Personalized dental restoration intelligent design method based on generative adversarial network

Abstract

The invention belongs to the technical field of computer aided design, and provides a generative adversarial network-based personalized dental restoration intelligent design method, which comprises the following steps of: marking a difference set part connected with a temporary restoration region in a wearing three-dimensional model as a carrying model; carrying out combined Boolean operation on the temporary restoration body area and the undertaking model to obtain a combined part, and recording the combined part as a stress model; identifying a stress shielding area on the stress model according to the stress nephogram; performing elastic correction on the stress shielding area of the temporary restoration area to obtain a three-dimensional model of the dental restoration; the wall thickness of the strong sub-region with an unstable stress path caused by overall stress shielding is corrected through the wall thickness difference value of the corresponding elastic correction weak sub-region, the problem that overall stress shielding is converted into a local stress peak value is reduced, stress transmission is uniform, and the problem that stability fluctuates due to the fact that the stress path changes along with time is solved.

Description

Technical Field

[0001] This invention belongs to the fields of computer-aided design and image generation technology, and specifically relates to a personalized intelligent design method for dental prostheses based on generative adversarial networks. Background Technology

[0002] In prosthodontic treatment, after the implanted implant has completed osseointegration, a temporary prosthesis is usually fabricated based on the shape of the soft tissue surrounding the implant. This temporary prosthesis is then shaped to create the appropriate peri-implant soft tissue profile. By continuously adjusting the transgingival contour of the temporary prosthesis, a suitable peri-implant soft tissue profile is achieved, after which the final prosthesis is designed. The transgingival portion of the final prosthesis must have a consistent three-dimensional shape with the transgingival contour shaped by the temporary prosthesis. Implanting a high-rigidity and high-strength prosthesis to fix the transgingival contour provides a stable mechanical environment for the soft tissue, preventing further damage and facilitating normal occlusal activity during the prosthodontic process.

[0003] However, because the stiffness of dental prostheses is much greater than that of the alveolar bone and the tissues at the gingival contour, a stress shielding effect occurs: the alveolar bone and tissues at the gingival contour remain at a low stress level for a long period. In the middle and later stages of dental prosthesis treatment, the bone tissue may experience incomplete marginal healing, excessive marginal bone resorption, and malocclusion due to insufficient mechanical stimulation. Generally, the longer the dental prosthesis is in place, the worse the mechanical properties of the hyperplastic alveolar bone tissue becomes. Summary of the Invention

[0004] This invention aims to at least partially address one of the technical problems in the related art. Therefore, the object of this invention is to propose a personalized intelligent design method for dental prostheses based on generative adversarial networks, to solve one or more technical problems existing in the prior art, and at least provide a beneficial option or create favorable conditions.

[0005] To achieve the above objectives, this invention proposes a personalized intelligent design method for dental prostheses based on generative adversarial networks, the method comprising the following steps: S100: The patient's oral cavity is imaged in three dimensions and reconstructed in three dimensions to obtain a three-dimensional model of the patient wearing a temporary prosthesis. S200: The patient's oral cavity interior is imaged in three dimensions and reconstructed in three dimensions to obtain an empty three-dimensional model when the temporary prosthesis is removed; S300 spatially registers the worn 3D model and the unloaded 3D model, resulting in a worn registration model and an unloaded registration model, respectively. S400, perform a Boolean operation to subtract the worn registration model and the unloaded registration model to obtain the difference set, and mark the difference set that is connected to the temporary prosthesis area in the worn 3D model as the receiving model; S500, the Boolean operation that combines the temporary repair area with the receiving model yields the combined part, which is denoted as the stress model; S600 performs finite element analysis on the stress model to obtain a stress cloud map, and identifies the stress-shielding area on the stress model based on the stress cloud map. S700 uses elastic correction of the stress-shielding area of ​​the temporary restoration to obtain a three-dimensional model of the oral restoration.

[0006] Preferably, in S700, the method for elastically correcting the stress-shading area of ​​the temporary prosthesis region to obtain a three-dimensional model of the oral prosthesis is as follows: the stress-shading area of ​​the temporary prosthesis region is reconstructed according to the trained generative adversarial network (GAN) to obtain a three-dimensional model of the oral prosthesis.

[0007] Preferably, the method for reconstructing the three-dimensional model of the oral prosthesis from the stress-masking region of the temporary prosthesis area using a trained generative adversarial network (GAN) is as follows: Form a paired dataset {original model, stress occlusion label, target model}; Align the 3D model and stress shielding labels of the temporary restoration area to ensure spatial consistency; The 3D model of the temporary restoration area was converted into a voxel mesh (resolution 0.1 mm³). Building a Generative Adversarial Network (GAN): Includes: generator and discriminator: Generator (G): 3D U-Net, input is the original model + stress occlusion label, output is the reconstructed model; Discriminator (D): 3D CNN, input is real / generative model, output is real / false probability; Training a Generative Adversarial Network (GAN): Pre-training G uses only L_L1 (100 rounds); Jointly train G and D, gradually adding L_stress (learning rate 1e-4, batch size 8); Input the patient's temporary prosthesis model and stress occlusion mask into the trained generative adversarial network (GAN); A three-dimensional model of an oral prosthesis is generated using a generative adversarial network (GAN).

[0008] Furthermore, in S100, three-dimensional imaging of the inside of the patient's mouth when wearing a temporary prosthesis is performed by performing a three-dimensional imaging scan of the inside of the patient's mouth when wearing a temporary prosthesis using oral CBCT, and obtaining a three-dimensional model of the patient wearing the prosthesis through three-dimensional reconstruction.

