A Bionic Design Method of Intervertebral Fusion Cage Based on Lumbar Statistical Shape Model

Abstract

A bionic design method for intervertebral fusion cages based on a lumbar spine statistical shape model, including medical CT data segmentation and preprocessing, automatic construction of a lumbar spine statistical shape model database, classification analysis of the statistical shape model database, and bionic design of intervertebral fusion cages. This method can automatically generate a lumbar spine statistical shape model database and quickly and effectively generate a large number of three-dimensional lumbar spine shape models. By extracting lumbar spine shape features and performing classification analysis and bionic design, intervertebral fusion cages that can meet a larger population range and are more in line with the individual lumbar spine shape are obtained.

Description

Technical Field

[0001] This invention patent belongs to the field of bionic implant device design, and specifically relates to a bionic design method of an intervertebral fusion cage based on a lumbar spine statistical shape model. Background Art

[0002] Low back pain is a common lumbar spine disease, generally caused by lumbar spine degenerative changes and specific occupational habits, which lead to changes in the lumbar spine anatomical structure and mechanical properties.

[0003] Intervertebral fusion is currently the most common and effective way to treat low back pain. In intervertebral fusion, an intervertebral fusion cage is used to fuse vertebrae to maintain the stability of the lumbar spine, reduce the pain of patients with low back pain, and play an important role in supporting bone growth and maintaining the normal physiological height of the lumbar spine.

[0004] Literature research shows that the current development of intervertebral fusion cages mostly focuses on adjusting the porosity and materials of the intervertebral fusion cages to reduce the subsidence amount after implantation, so as to achieve a better fusion effect.

[0005] Clinically, the selection of intervertebral fusion cages is generally made by experienced orthopedic surgeons who select a more appropriate size of intervertebral fusion cage from a limited number of commercial series of intervertebral fusion cages according to the medical images of patients.

[0006] Due to large individual differences, the shapes of currently widely used intervertebral fusion cages in clinics cannot well match the upper and lower endplates and vertebrae of patients, resulting in uneven stress on the upper and lower endplates of the patient's lumbar spine, affecting the fusion effect, and easily leading to problems such as subsidence after fusion surgery.

[0007] In recent years, some researchers have begun to combine 3D printing technology to customize intervertebral fusion cages based on specific patients to meet the fitting of the intervertebral fusion cage shape with the upper and lower endplates of the vertebrae of specific patients. However, the steps of customizing intervertebral fusion cages using this method are relatively cumbersome, with high labor and time costs, so it is difficult to form a large-scale application.

[0008] The statistical shape model is an effective method to describe the shape changes of objects. The statistical shape model is established based on the principle that when there is a large sample of objects, their shape changes follow a Gaussian distribution, and the change law of the object shape can be described by combining the average shape model and the variation mode.

[0009] The emergence of the statistical shape model enables the effective and rapid generation of a large number of object shape models, facilitating the extraction of object shape change characteristics, and providing a new design idea for designing intervertebral fusion cages that better meet a wide range of groups and are more in line with the individual lumbar spine shape. Summary of the Invention

[0010] In view of the technical problem that the current design of intervertebral fusion cages is difficult to meet the needs of a larger population and conform to the shape of individual lumbar vertebrae at the same time, the present invention provides a bionic design method for intervertebral fusion cages based on a lumbar statistical shape model.

[0011] This method can effectively and automatically construct a lumbar statistical shape model database, and perform bionic design of intervertebral fusion cages based on the lumbar statistical shape model database, so as to design bionic intervertebral fusion cages that meet a larger population and are more in line with the shape of individual lumbar vertebrae.

