A Digital Design Method for Dental Prostheses Based on AI

Abstract

This invention discloses an AI-based digital design method for dental prostheses, comprising: simultaneously acquiring three-dimensional point cloud data of the patient's oral surface, three-dimensional data of skeletal structure, and dynamic occlusal motion trajectory and force data; fusing the three-dimensional point cloud data of the oral surface, the three-dimensional data of skeletal structure, and the dynamic occlusal motion trajectory and force data through a deep learning-based multi-stage registration algorithm to generate a comprehensive three-dimensional model of oral soft and hard tissues and dynamic occlusion with static geometric accuracy and dynamic occlusal function; based on the comprehensive three-dimensional model, performing key point identification, automatic coordinate system alignment, and key feature extraction of the point cloud from the three-dimensional dental model through an AI intelligent network. This invention achieves breakthrough improvements in static accuracy, dynamic function, and clinical application through multi-source data AI fusion, biomimetic structural design, and mechanical closed-loop optimization, providing a solution for digital design of dental prostheses.

Description

Technical Field

[0001] This invention relates to the field of dental prosthesis technology, and in particular to a digital design method for dental prostheses based on AI intelligence. Background Technology

[0002] Current removable partial denture manufacturing mainly relies on the traditional route of "alginate impression - plaster casting - hand wax model". The static accuracy of the model can only reach 0.1mm, and it completely ignores the patient's soft tissue thickness and elasticity, individual jawbone density, nerve canal orientation and dynamic occlusal trajectory, resulting in a high clinical revision rate.

[0003] While intraoral scanning combined with 3D printing has emerged in recent years as a digital solution to replace physical impressions, it still has significant drawbacks. Because soft tissues such as the gums, hard tissues of the teeth and jawbone, and dynamic occlusion data are heterogeneous data of different dimensions, current technologies lack effective multi-source fusion mechanisms. This leads to large cumulative errors during registration, making it difficult to generate a comprehensive model that combines static geometric accuracy with dynamic occlusion function. Furthermore, the digital design process heavily relies on technicians manually drawing denture boundaries; software cannot automatically identify key points and extract precise segmentation boundaries based on three-dimensional dental tissue, often resulting in insufficient placement space or invasion of physiological structures. In addition, denture fabrication still uses homogeneous materials and a single structural design; the denture base, artificial teeth, and connectors lack biomimetic mechanical transitions, making it difficult to match the complex stress environment within the oral cavity.

[0004] Therefore, there is an urgent need for a digital method that can simultaneously process multi-source heterogeneous data, automatically define design boundaries, and realize intelligent design of biomimetic structures to solve the problems of low precision, poor fit, and easy damage of existing dentures. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a digital design method for dental prostheses based on AI intelligence.

[0006] To achieve the above objectives, the present invention provides an AI-based digital design method for dental prostheses, S1: Simultaneously collecting three-dimensional point cloud data of the patient's oral surface, three-dimensional data of bone tissue structure, and dynamic occlusal motion trajectory and force value data to obtain basic design parameter data for the prosthesis, and performing data preprocessing on the basic design parameter data for the prosthesis. S2: The preprocessed denture basic design parameter data are fused using a deep learning-based multi-stage registration algorithm to obtain a spatiotemporally aligned oral soft and hard tissue-dynamic occlusion dataset, and a comprehensive three-dimensional model with static geometric accuracy and dynamic occlusion function is reconstructed based on the dataset. S3: Based on the comprehensive 3D model, key point identification, automatic coordinate system alignment and key feature extraction of point cloud are performed on the patient's 3D dental and jaw tissue, and the point cloud classification prediction results are output to form the denture segmentation boundary prediction. S4: Based on the segmentation boundary prediction results, perform three-dimensional intelligent modeling and filling of the denture to generate an initial three-dimensional denture model. Optimize the initial three-dimensional denture model through a segmentation refinement algorithm. Combined with biomimetic structural parametric modeling, generate a three-dimensional denture model with biomimetic structural features.

[0007] Furthermore, the data accuracy of the integrated 3D model is evaluated through two dimensions: geometric deviation and dynamic accuracy. Geometric deviation is calculated by measuring the average distance error between the integrated 3D model and the high-precision oral reference point cloud data acquired preoperatively. An assessment was conducted, in which... , where N is the number of sampling points ≥ 1000, evenly distributed in the edentulous area and adjacent teeth; , , These are the three-dimensional coordinates of the sampling points in the model; The coordinates of the actual oral cavity sampling points obtained by a coordinate measuring machine are given, with an accuracy of 0.001 mm. 'i' is the index, representing the coordinates of the i-th sampling point, 'm' represents the model value, and 't' represents the actual measured value. , , () represents the three-dimensional coordinates of the i-th sampling point in the model, and the corresponding... The superscript 't' in the text represents the actual measured value, that is, the coordinates of the actual oral sampling point obtained by the coordinate measuring machine. It is required that... ≤0.005mm; Dynamic accuracy is evaluated by calculating the degree of overlap, R, between the bite trajectory simulated by the integrated three-dimensional model and the dynamic bite movement trajectory actually collected from the patient. ,in: The overlap length between the simulated bite trajectory of the 3D model and the actual collected trajectory; The total length of the actual bite trajectory is required to be R≥98%.

[0008] Furthermore, the three-dimensional point cloud data of the oral cavity surface is acquired by an intraoral scanner with a spatial resolution of 0.01 mm. The three-dimensional data of the skeletal tissue structure was acquired by cone-beam CT scanning. The three-dimensional data of the skeletal tissue structure includes alveolar bone height, bone density data and nerve canal location data. The interslice spacing of the tomographic image reconstructed after cone-beam CT scanning is 0.1 mm. The dynamic occlusal motion trajectory and force data are collected by an intraoral dynamic occlusal recorder with a sampling frequency of 1000Hz, covering three motion states: centric occlusion, protruding occlusion, and lateral occlusion.

