A modeling method for a skeletal statistical model

Abstract

The present invention provides a modeling method for a skeletal statistical model, characterized in that it comprises the following steps: step S1, collecting multiple skeletal images and bone quality data. Step S2, using image reconstruction technology to perform grid processing on multiple skeletal images to obtain multiple skeletal network models. Step S3, sorting and classifying multiple skeletal network models to obtain skeletal template models and skeletal sample models, and then aligning and gridding to obtain skeletal template model grids. Step S4, performing deformation registration on the skeletal sample model and the skeletal template model grid to obtain a deformed registered surface grid and a deformed registered volume grid. Step S5, assigning bone quality data to grid nodes on the deformed registered surface grid and the deformed registered volume grid to obtain a training sample model. Step S6, performing principal component analysis on the training sample model using a principal component analysis method to obtain a skeletal statistical model.

Description

Technical Field

[0001] The invention relates to a modeling method of a skeleton statistical model. Background Art

[0002] At present, some orthopedic implant devices, such as bone plates, intramedullary nails, distractors, acetabular cups, etc., complete parameter design based on a single or a limited number of bone model cases. A few typical cases do have more direct clinical significance, but because it is impossible to effectively evaluate whether the device is suitable for a specific population, the device design cannot be promoted.

[0003] In the current finite element analysis and optimization design of bone models, the bone model mesh material distribution often uses an eggshell model to simulate the real bone mass distribution, that is, the outer mesh is cortical bone, which has a large bone density and strength value, and the inner mesh is cancellous bone, which has a small bone density and strength value. This method is somewhat different from the real bone mass distribution.

[0004] Patent CN104200524A discloses a three-dimensional mesh bone model averaging method for bone plate design, and proposes that the bone plate completed on the average model has the characteristic of being more closely fitting the fracture site. This method has the following problems: the average model is a model obtained by averaging all sample bone models after registration, and it is a single fixed model. Using a single fixed model to adapt to the fracture site where the anatomical structure is prone to variation still results in poor fitting accuracy. Summary of the invention

[0005] In order to solve the above problems, a modeling method of a skeleton statistical model is provided, and the present invention adopts the following technical solutions:

[0006] The present invention provides a modeling method of a bone statistical model, which is used for meshing and deforming a bone image to obtain a bone statistical model X, and is characterized in that it includes the following steps: step S1, collecting multiple bone images and bone quality data. Step S2, using image reconstruction technology to mesh multiple bone images to obtain multiple bone network models. Step S3, sorting and classifying multiple bone network models to obtain a bone template model and a bone sample model, and then aligning and meshing to obtain a bone template model grid. Step S4, deforming and registering the bone sample model and the bone template model grid to obtain a deformed registered surface grid and a deformed registered volume grid corresponding to each bone sample model. Step S5, assigning bone quality data to the grid nodes on the deformed registered surface grid and the deformed registered volume grid to obtain a training sample model. Step S6, using the principal component analysis method to perform principal component analysis on the training sample model to obtain the bone statistical model X. Among them, the network nodes on the skeletal network model have a geometric topological connection relationship. The skeletal network model includes surface meshes and volume meshes. The bone quality data includes bone density, bone strength, and Poisson's ratio. The specific expression of the skeletal statistical model X is: In the formula, is the average model of all bone sample models after mesh deformation and registration, ξ i is the direction of the i-th principal component, c i is the weight deformation coefficient corresponding to the direction of the i-th principal component, -3σ i ≤c i ≤3σ i , σ i is the standard deviation.

[0007] The modeling method of a skeleton statistical model provided by the present invention can also have such technical features, wherein step S3 comprises the following steps: step S3-1, sorting a plurality of skeleton network models according to the bone curvature radius, taking the skeleton network model with the median of the sorting as the skeleton template model, and the remaining skeleton network models as the skeleton sample models. Step S3-2, aligning all the skeleton sample models with the skeleton template model using the nearest point iterative rigid registration algorithm. Step S3-3, dividing the skeleton template model into a uniform grid size according to the geometric topological connection relationship to obtain a skeleton template model grid.

[0008] A modeling method for a skeleton statistical model provided by the present invention may also have such technical features, wherein step S4 comprises the following steps: step S4-1, using a mid-plane mesh deformation algorithm to obtain a mid-plane mesh between the skeleton template model mesh and the face mesh of each skeleton sample model. Step S4-2, using a single K nearest neighbor algorithm, searching for the nearest mesh node of the corresponding mid-plane mesh for each network node on the skeleton template model mesh. Step S4-3, using a feature vector amplification algorithm, obtaining a deformed and registered face mesh. Step S4-4, using the face mesh deformation registration result as an external constraint for the volume mesh deformation of the skeleton template model mesh, and using a volume mesh deformation algorithm to obtain a deformed and registered volume mesh.