[0012] The specific technical solution of the present invention is as follows:

[0013] As Figure 1 shown, a bionic design method for intervertebral fusion cages based on a lumbar statistical shape model comprises the following steps:

[0014] 1) Data preprocessing: Segment N human lumbar vertebrae from medical CT data, reconstruct the three-dimensional shape model of the lumbar vertebrae, output the STL file, and perform mesh simplification and optimization processing on the mesh of the three-dimensional shape model of the lumbar vertebrae as the lumbar training set;

[0015] 2) Apply relevant algorithms and write programs to automatically construct a lumbar statistical shape model database, and the specific steps are as follows:

[0016] 2.01) Arbitrarily select a three-dimensional shape model of a lumbar vertebra in the training set as the template model, and align the remaining three-dimensional shape models of the lumbar vertebrae to the template model through the iterative closest point algorithm;

[0017] 2.02) This algorithm makes two point sets coincide as much as possible through transformations such as rotation and translation, and reduces the shape error caused by rotation, translation, etc.;

[0018] 2.03) Assume that the point set to be aligned is the template point set is By calculating the objective function whether the result reaches the threshold is used as an index to judge whether to stop the iteration. Among them, M and X are point sets, is a point in the point set, T is the translation matrix, R is the rotation matrix, and N m is the number of points in the point set to be aligned;

[0019] 2.04) In order to unify the three-dimensional shape models of the lumbar vertebrae with different numbers of points into the same number of points and establish the corresponding relationship between points, a non-rigid iterative closest point algorithm is adopted to register the template model to the aligned three-dimensional shape model of the lumbar vertebrae;

[0020] 2.05) On the basis of the rigid transformation, the algorithm adds an affine transformation, so that the template model is deformed onto the lumbar three-dimensional shape model to be registered, completing the registration process, establishing the corresponding relationship between points of the template model and the lumbar three-dimensional shape model to be registered, and obtaining a new lumbar training set;

[0021] 2.06) Perform principal component analysis on the registered lumbar training set:

[0022] 2.07) Calculate the average shape model of the lumbar training set: n represents the number of lumbar three-dimensional shape models in the lumbar training set, and X i represents the lumbar three-dimensional shape model;

[0023] 2.08) Calculate the covariance matrix of the lumbar training set:

[0024] 2.09) Find the eigenvectors and eigenvalues of the covariance matrix Sφ i = λ i φ i , where λ i represents the eigenvalue, and φ i represents the eigenvector. The eigenvector represents the main mode of shape change, the eigenvalue represents the variance above the corresponding principal component, and the larger the eigenvalue, the more data is retained in the direction of its corresponding eigenvector;

[0025] 2.10) The lumbar statistical shape model database can be expressed as t represents the first t eigenvalues after sorting the eigenvalues from large to small. The shape parameter b i is independent and follows a Gaussian distribution of (0, λ i ). To control the range of shape change, generally take

[0026] 2.11) Execute the written automated program to automatically generate the lumbar statistical shape model database;

[0027] 3) Conduct classification analysis on the lumbar statistical shape model database to form sub-databases of lumbar statistical shape models, and calculate the average shape model for each sub-database:

[0028] 3.01) Obtain the median sagittal plane of each shape model in the lumbar statistical shape model database through the mirror method;

[0029] 3.02) In the mirror method, the lumbar three-dimensional shape model needs to be mirrored with any sagittal plane to obtain a new model. The new model and the original model are aligned through the iterative closest point algorithm, and the symmetric plane of the two aligned models is obtained as the median sagittal plane;

[0030] 3.03) Extract the required lumbar anatomical features, i.e., measure the disc height and segmental lordosis angle on the sagittal plane of each three-dimensional lumbar shape model in the lumbar statistical shape model database;

[0031] 3.04) Classify and analyze the lumbar statistical shape model database based on the disc height and segmental lordosis angle, and classify the three-dimensional lumbar shape models with different disc heights and different segmental lordosis angles;

[0032] 3.05) After classification, form multiple sub-databases and calculate the average shape model of the sub-databases, that is As the three-dimensional lumbar shape model for bionic design, j is the number of three-dimensional lumbar shape models in each sub-database, and X i represents the three-dimensional lumbar shape model in the sub-database;

[0033] 4) Conduct bionic design of the intervertebral fusion cage based on the average shape model of each sub-database:

[0034] 4.01) Extract the design surfaces required for design, and extract the upper endplates and lower endplates of two adjacent vertebrae of the average shape model of each sub-database;

[0035] 4.02) Based on the extracted design surfaces, conduct bionic design to form a series of intervertebral fusion cages with bionic shapes.

[0036] The present invention mainly includes four processes: medical CT data segmentation and preprocessing, automatic construction of the lumbar statistical shape model database, classification and analysis of the statistical shape model database, and bionic design of the intervertebral fusion cage.