[0009] Furthermore, data preprocessing is performed on the aforementioned denture basic design parameter data, including the following steps: A global reference coordinate system is established based on the three-dimensional point cloud data of the oral cavity surface. Rigid body transformation registration is performed on the cone-beam CT data, and the dynamic occlusal motion trajectory and force data are resampled according to the sampling timestamp to unify the spatial scale and temporal dimension of multi-source heterogeneous data. For the three-dimensional point cloud data of the oral cavity surface, normal consistency filtering is used to remove flying points, and Gaussian filtering is used for smoothing and noise reduction. For the three-dimensional data of bone tissue structure, median filtering is performed based on the HU threshold range to suppress ring artifacts. For the dynamic occlusal motion trajectory and force data, sliding window moving average filtering is used to eliminate instantaneous interference force values. Based on prior information about the anatomical structure of the edentulous area, non-target areas in the maxillary sinus, mandibular canal, and distal free soft tissue are automatically eliminated, retaining only the effective design domain data within the edentulous area and the range of the three adjacent teeth on each side.

[0010] Furthermore, in step S2, the multi-stage registration algorithm is used to achieve multi-source data fusion, and its processing flow includes: The point cloud density of the preprocessed oral cavity surface point cloud data was reduced to 500 points / mm². The voxel data obtained after the cone-beam CT device performs three-dimensional X-ray scanning of the patient's maxillofacial region is segmented by HU value thresholding. First, the bone voxels are picked out using the HU threshold, and then the MarchingCubes algorithm is used to directly generate triangular meshes on the bone surface, that is, to extract the three-dimensional model of the jawbone. Using the occlusal surface feature points of adjacent teeth as a reference, the preprocessed three-dimensional point cloud data of the oral cavity surface and the generated three-dimensional model of the jawbone are coarsely registered using the ICP algorithm, with an error ≤0.05mm; A deep learning model based on PointNet++ is used to perform fine registration of the three-dimensional point cloud data of the oral cavity surface after coarse registration with the three-dimensional model of the jawbone. The 128-dimensional feature vector is fused, and the key areas within the effective design domain are focused through the attention mechanism to achieve a registration error of ≤0.005mm. The preprocessed dynamic occlusal motion trajectory and force data are mapped to the finely registered 3D model through the timestamp to obtain a spatiotemporally aligned oral soft and hard tissue-dynamic occlusal dataset.

[0011] Further, step S3 includes the following steps: Based on the comprehensive three-dimensional model, anatomical landmarks of the cusps, central fossa and alveolar crest of the three-dimensional dental and maxillary tissues are identified, and a local coordinate system is constructed using the anatomical landmarks to automatically orient the three-dimensional dental and maxillary tissues in the coordinate system. On the three-dimensional dental and jaw tissue point cloud of the patient automatically aligned by the coordinate system, the dental arch curve and the key curve of the cervical margin are fitted, and the curvature, normal vector and neighborhood density are calculated point by point to extract 128-dimensional key feature vectors. The 128-dimensional key feature vector is input into a deep learning model based on the PointNet++ architecture, and the output is a point-by-point classification label, which includes at least the basement region, the artificial tooth region, and the connector region. Based on the point-by-point classification labels, the boundaries at the intersections of different category regions are extracted to generate denture segmentation boundary prediction results.

[0012] Further, step S4 includes: Based on the initial denture segmentation boundary and the denture segmentation boundary prediction result, the three-dimensional intelligent modeling of the denture is filled to generate the initial three-dimensional denture model. Based on the initial three-dimensional model of the denture, a graph cut algorithm is used to smooth the outer surface of the initial three-dimensional model of the denture to eliminate jagged edges. The point cloud classification labels are mapped to weighted graph nodes, and an energy function containing boundary smoothing and region consistency terms is defined. The optimal boundary is solved by the minimum cut / maximum flow algorithm to achieve high-precision fitting between the segmentation boundary and the natural tooth jaw anatomical interface. Based on the initial three-dimensional model of the denture after high-precision fitting, and combined with biomimetic structural parametric modeling, a three-dimensional model of the denture with biomimetic structural features is generated.

[0013] Furthermore, the three-dimensional intelligent modeling of the denture includes the denture base, the artificial tooth, and the connector. Based on the optimized denture basic model, biomimetic structural modeling is performed to obtain a three-dimensional denture model with biomimetic structural features, specifically including: The denture base structure was modeled, and a gradient pore structure design was adopted, defining the pore size as 50-100 μm near the mucosa, 200-300 μm in the middle layer, and 300-500 μm on the outer layer. Multi-layer biomimetic modeling was performed on the artificial tooth portion, and the pit and fissure structure of natural teeth was replicated on the crown surface. The dentin layer was designed with a gradient in elastic modulus, decreasing from 80 GPa on the enamel side to 20 GPa on the dentin side. The gradient formula is as follows: ,in It is a gradual function of the elastic modulus of the dentin layer of artificial teeth, representing the elastic modulus at a distance z from the enamel side. It is the distance from the enamel side to the dentin side. The thickness of the dentin layer. A biomimetic transition model was created for the connection between the denture base and the artificial tooth, employing a biomimetic tendon-bone connection structure. The elastic modulus gradually changes from 30 GPa at the denture base end to 60 GPa at the artificial tooth end, with a transition length of 3-5 mm. The transition formula is as follows: Where E(t) is the gradual function of the elastic modulus of the connected body, The distance from the base of the denture to the end of the artificial tooth. The length of the transition section. Based on the above parameters, a three-dimensional model of the denture with biomimetic structural features is generated.