[0009] The modeling method of a bone statistical model provided by the present invention may also have such a technical feature, wherein the principal component analysis method in step S6 includes the following steps: step S6-1, constructing a training matrix TD according to the training sample model. Step S6-2, performing standardization processing on the training matrix TD to obtain a standardized training matrix. Step S6-3, performing principal component analysis on the standardized training matrix to obtain an average model m principal component directions ξ i and standard deviation σ i Step S6-4, obtain the skeleton statistical model X according to the linear combination, and the specific expression of the training matrix TD is: TD = [X1, X2, ..., X i …X N ] T ∈R N×4n , X i ={x i,1 ,x i,2 ,…x i,j ,…x i,n ,y i,1 ,y i,2 ,…y i,j ,…y i,n ,z i,1 ,z i,2 ,…z i,j ,…z i,n ,ρ i,1 ,ρ i,2 ,…ρ i,j ,…ρ i,n ,E i,1 ,E i,2 ,…E i,j ,…E i,n} 1×4n , where N is the number of training samples, n is the number of grid nodes for each training sample, R is the latitude symbol of the training matrix TD, and X iis the i-th row of the TD matrix, indicating the training sample model corresponding to the i-th skeleton sample model, (x i,j ,y i,j ,z i,j ) is the coordinate of the jth grid node on the i-th bone sample model, ρ i,j is the bone density corresponding to the jth grid node on the i-th bone sample model, E i,j is the corresponding bone strength of the jth mesh node on the i-th bone sample model.

[0010] Function and Effect of the Invention

[0011] According to a modeling method of a bone statistical model of the present invention, machine learning is first performed from a large number of bone images, and the external geometric shape of bones and the internal bone density, bone strength distribution and other characteristics of the same population are used to obtain a bone statistical model containing grid space coordinates and bone density and bone strength distribution. This eliminates the need to simulate or simplify the bone model material, saves grid material allocation work, and improves the accuracy of the finite element analysis model.

[0012] Secondly, the bone statistical model in the present invention performs surface mesh deformation registration and volume mesh deformation registration through certain constrained deformation, which fits the actual bone anatomical structure of any patient to the greatest extent, and can obtain the patient's bone model without the need for tedious modeling of the patient's bones. The bone statistical model is applicable to fracture site reconstruction, bone defect repair, optimized design of personalized implant devices in direct contact with bones, and rehabilitation prostheses and protective gear design that indirectly interact with bones. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a flow chart of a modeling method of a skeletal statistical model in an embodiment of the present invention;

[0014] Figure 2 is a schematic diagram of a modeling method of a skeleton statistical model in an embodiment of the present invention;

[0015] Figure 3 is a schematic diagram of the alignment result between the skeleton sample model and the skeleton template model in an embodiment of the present invention;

[0016] Figure 4 is a schematic diagram of the result of redrawing and dividing the mesh of the skeleton template model in an embodiment of the present invention;

[0017] Figure 5 is a schematic diagram of the position of a plane mesh in a skeleton in an embodiment of the present invention within a cutting plane;

[0018] Figure 6 is a schematic diagram of the result of fitting a skeleton sample model with a deformed surface mesh in an embodiment of the present invention;

[0019] Figure 7 is a schematic diagram comparing the volume mesh of the skeleton template model before and after deformation in an embodiment of the present invention;

[0020] Figure 8 This is a schematic diagram of the distribution of the anatomical structure abnormality of the proximal left humerus of a male affected by the first principal component of the skeleton template model in an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, a modeling method of a bone statistical model of the present invention is specifically described below in conjunction with embodiments and drawings.

[0022] <Example>

[0023] Figure 1 is a flow chart of a modeling method of a skeletal statistical model in an embodiment of the present invention; Figure 2 It is a schematic diagram of a modeling method of a skeletal statistical model in an embodiment of the present invention.

[0024] like Figure 1 and Figure 2 As shown, this embodiment provides a modeling method for a skeletal statistical model, comprising the following steps:

[0025] Step S1, collecting multiple bone images and bone quality data.

[0026] In this embodiment, skeletal images of the proximal left humerus of 9 males were collected.

[0027] Step S2, using image reconstruction technology to perform mesh processing on 9 bone images to obtain 9 bone network models. 9 male left humerus proximal bone models are used as training samples of the statistical model. The input sample size is not limited to 9, and is at least 5 bone model samples. When the input sample size is too small, the statistical model obtained by training is not statistically significant.