[0009] Furthermore, in S200, three-dimensional imaging of the inside of the patient's mouth when the temporary prosthesis is removed is performed by performing a three-dimensional imaging scan of the inside of the patient's mouth when the temporary prosthesis is removed using oral CBCT, and a three-dimensional model of the patient wearing the prosthesis is obtained through three-dimensional reconstruction.

[0010] Preferably, the three-dimensional reconstruction method is to obtain a three-dimensional image by performing three-dimensional reconstruction using the three-dimensional reconstruction function of Mimics software.

[0011] Preferably, the oral CBCT is a three-in-one oral CBCT machine.

[0012] Furthermore, in S300, the spatial registration method is the SIFT matching method.

[0013] Furthermore, in S400, the method for obtaining the difference set by subtracting the worn registration model and the unloaded registration model using a Boolean operation is as follows: subtract the part that overlaps with the unloaded registration model from the worn registration model to obtain the difference set.

[0014] Furthermore, in S500, the method for performing a Boolean operation to combine the temporary restoration area and the receiving model to obtain the combined part, denoted as the stress model, is as follows: the two three-dimensional models, the temporary restoration area and the receiving model, are merged, and the intersecting parts are deleted, thereby merging into a combined part, denoted as the stress model.

[0015] Furthermore, in S600, the method for obtaining stress contour maps by performing finite element analysis on the stress model includes: obtaining stress contour maps by performing finite element analysis on the stress model using the open-source nonlinear FEBio / FEBio Studio finite element analysis software package.

[0016] Specifically, methods for obtaining stress contour plots through finite element analysis include: Import the 3D model of the force model; Define material properties: dental prosthesis material (zirconia or alumina), E (Young's modulus) = 200 GPa (zirconia), ν (Poisson's ratio) = 0.3; Create a finite element mesh: Set the global finite element size (0.5 mm, Jacobian matrix > 0.7); Set boundary conditions and loads: Fixed constraints: bottom of the restoration or contact surface of adjacent teeth; Occlusal load: occlusal force of 300 N on the occlusal surface; Contact surface setup: sliding friction between the dental prosthesis and alveolar bone, and between the prosthesis and the opposing tooth, with a friction coefficient µ=0.2~0.5 (typical value between dental materials). Configure the solution control parameters: (1) Analysis step configuration: total time = 1, number of steps = 100 (quasi-static analysis); (2) Solver parameter configuration: Perform large deformation analysis using the Newton-Raphson iterative method and adjust the convergence tolerance (e.g., convergence tolerance rtol=4); Generates a visual stress cloud map, including: comprehensive stress assessment and tensile / compressive stress separation display.

[0017] Among them, the visual stress cloud map under the applied peak stress load is used as the stress cloud map.

[0018] Furthermore, in S600, the method for identifying stress-obstructed regions on the stress model based on the stress cloud map is as follows: the stress cloud map is grayscaled and then segmented using the watershed algorithm to obtain multiple sub-regions; sub-regions whose maximum stress value (the maximum stress value borne by all points in the sub-region) is less than the low stress threshold are marked as stress-obstructed regions (the low stress threshold is set to 1 MPa to 5 MPa).

[0019] Preferably, in S600, the method for identifying stress-shielding areas on the stress model based on the stress cloud diagram is as follows: The stress cloud map is converted to grayscale and then segmented using the watershed algorithm to obtain multiple sub-regions. The average grayscale of each sub-region is taken as the low-stress grayscale. Sub-regions where the average grayscale of all points in each sub-region is greater than the low-stress grayscale are defined as stress occlusion regions.

[0020] In the grayscale stress cloud map, the low-stress areas appear darker.

[0021] The stress shielding effect occurs in stress-shielded areas. This effect leads to a problem during alveolar bone reconstruction: the stiffer implant (restor) and the less stiff bone tissue share the load. Under uniform strain boundaries, the stress borne by the bone tissue is much less than that borne by the implant. According to Wolff's Law, bone growth and remodeling adapt to mechanical stress stimuli; that is, bone strengthens its structure in areas subjected to load pressure, while gradually decreasing its strength in areas lacking stress. Existing technologies, such as the Chinese invention patent with publication number CN115105266A, use finite element analysis and topology-optimized gradient hole structures to reduce peak stress and uniform stress, avoiding stress concentration. They also use bioceramic alumina to reduce the elastic modulus or stiffness of the implant, thus avoiding the stress shielding effect. However, existing technologies present the following problems: While porous structures reduce modulus, they also significantly reduce the load-bearing area under stress, making lattice nodes and intersecting members "natural gaps" for fatigue crack initiation. Under cyclic loading during oral chewing, fatigue often becomes the failure mode before static strength. Gradient-pore structures, under stress, often create abrupt changes in equivalent cross-sections in areas of fastest gradient change, where stress may shift from "overall stress shielding" to "local stress peaks," especially noticeable at the neck, thread roots, and the boundary between the porous structure and the dense region. Furthermore, although gradient-pore structures can promote cell and blood vessel attachment, bone ingrowth exhibits significant temporal and spatial heterogeneity: areas closer to cortical bone and with better blood supply are more prone to ingrowth; deep porous regions may remain "partially filled" for extended periods. This can cause the interfacial stiffness of the restoration to change nonlinearly over time under long-term occlusal loads. The stress path changes continuously from the early to the middle to the late stages, resulting in stability fluctuations. Furthermore, reducing the elastic modulus or stiffness of the implant will increase the deformation of the implant under occlusal loads, making it difficult for the bone-implant interface to form a stable osseointegration, and even causing marginal bone resorption. This is because reducing stiffness will transfer more load to the alveolar bone tissue, but the stress transmission mode is uneven: if the osseointegration surface area increases, the load may be mainly concentrated in the alveolar bone, which may cause excessive local stress, leading to abnormal bone remodeling or marginal bone resorption.