[0014] Furthermore, the three-dimensional model of the denture is imported into the finite element simulation platform in STL format with a precision of 0.001mm, and the material properties of each part of the denture are configured: The base portion is made of PEEK-HA gradient composite material, with the hydroxyapatite (HA) mass fraction decreasing from 30% on the mucosal side to 10% on the outer side; the enamel layer of the artificial tooth portion is made of zirconia ceramic, and the dentin layer is made of PMMA-nano SiO2-HA composite resin; the connector portion is made of titanium alloy-PEEK composite structure, with the titanium alloy surface roughness set to Ra1.0-2.0μm; Three typical clinical dynamic loading conditions—centric occlusion, protruding occlusion, and lateral occlusion—were applied, and finite element mechanical calculations were performed in parallel to verify that the following three indicators were simultaneously satisfied: (a) Maximum equivalent stress σ of the Keetor max ≤80MPa; (b) Stress uniformity coefficient C of artificial tooth occlusal surface v ≤0.3, where C v =σ std / σ mean , σ mean σ represents the average stress at the interlocking joint surfaces. std (c) The stress concentration factor α of the connection body is ≤1.5, where α = σ peak / σ nom , σ peak σ is the maximum local stress of the connector. nom For the nominal stress in the far field; If any one of the indicators fails to meet the requirements, the parameter iteration correction process will be automatically triggered: adjust at least one of the following: the hierarchical pore diameter of the base plate gradient pores, the gradient slope of the elastic modulus of the artificial tooth, or the length of the transition section of the connector; regenerate the three-dimensional model of the denture and repeat the above simulation verification steps until all three indicators meet the requirements; use the finally qualified three-dimensional model of the denture as the direct digital data source for selective laser sintering, photopolymerization stereolithography, and laser cladding additive manufacturing.

[0015] Furthermore, the process of generating a three-dimensional model of a denture with biomimetic structural features by combining biomimetic structural parametric modeling specifically includes: Based on the three-dimensional data of the skeletal tissue structure in the comprehensive three-dimensional model, the density distribution characteristics of alveolar bone in the edentulous area are obtained. Based on the density distribution characteristics, the gradient porosity parameters of the denture base, the elastic modulus gradient parameters of the artificial tooth, and the transition stiffness parameters of the connector are determined. Based on the gradient porosity parameters, elastic modulus gradient parameters, and transition stiffness parameters, the internal structure of the initial three-dimensional denture model after high-precision fitting is reconstructed to generate a three-dimensional denture model with biomimetic structural features.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention, in the process of denture design, differs from existing technologies that only collect oral surface data or simply overlay CBCT data. This invention adds dynamic occlusal motion trajectory and force value data to the basic denture design parameter data, and based on a multi-stage registration algorithm of deep learning, integrates oral surface point clouds, jawbone tissue structure, and dynamic occlusal motion trajectory in the same spatiotemporal coordinate system to construct a comprehensive three-dimensional model that combines static geometric accuracy and dynamic occlusal function. Then, the comprehensive three-dimensional model is used to identify the denture segmentation boundary, thereby generating a three-dimensional denture model with a biomimetic structure. Because the factors considered in the construction of the three-dimensional denture model are relatively comprehensive, including both static geometric data and dynamic occlusal data, the occlusal interference caused by data fragmentation is fundamentally eliminated. Breakthroughs are achieved in accuracy, function, comfort, efficiency, and lifespan, and the risk of clinical revision is greatly reduced. Attached Figure Description

[0017] To more clearly illustrate the solutions in this invention, the accompanying drawings used in the description of the embodiments of this invention will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the process provided by the present invention; Figure 2 This is a schematic diagram of the multi-stage registration algorithm provided by the present invention. Detailed Implementation

[0019] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The terms "comprising" and "having," and any variations thereof, used in the specification, claims, and accompanying drawings of this invention are intended to cover non-exclusive inclusion. The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a particular order.

[0021] Please see Figure 1-2 This invention provides an AI-based digital design method for dental prostheses, comprising the following steps: S1: Simultaneously collect the three-dimensional point cloud data of the patient's oral cavity surface, the three-dimensional data of the bone tissue structure, and the dynamic occlusal motion trajectory and force data to obtain the basic design parameters data of the denture, and perform data preprocessing on the basic design parameters data of the denture. S2: The preprocessed denture basic design parameter data are fused using a deep learning-based multi-stage registration algorithm to obtain a spatiotemporally aligned oral soft and hard tissue-dynamic occlusion dataset, and a comprehensive three-dimensional model with static geometric accuracy and dynamic occlusion function is reconstructed based on the dataset. S3: Based on the comprehensive 3D model, key point identification, automatic coordinate system alignment and key feature extraction of point cloud are performed on the patient's 3D dental and jaw tissue, and the point cloud classification prediction results are output to form the denture segmentation boundary prediction. S4: Based on the segmentation boundary prediction results, perform three-dimensional intelligent modeling and filling of the denture to generate an initial three-dimensional denture model. Optimize the initial three-dimensional denture model through a segmentation refinement algorithm. Combined with biomimetic structural parametric modeling, generate a three-dimensional denture model with biomimetic structural features.