[0028] Step S3, sorting and classifying the multiple skeleton network models to obtain skeleton template models and skeleton sample models, and then aligning and meshing them to obtain skeleton template model meshes.

[0029] Step S3 includes the following steps:

[0030] Step S3-1, sorting multiple skeleton network models according to the bone curvature radius, taking the skeleton network model with the median in the sorting as the skeleton template model, and the remaining skeleton network models as the skeleton sample models.

[0031] The selection of the bone template model among all the above-mentioned proximal humerus bone models is not limited to the method of sorting the anatomical measurement values ​​shown in the embodiment. Technicians in this field may also adopt other forms of bone template model selection methods, including arbitrarily specifying a bone model as the bone template model.

[0032] Step S3-2, using the closest point iterative rigid registration algorithm, align all the skeleton sample models with the skeleton template model.

[0033] Figure 3 It is a schematic diagram of the alignment result between the skeleton sample model and the skeleton template model in an embodiment of the present invention.

[0034] like Figure 3 As shown, all bone sample models are aligned with the bone template model.

[0035] Step S3-3, dividing the skeleton template model into grids of uniform size according to the geometric topological connection relationship to obtain a skeleton template model grid.

[0036] Figure 4 It is a schematic diagram of the result of redrawing and dividing the mesh of the skeleton template model in an embodiment of the present invention.

[0037] like Figure 4 As shown, in this embodiment, the skeleton template model is divided into triangular tetrahedral meshes of uniform size, and the meshes include surface meshes and volume meshes. The mesh size is selected to be 1mm-1.5mm.

[0038] The triangular tetrahedral mesh used in this embodiment is used to divide the skeleton template mesh into uniform sizes. Other meshes with certain geometric topological connection relationships, such as quadrilateral hexahedral meshes, may also be used.

[0039] Step S4, performing deformation registration on the skeleton sample model and the skeleton template model meshes to obtain a deformation registered surface mesh and a deformation registered volume mesh corresponding to each skeleton sample model.

[0040] Step S4 includes the following steps:

[0041] Step S4-1, using a mid-plane mesh deformation algorithm, obtain a mid-plane mesh between the skeleton template model mesh and the surface mesh of each skeleton sample model.

[0042] Figure 5 It is a schematic diagram of the position of the plane mesh in the skeleton within the cutting plane in an embodiment of the present invention.

[0043] like Figure 5 As shown, the midplane mesh is between the skeleton template model mesh and the face mesh of each skeleton sample model.

[0044] Step S4-2, using a single K-nearest neighbor algorithm, searches for the nearest grid node of the corresponding midplane grid for each network node on the skeleton template model grid.

[0045] Step S4-3, using a feature vector magnification algorithm to obtain a deformable registered surface mesh.

[0046] Figure 6 It is a schematic diagram of the result of fitting a skeleton sample model with a deformed registered surface mesh in an embodiment of the present invention.

[0047] like Figure 6 As shown, the skeleton sample model and the deformed registered surface mesh fit each other.

[0048] Step S4-4, using the surface mesh deformation registration result as an external constraint for the volume mesh deformation of the skeleton template model mesh, and using a volume mesh deformation algorithm to obtain a deformed and registered volume mesh.

[0049] Figure 7 3 is a schematic diagram comparing the volume mesh of the skeleton template model before and after deformation in an embodiment of the present invention.

[0050] like Figure 7 As shown, the surface mesh deformation registration result is an external constraint for the volume mesh deformation registration.

[0051] Step S5, assigning bone mass data to the mesh nodes on the deformable registered surface mesh and the deformable registered volume mesh to obtain a training sample model.

[0052] Step S6, using principal component analysis to perform principal component analysis on the training sample model to obtain a skeleton statistical model X.

[0053] Step S6 includes the following steps:

[0054] Step S6-1, construct a training matrix TD according to the training sample model.

[0055] The specific expression of the training matrix TD is:

[0056] TD=[X1,X2,…,X i …X N ] T ∈R N×4n ,

[0057] X i ={x i,1 ,x i,2 ,…x i,j ,…x i,n ,y i,1 ,y i,2 ,…y i,j ,…y i,n ,zi,1 ,z i,2 ,…z i,j ,…z i,n ,

[0058] ρ i,1 ,ρ i,2 ,…ρ i,j ,…ρ i,n ,E i,1 ,E i,2 ,…E i,j ,…E i,n} 1×4n ,

[0059] Where N is the number of training samples, n is the number of grid nodes for each training sample, R is the latitude symbol of the training matrix TD, and X i is the i-th row of the TD matrix, indicating the training sample model corresponding to the i-th skeleton sample model, (x i,j ,y i,j ,z i,j ) is the coordinate of the jth grid node on the i-th bone sample model, ρ i,j is the bone density corresponding to the jth grid node on the i-th bone sample model, E i,j is the corresponding bone strength of the jth mesh node on the i-th bone sample model.

