Depth variation reasoning-combined edentulous jaw dental implant positioning method

Abstract

The invention discloses an edentulous jaw dental implant positioning method combined with depth variation reasoning. The method comprises the following steps: S1, generating optimized medical image data; s2, generating a jaw bone three-dimensional reconstruction model of the edentulous jaw patient; s3, generating anatomical feature data; s4, generating a preliminary implant positioning scheme; s5, generating a local bone reconstruction simulation result; s6, performing dynamic interaction on a local bone reconstruction simulation result and the preliminary implant positioning scheme, optimizing the preliminary implant positioning scheme through a depth variation inference model, and generating an implant optimized positioning scheme including mechanical stability and biomechanical performance evaluation; and S7, based on the implant optimization positioning scheme, generating implant guide plate design data and surgical navigation data in combination with the jaw bone three-dimensional reconstruction model of the patient. According to the method, the final implant positioning point obtains a globally optimal solution in the aspects of long-term stability and biological adaptability.

Description

Technical Field

[0001] The present invention relates to the technical field of dentistry, and in particular, to a method for positioning edentulous jaw dental implants combined with deep variational inference. Background Art

[0002] With the development of digital medicine and computer-aided design technology, implant-guided surgery is increasingly widely used in the field of oral implantology. Especially in edentulous patients, the accuracy of implant positioning is directly related to the postoperative mechanical stability and long-term biological adaptability. At present, in actual clinical practice, the positioning design of implants mainly relies on CBCT image data, and professional physicians manually determine the implantation area and formulate implant plans by combining experience with two-dimensional or three-dimensional images. However, relying on empirical judgment has prominent problems such as strong subjectivity, lack of quantitative indicators, and weak ability to estimate local bone mass changes.

[0003] Some existing studies have tried to introduce deep learning algorithms to segment and model medical images to achieve implant-assisted positioning, but generally ignore the spatial uncertainty of bone mass and bone quality distribution in the jawbone structure, and cannot fully evaluate the biomechanical reliability of the implant area. In addition, most of the existing auxiliary design tools are based on static geometric models and fail to simulate the dynamic response process of bone tissue after implant placement, such as bone remodeling or resorption problems caused by local stress, thus limiting the ability to predict the long-term effects of implant plans.

[0004] Especially in the edentulous jaw scenario, the jawbone structure degenerates severely and the anatomical landmark points are significantly missing, resulting in large errors in traditional image segmentation and implant position evaluation techniques in key structure recognition, bone density evaluation, and implant biomechanical simulation, affecting the scientificity of preoperative design and the stability of postoperative effects. Therefore, there is an urgent need for a new positioning method that can integrate fine modeling of medical images, uncertainty estimation, and biomechanical dynamic simulation to improve the intelligent, precise, and individualized level of edentulous jaw implant design. Summary of the Invention

[0005] An object of the present invention is to propose a method for positioning edentulous jaw dental implants combined with deep variational inference, which enables the final implant positioning points to obtain a global optimal solution in terms of long-term stability and biological adaptability.

[0006] A method for positioning edentulous jaw dental implants combined with deep variational inference according to an embodiment of the present invention includes the following steps:

[0007] S1. Obtain medical image data of an edentulous patient and perform preprocessing to generate optimized medical image data;

[0008] S2. Based on the optimized medical image data, use a deep learning model to perform three-dimensional segmentation of medical images, extract the bone mass distribution, bone quality information, and key anatomical structure marker points in the jawbone region, and generate a three-dimensional reconstruction model of the jawbone for edentulous patients;

[0009] S3. Through a deep variational inference model, perform distribution estimation of anatomical information on the three-dimensional reconstruction model of the jawbone, establish an uncertainty quantification model for bone mass distribution and bone quality parameters, and generate anatomical feature data;

[0010] S4. Use the anatomical feature data combined with implant mechanical parameters to calculate the initial implantation position, angle, and depth of the implant in the jawbone region, and generate an initial implant positioning plan;

[0011] S5. Based on the initial implant positioning plan, establish a dynamic simulation model for local bone mass reconstruction after implant implantation, simulate the stress distribution, bone strain, and bone quality change process of local bone tissue after implant implantation, and generate a local bone mass reconstruction simulation result;

[0012] S6. Perform dynamic interaction between the local bone mass reconstruction simulation result and the initial implant positioning plan, optimize the initial implant positioning plan through a deep variational inference model, and generate an optimized implant positioning plan including mechanical stability and biomechanical performance evaluation;

[0013] S7. Based on the optimized implant positioning plan, generate implant guide design data and surgical navigation data in combination with the three-dimensional reconstruction model of the patient's jawbone.

[0014] Optionally, S1 includes the following steps:

[0015] S11. Collect medical image data of edentulous patients. The medical image data is a continuous cross-sectional image sequence obtained based on cone beam computed tomography technology, and define the original image data set:

[0016] I raw ={I t ∣t=1,2,…,T};

[0017] Where, I t represents the t-th layer cross-sectional image, T is the total number of cross-sectional image layers, and I raw represents the original image data set;

[0018] S12. Perform noise removal processing on the original image data set I raw , perform gray-scale homogenization processing on the denoised image data set, map the gray-scale values between different cross-sectional images to a unified gray-scale interval, perform spatial resolution correction processing on the cross-sectional image sequence after gray-scale homogenization processing, resample the voxels of each axis of the cross-sectional image to the set spatial resolution, and generate optimized medical image data Iopt 。

[0019] Optionally, S2 includes the following steps:

[0020] S21. Input the optimized medical image data I opt into a multi-task deep learning model for edentulous jaw dental implant positioning design . The multi-task deep learning model includes an encoder module E(·) and two branch decoders. One branch of the two branch decoders is a jaw region segmentation decoder based on the attention mechanism. The output of the jaw region segmentation decoder is a jaw segmentation map S jaw (x, y, z), representing the pixel probability of the jaw region at the three-dimensional coordinates (x, y, z). The other branch is an anatomical key landmark detection decoder, and the output of the anatomical key landmark detection decoder is an anatomical heat map H anat (x, y, z) and the bone density feature map B den (x, y, z);

[0021] S22. Introduce an attention module in the jaw region segmentation decoder to fuse the encoder features E(I opt ) and the predefined jaw shape prior. The jaw shape prior is a jaw shape template constructed based on clinical statistical data to optimize the segmentation accuracy of the jaw region;

[0022] S23. In the anatomical key landmark detection decoder, use the heat map regression technique to post-process the anatomical heat map H anat (x, y, z) to extract the set P anat :

[0023]

[0024] where p i =(x i , y i , z i ) represents the coordinates of the i-th anatomical landmark in three-dimensional space, Ω i represents the local search area of the i-th landmark, and N p is the total number of anatomical landmarks;

[0025] S24. Integrate the jaw segmentation map S jaw , the bone density feature map B den (x, y, z) and the set P anat of key anatomical structure landmarks, and adopt a fusion function F fusion (·; α), where α is a fusion weight parameter with a value range of 0 < α < 1, used to balance the segmentation accuracy and the anatomical positioning accuracy, and generate a three-dimensional reconstruction model M jaw, the three-dimensional reconstruction model of the jawbone simultaneously reflects the jawbone morphology, bone mass distribution, and anatomical reference information.