[0022] In the aforementioned technical solution, the three-dimensional point cloud data of the oral cavity surface is obtained by directly scanning the patient's oral cavity with an intraoral scanner. The spatial resolution of the scan is 0.01 mm, improving the accuracy of the data. This three-dimensional point cloud data of the oral cavity surface typically represents millions of tiny coordinate points (X, Y, Z), accurately recording the gingival morphology, the shape of the remaining teeth, the inclination angle of adjacent teeth, and the atrophy of the alveolar ridge in the edentulous area (such as a blade-shaped ridge or a low-lying ridge). The three-dimensional point cloud data of the oral cavity surface determines the arrangement of the prosthetic teeth and the edge end line (cervical margin) of the denture base. If the data is inaccurate, the denture base may compress the gums or cause food impaction due to poor sealing. Simultaneously, the tightness of contact between the denture and the adjacent teeth on both sides also affects food impaction and denture stability.

[0023] In the aforementioned technical solution, three-dimensional data of the skeletal structure is acquired through cone-beam computed tomography (CBCT). This data includes alveolar bone height, bone density, and nerve canal location. The interslice spacing of the reconstructed tomographic images from the CBCT scan is 0.1 mm. The alveolar bone height provides a basic reference dimension for denture design, adapting to the specific patient's oral environment. The nerve canal location data includes the position of the mandibular nerve canal and the height of the maxillary sinus floor. The design must avoid the mandibular nerve canal (inferior alveolar neurovascular bundle) and the maxillary sinus to prevent nerve damage or sinus penetration during surgery or when the denture is worn. Bone density data provides mechanical support and directly affects the extension range and thickness of the denture base. If there is severe alveolar bone resorption (low bone density) in the edentulous area, the saddle area of ​​the denture base needs to be enlarged during design to distribute occlusal forces and prevent further alveolar bone resorption.

[0024] In the above technical solution, dynamic occlusal motion trajectory and force data are collected using an intraoral dynamic occlusal recorder at a sampling frequency of 1000Hz, covering three motion states: centric occlusion, protruding occlusion, and lateral occlusion. This includes the mandibular opening and closing trajectory (sagittal, vertical, and lateral displacement), the protruding guidance path, and the contact force values ​​of each tooth position measured by the T-Scan system (e.g., the first molar bears 50N during chewing, and the incisors bear 10N). In areas with high force data (primary load-bearing areas), the thickness of the prosthesis connector and base needs to be increased, or a higher-strength biomimetic structure (such as internal reinforcing ribs) needs to be used to prevent breakage. The dynamic occlusal motion trajectory and force data can determine the cusp inclination and tooth arrangement curve shape of the artificial teeth, providing dynamic reference value.

[0025] The working principle of this invention lies in combining three reference factors—static morphology, dynamic function, and biomechanics—during the denture design process, breaking the limitation of traditional denture design which relies on only a portion of reference data. Specifically, it achieves automated semantic fusion and reconstruction of multi-source heterogeneous data through deep learning algorithms. First, a multi-stage registration algorithm is used to perform feature-level fusion of preprocessed oral surface point clouds, skeletal tissue structure data, and dynamic occlusal trajectories, generating a spatiotemporally aligned comprehensive 3D model, thus solving the problem of cumulative errors in spatial registration of data from different dimensions. Based on this comprehensive 3D model, 128-dimensional key feature vectors are extracted, and the PointNet++ deep learning model is used to perform point-by-point semantic segmentation of the 3D dental and jaw tissues, automatically identifying the denture base area, artificial tooth area, and connector area, thereby transforming disordered point cloud data into anatomically meaningful classification labels. Finally, based on the placement space determined by the classification labels, energy function optimization in the segmentation refinement algorithm achieves high-precision fitting between the denture geometry and the natural dental and jaw anatomical interface, and a solid model with gradient porosity and gradually changing elastic modulus is generated according to biomimetic structural parameterization rules.

[0026] Compared with existing technologies, this invention achieves precise coordination between oral soft tissues, hard tissues, and dynamic occlusal function through a multi-stage registration algorithm based on deep learning. This overcomes the shortcomings of traditional static models in simulating dynamic occlusal trajectories and significantly reduces the clinical revision rate. Simultaneously, it utilizes a deep learning model to automatically extract key point cloud features and perform classification prediction, replacing the inefficient manual boundary drawing method in traditional CAD design. This ensures the accuracy of the denture placement space definition and the consistency of design standards. Furthermore, by constructing gradient porosity and a gradually changing elastic modulus structure within the denture using biomimetic parametric modeling technology, it effectively solves the problems of denture base fracture and compression absorption of the remaining alveolar ridge caused by stress concentration in traditional homogeneous material dentures, significantly improving the biomechanical compatibility and long-term stability of the restoration.

[0027] Furthermore, in step S1, data preprocessing is performed on the denture basic design parameter data, including the following steps: A global reference coordinate system is established based on the three-dimensional point cloud data of the oral cavity surface. Rigid body transformation registration is performed on the cone-beam CT data, and the dynamic occlusal motion trajectory and force data are resampled according to the sampling timestamp to unify the spatial scale and temporal dimension of multi-source heterogeneous data. For the three-dimensional point cloud data of the oral cavity surface, normal consistency filtering is used to remove flying points, and Gaussian filtering is used for smoothing and noise reduction. For the three-dimensional data of bone tissue structure, median filtering is performed based on the HU threshold range to suppress ring artifacts. For the dynamic occlusal motion trajectory and force data, sliding window moving average filtering is used to eliminate instantaneous interference force values. Based on prior anatomical information of the edentulous area, non-target regions in the maxillary sinus, mandibular canal, and distal free soft tissue are automatically eliminated, retaining only the effective design domain data within the edentulous area and the three adjacent teeth on each side. By limiting the effective design domain to the edentulous area and the three adjacent teeth on each side, more than 90% of redundant data processing is reduced, significantly shortening model training and inference time.