[0060] Step S6-2, normalize the training matrix TD to obtain a standardized training matrix.

[0061] Before the principal component analysis is performed on the training matrix TD in this embodiment, a standard scaling is first performed on each column of the TD matrix so that the mean value of each corresponding feature column is 0 and the variance is 1.

[0062] Step S6-3, perform principal component analysis on the standardized training matrix to obtain the average model m principal component directions ξ i and standard deviation σ i .

[0063] In this embodiment, the principal component analysis is performed on the TD matrix, and the cumulative variance of the principal component direction is selected to be ≥ 95%, and the average model of the statistical model and the seven principal component directions ξ i and standard deviation σ i The size, standard deviation and standard deviation ratio of the seven principal components are shown in Table 1. The variance ratio corresponding to each principal component direction means how much training sample information is retained. The larger the variance ratio, the more original training sample information is retained. The seven principal component directions cumulatively retain 100% of the training sample information.

[0064] Table 1 Standard deviation and proportion of principal component directions of the statistical model of male left humerus

[0065]

[0066] Step S6-4, obtaining the skeleton statistical model X according to the linear combination.

[0067] According to the average model 7 principal component directions ξ i and standard deviation σ i The three are linearly combined to obtain the bone statistical model X, which is specifically expressed as:

[0068]

[0069] In the formula, is the average model of all bone sample models after mesh deformation and registration, ξ i is the direction of the i-th principal component, c i is the weight deformation coefficient corresponding to the direction of the i-th principal component, -3σ i ≤c i ≤3σ i , σ i is the standard deviation.

[0070] In this embodiment, there are 7 principal component directions, but the number of principal component directions is related to the sample size of the skeleton statistical model training and the internal similarity and correlation characteristics of the skeleton sample model. The corresponding standard deviation value also changes accordingly.

[0071] For a specific population of bones, the anatomical variation of the bones conforms to the normal distribution, within [-3σ i ,3σ i The probability density in the interval is 0.9974. The direction of the principal component is in [-3σ i ,3σ i ] significantly affects the distribution of anatomical structural variability of the proximal left humerus in men. Each principal component can independently affect the changes in the statistical model, and the change trends are different.

[0072] Figure 8 This is a schematic diagram of the distribution of the anatomical structure abnormality of the proximal left humerus of a male affected by the first principal component of the skeleton template model in an embodiment of the present invention.

[0073] like Figure 8 As shown, taking the first principal component ξ1 as an example, the first principal component ξ1 affects the scaling deformation of the humeral head. In addition, the second principal component ξ2 affects the relative position deflection of the humeral shaft and the humeral head, the third principal component ξ3 affects the regional changes of the large and small tubercles, the fourth principal component ξ4 affects the radial scaling of the humeral shaft, and the fifth to seventh principal components ξ5, ξ6, and ξ7 can affect the changes in certain trends of the statistical model.

[0074] This embodiment adopts the principal component analysis method to build the statistical model, and other methods such as the least squares method may also be used to obtain results similar to those of this embodiment.

[0075] This embodiment provides a method for modeling a skeleton statistical model with bone quality information, and a method without bone quality information, i.e., a statistical shape model, can also be selected according to actual needs. In the process of training the model, this embodiment can ignore the link of bone deformation mesh material allocation, and finally train to obtain a statistical shape model; it can also include the link of bone deformation mesh material allocation in the process of training the model, and finally train to obtain a statistical shape and strength model.

[0076] Example Function and Effect

[0077] According to a modeling method of a bone statistical model provided in this embodiment, machine learning is first performed from a large number of bone images, and the external geometric shape of bones and internal bone density, bone strength distribution and other characteristics of the same population are used to build a bone statistical model. By training bones in the bone statistical model, the obtained bone statistical model contains grid space coordinates and bone density, bone strength distribution, and there is no need to simulate or simplify the bone model material, which can save grid material allocation work and improve the accuracy of the finite element analysis model.