[0026] Optionally, S3 includes the following steps:

[0027] S31. Input the three-dimensional reconstruction model M of the jawbone jaw into the anatomical information distribution estimation model, and the anatomical information distribution estimation model is a variational auto-encoding network model f VI (·; φ) based on deep variational inference. The anatomical information distribution estimation model includes an encoding network and a decoding network. The encoding network maps the three-dimensional reconstruction model M of the jawbone jaw to the latent variable space to generate the latent variable z latent , which is used to represent the potential anatomical feature distribution of the edentulous jawbone structure; the decoding network reconstructs the jawbone structure information based on the latent variable z latent to simulate the jawbone reconstruction distribution under different structural variations, reflecting the structural uncertainty of the three-dimensional reconstruction model M of the jawbone jaw ;

[0028] S32. Based on the multi-sample set output by the latent variable z latent , reconstruct the bone mass and bone quality parameters to generate the bone mass distribution probability map P vol (x, y, z) and the bone quality distribution probability map P qual (x, y, z). Among them, the bone mass distribution probability map P vol (x, y, z) represents the probability value of the bone density volume distribution at the three-dimensional coordinates (x, y, z), and the bone quality distribution probability map P qual (x, y, z) represents the probability value of the bone quality grade distribution at the three-dimensional coordinates (x, y, z). Both probability values are defined in the interval [0, 1], and are used to quantitatively characterize the uncertainty degree of the bone mass and bone quality information in this area;

[0029] S33. Finally, output the bone mass distribution probability map P vol (x, y, z) and the bone quality distribution probability map P qual (x, y, z) as anatomical feature data, including the uncertainty information of the bone mass and bone quality distribution in different spatial positions of the edentulous jawbone area.

[0030] Optionally, S3 also includes training and optimizing the model parameters φ of the anatomical information distribution estimation model f VI (·; φ). A variational lower bound loss objective function containing a likelihood term and a regularization term is constructed using a loss function based on the variational inference principle. By maximizing the reconstruction possibility of the anatomical information distribution estimation model for the original three-dimensional reconstruction model M of the jawbone jaw , while minimizing the Kullback-Leibler divergence between the latent variable distribution and the preset prior distribution.

[0031] Optionally, S4 includes the following steps:

[0032] S41. Input the bone mass distribution probability map P vol (x, y, z) and the bone quality distribution probability map P qual (x, y, z) into the implant site evaluation model to construct the multi-factor fitness function F score (x, y, z) for evaluating the adaptability of the three-dimensional coordinate point (x, y, z) as the initial implant point of the implant, defined as:

[0033] F score (x, y, z) = λ1·P vol (x, y, z) + λ2·P qual (x, y, z);

[0034] where λ1 and λ2 are the evaluation weight coefficients of bone mass and bone quality;

[0035] S42. Set the set of implant structure parameters I = {L, D, θ axis , R}, where L is the implant length, D is the implant diameter, θ axis is the angle between the implant axis and the normal of the jaw surface, and R is the minimum contact radius between the implant and the jaw; based on the fitness function and the structure parameters, construct the initial positioning optimization problem and solve the implant center position (x * , y * , z * ), the implant angle θ * and the implant depth d * :

[0036]

[0037] where S mech (·) is the initial mechanical stability scoring function of the implant, used to measure the influence of the implant parameters on the implant stability;

[0038] S43. According to the values of the bone mass distribution probability map, the bone quality distribution probability map and the initial mechanical stability scoring function of the implant, output the initial implant positioning plan P init .

[0039] Optionally, the determination rule of the initial implant positioning plan P init is as follows:

[0040] If the bone mass distribution probability map P vol (x, y, z) ≥ 0.7, the bone quality distribution probability map P qual (x, y, z) ≥ 0.6 and the value of the initial mechanical stability scoring function of the implant Smech If (x, y, z, θ, d; I) ≥ 0.8, it is determined as the preferred implantation area;

[0041] If the bone mass distribution probability map 0.5 ≤ P vol (x, y, z) < 0.7 or the bone quality distribution probability map 0.4 ≤ P qual (x, y, z) < 0.6 and the value S of the initial mechanical stability scoring function of the implant mech (x, y, z, θ, d; I) ≥ 0.6, it is determined as the alternative implantation area;

[0042] If the bone mass distribution probability map P vol (x, y, z) < 0.5 or the bone quality distribution probability map P qual (x, y, z) < 0.4 or the value S of the initial mechanical stability scoring function of the implant mech (x, y, z, θ, d; I) < 0.6, it is determined as the prohibited implantation area;

[0043] Combine the implant point (x * , y * , z * ) with the highest fitness score in the preferred implantation area, the angle θ * and the depth d * to form the initial implant positioning plan P init .

[0044] Optionally, the S5 includes the following steps:

[0045] S51. Based on the initial implant positioning plan P init ={(x * , y * , z * ), θ * , d * , I} and the three-dimensional reconstruction model M of the jawbone jaw Establish a local simulation area Ω in the area where the implant is to be implanted local , and the local simulation area is centered on the initial implant point (x * , y * , z * ) and includes the contact area of the implant to be implanted in the jawbone and its adjacent tissues;

[0046] S52. In the local simulation area Ω localA coupled mechanical model of the implant-jawbone tissue is established, and the three-dimensional finite element method is used to calculate the stress distribution and bone strain caused by implant placement in a local area. The stress distribution is described by stress components in each direction, including the stress component from the i-axis direction to the j-axis direction; the bone strain is the local deformation in the corresponding direction, which is used to represent the physical response of the bone tissue under the action of the implant; the material properties of the bone tissue in the area are modeled using a spatially non-uniform elastic stiffness tensor.