[0037] The present invention first preprocesses the collected medical CT images, and automatically establishes a lumbar statistical shape model database through processes such as alignment, registration, and principal component analysis;

[0038] Classify and analyze the established lumbar statistical shape models, form multiple sub-databases and calculate the average shape model of the sub-databases;

[0039] Finally, conduct bionic design of the intervertebral fusion cage based on the solved average lumbar shape model of the sub-database.

[0040] Advantages of the present invention:

[0041] 1. Automatically establish a lumbar statistical shape model database, which can automatically, quickly and effectively generate a large number of three-dimensional lumbar shape models, facilitating the extraction of lumbar shape features and effectively saving time and labor costs;

[0042] 2. Based on the lumbar spine statistical shape model, through classification analysis and bionic design, an intervertebral fusion device that can meet a larger group range and is more in line with the individual lumbar spine shape is designed, significantly improving the fitting of the fusion device to the patient's vertebrae, and thus helping to reduce the risks of loosening and subsidence. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flowchart of the present invention;

[0044] Figure 2 is the rendering after alignment by the iterative algorithm for the closest points of two L4-L5 lumbar vertebrae and registration by the non-rigid iterative closest point algorithm;

[0045] Figure 3 is the rendering of the change under the first principal component of the L4-L5 lumbar vertebrae;

[0046] Figure 4 is a schematic diagram for measuring the disc height and segmental lordosis angle on the mid-sagittal plane of the three-dimensional shape model of the L4-L5 lumbar vertebrae;

[0047] Figure 5 is a schematic diagram for designing an anterior interbody fusion device with a bionic shape based on a sub-database. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] The following will further describe in detail the specific embodiments of the present invention with reference to the accompanying drawings:

[0049] In the specific embodiments, taking the L4-L5 vertebrae as an example, a statistical shape model thereof is established, and an anterior interbody fusion device is designed based on this statistical shape model;

[0050] Thirty volunteers' lumbar spine medical CT images are segmented to reconstruct the three-dimensional shape model of the L4-L5 lumbar vertebrae, and an STL format file is output;

[0051] The obtained three-dimensional shape model of the L4-L5 lumbar vertebrae is further processed to simplify and optimize the mesh, so that the number of vertices is kept between 4000 and 5000 to improve the efficiency of subsequent algorithms;

[0052] Through preprocessing, a training set containing 30 three-dimensional shape model samples of the L4-L5 lumbar vertebrae is obtained;

[0053] The three-dimensional shape model samples of the L4-L5 lumbar vertebrae in the training set can be expressed as:

[0054] X i =(x i1 ,y i1 ,z i1 ,…,x in ,y in ,z in );, x in, y in , z in represent the coordinates of points in the sample;

[0055] The training set of L4-L5 lumbar spine can be expressed as: X = (X1, X2, …, X 30 );

[0056] Randomly select an L4-L5 lumbar spine three-dimensional shape model X1 from the training set of L4-L5 lumbar spine three-dimensional shape models as the template model, and align the remaining L4-L5 lumbar spine three-dimensional shape models with the template model through the iterative closest point algorithm;

[0057] This process is to minimize the position error caused by different people and different shooting environments as much as possible;

[0058] The updated training set of the aligned L4-L5 lumbar spine three-dimensional shape model is: X = (X1, X2, …, X 30 );

[0059] After alignment, register the template model X1 to the training set of the aligned L4-L5 lumbar spine three-dimensional shape model through the non-rigid iterative closest point algorithm;

[0060] Through the registration of the non-rigid iterative closest point algorithm, this process obtains a training set of L4-L5 lumbar spine three-dimensional shape models with the same number of points and corresponding points;

[0061] After registration, update the training set of the L4-L5 lumbar spine three-dimensional shape model to: X = (X1, X2, …, X 30 );

[0062] Figure 2 Shows the alignment and registration effect diagram of one of the L4-L5 lumbar spine three-dimensional shape models and the template model;

[0063] Perform principal component analysis on the training set of the L4-L5 lumbar spine three-dimensional shape model:

[0064] Calculate the average shape model of the training set of the L4-L5 lumbar spine three-dimensional shape model:

[0065] Calculate the covariance matrix of the training set of the L4-L5 lumbar spine three-dimensional shape model:

[0066] Calculate the eigenvalues and eigenvectors of the covariance matrix: Sφ i = λ i φ i , λ i represents the eigenvalue of the covariance matrix, arrange the eigenvalues in descending order, φ i represents the eigenvector;

[0067]

[0068] Let \(m\) be the number of variation patterns, \(k\) represent the first \(k\) variation patterns, and \(P\) represent the probability that the first \(k\) variation patterns account for the total variation patterns, with the value \(P = 95\%\).