[0078] Secondly, the bone statistical model in this embodiment performs surface mesh deformation registration and volume mesh deformation registration through certain constrained deformation, which fits the actual bone anatomical structure of any patient to the greatest extent, and can obtain the patient's bone model without tedious modeling of the patient's bones. The bone statistical model is applicable to fracture reconstruction, bone defect repair, optimization design of personalized implant devices in direct contact with bones, and rehabilitation prosthesis and protective gear design that indirectly interact with bones.

[0079] The above embodiments are only used to illustrate specific implementation modes of the present invention, and the present invention is not limited to the description scope of the above embodiments.

Claims

1. A method for modeling a skeleton statistical model, for performing meshing and deformation registration on a skeleton image to obtain a skeleton statistical model X, characterized in that: The following steps are involved: Step S1, collecting multiple bone images and bone quality data; Step S2, using image reconstruction technology to perform grid processing on the plurality of bone images to obtain a plurality of bone network models; Step S3, sorting and classifying the plurality of skeleton network models to obtain skeleton template models and skeleton sample models, and then aligning and meshing them to obtain skeleton template model meshes; Step S4, performing deformation registration on the skeleton sample model and the skeleton template model mesh to obtain a deformation registered surface mesh and a deformation registered volume mesh corresponding to each skeleton sample model; Step S5, assigning the bone mass data to the grid nodes on the deformable registered surface grid and the deformable registered volume grid to obtain a training sample model; Step S6, using principal component analysis to perform principal component analysis on the training sample model to obtain the bone statistical model X, The network nodes on the skeleton network model have a geometric topological connection relationship. The skeleton network model includes a surface mesh and a volume mesh. The bone quality data include bone density, bone strength, and Poisson's ratio. The specific expression of the bone statistical model X is: In the formula, is the average model of all bone sample models after mesh deformation and registration, ξ i is the direction of the i-th principal component, c i is the weight deformation coefficient corresponding to the i-th principal component direction, -3σ i ≤c i ≤3σ i , σ i is the standard deviation.

2. A method for modeling a skeletal statistical model according to claim 1, Features: Wherein, the step S3 comprises the following steps: Step S3-1, sorting the plurality of skeleton network models according to the bone curvature radius, taking the skeleton network model with the median of the sorting as the skeleton template model, and the remaining skeleton network models as the skeleton sample models; Step S3-2, aligning all the bone sample models with the bone template model using a closest point iterative rigid registration algorithm; Step S3-3, dividing the skeleton template model into grids of uniform size according to the geometric topological connection relationship to obtain the skeleton template model grid.

3. A method for modeling a skeletal statistical model according to claim 1, Features: Wherein, the step S4 comprises the following steps: Step S4-1, using a mid-plane mesh deformation algorithm to obtain a mid-plane mesh between the skeleton template model mesh and the surface mesh of each of the skeleton sample models; Step S4-2, using a single K nearest neighbor algorithm, searching for the nearest grid node corresponding to the midplane grid for each network node on the skeleton template model grid; Step S4-3, using a feature vector magnification algorithm to obtain a deformable registered surface mesh; Step S4-4, using the surface mesh deformation registration result as an external constraint for the volume mesh deformation of the skeleton template model mesh, and using a volume mesh deformation algorithm to obtain a deformed and registered volume mesh.

4. A method for modeling a skeletal statistical model according to claim 1, Features: The principal component analysis method in step S6 includes the following steps: Step S6-1, constructing a training matrix TD according to the training sample model; Step S6-2, normalizing the training matrix TD to obtain a standardized training matrix; Step S6-3, performing principal component analysis on the standardized training matrix to obtain an average model m principal component directions ξ i and standard deviation σ i ; Step S6-4, obtaining the skeleton statistical model X according to the linear combination, The specific expression of the training matrix TD is: TD=[X1,X2,…,X i …X N ] T ∈R N×4n , X i {x i,1 ,x i,2 ,…x i,j ,…x i,n ,y i,1 ,y i,2 ,…y i,j ,…y i,n ,z i,1 ,z i,2 ,...z i,j ,...z i,n ,ρ i,1 ,ρ i,2 ,…ρ i,j ,…ρ i,n ,E i,1 ,E i,2 ,…E i,j ,…E i,n } 1×4n , Where N is the number of training samples, n is the number of grid nodes for each training sample, R is the latitude symbol of the training matrix TD, and X i is the i-th row of the TD matrix, indicating the training sample model corresponding to the i-th skeleton sample model, (x i,j ,y i,j ,z i,j ) is the coordinate of the jth grid node on the i-th bone sample model, ρ i,j is the bone density corresponding to the jth grid node on the i-th bone sample model, E i,j is the corresponding bone strength of the j-th mesh node on the i-th bone sample model.

Images