[0047] S53. Establish a biological response modeling module for the local bone tissue after implant placement in the local simulation area. The biological response modeling module is used to dynamically simulate the bone remodeling process, construct a bone mass change function Q(x, y, z, t) of the bone tissue per unit volume in the time dimension. The bone mass change function is driven by the equivalent stress. The higher the equivalent stress, the faster the bone remodeling rate; at the same time, a natural bone degradation factor is introduced to control the bone resorption rate and simulate the biological changes of the bone mass during the time evolution process.

[0048] S54. Iteratively solve the bone tissue stress, bone tissue strain, and bone mass change in the simulation area with time as the dimension to obtain the stress field distribution map, strain field distribution map, and bone remodeling image at multiple time nodes, which respectively describe the mechanical stimulus distribution, tissue deformation response, and dynamic update characteristics of bone mass of the local bone tissue after implant placement.

[0049] S55. Output the local bone remodeling simulation result set R sim , the local bone remodeling simulation result set includes the bone tissue stress distribution map Σ(x, y, z, t), bone tissue strain map E(x, y, z, t), and bone remodeling map Q(x, y, z, t) in the three-dimensional space and time dimension, which are used to comprehensively evaluate the mechanical effects and bone tissue responses generated by the implant on the local area of the jawbone under the preliminary positioning scheme.

[0050] Optionally, the S6 includes the following steps:

[0051] S61. Input the local bone remodeling simulation result set R sim and the preliminary implant positioning scheme P init into the implant positioning optimization module to construct a joint optimization objective function. The joint optimization objective function comprehensively considers the local stress distribution, bone strain response, and bone remodeling trend, and generates an implant mechanical stability index S stab and a biomechanical adaptability index S bio , and jointly evaluate the preliminary implant parameters.

[0052] S62. Introduce the joint optimization objective function F joint during the implant positioning optimization process. The joint optimization objective function comprehensively measures the bone mass support, bone mass response, implant mechanical stability, and long-term bone remodeling performance of the optimized positioning point:

[0053] F joint = γ1·S score + γ2·S stab + γ3·S bio ;

[0054] Wherein, S score represents the comprehensive bone mass and bone quality score corresponding to the optimized positioning point, S stab represents the mechanical stability score, S bio represents the bone mass dynamic performance score under simulation prediction, and γ1, γ2, and γ3 are weight coefficients;

[0055] S64. Based on the scoring results of the joint optimization objective function and the variational inference optimization results, screen the optimization positioning parameter group (x 最终 , y 最终 , z 最终 ), θ 最终 , d 最终 with the highest score, and output the implant optimization positioning scheme P opt :

[0056] If the optimized scoring function value F joint ≥ 0.8, and the bone mass reconstruction score S bio ≥ 0.75, then it is determined as an optimal implantation scheme;

[0057] If the optimized scoring function value 0.6 ≤ F joint < 0.8 or the bone mass reconstruction score 0.6 ≤ S bio < 0.75, then it is determined as an available implantation scheme;

[0058] If the optimized scoring function value F joint < 0.6 or the bone mass reconstruction score S bio < 0.6, then it is determined as a non-recommended implantation scheme;

[0059] Take the optimal point in the optimal implantation scheme as the final output, and construct the implant optimization positioning scheme P opt = {(x 最终 , y 最终 , z 最终 ), θ 最终 , d 最终 , I}.

[0060] The beneficial effects of the present invention are:

[0061] (1) The present invention inputs the three-dimensional reconstruction model of the edentulous jaw into a deep variational inference model based on a variational autoencoder network to obtain multiple-sample simulation results in the latent variable space, and then outputs a bone mass distribution probability map and a bone quality distribution probability map to reflect the spatial structure variability and modeling confidence of bone tissue in different regions, which can effectively avoid misjudging the implantation positions in high-noise or structurally ambiguous regions.

[0062] (2) On the basis of the preliminary implant plan, the present invention further introduces three-dimensional finite element stress field modeling and biological bone quality reconstruction reaction simulation. By simulating the stress distribution, bone tissue strain, and bone quality dynamic evolution at different time nodes, it outputs the spatio-temporal change trend of local bone quality response, feeds the results back to the implant positioning optimization module, and conducts a combined score with the preliminary plan to form an iterative closed-loop optimization mechanism, so that the final implant positioning point achieves a global optimal solution in terms of long-term stability and biological adaptability.

[0063] (3) The present invention constructs an encoder-decoder segmentation network integrating an attention mechanism, introduces an anatomical morphology prior template generated based on clinical statistical data for guidance, improves the response ability of the segmented region to key structures, extracts the key point coordinates by combining heat map regression and local search strategies, and generates a complete three-dimensional jaw reconstruction model in cooperation with the bone density distribution mapping, significantly improving the anatomical positioning accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0065] Figure 1 is a flowchart of a method for positioning edentulous jaw dental implants combining deep variational inference proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0066] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0067] Refer to Figure 1 , a method for positioning edentulous jaw dental implants combining deep variational inference, includes the following steps:

[0068] S1. Obtain the medical image data of the edentulous jaw patient and perform preprocessing to generate optimized medical image data;

[0069] S2. Based on the optimized medical image data, use a deep learning model to perform three-dimensional segmentation of the medical image, extract the bone mass distribution, bone quality information, and key anatomical structure marker points in the jaw region, and generate a three-dimensional reconstruction model of the jaw of the edentulous jaw patient;

[0070] S3. Estimate the distribution of anatomical information for the three-dimensional reconstruction model of the jawbone through a deep variational inference model, establish an uncertainty quantification model for bone mass distribution and bone quality parameters, and generate anatomical feature data;

[0071] S4. Combine the anatomical feature data with the mechanical parameters of the implant, calculate the preliminary implantation position, angle and depth of the implant in the jawbone area, and generate a preliminary implant positioning plan;

[0072] S5. Based on the preliminary implant positioning plan, establish a dynamic simulation model for local bone mass reconstruction after implant implantation, simulate the stress distribution, bone strain and bone quality change process of local bone tissue after implant implantation, and generate a local bone mass reconstruction simulation result;

[0073] S6. Dynamically interact the local bone mass reconstruction simulation result with the preliminary implant positioning plan, optimize the preliminary implant positioning plan through the deep variational inference model, and generate an optimized implant positioning plan including mechanical stability and biomechanical performance evaluation;

[0074] S7. Based on the optimized implant positioning plan, generate implant guide design data and surgical navigation data in combination with the three-dimensional reconstruction model of the patient's jawbone.