[0028] The principle and technical effect of the above technical solution are as follows: By establishing a standardized preprocessing mechanism for constructing multi-source heterogeneous data, noise and dimensionality differences at the data acquisition end are eliminated, providing high-quality input for subsequent deep learning registration. First, by establishing a global reference coordinate system and resampling dynamic occlusal data according to the sampling timestamp, precise alignment of oral soft tissue, jawbone hard tissue, and occlusal motion trajectory in the temporal and spatial dimensions is achieved, solving the registration drift problem caused by the lack of unified spatiotemporal references for multi-source data. Second, differentiated filtering strategies are implemented for data sources with different physical characteristics: normal consistency filtering is used to remove floating noise in oral scan data, HU threshold and median filtering is used to suppress ring artifacts in CBCT data, and sliding window filtering is used to smooth instantaneous fluctuations in occlusal force, thereby maximizing the signal-to-noise ratio while preserving anatomical features and ensuring the authenticity of anatomical morphology in the comprehensive 3D model. Finally, based on prior information about the anatomical structure of the edentulous area, the data space is trimmed, eliminating non-target areas such as the maxillary sinus and mandibular canal. Computational resources are then focused on the effective design domain of the edentulous area and its adjacent teeth, significantly reducing the complexity and computational load of feature extraction in subsequent deep learning models. This technical solution, through preprocessing of the basic denture design parameter data, significantly improves the accuracy and efficiency of subsequent denture design.

[0029] Furthermore, in step S2, the multi-stage registration algorithm is used to achieve multi-source data fusion, and its processing flow includes: The point cloud density of the preprocessed oral cavity surface point cloud data was reduced to 500 points / mm². The voxel data obtained after the cone-beam CT device performs three-dimensional X-ray scanning of the patient's maxillofacial region is segmented by HU value thresholding. First, the bone voxels are picked out using the HU threshold, and then the MarchingCubes algorithm is used to directly generate triangular meshes on the bone surface, that is, to extract the three-dimensional model of the jawbone. Using the occlusal surface feature points of adjacent teeth as a reference, the preprocessed three-dimensional point cloud data of the oral cavity surface and the generated three-dimensional model of the jawbone are coarsely registered using the ICP algorithm, with an error ≤0.05mm; A deep learning model based on PointNet++ is used to perform fine registration of the three-dimensional point cloud data of the oral cavity surface after coarse registration with the three-dimensional model of the jawbone. The 128-dimensional feature vector is fused, and the key areas within the effective design domain are focused through the attention mechanism to achieve a registration error of ≤0.005mm. The preprocessed dynamic occlusal motion trajectory and force data are mapped to the finely registered 3D model through the timestamp to obtain a spatiotemporally aligned oral soft and hard tissue-dynamic occlusal dataset.

[0030] The principle and effect of the above technical solution are as follows: First, by reducing the density of the preprocessed oral surface point cloud, the data dimensionality is reduced while preserving anatomical features. Then, the Marching Cubes algorithm is used to extract a three-dimensional jawbone model from CBCT voxel data, establishing a baseline framework for rigid registration. Subsequently, using the occlusal surface feature points of adjacent teeth as anchor points, the ICP algorithm is used to complete the initial rigid alignment of soft and hard tissues. Then, a 128-dimensional feature vector is extracted using a PointNet++-based deep learning model for fine registration. An attention mechanism is used to focus on the effective design domain of the edentulous area and its adjacent teeth, capturing the subtle nonlinear mapping relationship between soft tissue deformation and bone structure, thereby improving the registration accuracy to the micrometer level. Finally, based on the timestamp mapping relationship established in the preprocessing stage, the dynamic occlusal motion trajectory and force value data are accurately superimposed onto the finely registered static model to construct a comprehensive dataset with spatiotemporal alignment.

[0031] This invention effectively addresses the shortcomings of traditional single-registration methods in cross-modal data fusion by employing a multi-stage registration algorithm, which suffers from insufficient accuracy or poor robustness. By combining a hybrid registration strategy of geometric feature points and deep learning feature vectors, this invention overcomes the interference of physiological displacement of oral soft tissues during chewing on registration accuracy, ensuring a high degree of fit between the denture base and the mucosal supporting tissues. Simultaneously, dynamic occlusal data is mapped to a static 3D model in real time, resulting in a comprehensive 3D model that not only includes static anatomical morphology but also fully reproduces the patient's personalized occlusal movement patterns and force distribution. This provides a high-fidelity digital twin foundation for subsequent biomimetic design and dynamic functional verification of the denture, fundamentally reducing the probability of clinical occlusal interference and prosthesis failure caused by model distortion.

[0032] Furthermore, the accuracy of the integrated 3D model is evaluated through two dimensions: geometric deviation and dynamic accuracy. Geometric deviation is calculated by measuring the average distance error between the integrated 3D model and the high-precision oral reference point cloud data acquired preoperatively. An assessment was conducted, in which... , where N is the number of sampling points ≥ 1000, evenly distributed in the edentulous area and adjacent teeth; , , These are the three-dimensional coordinates of the sampling points in the model; The coordinates of the actual oral cavity sampling points obtained by a coordinate measuring machine are given, with an accuracy of 0.001 mm. 'i' is the index, representing the coordinates of the i-th sampling point, 'm' represents the model value, and 't' represents the actual measured value. , , () represents the three-dimensional coordinates of the i-th sampling point in the model, and the corresponding... The superscript 't' in the text represents the actual measured value, that is, the coordinates of the actual oral sampling point obtained by the coordinate measuring machine. It is required that... ≤0.005mm; Dynamic accuracy is evaluated by calculating the degree of overlap, R, between the bite trajectory simulated by the integrated three-dimensional model and the dynamic bite movement trajectory actually collected from the patient. ,in: The overlap length between the simulated bite trajectory of the 3D model and the actual collected trajectory; The total length of the actual bite trajectory is required to be R≥98%.

