakage early warning system based on magnetic resonance imaging using the method of any one of claims 1-7, characterized in that, The system includes: The preprocessing module is used to acquire MRI image data of the vertebral body and preprocess the MRI image data to obtain standard image data. The segmentation module is used to segment standard image data using Attention U-Net to extract vertebral bone and soft tissue parts and generate a three-dimensional geometric model. The calculation module is used to construct a fluid-structure coupled finite element model for simulation calculation based on bone cement injection parameters and a three-dimensional geometric model; The processing module is used to input the simulation results into the multi-branch fusion network model for processing and output the vertebral body bone cement leakage results.

Images