[0075] In this embodiment, S1 includes the following steps:

[0076] S11. Collect the medical image data of edentulous patients. The medical image data is a continuous cross-sectional image sequence obtained based on cone beam computed tomography technology, and define the original image data set:

[0077] I raw ={I t ∣t = 1, 2, …, T};

[0078] Among them, I t represents the t-th layer cross-sectional image, T is the total number of cross-sectional image layers, and I raw represents the original image data set;

[0079] S12. Perform noise removal processing on the original image data set I raw , perform gray-scale normalization processing on the denoised image data set, map the gray-scale values between different cross-sectional images to a unified gray-scale interval, perform spatial resolution correction processing on the cross-sectional image sequence after gray-scale normalization processing, resample the voxels of each axis of the cross-sectional image to the set spatial resolution, and generate optimized medical image data I opt .

[0080] In this embodiment, S2 includes the following steps:

[0081] S21. Input the optimized medical image data I opt into a multi-task deep learning model for edentulous jaw dental implant positioning design where the multi-task deep learning model includes an encoder module E(·) and two branch decoders. One of the two branch decoders is a jaw region segmentation decoder based on an attention mechanism, and the output of the jaw region segmentation decoder is a jaw segmentation map S jaw (x, y, z), representing the pixel probability of the jaw region at the three-dimensional coordinates (x, y, z). The other branch is an anatomical key landmark detection decoder, and the output of the anatomical key landmark detection decoder is an anatomical heat map H anat (x, y, z) and the bone density feature map B den (x, y, z);

[0082] S22. Introduce an attention module in the jaw region segmentation decoder to fuse the encoder features E(I opt ) and the predefined jaw shape prior. The jaw shape prior is a jaw shape template constructed based on clinical statistical data to optimize the segmentation accuracy of the jaw region;

[0083] S23. In the anatomical key landmark detection decoder, use heat map regression technology to post-process the anatomical heat map H anat (x, y, z) to extract the set P anat :

[0084]

[0085] where p i =(x i , y i , z i ) represents the coordinates of the i-th anatomical landmark in three-dimensional space, Ω i represents the local search area of the i-th landmark, and N p is the total number of anatomical landmarks;

[0086] S24. Integrate the jaw segmentation map S jaw , the bone density feature map B den (x, y, z) and the set P anat of key anatomical structure landmarks, and adopt a fusion function F fusion (·; α), where α is a fusion weight parameter with a value range of 0 < α < 1, used to balance the segmentation accuracy and the anatomical positioning accuracy, and generate a three-dimensional reconstruction model M jaw of the edentulous jaw patient. The three-dimensional reconstruction model of the jaw simultaneously reflects the jaw shape, bone mass distribution, and anatomical reference information.

[0087] In this embodiment, S3 includes the following steps:

[0088] S31. Input the three-dimensional reconstruction model M of the jawbone jaw into the anatomical information distribution estimation model, which is a variational auto-encoder network model f based on deep variational inference VI (·; φ). The anatomical information distribution estimation model includes an encoding network and a decoding network. The encoding network maps the three-dimensional reconstruction model M of the jawbone jaw to the latent variable space to generate the latent variable z latent , which is used to represent the potential anatomical feature distribution of the edentulous jawbone structure; the decoding network reconstructs the jawbone structure information based on the latent variable z latent to simulate the jawbone reconstruction distribution under different structural variations and reflect the structural uncertainty of the three-dimensional reconstruction model M of the jawbone jaw ;

[0089] S32. Based on the multi-sample set output by the latent variable z latent , reconstruct the bone mass and bone quality parameters to generate the bone mass distribution probability map P vol (x, y, z) and the bone quality distribution probability map P qual (x, y, z). Among them, the bone mass distribution probability map P vol (x, y, z) represents the probability value of the bone density volume distribution at the three-dimensional coordinates (x, y, z), and the bone quality distribution probability map P qual (x, y, z) represents the probability value of the bone quality grade distribution at the three-dimensional coordinates (x, y, z). Both probability values are defined in the interval [0, 1] and are used to quantitatively characterize the uncertainty degree of the bone mass and bone quality information in this area;

[0090] S33. Finally, output the bone mass distribution probability map P vol (x, y, z) and the bone quality distribution probability map P qual (x, y, z) as anatomical feature data, which includes the uncertainty information of the bone mass and bone quality distribution in the edentulous jawbone area at different spatial positions.

[0091] In this embodiment, step S3 also includes training and optimizing the model parameters φ of the anatomical information distribution estimation model f VI (·; φ). A variational lower bound loss objective function including a likelihood term and a regularization term is constructed by using a loss function based on the variational inference principle. By maximizing the reconstruction possibility of the anatomical information distribution estimation model for the original three-dimensional reconstruction model M of the jawbone jaw , and at the same time minimizing the Kullback-Leibler divergence between the latent variable distribution and the preset prior distribution.

[0092] In this embodiment, S4 includes the following steps:

[0093] S41. Input the bone mass distribution probability map P vol(x, y, z) and the bone mass distribution probability map P qual Input (x, y, z) into the implant site evaluation model to construct the multi-factor fitness function F of the jawbone region score (x, y, z), which is used to evaluate the adaptability of the three-dimensional coordinate point (x, y, z) as the initial implant point of the implant, and is defined as:

[0094] F score F(x, y, z) = λ1·P(x, y, z) + λ2·P(x, y, z); vol (x, y, z)+λ2·P qual (x,y,z);

[0095] Among them, λ1 and λ2 are the evaluation weight coefficients of bone mass and bone quality;

[0096] S42. Set the set of implant structure parameters I = {L, D, θ axis , R}, where L is the implant length, D is the implant diameter, θ axis is the angle between the implant axis and the normal of the jawbone surface, and R is the minimum contact radius between the implant and the jawbone; Based on the fitness function and the structure parameters, construct the preliminary positioning optimization problem and solve the implant center position (x * , y * , z * ), implant angle θ * and implant depth d * :

[0097]

[0098] Among them, S(·) is the preliminary mechanical stability scoring function of the implant, which is used to measure the influence of the implant parameters on the implant stability; mech (·) is the preliminary mechanical stability scoring function of the implant, which is used to measure the influence of the implant parameters on the implant stability;

[0099] S43. According to the values of the bone mass distribution probability map, the bone quality distribution probability map and the preliminary mechanical stability scoring function of the implant, output the preliminary implant positioning plan P init .