[0022] Therefore, this application modifies the stress shielding area of ​​the oral prosthesis using the following method to address the issues of shifting from "overall stress shielding" to "local stress peaks," uneven stress transmission, and stability fluctuations caused by changes in the stress path over time. The specific method is as follows: Preferably, in S700, the method for obtaining a three-dimensional model of the oral prosthesis by elastically correcting the stress-shielding area of ​​the temporary prosthesis region is as follows: The stress cloud maps under strong and weak stresses are converted to grayscale and then segmented using the watershed algorithm to obtain multiple sub-regions. The sub-regions inside the stress occlusion area in the strong stress cloud map are denoted as strong sub-regions, and the sub-regions inside the stress occlusion area in the weak stress cloud map are denoted as weak sub-regions. The point with the highest stress value in each weak subregion is mapped to the position on the force model as the weak peak point, and the point with the lowest stress value is mapped to the position on the force model as the weak valley point; the point with the highest stress value in each hadron region is mapped to the position on the force model as the strong peak point, and the point with the lowest stress value is mapped to the position on the force model as the strong valley point. Iterate through all hadron regions and determine whether the stress paths at their positions on the stress model are stable. The position of the hadron region with unstable stress path on the temporary restoration region is elastically corrected to obtain a three-dimensional model of the oral restoration.

[0023] The specific method for determining whether the stress path is stable is as follows: take the traversed hadron region as the current hadron region; take the weak region corresponding to the weak peak point of the weak subregion with the shortest Euclidean distance to the strong peak point of the current hadron region as the current mapped weak subregion. The strong peak point and strong valley point of the current hadron region are the strong conduction line; the weak peak point and weak valley point of the current mapped weak sub-region are the weak conduction line. If the strong and weak conduction lines are parallel, the stress path is considered stable; otherwise, the stress path is considered unstable.

[0024] The above method determines whether the stress path is affected based on the stress cloud diagrams corresponding to strong and weak loads. Since stress changes linearly under extreme occlusal loads, if the strong and weak conduction lines are parallel, the stress path is considered to be stable; otherwise, it is considered to have fluctuated. Such fluctuations can lead to difficulties in forming stable osseointegration at the bone-implant interface, and may even cause marginal bone resorption in the future. Therefore, the above method needs to identify the area of ​​the bone-implant interface affected by the stress path for subsequent correction to avoid fluctuations. However, when the stability of the stress path fluctuates, the stress is not linearly changing. Therefore, the strong and weak conduction lines located by the above method cannot accurately determine the stability of the stress path. Therefore, this application proposes the following preferred solution: Preferably, the specific method for determining whether the stress path is stable is as follows: take the traversed hadron region as the current hadron region; take the weak subregion corresponding to the weak peak point of the weak subregion with the shortest Euclidean distance from the strong peak point corresponding to the current hadron region as the current mapped weak subregion; and record the weak peak point corresponding to the current mapped weak subregion as the reference point. The hadron regions are sorted from closest to furthest according to the Euclidean distance of their corresponding dominant peak points from the reference point; the stress value of the dominant peak point corresponding to the current hadron region is denoted as Strong. i Strong i Let be the stress value of the dominant peak point corresponding to the i-th hadron region; i is the index of the current hadron region among all hadron regions. If Strong i-1 >Strong i And Strong i <Strong i+1If the strong peak point and strong valley point of the current hadron region are connected, then the strong transmission line is formed by connecting the weak peak point and weak valley point of the current mapped weak subregion; otherwise, calculate the transmission wave intensity of each hadron region, and connect the strong peak point and strong valley point of the hadron region with the smallest transmission wave intensity to form the strong transmission line, and connect the weak peak point and weak valley point of the current mapped weak subregion to form the weak transmission line. If the strong and weak conduction lines are parallel, the stress path is considered stable; otherwise, the stress path is considered unstable.

[0025] The method for calculating the propagation wave intensity in the hadron region is as follows: the hadron region whose stress propagation intensity is to be calculated is numbered j in each hadron region; The sum of the stress values ​​corresponding to the strong peak values ​​of all hadron regions from the current hadron region to the j-th hadron region is the cumulative stress transfer fluctuation value; the stress value of the reference point is used as the reference stress value. The stress transmission fluctuation intensity of the hadron region for which the stress transmission intensity is to be calculated is the ratio of the reference stress value to the cumulative value of stress transmission fluctuation.

[0026] Among them, the above method can replace the hadron region with the smallest transmitted fluctuation intensity obtained by screening the ratio of the cumulative sum to the individual points when the stress path is unstable and the stress is not changing linearly. This prevents the hadron region with stability fluctuation from being missed due to the nonlinear transmission of stress, and improves the accuracy of the stability fluctuation region screening.

[0027] Among them, the stress cloud diagram under strong stress is the stress cloud diagram when the meshing load is set to 300 N; the stress cloud diagram under weak stress is the stress cloud diagram when the meshing load is set to 40 N.

[0028] Preferably, the high-stress stress cloud map is a visual stress cloud map obtained when the peak stress load is applied; the low-stress stress cloud map is a visual stress cloud map obtained when a single stress load is applied.