[0033] The aforementioned technical solution constructs a dual-dimensional accuracy evaluation mechanism encompassing static geometry and dynamic function by calculating the average distance error between the integrated 3D model and the preoperatively acquired high-precision oral reference point cloud data. This ensures that the generated integrated 3D model can faithfully reproduce the actual state of the patient's oral cavity. In the static dimension, the geometric accuracy of the model is quantified by calculating the Euclidean distance error between the sampling points on the surface of the integrated 3D model and the actual oral reference point cloud. At least 1000 evenly distributed points in the edentulous area and adjacent tooth areas are selected, and measured data obtained using a high-precision coordinate measuring machine (CMM) are used as ground truth to calculate the average distance error. This indicator directly reflects the model's accuracy in replicating anatomical morphology at the micrometer scale. In the dynamic dimension, the overlap R between the simulated occlusal trajectory and the actual acquired trajectory is calculated to quantify the kinematic accuracy of the model. The motion path of the virtual model is spatiotemporally aligned with the dynamic occlusal trajectory acquired clinically, and the proportion of the overlap length to the total trajectory length is calculated. This indicator reflects the physiological compatibility of the model when simulating mandibular opening and closing, protrusion, and lateral movements, thereby ensuring that the data accuracy of the comprehensive three-dimensional model itself meets the requirements and provides a model basis for the subsequent design of dentures.

[0034] Furthermore, step S3 includes the following steps: Based on the comprehensive three-dimensional model, anatomical landmarks of the cusps, central fossa and alveolar crest of the three-dimensional dental and maxillary tissues are identified, and a local coordinate system is constructed using the anatomical landmarks to automatically orient the three-dimensional dental and maxillary tissues in the coordinate system. On the three-dimensional dental and jaw tissue point cloud of the patient automatically aligned by the coordinate system, the dental arch curve and the key curve of the cervical margin are fitted, and the curvature, normal vector and neighborhood density are calculated point by point to extract 128-dimensional key feature vectors. The 128-dimensional key feature vector is input into a deep learning model based on the PointNet++ architecture, and the output is a point-by-point classification label, which includes at least the basement region, the artificial tooth region, and the connector region. Based on the point-by-point classification labels, the boundaries at the intersections of different category regions are extracted to generate denture segmentation boundary prediction results.

[0035] The principle and effect of the above technical solution are as follows: First, based on a comprehensive 3D model, key anatomical landmarks such as cusps, central fossa, and alveolar ridge crest are identified, and a local coordinate system is constructed to achieve automatic coordinate system alignment, eliminating patient positional differences during imaging and establishing a standardized anatomical reference benchmark. Then, the dental arch curve and cervical margin are fitted onto the aligned point cloud, and geometric features such as curvature, normal vector, and neighborhood density are extracted point by point to form a 128-dimensional key feature vector. This vector not only contains spatial location information of the points but also contains local topological structural features. Finally, a deep learning model based on the PointNet++ architecture is used to perform semantic reasoning on these high-dimensional feature vectors, automatically distinguishing the denture base, artificial teeth, and connector regions. A boundary extraction algorithm transforms discrete classification labels into continuous geometric contours, thereby accurately defining the restorative space of the tooth.

[0036] Therefore, compared with existing denture design methods, this invention, by introducing anatomical landmark-driven automatic alignment and deep learning semantic segmentation, completely eliminates the reliance on technicians' manual boundary drawing experience in traditional CAD design. Utilizing 128-dimensional feature vectors to capture the microscopic geometric characteristics of the dentition and jaw tissues, combined with the powerful feature learning capabilities of the PointNet++ model, it can accurately identify complex edentulous morphology, ensuring a high degree of conformity between the denture segmentation boundary and the natural dentition and jaw anatomical interfaces (such as the cervical margin and alveolar ridge), effectively avoiding the jagged edges or intrusion into physiologically restricted areas that are prone to occur in boundary recognition by traditional algorithms. Simultaneously, the segmentation boundary prediction results generated based on point-by-point classification labels provide precise spatial constraints for subsequent three-dimensional intelligent modeling of the denture, ensuring the morphological accuracy and biomechanical rationality of the restoration, significantly improving the initial stability and long-term prognosis after denture placement.

[0037] Further, step S4 includes: Based on the initial denture segmentation boundary and the denture segmentation boundary prediction result, the three-dimensional intelligent modeling of the denture is filled to generate the initial three-dimensional denture model. Based on the initial three-dimensional model of the denture, a graph cut algorithm is used to smooth the outer surface of the initial three-dimensional model of the denture to eliminate jagged edges. The point cloud classification labels are mapped to weighted graph nodes, and an energy function containing boundary smoothing and region consistency terms is defined. The optimal boundary is solved by the minimum cut / maximum flow algorithm to achieve high-precision fitting between the segmentation boundary and the natural tooth jaw anatomical interface. Based on the initial three-dimensional model of the denture after high-precision fitting, and combined with biomimetic structural parametric modeling, a three-dimensional model of the denture with biomimetic structural features is generated.