[0100] In this embodiment, the determination rule of the preliminary implant positioning plan P init is as follows:

[0101] If the bone mass distribution probability map P(x, y, z) ≥ 0.7, the bone quality distribution probability map P(x, y, z) ≥ 0.6 and the value of the preliminary mechanical stability scoring function of the implant S(x, y, z, θ, d; I) ≥ 0.8, then it is determined as the preferred implant area; vol (x,y,z)≥0.7, the bone quality distribution probability map P qual (x,y,z)≥0.6 and the value of the preliminary mechanical stability scoring function of the implant S mech (x,y,z,θ,d;I)≥0.8, then it is determined as the preferred implant area;

[0102] If the bone mass distribution probability map 0.5 ≤ P vol(x, y, z) < 0.7 or the bone mass distribution probability map 0.4 ≤ P qual (x, y, z) < 0.6 and the value S of the implant preliminary mechanical stability scoring function mech (x, y, z, θ, d; I) ≥ 0.6, then it is determined as an alternative implant area;

[0103] If the bone mass distribution probability map P vol (x, y, z) < 0.5 or the bone mass distribution probability map P qual (x, y, z) < 0.4 or the value S of the implant preliminary mechanical stability scoring function mech (x, y, z, θ, d; I) < 0.6, then it is determined as a prohibited implant area;

[0104] The implant point (x * , y * , z * ) with the highest fitness score in the preferred implant area, the angle θ * and the depth d * are combined into the implant preliminary positioning scheme P init .

[0105] In this embodiment, S5 includes the following steps:

[0106] S51. Based on the implant preliminary positioning scheme P init ={(x * , y * , z * ), θ * , d * , I} and the jawbone three-dimensional reconstruction model M jaw A local simulation area Ω is established in the implant proposed area local , and the local simulation area is centered on the implant preliminary implantation point (x * , y * , z * ), and includes the contact area of the proposed implant in the jawbone and its adjacent tissues;

[0107] S52. Establish an implant-jawbone tissue coupled mechanical model in the local simulation area Ω local , and use the three-dimensional finite element method to calculate the stress distribution and bone strain caused by the implant in the local area. The stress distribution is described by the stress components in each direction, including the stress component from the i-axis direction to the j-axis direction; the bone strain is the local deformation in the corresponding direction, which is used to represent the physical response of the bone tissue under the action of the implant; the material properties of the bone tissue in the area are modeled using a spatially non-homogeneous elastic stiffness tensor;

[0108] S53. Establish a bioreaction modeling module for the local bone tissue after implant placement in the local simulation area. The bioreaction modeling module is used to dynamically simulate the bone mass reconstruction process, construct a bone mass change function Q(x, y, z, t) of unit volume bone tissue in the time dimension. The bone mass change function is driven by the equivalent stress. The higher the equivalent stress, the faster the bone mass reconstruction rate. At the same time, introduce a natural bone degradation factor to control the bone resorption rate and simulate the biological changes of bone mass in the time evolution process.

[0109] S54. Iteratively solve the bone tissue stress, bone tissue strain, and bone mass change amount in the simulation area with time as the dimension to obtain the stress field distribution map, strain field distribution map, and bone mass reconstruction image at multiple time nodes, which respectively describe the mechanical stimulation distribution, tissue deformation response, and dynamic update characteristics of bone mass of the local bone tissue after implant placement.

[0110] S55. Output the local bone mass reconstruction simulation result set R sim , and the local bone mass reconstruction simulation result set includes the bone tissue stress distribution map Σ(x, y, z, t), bone tissue strain map E(x, y, z, t), and bone mass reconstruction map Q(x, y, z, t) in the three-dimensional space and time dimension, which are used to comprehensively evaluate the mechanical effects and bone tissue responses generated by the implant on the local area of the jawbone under the preliminary positioning scheme.

[0111] In this embodiment, S6 includes the following steps:

[0112] S61. Input the local bone mass reconstruction simulation result set R sim and the preliminary implant positioning scheme P init into the implant positioning optimization module to construct a joint optimization objective function. The joint optimization objective function comprehensively considers the local stress distribution, bone strain response, and bone mass reconstruction trend, and generates an implant mechanical stability index S stab and a biomechanical adaptability index S bio , and jointly evaluate the preliminary implant parameters.

[0113] S62. Introduce the joint optimization objective function F joint in the implant positioning optimization process. The joint optimization objective function comprehensively measures the bone mass support, bone mass response, implant mechanical stability, and long-term bone mass reconstruction performance of the optimized positioning point:

[0114] F joint =γ1·S score +γ2·S stab +γ3·S bio ;

[0115] Among them, S score represents the comprehensive score of bone mass and bone quality corresponding to the optimized positioning point, and S stabDenote the mechanical stability score as S bio Denote the bone mass dynamic performance score under simulation prediction as γ1, γ2, γ3 are weight coefficients;

[0116] S64. Based on the scoring results of the joint optimization objective function and the variational inference optimization results, screen the optimization positioning parameter group (x 最终 , y 最终 , z 最终 ), θ 最终 , d 最终 with the highest score, and output the optimized implant positioning scheme P opt :

[0117] If the optimized scoring function value F joint ≥0.8, and the bone mass reconstruction score S bio ≥0.75, then it is determined as an optimal implantation scheme;

[0118] If the optimized scoring function value 0.6 ≤ F joint <0.8 or the bone mass reconstruction score 0.6 ≤ S bio <0.75, then it is determined as an available implantation scheme;

[0119] If the optimized scoring function value F joint <0.6 or the bone mass reconstruction score S bio <0.6, then it is determined as a non-recommended implantation scheme;

[0120] Take the optimal point in the optimal implantation scheme as the final output, and construct the optimized implant positioning scheme P opt ={(x 最终 , y 最终 , z 最终 ), θ 最终 , d 最终 , I}.

[0121] Example 1:

[0122] The Digital Dental Center of the Ninth People's Hospital in City A received a male patient, Wang, who had been completely edentulous in the mandible for more than 7 years, resulting in severely limited chewing function. At the same time, the lower part of the face collapsed, affecting the quality of life. Wang hoped to undergo mandibular implant restoration treatment. However, due to severe bone degeneration over the years, traditional empirical method evaluation showed that the safe distance between the mandibular nerve canal and the top of the bone ridge was less than 6 mm, presenting a relatively high implantation risk.