[0029] Furthermore, methods for elastically correcting the position of the hadron region with unstable stress path on the temporary prosthesis region to obtain a three-dimensional model of the oral prosthesis include: The positions of the hadron regions with unstable stress paths on the temporary repair body region are elastically corrected sequentially. The specific method is as follows: The corresponding position of the hadron region with unstable stress path on the temporary repair body region is the elastic correction sub-region; The weak subregion corresponding to the weak peak point of the weak subregion with the shortest Euclidean distance to the strong peak point corresponding to the elastic correction subregion is the corresponding position of the weak subregion on the temporary repair body region. Let the wall thickness at the weakest peak point of the elastically modified weak subregion in the stress model be the peak modified wall thickness, and the wall thickness at the weakest valley point be the valley modified wall thickness; let the absolute value of the difference between the peak modified wall thickness and the valley modified wall thickness be the modified wall thickness difference. Increase the wall thickness difference of the elastic correction sub-region by adjusting the wall thickness difference; Obtain a three-dimensional model of the dental prosthesis.

[0030] The beneficial effects are as follows: by correcting the wall thickness difference of the weak subregion with corresponding elastic correction, the wall thickness of the hadron region that is unstable due to stress path caused by "overall stress shielding" is corrected, thereby reducing the problem of "overall stress shielding" turning into "local stress peak" and making stress transmission uniform.

[0031] The beneficial effects of this invention are as follows: This invention provides a personalized intelligent design method for dental prostheses based on generative adversarial networks. By correcting the wall thickness difference of the corresponding elastically modified weak subregions, the wall thickness of the strong subregions that are unstable due to "overall stress shielding" is corrected. This reduces the problem of "local stress peak" from "overall stress shielding", making the stress transmission uniform and the stability fluctuation caused by the change of stress path over time. Attached Figure Description

[0032] The above and other features of the present invention will become more apparent from the detailed description of the embodiments shown in conjunction with the accompanying drawings. In the accompanying drawings, the same reference numerals denote the same or similar elements. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort. In the drawings: Figure 1 The diagram shows a flowchart of a personalized intelligent design method for dental prostheses based on generative adversarial networks. Detailed Implementation

[0033] The following will provide a clear and complete description of the concept, specific structure, and technical effects of the present invention in conjunction with the embodiments and accompanying drawings, so as to fully understand the purpose, solution, and effects of the present invention. It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other.

[0034] like Figure 1 The diagram shows a flowchart of a personalized intelligent design method for dental prostheses based on generative adversarial networks. The following section will combine... Figure 1This paper describes a personalized intelligent design method for dental prostheses based on generative adversarial networks according to an embodiment of the present invention. The method includes the following steps: Example 1: S100: The patient's oral cavity is imaged in three dimensions and reconstructed in three dimensions to obtain a three-dimensional model of the patient wearing a temporary prosthesis. S200: The patient's oral cavity interior is imaged in three dimensions and reconstructed in three dimensions to obtain an empty three-dimensional model when the temporary prosthesis is removed; S300 spatially registers the worn 3D model and the unloaded 3D model, resulting in a worn registration model and an unloaded registration model, respectively. S400, perform a Boolean operation to subtract the worn registration model and the unloaded registration model to obtain the difference set, and mark the difference set that is connected to the temporary prosthesis area in the worn 3D model as the receiving model; S500, the Boolean operation that combines the temporary repair area with the receiving model yields the combined part, which is denoted as the stress model; S600 performs finite element analysis on the stress model to obtain a stress cloud map, and identifies the stress-shielding area on the stress model based on the stress cloud map. S700 uses elastic correction of the stress-shielding area of ​​the temporary restoration to obtain a three-dimensional model of the oral restoration.

[0035] Preferably, in S700, the method for elastically correcting the stress-shading area of ​​the temporary prosthesis region to obtain a three-dimensional model of the oral prosthesis is as follows: the stress-shading area of ​​the temporary prosthesis region is reconstructed according to the trained generative adversarial network (GAN) to obtain a three-dimensional model of the oral prosthesis.

[0036] Preferably, the method for reconstructing the three-dimensional model of the oral prosthesis from the stress-masking region of the temporary prosthesis area using a trained generative adversarial network (GAN) is as follows: Form a paired dataset {original model, stress occlusion label, target model}; Align the 3D model and stress shielding labels of the temporary restoration area to ensure spatial consistency; The 3D model of the temporary restoration area was converted into a voxel mesh (resolution 0.1 mm³). Building a Generative Adversarial Network (GAN): Includes: generator and discriminator: Generator (G): 3D U-Net, input is the original model + stress occlusion label, output is the reconstructed model; Discriminator (D): 3D CNN, input is real / generative model, output is real / false probability; Training a Generative Adversarial Network (GAN): Pre-training G uses only L_L1 (100 rounds); Jointly train G and D, gradually adding L_stress (learning rate 1e-4, batch size 8); Input the patient's temporary prosthesis model and stress occlusion mask into the trained generative adversarial network (GAN); A three-dimensional model of an oral prosthesis is generated using a generative adversarial network (GAN).

[0037] Furthermore, in S100, three-dimensional imaging of the inside of the patient's mouth when wearing a temporary prosthesis is performed by performing a three-dimensional imaging scan of the inside of the patient's mouth when wearing a temporary prosthesis using oral CBCT, and obtaining a three-dimensional model of the patient wearing the prosthesis through three-dimensional reconstruction.