[0038] The above technical solution firstly involves three-dimensional intelligent modeling and filling based on the initial predicted denture segmentation boundaries, transforming the two-dimensional planar defined restoration space into a closed three-dimensional solid prototype. Subsequently, a graph cut algorithm is used to transform the optimization problem of the three-dimensional model's outer surface into an energy minimization problem: the point cloud classification labels of the model surface are mapped to weighted graph nodes, and an energy function is defined that includes boundary smoothness terms (ensuring surface smoothness) and region consistency terms (ensuring volume stability). This function is iteratively solved using a minimum cut / maximum flow algorithm, dynamically adjusting the positions of the nodes on the model's outer surface to eliminate jagged edges and achieve a high-precision microscopic fit with the natural dentition anatomical interface without altering the overall morphology. Finally, based on the optimized geometry, biomimetic structural parametric modeling is driven by occlusal force data, generating gradient porosity and a gradually changing elastic modulus structure within the solid that conforms to biomechanical distribution.

[0039] This invention effectively solves the problem of morphological distortion or collapse of anatomical landmarks (such as the cervical margin) caused by traditional mesh smoothing algorithms through energy optimization by introducing a graph-cut algorithm, ensuring the sealing and fit of the denture base edges. Simultaneously, it extends the high-precision fit between the segmentation boundary and natural tissue from the macro-geometric level to the micro-surface level, significantly improving denture retention and wearing comfort. Furthermore, by combining biomimetic structural parametric modeling, it breaks through the limitations of traditional homogeneous material design for dentures, enabling the generated three-dimensional denture model to possess differentiated mechanical properties in the base, artificial teeth, and connector areas. This allows for more effective dispersion of masticatory stress, reducing pressure absorption on the remaining alveolar ridge, thereby significantly improving the fatigue resistance and long-term reliability of the restoration.

[0040] Furthermore, the process of generating a three-dimensional model of a denture with biomimetic structural features by combining biomimetic structural parametric modeling specifically includes: Based on the three-dimensional data of the skeletal tissue structure in the comprehensive three-dimensional model, the density distribution characteristics of alveolar bone in the edentulous area are obtained. Based on the density distribution characteristics, the gradient porosity parameters of the denture base, the elastic modulus gradient parameters of the artificial tooth, and the transition stiffness parameters of the connector are determined. Based on the gradient porosity parameters, elastic modulus gradient parameters, and transition stiffness parameters, the internal structure of the initial three-dimensional denture model after high-precision fitting is reconstructed to generate a three-dimensional denture model with biomimetic structural features.

[0041] The principle and technical effect of the above scheme are as follows: First, the HU value distribution characteristics of the alveolar ridge and surrounding supporting bone in the edentulous area are extracted from the comprehensive three-dimensional model, and the mechanical support level is divided according to the bone mineral density. Then, a parametric modeling engine is used to transform the density data into specific geometric constraints: for sparse bone areas, a gradient lattice structure with high porosity and low elastic modulus is automatically matched to increase the contact area and buffer stress; for areas with sufficient bone, a dense structure with low porosity and high stiffness is matched to resist high-intensity chewing forces. Through this real-time mapping between density and stiffness, a biomimetic mechanical transmission path with a gradient change from the mucosal side of the denture base to the occlusal surface is reconstructed inside the denture, simulating the absorption and dispersion function of the natural periodontal ligament for occlusal forces.

[0042] Compared to existing technologies, this invention addresses the clinical complications caused by the singular mechanical properties of traditional homogeneous dentures through biomimetic parametric modeling based on bone density. By customizing gradient porosity and elastic modulus according to individual bone volume, it effectively eliminates stress concentration at the denture-tissue interface, significantly reducing the risk of pressure pain and residual alveolar ridge resorption caused by excessive local pressure, thus extending the lifespan of the denture. Simultaneously, this data-driven precision design replaces traditional experience-based manual adjustments, ensuring consistency and scientific rigor in design standards across different cases. This allows the generated three-dimensional denture model to achieve not only anatomical biomimicry but also biomechanical biomimicry, significantly improving the long-term prognosis of the restoration.

[0043] Furthermore, the three-dimensional intelligent modeling of the denture includes the denture base, the artificial tooth, and the connector. Based on the optimized denture basic model, biomimetic structural modeling is performed to obtain a three-dimensional denture model with biomimetic structural features, specifically including: The denture base structure was modeled, and a gradient pore structure design was adopted, defining the pore size as 50-100 μm near the mucosa, 200-300 μm in the middle layer, and 300-500 μm on the outer layer. Multi-layer biomimetic modeling was performed on the artificial tooth portion, and the pit and fissure structure of natural teeth was replicated on the crown surface. The dentin layer was designed with a gradient in elastic modulus, decreasing from 80 GPa on the enamel side to 20 GPa on the dentin side. The gradient formula is as follows: ,in It is a gradual function of the elastic modulus of the dentin layer of artificial teeth, representing the elastic modulus at a distance z from the enamel side. It is the distance from the enamel side to the dentin side. The thickness of the dentin layer. A biomimetic transition model was created for the connection between the denture base and the artificial tooth, employing a biomimetic tendon-bone connection structure. The elastic modulus gradually changes from 30 GPa at the denture base end to 60 GPa at the artificial tooth end, with a transition length of 3-5 mm. The transition formula is as follows: Where E(t) is the gradual function of the elastic modulus of the connected body, The distance from the base of the denture to the end of the artificial tooth. The length of the transition section. Based on the above parameters, a three-dimensional model of the denture with biomimetic structural features is generated.