[0123] The oral implant center of the hospital decided to introduce this invention into actual clinical practice for the first time to explore the feasibility and optimization ability of the technology under extreme conditions.

[0124] Wang was scanned by CBCT in the radiology room using a Planmeca ProMax 3D device, obtaining 512 images with a single-frame resolution of 672×672 pixels and a voxel spacing of 0.3 mm. The original image dataset Iraw was loaded into the server by the preprocessing module for automated processing.

[0125] Denoising used a multi-scale denoising algorithm based on wavelet transform. A metal artifact was detected in the 212th layer image. After processing, the peak signal-to-noise ratio increased from 22.8 to 27.3. Subsequently, gray-level equalization and resolution resampling were performed, and the spatial voxel was resampled to 0.25 mm cube to generate the optimized dataset Iopt, which was written into the project directory named "WANG_SMILE_20241113".

[0126] The deep segmentation model started running with a model version of seg-jaw-v3.2.4, loading the pre-trained parameters jaw_2023_finetune.ckpt. The anatomical structure landmark points identified by the model included the left and right mandibular foramens, the mandibular nerve canal, the mandibular ridge, and the anterior basal bone. A total of 17 structure points were identified. The spatial coordinates of the left mandibular foramen were (x = 126.5, y = 204.8, z = 77.3). The nerve path was almost completely consistent with the doctor's observation, and the spatial error was less than 1.2 mm. There were significant differences in bone density characteristics in the left and right premolar areas. The system output the three-dimensional reconstruction model of the jaw `Mjaw` and marked the bone density score maps of each region.

[0127] The model "vvae-jaw-uncertainty-v2.1.7" was activated, using a 128-dimensional latent variable space to perform distribution inference on Mjaw in 32 batches and 6000 iterations. The bone volume distribution probability Pvol(x,y,z) in the left first premolar area of the mandible was 0.83, and the bone quality probability Pqual(x,y,z) was 0.72. The system scored and marked it as the "preferred implantation area".

[0128] The preliminary positioning evaluation parameters are as follows: the parameters of the implant to be implanted: length 10 mm, diameter 3.6 mm; the initial angle range: [45°, 90°]; the depth search range: [6 mm, 12 mm].

[0129] The algorithm completed the three-dimensional space search within 95 seconds, outputting the preliminary positioning point: (x = 129.2, y = 210.6, z = 74.8), the angle θ = 72.5°, the depth d = 9.8 mm, and the preliminary mechanical score Smech = 0.86.

[0130] The simulation model ran on the Ansys platform, setting a simulation period of 12 weeks, and outputting stress, strain, and bone quality maps every week.

[0131] The simulation data for the 3rd week shows that the maximum stress is 9.1 MPa; the local maximum strain is 0.0041; the osteophyte prediction diagram shows that a bone structure strengthening area is formed at x = 129.0, y = 211.0 starting from the 6th week, and the value of Q(x, y, z, t) increases by 18%.

[0132] In the 12th week, the volume of the bone reconstruction area is 123.6 mm 3 , and the predicted bone integration score Sbio = 0.78, which is about 23% higher than that of the traditional non - simulated scheme.

[0133] To verify the advantages of this technology, the hospital conducted a comparative study among 10 edentulous patients during the same period (traditional experience manual design group vs. the method group of the present invention), and the main data are as follows:

[0134] Table 1 Comparative data between the traditional experience manual design group and the method group of the present invention

[0135]

[0136] Wang finally completed the implantation surgery. A total of 4 mandibular implants were implanted using the positioning points determined by the method of the present invention. The postoperative CBCT re - examination showed that the deviation of each implantation point from the estimated plan position was within ±1.1 mm. There was no obvious stress concentration after the operation, and no patient complained of discomfort. The physician team submitted this plan as a demonstration case for the upgrade of the hospital's digital implant navigation system and planned to promote it to the high - risk elderly edentulous population.

[0137] Example 1 fully verifies the systematic advantages of the present invention in modeling uncertain anatomical structures, mechanical simulation optimization, and intelligent closed - loop implant path reasoning in real - world complex scenarios, providing a replicable intelligent reference path for edentulous implant design.

[0138] The present invention inputs the three - dimensional reconstruction model of the edentulous jawbone into a deep variational inference model based on a variational auto - encoder network to obtain multiple - sample simulation results in the latent variable space, and then outputs the bone mass distribution probability map and the bone quality distribution probability map to reflect the spatial structure variability and modeling confidence of bone tissues in different regions. It can effectively avoid misjudging the implantation positions in high - noise or structure - blurred regions and can also provide quantitative support for subsequent biomechanical optimization.

[0139] Based on the preliminary implant plan, the present invention further introduces three - dimensional finite - element stress field modeling and biological bone reconstruction reaction simulation. By simulating the stress distribution, bone tissue strain, and bone quality dynamic evolution at different time nodes, it outputs the spatio - temporal change trend of local bone quality response, feeds the results back to the implant positioning optimization module, and conducts a joint scoring with the preliminary plan to form an iterative closed - loop optimization mechanism, enabling the final implant positioning points to achieve a global optimal solution in terms of long - term stability and biological adaptability.

[0140] The present invention constructs an encoder-decoder segmentation network integrating an attention mechanism, introduces an anatomical morphological prior template generated based on clinical statistical data for guidance, enhances the response ability of the segmentation region to key structures, combines heatmap regression with a local search strategy to extract key point coordinates, and cooperates with bone density distribution mapping to generate a complete three-dimensional jawbone reconstruction model, significantly improving the accuracy of anatomical positioning.

[0141] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered within the protection scope of the present invention.