[0038] Furthermore, in S200, three-dimensional imaging of the inside of the patient's mouth when the temporary prosthesis is removed is performed by performing a three-dimensional imaging scan of the inside of the patient's mouth when the temporary prosthesis is removed using oral CBCT, and a three-dimensional model of the patient wearing the prosthesis is obtained through three-dimensional reconstruction.

[0039] Preferably, the three-dimensional reconstruction method is to obtain a three-dimensional image by performing three-dimensional reconstruction using the three-dimensional reconstruction function of Mimics software.

[0040] Preferably, the oral CBCT is a three-in-one oral CBCT machine.

[0041] Furthermore, in S300, the spatial registration method is the SIFT matching method.

[0042] Furthermore, in S400, the method for obtaining the difference set by subtracting the worn registration model and the unloaded registration model using a Boolean operation is as follows: subtract the part that overlaps with the unloaded registration model from the worn registration model to obtain the difference set.

[0043] Furthermore, in S500, the method for performing a Boolean operation to combine the temporary restoration area and the receiving model to obtain the combined part, denoted as the stress model, is as follows: the two three-dimensional models, the temporary restoration area and the receiving model, are merged, and the intersecting parts are deleted, thereby merging into a combined part, denoted as the stress model.

[0044] Furthermore, in S600, the method for obtaining stress contour maps by performing finite element analysis on the stress model includes: obtaining stress contour maps by performing finite element analysis on the stress model using the open-source nonlinear FEBio / FEBio Studio finite element analysis software package.

[0045] Specifically, methods for obtaining stress contour plots through finite element analysis include: Import the 3D model of the force model; Define material properties: Dental prosthesis material (zirconia), E (Young's modulus) = 200 GPa (zirconia), ν (Poisson's ratio) = 0.3; Create a finite element mesh: Set the global finite element size (0.5 mm, Jacobian matrix > 0.7); Set boundary conditions and loads: Fixed constraint: contact surface between the bottom of the restoration and adjacent teeth; Occlusal load: occlusal force of 300 N on the occlusal surface; Contact surface setup: sliding friction between the dental prosthesis and alveolar bone, and between the prosthesis and the opposing tooth, with a friction coefficient µ=0.2~0.5 (typical value between dental materials). Configure the solution control parameters: (1) Analysis step configuration: total time = 1, number of steps = 100 (quasi-static analysis); (2) Solver parameter configuration: Perform large deformation analysis using the Newton-Raphson iterative method and adjust the convergence tolerance (e.g., convergence tolerance rtol=4); Generates a visual stress cloud map, including: comprehensive stress assessment and tensile / compressive stress separation display.

[0046] Among them, the visual stress cloud map under the applied peak stress load is used as the stress cloud map.

[0047] Furthermore, in S600, the method for identifying stress-obstructed regions on the stress model based on the stress cloud map is as follows: the stress cloud map is grayscaled and then segmented using the watershed algorithm to obtain multiple sub-regions; sub-regions whose maximum stress value (the maximum stress value borne by all points in the sub-region) is less than the low stress threshold are marked as stress-obstructed regions (the low stress threshold is set to 5MPa).

[0048] Preferably, in S600, the method for identifying stress-shielding areas on the stress model based on the stress cloud diagram is as follows: The stress cloud map is converted to grayscale and then segmented using the watershed algorithm to obtain multiple sub-regions. The average grayscale of each sub-region is taken as the low-stress grayscale. Sub-regions where the average grayscale of all points in each sub-region is greater than the low-stress grayscale are defined as stress occlusion regions.

[0049] Among them, the stress cloud diagram under strong stress is the stress cloud diagram when the meshing load is set to 300 N; the stress cloud diagram under weak stress is the stress cloud diagram when the meshing load is set to 40 N.

[0050] Preferably, the high-stress stress cloud map is a visual stress cloud map obtained when the peak stress load is applied; the low-stress stress cloud map is a visual stress cloud map obtained when a single stress load is applied.

[0051] Furthermore, methods for elastically correcting the position of the hadron region with unstable stress path on the temporary prosthesis region to obtain a three-dimensional model of the oral prosthesis include: The positions of the hadron regions with unstable stress paths on the temporary repair body region are elastically corrected sequentially. The specific method is as follows: The corresponding position of the hadron region with unstable stress path on the temporary repair body region is the elastic correction sub-region; The weak subregion corresponding to the weak peak point of the weak subregion with the shortest Euclidean distance to the strong peak point corresponding to the elastic correction subregion is the corresponding position of the weak subregion on the temporary repair body region. Let the wall thickness at the weakest peak point of the elastically modified weak subregion in the stress model be the peak modified wall thickness, and the wall thickness at the weakest valley point be the valley modified wall thickness; let the absolute value of the difference between the peak modified wall thickness and the valley modified wall thickness be the modified wall thickness difference. Increase the wall thickness difference of the elastic correction sub-region by adjusting the wall thickness difference; Obtain a three-dimensional model of the dental prosthesis.

[0052] Preferably, it also includes importing the three-dimensional model of the dental prosthesis into 3D printing layout and slicing software.

[0053] Based on the 3D printing process parameter library, the dental prostheses are classified, pre-annotated, and locally optimized. The local optimization includes posture adjustment and support addition. The components of the classified, annotated, and locally optimized dental prostheses are sliced ​​into layers, and the scanning path and scanning process parameters of each layer are planned. The process code for generating the layered slicing of the 3D model of the dental prosthesis is then input into the control software of the 3D printer, and the dental prosthesis is printed according to the layered slicing code.