[0069] In this example, when \(k = 8\), the proportion is \(96.7\%\), which is greater than \(95\%\), so \(k = 8\) is taken.

[0070] The statistical shape model database of the L4 - L5 lumbar vertebrae can be expressed as:

[0071] Run the program for automatically establishing the statistical shape database of the lumbar vertebrae to automatically generate the statistical shape model database of the L4 - L5 lumbar vertebrae.

[0072] Figure 3 It is the effect diagram of variation under the first principal component of the L4 - L5 lumbar vertebrae.

[0073] Classify and analyze the obtained statistical shape model database of the L4 - L5 lumbar vertebrae.

[0074] This process involves measuring the disc height and the segmental lordosis angle on the mid - sagittal plane of the three - dimensional shape model of the L4 - L5 lumbar vertebrae.

[0075] Obtain the mid - sagittal plane of the three - dimensional shape model of the L4 - L5 lumbar vertebrae through the mirror method.

[0076] In the mirror method, the three - dimensional shape model of the L4 - L5 lumbar vertebrae needs to be mirrored with respect to an arbitrary sagittal plane to obtain a new model. The new model and the original model are aligned through the iterative closest point algorithm; the two aligned models are regarded as a whole, and its symmetry plane, that is, the mid - sagittal plane, is obtained.

[0077] Extract the required anatomical features of the L4 - L5 lumbar vertebrae, that is, measure the disc height and the segmental lordosis angle on the mid - sagittal plane of the three - dimensional shape model of the L4 - L5 lumbar vertebrae, and classify and analyze the measured statistical shape model database of the L4 - L5 lumbar vertebrae according to the segmental lordosis angle and the disc height size to obtain different sub - databases.

[0078] The difference between different sub - databases lies in the different segmental lordosis angles and disc heights, and calculate the average shape model of the sub - databases.

[0079] Figure 4 It is a schematic diagram for measuring the disc height and the segmental lordosis angle on the mid - sagittal plane of the three - dimensional shape model of the L4 - L5 lumbar vertebrae.

[0080] Based on the average shape model of the sub-module database obtained, a bionic design of the intervertebral fusion cage is carried out. The lower endplate surface of the L4 vertebra and the upper endplate surface of the L5 vertebra are extracted as its design surfaces;

[0081] Based on these design surfaces, a bionic design of the intervertebral fusion cage is carried out to design an anterior intervertebral fusion cage that can meet a larger group range and is more in line with the individual lumbar spine shape.

[0082] Figure 5 It is an anterior intervertebral fusion cage with a bionic shape designed based on the average shape model of one of the sub-databases.

[0083] In summary, the present invention can automatically establish a lumbar spine statistical shape model, and through classification analysis, carry out a bionic design of the intervertebral fusion cage based on the lumbar spine statistical shape model, and illustrate the principle and effect of the present invention through specific examples.