Claims

1. A method for positioning edentulous jaw dental implants combining deep variational inference, characterized in that, The method includes the following steps: S1. Obtain the medical image data of the edentulous patient and perform preprocessing to generate optimized medical image data; S2. Based on the optimized medical image data, use a deep learning model to perform three-dimensional segmentation of the medical image, extract the bone mass distribution, bone quality information, and key anatomical structure marker points in the jawbone area, and generate a three-dimensional reconstruction model of the jawbone of the edentulous patient; S3. Through a deep variational inference model, perform distribution estimation of the anatomical information of the three-dimensional reconstruction model of the jawbone, establish an uncertainty quantification model for the bone mass distribution and bone quality parameters, and generate anatomical feature data; S4. Use the anatomical feature data combined with the implant mechanical parameters to calculate the preliminary implantation position, angle, and depth of the implant in the jawbone area, and generate a preliminary implant positioning plan; S5. Based on the preliminary implant positioning plan, establish a dynamic simulation model for local bone mass reconstruction after implant implantation, simulate the stress distribution, bone strain, and bone quality change process of the local bone tissue after implant implantation, and generate a local bone mass reconstruction simulation result; S6. Perform dynamic interaction between the local bone mass reconstruction simulation result and the preliminary implant positioning plan, and optimize the preliminary implant positioning plan through a deep variational inference model to generate an optimized implant positioning plan including mechanical stability and biomechanical performance evaluation; S7. Based on the optimized implant positioning plan, generate implant guide design data and surgical navigation data in combination with the three-dimensional reconstruction model of the patient's jawbone.

2. The method for positioning edentulous jaw dental implants combined with deep variational inference according to claim 1, characterized in that The S1 includes the following steps: S11. Collect the medical image data of the edentulous patient. The medical image data is a sequence of continuous cross-sectional images obtained based on cone beam computed tomography technology, and define the original image dataset: I raw = {I t | t = 1, 2, …, T}; Among them, I t represents the cross-sectional image of the t-th layer, T is the total number of cross-sectional image layers, and I raw represents the original image data set; S12. Perform noise removal processing on the original image dataset I raw Perform gray-scale homogenization processing on the denoised image dataset, map the gray-scale values between different cross-sectional images to a unified gray-scale interval, perform spatial resolution correction processing on the sequence of cross-sectional images after gray-scale homogenization processing, resample the voxels of each axis of the cross-sectional image to the set spatial resolution, and generate the optimized medical image data I opt .

3. A method for positioning edentulous jaw dental implants combined with deep variational inference according to claim 2, wherein, The S2 includes the following steps: S21. Input the optimized medical image data I opt into a multi-task deep learning model for edentulous implant positioning design wherein the multi-task deep learning model includes an encoder module E(·) and two branch decoders. One of the two branch decoders is a jaw region segmentation decoder based on an attention mechanism, and the output of the jaw region segmentation decoder is a jaw segmentation map S jaw (x, y, z), representing the pixel probability of the jaw region at the three-dimensional coordinates (x, y, z). The other branch is an anatomical key landmark detection decoder, and the output of the anatomical key landmark detection decoder is an anatomical heat map H anat (x, y, z) and a bone density feature map B den (x, y, z); S22. Introduce an attention module in the jaw region segmentation decoder to fuse the encoder feature E(I opt ), and the predefined prior of jaw morphology, where the prior of jaw morphology is a jaw morphology template constructed based on clinical statistical data, to optimize the segmentation accuracy of the jaw region; S23. In the anatomical key landmark detection decoder, the anatomical heatmap H anat (x, y, z) is post-processed using heatmap regression technology to extract the set P of key anatomical structure landmarks anat : where p i = (x i , y i , z i ) represents the coordinates of the i-th anatomical landmark point in three-dimensional space, Ω i represents the local search area of the i-th landmark point, and N p is the total number of anatomical landmark points; S24. Integrate the jawbone segmentation map S jaw , the bone density feature map B den (x, y, z) and the key anatomical structure marker point set P anat , and adopt the fusion function F fusion (·; α), where α is the fusion weight parameter, and the value range is 0 < α < 1, which is used to balance the segmentation accuracy and the anatomical positioning accuracy, and generate the three-dimensional reconstruction model M of the jawbone of the edentulous patient jaw . The three-dimensional reconstruction model of the jawbone reflects the jawbone morphology, bone mass distribution and anatomical reference information at the same time.

4. A method for positioning edentulous jaw dental implants combined with deep variational inference according to claim 3, characterized in that The S3 includes the following steps: S31. Input the three-dimensional reconstruction model M of the jawbone jaw into the anatomical information distribution estimation model, which is a variational auto-encoding network model f based on deep variational inference VI (·; φ). The anatomical information distribution estimation model consists of two parts: an encoding network and a decoding network. The encoding network maps the three-dimensional reconstruction model M of the jawbone jaw to the latent variable space to generate a latent variable z latent , which is used to represent the potential anatomical feature distribution of the edentulous jawbone structure; the decoding network reconstructs the jawbone structure information based on the latent variable z latent to simulate the jawbone reconstruction distribution under different structural variations and reflect the structural uncertainty of the three-dimensional reconstruction model M of the jawbone jaw . S32. Based on the latent variable z latent For the output multi-sample set, reconstruct the bone mass and bone quality parameters to generate a bone mass distribution probability map P vol (x, y, z) and a bone quality distribution probability map P qual (x, y, z), where the bone mass distribution probability map P vol (x, y, z) represents the probability value of the bone density volume distribution at the three-dimensional coordinates (x, y, z), and the bone quality distribution probability map P qual (x, y, z) represents the probability value of the bone quality grade distribution at the three-dimensional coordinates (x, y, z). Both probability values are defined in the interval [0, 1] and are used to quantitatively characterize the uncertainty degree of the bone mass and bone quality information in this region; S33. Finally, output the bone mass distribution probability map P vol (x, y, z) and the bone mass distribution probability map P qual (x, y, z), as anatomical feature data, contains the uncertainty information of bone mass and bone quality distribution in different spatial positions of the edentulous jaw region.

5. A method for positioning edentulous jaw dental implants combined with deep variational inference according to claim 4, characterized in that Step S3 also includes training and optimizing the model parameters φ of the anatomical information distribution estimation model f VI (·; φ), constructing a variational lower bound loss objective function including a likelihood term and a regularization term using a loss function based on the variational inference principle, and maximizing the reconstruction possibility of the original jaw three-dimensional reconstruction model M jaw by the anatomical information distribution estimation model, while minimizing the Kullback-Leibler divergence between the latent variable distribution and the preset prior distribution.​​​​ 6. A method for positioning edentulous jaw dental implants combining deep variational inference, characterized in that The S4 includes the following steps: S41. Input the bone mass distribution probability map P vol (x, y, z) and the bone mass distribution probability map P qual (x, y, z) into the implant site evaluation model to construct the multi-factor fitness function F score (x, y, z) of the jawbone region, which is used to evaluate the adaptability of the three-dimensional coordinate point (x, y, z) as the preliminary implant point of the implant, and is defined as: F score (x, y, z) = λ1·P vol (x, y, z) + λ2·P qual (x, y, z); Wherein, λ1 and λ2 are the evaluation weight coefficients of bone mass and bone quality; S42. Set the set of implant structure parameters I = {L, D, θ axis , R}, where L is the implant length, D is the implant diameter, θ axis is the angle between the implant axis and the normal of the jaw surface, and R is the minimum contact radius between the implant and the jaw; based on the fitness function and the structure parameters, construct a preliminary positioning optimization problem to solve the implant center position (x * , y * , z * ), the implant angle θ * and the implant depth d * : Among them, S mech (·) is the initial mechanical stability scoring function of the implant, which is used to measure the influence of the implant parameters on the implant stability; S43. Output the preliminary implant positioning plan P based on the bone mass distribution probability map, bone quality distribution probability map, and the value of the preliminary implant mechanical stability scoring function init .