[0054] Example 2: Example 2 replaces the method of elastically correcting the stress shielding area of ​​the temporary prosthesis region to obtain a three-dimensional model of the oral prosthesis, based on Example 1, with the following method: Preferably, in S700, the method for obtaining a three-dimensional model of the oral prosthesis by elastically correcting the stress-shielding area of ​​the temporary prosthesis region is as follows: The stress cloud maps under strong and weak stresses are converted to grayscale and then segmented using the watershed algorithm to obtain multiple sub-regions. The sub-regions inside the stress occlusion area in the strong stress cloud map are denoted as strong sub-regions, and the sub-regions inside the stress occlusion area in the weak stress cloud map are denoted as weak sub-regions. The point with the highest stress value in each weak subregion is mapped to the position on the force model as the weak peak point, and the point with the lowest stress value is mapped to the position on the force model as the weak valley point; the point with the highest stress value in each hadron region is mapped to the position on the force model as the strong peak point, and the point with the lowest stress value is mapped to the position on the force model as the strong valley point. Iterate through all hadron regions and determine whether the stress paths at their positions on the stress model are stable. The position of the hadron region with unstable stress path on the temporary restoration region is elastically corrected to obtain a three-dimensional model of the oral restoration.

[0055] The specific method for determining whether the stress path is stable is as follows: take the traversed hadron region as the current hadron region; take the weak region corresponding to the weak peak point of the weak subregion with the shortest Euclidean distance to the strong peak point of the current hadron region as the current mapped weak subregion. The strong peak point and strong valley point of the current hadron region are the strong conduction line; the weak peak point and weak valley point of the current mapped weak sub-region are the weak conduction line. If the strong and weak conduction lines are parallel, the stress path is considered stable; otherwise, the stress path is considered unstable.

[0056] Example 3: Example 3, based on Example 2, replaces the specific method for determining whether the stress path is stable with the following method: Preferably, the specific method for determining whether the stress path is stable is as follows: take the traversed hadron region as the current hadron region; take the weak subregion corresponding to the weak peak point of the weak subregion with the shortest Euclidean distance from the strong peak point corresponding to the current hadron region as the current mapped weak subregion; and record the weak peak point corresponding to the current mapped weak subregion as the reference point. The hadron regions are sorted from closest to furthest according to the Euclidean distance of their corresponding dominant peak points from the reference point; the stress value of the dominant peak point corresponding to the current hadron region is denoted as Strong. i Strong i Let be the stress value of the dominant peak point corresponding to the i-th hadron region; i is the index of the current hadron region among all hadron regions. If Strong i-1 >Strong i And Strong i <Strong i+1If the strong peak point and strong valley point of the current hadron region are connected, then the strong transmission line is formed by connecting the weak peak point and weak valley point of the current mapped weak subregion; otherwise, calculate the transmission wave intensity of each hadron region, and connect the strong peak point and strong valley point of the hadron region with the smallest transmission wave intensity to form the strong transmission line, and connect the weak peak point and weak valley point of the current mapped weak subregion to form the weak transmission line. If the strong and weak conduction lines are parallel, the stress path is considered stable; otherwise, the stress path is considered unstable.

[0057] The method for calculating the propagation wave intensity in the hadron region is as follows: the hadron region whose stress propagation intensity is to be calculated is numbered j in each hadron region; The sum of the stress values ​​corresponding to the strong peak values ​​of all hadron regions from the current hadron region to the j-th hadron region is the cumulative stress transfer fluctuation value; the stress value of the reference point is used as the reference stress value. The stress transmission fluctuation intensity of the hadron region for which the stress transmission intensity is to be calculated is the ratio of the reference stress value to the cumulative value of stress transmission fluctuation.

[0058] Comparative example: A method for designing dental prostheses includes the following steps: The digital information of the patient's intraoral dentition and implant sites when wearing temporary prostheses is processed to obtain a dentition model; The three-dimensional morphological information of the temporary prosthesis obtained after the patient's body was removed was processed to obtain the prosthesis model; The dental arch model and the restoration model are spatially registered to obtain a registration model; the registration model includes the registered dental arch model and the restoration model. The temporary restoration data in the dentition model of the registration model is removed, and the dentition model and the restoration model with the temporary restoration data removed are merged to obtain a fused model; the fused model includes: dentition, implant site information and three-dimensional morphological information of the transgingival contour.

[0059] Spatial registration of the dental arch model and the prosthesis model to obtain a registration model specifically includes: data conversion of the dental arch model and the prosthesis model to obtain the data-converted dental arch model and oral prosthesis model; Obtain the common features of the dental arch model and the restoration model after data conversion; Using surface registration algorithms and common features, the spatial coordinate systems of the dental arch model and the restoration model after data transformation are aligned to obtain a registration model.

[0060] The oral prosthesis model includes the shape of the crown portion and the perforated portion of the temporary prosthesis.

[0061] The remaining details of the comparative example are as described in Example 1 of Chinese Patent Publication No. CN118351154A; The three-dimensional models of the oral prostheses obtained from Examples 1, 2, 3, and the comparative design were imported into the 3D printing layout and slicing software. The oral prostheses were classified and pre-annotated and locally optimized according to the 3D printing process parameter library. The local optimization included posture adjustment and support addition. The components of the classified and locally optimized oral prostheses were sliced ​​layer by layer, and the scanning path and scanning process parameters of each layer were planned. The process code for generating the layered slicing of the 3D model of the dental prosthesis is then input into the control software of the 3D printer, and the dental prosthesis is printed according to the layered slicing code.