Claims

1. A bionic design method of an intervertebral fusion cage based on a lumbar spine statistical shape model, characterized in that: The steps are as follows: 1) Data preprocessing: Segment N human lumbar vertebrae from medical CT data, reconstruct the three-dimensional shape model of the lumbar vertebrae, output an STL file, and perform mesh simplification and optimization on the mesh of the three-dimensional shape model of the lumbar vertebrae as the lumbar training set; 2) Apply relevant algorithms and write programs to automatically construct a lumbar statistical shape model database; 3) Classify and analyze the lumbar statistical shape model database to form sub-databases of lumbar statistical shape models, and calculate the average shape model for each sub-database; 4) Conduct bionic design of intervertebral fusion devices based on the average shape model of each sub-database; The steps of the classification and analysis of the lumbar statistical shape model database in 3) to form sub-databases of lumbar statistical shape models and calculate the average shape model for each sub-database are as follows: 3.01) Obtain the median sagittal plane of each shape model in the lumbar statistical shape model database through the mirror method; in the mirror method, the three-dimensional shape model of the lumbar vertebrae needs to be mirrored with any sagittal plane to obtain a new model. The new model and the original model are aligned through the iterative closest point algorithm, and the two aligned models are regarded as a whole to obtain their symmetric plane, that is, the median sagittal plane; 3.02) Extract the required lumbar anatomical features, that is, measure the disc height and segmental lordosis angle on the sagittal plane of each three-dimensional shape model of the lumbar vertebrae in the lumbar statistical shape model database; 3.03) Classify and analyze the lumbar statistical shape model database based on the disc height and segmental lordosis angle, and classify the three-dimensional shape models of the lumbar vertebrae with different disc heights and different segmental lordosis angles; 3.04) After classification and analysis, multiple sub-databases are formed, and the average shape model of the sub-databases is obtained, that is As the three-dimensional lumbar shape model for bionic design, j is the number of three-dimensional lumbar shape models in each sub-database, and X i represents the three-dimensional lumbar shape model in the sub-database.

2. The bionic design method of an interbody fusion cage based on a lumbar statistical shape model according to claim 1, wherein: The steps of the automatic construction of the lumbar statistical shape model database in step 2) are as follows: 2.01) Arbitrarily select a three-dimensional shape model of a lumbar vertebra in the training set as the template model, and align the remaining three-dimensional shape models of the lumbar vertebrae to the template model through the iterative closest point algorithm; 2.02) Through rotation and translation transformations, make the two point sets coincide as much as possible to reduce the shape error caused by rotation and translation; 2.03) Assume that the point set to be aligned is and the template point set is By calculating the objective function Whether the result reaches the threshold is used as an indicator to judge whether to stop iteration. Among them, M and X are point sets, is a point in the point set, T is the translation matrix, R is the rotation matrix, and N m is the number of points in the template point set; 2.04) In order to unify the three-dimensional shape models of the lumbar vertebrae with different numbers of points into the same number of points and establish the corresponding relationship between points, adopt the non-rigid iterative closest point algorithm to register the template model to the aligned three-dimensional shape models of the lumbar vertebrae; 2.05) Add an affine transformation on the basis of the rigid transformation, so that the template model deforms onto the three-dimensional shape model of the lumbar vertebra to be registered, complete the registration process, establish the corresponding relationship between points of the template model and the three-dimensional shape model of the lumbar vertebra to be registered, and obtain a new lumbar training set; 2.06) Perform principal component analysis on the registered lumbar training set: 2.07) Calculate the average shape model of the lumbar spine training set: n represents the number of three-dimensional shape models of the lumbar spine in the lumbar spine training set, and X i represents the three-dimensional shape model of the lumbar spine; 2.08) Calculate the covariance matrix of the lumbar spine training set: 2.09) Calculate the eigenvectors and eigenvalues of the covariance matrix Sφ i = λ i φ i , where λ i represents the eigenvalue, and φ i represents the eigenvector. The eigenvector represents the main mode of shape change, and the eigenvalue represents the variance above the principal component. The larger the eigenvalue, the greater the amount of data retained in the direction of its corresponding eigenvector; 2.10) The lumbar statistical shape model database can be represented as t represents the sorting of eigenvalues from largest to smallest, the first t eigenvalues, and the shape parameter b i Independent and obeying the Gaussian distribution of (0, λ i ), to control the range of shape changes, take 2.11) Execute the written automatic program to automatically generate a lumbar statistical shape model database.

3. A biomimetic design method for an interbody fusion device based on a lumbar statistical shape model according to claim 1, characterized in that: The steps of the bionic design of intervertebral fusion devices based on the average shape model of each sub-database in 4) are as follows: 4.01) Extract the design surfaces required for the design, and extract the upper endplate surface and the lower endplate surface of two adjacent vertebrae of the average shape model of each sub-database; 4.02) Based on the extracted design surfaces, conduct bionic design to form a series of intervertebral fusion devices with bionic shapes.

Images