7. A method for positioning edentulous jaw dental implants combining deep variational inference according to claim 6, characterized in that, The preliminary implant positioning plan P init has the following judgment rules: If the bone mass distribution probability graph P vol (x, y, z) ≥ 0.7, the bone mass distribution probability graph P qual (x, y, z) ≥ 0.6 and the value S of the initial mechanical stability scoring function of the implant mech (x, y, z, θ, d; I) ≥ 0.8, then it is determined as the preferred implantation area; If the bone mass distribution probability map is such that 0.5 ≤ P vol (x, y, z) < 0.7 or the bone mass distribution probability map is such that 0.4 ≤ P qual (x, y, z) < 0.6 and the value S of the initial mechanical stability scoring function of the implant mech (x, y, z, θ, d; I) ≥ 0.6, then it is determined as an alternative implantation area; If the bone mass distribution probability graph P vol (x, y, z) < 0.5 or the bone mass distribution probability graph P qual (x, y, z) < 0.4 or the value S of the implant preliminary mechanical stability scoring function mech (x, y, z, θ, d; I) < 0.6, then it is determined as a prohibited implant area; The implant point (x * , y * , z * ) with the highest fitness score in the priority implantation area, the angle θ * and the depth d * are combined into the preliminary implant positioning plan P init .

8. A method for positioning edentulous jaw dental implants combining deep variational inference, characterized in that The S5 includes the following steps: S51. Based on the initial positioning plan P of the implant init ={(x * ,y * ,z * ),θ * ,d * ,I} and the three-dimensional reconstruction model M of the jawbone jaw Establish a local simulation area Ω in the area where the implant is to be implanted local , The local simulation area is centered on the initial implant insertion point (x * ,y * ,z * ), and includes the contact area of the implant to be implanted in the jawbone and its adjacent tissues; S52. In the local simulation region Ω local A coupled mechanical model of the implant-jawbone tissue is established, and the three-dimensional finite element method is used to calculate the stress distribution and bone strain caused by implant placement in the local region. The stress distribution is described by stress components in each direction, including the stress component from the i-axis direction to the j-axis direction; the bone strain is the local deformation in the corresponding direction, which is used to represent the physical response of the bone tissue under the action of the implant; the material properties of the bone tissue in the region are modeled using a spatially non-uniform elastic stiffness tensor; S53. Establish a biological response modeling module for the local bone tissue after implant implantation in the local simulation area. The biological response modeling module is used to dynamically simulate the bone mass reconstruction process, construct a bone quality change function Q(x, y, z, t) of the bone tissue per unit volume in the time dimension. The bone quality change function is driven by the equivalent stress. The higher the equivalent stress, the faster the bone mass reconstruction rate; at the same time, introduce a bone quality natural degradation factor to control the bone mass absorption rate and simulate the biological changes of the bone mass in the time evolution process; S54. Iteratively solve the bone tissue stress, bone tissue strain, and bone mass change amount in the simulation area with time as the dimension to obtain the stress field distribution map, strain field distribution map, and bone mass reconstruction image at multiple time nodes, which respectively describe the mechanical stimulation distribution, tissue deformation response, and bone mass dynamic update characteristics of the local bone tissue after implant implantation; S55. Output the local bone remodeling simulation result set R sim , where the local bone remodeling simulation result set includes the bone tissue stress distribution map Σ(x, y, z, t), the bone tissue strain map E(x, y, z, t), and the bone remodeling map Q(x, y, z, t) in the three-dimensional space and time dimension, and is used to comprehensively evaluate the mechanical effects and bone tissue responses generated by the implant on the local area of the jawbone under the preliminary positioning scheme.

9. A method for positioning edentulous jaw dental implants combined with deep variational inference, characterized in that The S6 includes the following steps: S61. Input the local bone reconstruction simulation result set R sim and the preliminary implant positioning scheme P init into the implant positioning optimization module to construct a combined optimization objective function. The combined optimization objective function comprehensively considers the local stress distribution, bone strain response, and bone reconstruction trend to generate the implant mechanical stability index S stab and the biomechanical adaptability index S bio , and jointly evaluate the preliminary implant parameters; S62. Introduce the combined optimization objective function F during the optimization process of implant positioning joint , and the combined optimization objective function comprehensively measures the bone mass support, bone quality response, implant mechanical stability, and long-term bone quality reconstruction performance of the optimized positioning point: F joint = γ1·S score + γ2·S stab + γ3·S bio ; Among them, S score represents the comprehensive bone mass and bone quality score corresponding to the optimized positioning point, S stab represents the mechanical stability score, S bio represents the bone mass dynamic performance score under simulation prediction, and γ1, γ2, γ3 are weighting coefficients; S64. Based on the scoring results and variational inference optimization results of the joint optimization objective function, select the optimization positioning parameter group (x 最终 , y 最终 , z 最终 ), θ 最终 , d 最终 with the highest score, and output the implant optimization positioning scheme P opt : If the optimized scoring function value F joint ≥ 0.8, and the bone remodeling score S bio ≥ 0.75, then it is determined as an optimal implantation plan; If the optimized scoring function value 0.6 ≤ F joint < 0.8 or the bone remodeling score 0.6 ≤ S bio < 0.75, it is determined as a usable implantation plan; If the optimized scoring function value F joint <0.6 or the bone remodeling score S bio <0.6, then it is determined that the implantation plan is not recommended; Taking the optimal point in the preferred implantation plan as the final output, construct the optimized positioning plan P of the implant opt ={(x 最终 , y 最终 , z 最终 ), θ 最终 , d 最终 , I}.

Images