[0062] Specific methods for printing dental prostheses: Take 50g of nano 3Y zirconium oxide powder, 10g of photosensitive resin, and 50g of deionized water; Add the prepared raw materials and water to the ball mill according to the ratio of raw materials to zirconia hollow balls = 1:1, grind for 12 hours, and discharge to obtain a slurry; stir and age the ground slurry at a temperature of 25°C for 24 hours; dehydrate the slurry to a moisture content of 25%; The layered slicing code of the three-dimensional model of the dental prosthesis is input into the control software of the 3D printer and into the 3D printer terminal. The dehydrated slurry is used as raw material and printed into shape by an inkjet 3D ceramic printer. The printed preform was dried at room temperature (25°C) for 36 hours. The resulting raw blank is white. It is immersed in a dyeing solution containing iron oxide for 30 seconds and then dried under a heat lamp. The stained dental prosthesis blank is sintered in a sintering furnace at a heating rate of 6℃ / min, with a maximum temperature of 1500℃ and a holding time of 2.5 hours. After glazing, it is sintered in a porcelain furnace to obtain the final dental prosthesis.

[0063] The dental prostheses obtained in Examples 1, 2, 3, and the comparative examples were tested using the Instron universal testing machine and the Instron electronic dynamic and static fatigue testing machine. Test conditions: Static compression test: Vertical loading to 800N, record peak stress and fracture load. Cyclic fatigue test: 50N~400N, 5×10 5 (Frequency 1.5Hz, simulating 5 years of chewing).

[0064] Experimental results data Example 1: Fracture load 715N, maximum peak stress 285MPa, stress uniformity 42±3MPa; displacement stability 15±2 (μm / 10) 5 Second-rate); Example 2: Fracture load 708N, maximum peak stress 275MPa, stress uniformity 38±3MPa; displacement stability 12±1 (μm / 10) 5 Second-rate); Example 3: Fracture load 723N, maximum peak stress 291MPa, stress uniformity 32±1MPa; displacement stability 10±1 (μm / 10 5 Second-rate); Comparative example: Fracture load 684 N, maximum peak stress 322 MPa, stress uniformity 58 ± 7 MPa; displacement stability 22 ± 3 (μm / 10) 5 Second-rate); Among them, stress uniformity is the value of stress cloud map distribution evaluated by standard deviation (SD); maximum stress peak is the maximum stress peak obtained by FEA finite element analysis combined with strain gauges (attached to the surface of the prosthesis); displacement stability is the amount of subsidence of the oral prosthesis under cyclic loading (unit: μm).

[0065] Analysis: Examples 1-3 significantly reduced peak stress and improved stress uniformity compared to the comparative examples. Examples 1-3 also showed reduced displacement fluctuations compared to the comparative examples, demonstrating that the method of this application can effectively suppress stress path drift. Among these, Example 3 performed best due to its improved accuracy in screening for stability fluctuation regions, resulting in further comprehensive stress dispersion in the product.

[0066] Although the invention has been described in considerable detail and particularly with regard to several of the described embodiments, it is not intended to limit itself to any of these details or embodiments or any particular embodiment, thereby effectively covering the intended scope of the invention. Furthermore, the invention has been described above with respect to embodiments foreseeable by the inventors in order to provide a useful description, and non-substantial modifications to the invention that have not yet been foreseen may still represent equivalent modifications.

Claims

1. A personalized intelligent design method for dental prostheses based on generative adversarial networks, characterized in that, The method includes the following steps: S100: The patient's oral cavity is imaged in three dimensions and reconstructed in three dimensions to obtain a three-dimensional model of the patient wearing a temporary prosthesis. S200: The patient's oral cavity interior is imaged in three dimensions and reconstructed in three dimensions to obtain an empty three-dimensional model when the temporary prosthesis is removed; S300 spatially registers the worn 3D model and the unloaded 3D model, resulting in a worn registration model and an unloaded registration model, respectively. S400, perform a Boolean operation to subtract the worn registration model and the unloaded registration model to obtain the difference set, and mark the difference set that is connected to the temporary prosthesis area in the worn 3D model as the receiving model; S500, the Boolean operation that combines the temporary repair area with the receiving model yields the combined part, which is denoted as the stress model; S600 performs finite element analysis on the stress model to obtain a stress cloud map, and identifies the stress-shielding area on the stress model based on the stress cloud map. S700 uses elastic correction of the stress-shielding area of ​​the temporary restoration to obtain a three-dimensional model of the oral restoration.

2. The intelligent design method for personalized dental prostheses based on generative adversarial networks according to claim 1, characterized in that, In S700, the method for elastically correcting the stress-shading area of ​​the temporary prosthesis region to obtain a three-dimensional model of the oral prosthesis is as follows: the stress-shading area of ​​the temporary prosthesis region is reconstructed using a trained generative adversarial network (GAN) to obtain a three-dimensional model of the oral prosthesis.

3. The intelligent design method for personalized dental prostheses based on generative adversarial networks according to claim 1, characterized in that, In S400, the method for obtaining the difference set by subtracting the worn registration model and the unloaded registration model using a Boolean operation is as follows: subtract the part that overlaps with the unloaded registration model from the worn registration model to obtain the difference set.

4. The intelligent design method for personalized dental prostheses based on generative adversarial networks according to claim 1, characterized in that, In S500, the method for performing a Boolean operation to combine the temporary restoration area and the receiving model to obtain the combined part, denoted as the stress model, is as follows: the two three-dimensional models, the temporary restoration area and the receiving model, are merged, and the intersecting parts are deleted, thus merging into a combined part, denoted as the stress model.

5. The intelligent design method for personalized dental prostheses based on generative adversarial networks according to claim 1, characterized in that, In S600, the method for identifying stress-occupying regions on the stress model based on stress cloud maps is as follows: the stress cloud map is converted to grayscale and then segmented using a watershed algorithm to obtain multiple sub-regions; the average grayscale of each sub-region is taken as the low-stress grayscale; and the sub-regions where the average grayscale of all points in each sub-region is greater than the low-stress grayscale are defined as stress-occupying regions.

6. The intelligent design method for personalized dental prostheses based on generative adversarial networks according to claim 1, characterized in that, In S700, the method for obtaining a three-dimensional model of the oral prosthesis by elastically correcting the stress-shielding area of ​​the temporary prosthesis region is as follows: The stress cloud maps under strong and weak stresses are converted to grayscale and then segmented using the watershed algorithm to obtain multiple sub-regions. The sub-regions inside the stress occlusion area in the strong stress cloud map are denoted as strong sub-regions, and the sub-regions inside the stress occlusion area in the weak stress cloud map are denoted as weak sub-regions. The point with the highest stress value in each weak subregion is mapped to the position on the force model as the weak peak point, and the point with the lowest stress value is mapped to the position on the force model as the weak valley point; the point with the highest stress value in each hadron region is mapped to the position on the force model as the strong peak point, and the point with the lowest stress value is mapped to the position on the force model as the strong valley point. Iterate through all hadron regions and determine whether the stress paths at their positions on the stress model are stable. The position of the hadron region with unstable stress path on the temporary restoration region is elastically corrected to obtain a three-dimensional model of the oral restoration.

7. The intelligent design method for personalized dental prostheses based on generative adversarial networks according to claim 6, characterized in that, The specific method for determining whether the stress path is stable is as follows: take the traversed hadron region as the current hadron region; take the weak subregion corresponding to the weak peak point of the weak subregion with the shortest Euclidean distance from the strong peak point of the current hadron region as the current mapped weak subregion; connect the strong peak point and the strong valley point of the current hadron region as the strong conduction line; connect the weak peak point and the weak valley point of the current mapped weak subregion as the weak conduction line. If the strong and weak conduction lines are parallel, the stress path is considered stable; otherwise, the stress path is considered unstable.

8. The intelligent design method for personalized dental prostheses based on generative adversarial networks according to claim 7, characterized in that, The method for determining whether the stress path is stable is replaced as follows: take the traversed hadron region as the current hadron region; take the weakest peak point corresponding to the weakest peak point of the weakest subregion with the shortest Euclidean distance from the strong peak point of the current hadron region as the current mapped weakest subregion; and denote the weakest peak point corresponding to the current mapped weakest subregion as the reference point. The hadron regions are sorted from closest to furthest according to the Euclidean distance of their corresponding dominant peak points from the reference point; the stress value of the dominant peak point corresponding to the current hadron region is denoted as Strong. i Strong i Let be the stress value at the dominant peak point corresponding to the i-th hadron region; i is the index of the current hadron region within the various hadron regions; If Strong i-1 >Strong i And Strong i <Strong i+1 If the strong peak point and strong valley point of the current hadron region are connected, then the strong transmission line is formed by connecting the weak peak point and weak valley point of the current mapped weak subregion; otherwise, calculate the transmission wave intensity of each hadron region, and connect the strong peak point and strong valley point of the hadron region with the smallest transmission wave intensity to form the strong transmission line, and connect the weak peak point and weak valley point of the current mapped weak subregion to form the weak transmission line. If the strong and weak conduction lines are parallel, the stress path is considered stable; otherwise, the stress path is considered unstable.

9. The intelligent design method for personalized dental prostheses based on generative adversarial networks according to claim 8, characterized in that, The method for calculating the propagation wave intensity in the hadron region is as follows: the hadron region whose stress propagation intensity is to be calculated is indexed as j in each hadron region; The sum of the stress values ​​corresponding to the strong peak values ​​of all hadron regions from the current hadron region to the j-th hadron region is called the cumulative stress transfer fluctuation value. The stress value at the reference point is used as the reference stress value; The stress transmission fluctuation intensity of the hadron region for which the stress transmission intensity is to be calculated is the ratio of the reference stress value to the cumulative value of stress transmission fluctuation.

10. The intelligent design method for personalized dental prostheses based on generative adversarial networks according to claim 6, characterized in that, Methods for obtaining a three-dimensional model of the oral prosthesis by elastically correcting the position of the hadron region with unstable stress path on the temporary prosthesis region include: The positions of the hadron regions with unstable stress paths on the temporary repair body region are elastically corrected sequentially. The specific method is as follows: The corresponding position of the hadron region with unstable stress path on the temporary repair body region is the elastic correction sub-region; The weak subregion corresponding to the weak peak point of the weak subregion with the shortest Euclidean distance to the strong peak point corresponding to the elastic correction subregion is the corresponding position of the weak subregion on the temporary repair body region. Let the wall thickness at the weakest peak point of the elastically modified weak subregion in the stress model be the peak modified wall thickness, and the wall thickness at the weakest valley point be the valley modified wall thickness; let the absolute value of the difference between the peak modified wall thickness and the valley modified wall thickness be the modified wall thickness difference. Increase the wall thickness difference of the elastic correction sub-region by adjusting the wall thickness difference; Obtain a three-dimensional model of the dental prosthesis.

Images