[0044] The working principle and technical effects of this invention are as follows: In the process of biomimetic structural modeling, this invention adopts a biomimetic parametric modeling mechanism with biomechanical gradient distribution, allocating material properties according to the differences in anatomical functions of different tissues. For the base portion, a gradient pore structure design is adopted. By defining small pores (50-100μm) near the mucosa side to promote soft tissue attachment and nutrient delivery, medium pores (200-300μm) in the middle layer provide stress buffering, and large pores (300-500μm) on the outer side ensure structural strength, thereby simulating the spongy support characteristics of natural alveolar bone. For the artificial tooth portion, the pits and fissures of natural teeth are replicated on the crown surface to restore chewing efficiency. At the same time, a gradual change in elastic modulus is implemented in the dentin layer. An exponential function is used to control the modulus to smoothly transition from 80GPa on the enamel side to 20GPa on the dentin side, simulating the hardness decay characteristics of natural tooth enamel transitioning to dentin. For the connecting part, a biomimetic tendon-bone connection structure is introduced. The elastic modulus is controlled by a linear gradient formula to gradually change from 30GPa at the base end to 60GPa at the artificial tooth end within a 3-5mm transition section, eliminating stress concentration points at the junction of the base and the artificial tooth.

[0045] Compared with existing technologies, this invention, through its gradient porosity and gradually changing elastic modulus design, completely solves the clinical problems caused by the mismatch of mechanical properties in traditional homogeneous material dentures. The gradient porosity structure of the denture base significantly increases the contact surface area between the base and the mucosa, enhancing atmospheric pressure and adsorption force, while avoiding alveolar ridge resorption caused by stress concentration. The gradually changing elastic modulus design of the artificial tooth effectively simulates the buffering function of the natural periodontal ligament, reducing the impact force on the abutment tooth and alveolar bone during chewing, preventing abutment tooth loosening or fracture. The biomimetic transition modeling of the connector eliminates the stress abrupt changes of traditional rigid connections, uniformly transmitting occlusal forces to the denture base support tissue, significantly improving the overall fatigue resistance and long-term reliability of the denture, achieving a leap from simple morphological restoration to functional biomimicry.

[0046] Furthermore, the three-dimensional model of the denture is imported into the finite element simulation platform in STL format with a precision of 0.001mm, and the material properties of each part of the denture are configured: The base portion is made of PEEK-HA gradient composite material, with the hydroxyapatite (HA) mass fraction decreasing from 30% on the mucosal side to 10% on the outer side; the enamel layer of the artificial tooth portion is made of zirconia ceramic, and the dentin layer is made of PMMA-nano SiO2-HA composite resin; the connector portion is made of titanium alloy-PEEK composite structure, with the titanium alloy surface roughness set to Ra1.0-2.0μm; Three typical clinical dynamic loading conditions—centric occlusion, protruding occlusion, and lateral occlusion—were applied, and finite element mechanical calculations were performed in parallel to verify that the following three indicators were simultaneously satisfied: (a) Maximum equivalent stress σ of the Keetor max ≤80MPa; (b) Stress uniformity coefficient C of artificial tooth occlusal surface v ≤0.3, where C v =σ std / σ mean , σ mean σ represents the average stress at the interlocking joint surfaces. std (c) The stress concentration factor α of the connection body is ≤1.5, where α = σ peak / σ nom , σ peak σ is the maximum local stress of the connector. nom For the nominal stress in the far field; If any one of the indicators fails to meet the requirements, the parameter iteration correction process will be automatically triggered: adjust at least one of the following: the hierarchical pore diameter of the base plate gradient pores, the gradient slope of the elastic modulus of the artificial tooth, or the length of the transition section of the connector; regenerate the three-dimensional model of the denture and repeat the above simulation verification steps until all three indicators meet the requirements; use the finally qualified three-dimensional model of the denture as the direct digital data source for selective laser sintering, photopolymerization stereolithography, and laser cladding additive manufacturing.

[0047] The principle and effect of the above technical solution are as follows: By importing the three-dimensional model of the prosthesis into a finite element simulation platform, the material properties are configured according to the differences in anatomical function: the base uses a PEEK-HA composite material with a gradient decreasing mass fraction of hydroxyapatite (HA) to simulate the bioactivity and stiffness transition of natural bone; the artificial tooth uses a layered structure of zirconium oxide and composite resin to replicate the wear resistance and cushioning characteristics of natural teeth; and the connector uses a titanium alloy-PEEK composite structure to balance strength and toughness. Based on this, three dynamic load conditions—centric, protruding, and lateral occlusion—are applied, and three core mechanical indicators are calculated and monitored in parallel: the structural strength of the base (σ... max ), uniformity of stress distribution on the occlusal surface (C vThe system automatically triggers a parameter iteration correction process if any indicator deviates from the safety threshold. This process involves adjusting the gradient pore level, the gradual slope of the elastic modulus, or the length of the transition section, regenerating the model, and iteratively verifying it until the mechanical response fully meets the physiological bearing standard.

[0048] Compared with existing technologies, this invention introduces finite element simulation and automatic parameter iterative correction to construct a closed-loop digital twin verification mechanism based on physical feedback, transforming the biomimetic design model into a clinically feasible manufacturing data source. Utilizing multi-condition dynamic load verification, it ensures the denture maintains mechanical stability under various complex occlusal movements, effectively avoiding common clinical risks such as denture base fracture, porcelain chipping of artificial teeth, and fatigue fracture of connectors. Simultaneously, by setting strict quantitative indicators (such as C...),... v (≤0.3 ensures uniform stress distribution, α≤1.5 limits local peak stress), transforming subjective tactile feedback into objective physical data, significantly improving the long-term success rate of restorations. The final compliant digital model can be directly used as a precise data source for additive manufacturing processes such as selective laser sintering and stereolithography, achieving seamless integration from digital design to intelligent manufacturing and ensuring a high degree of consistency in mechanical properties between the physical entity and the virtual design.

[0049] Based on the completed denture design, the three-dimensional model of the denture generated by this invention is formed through a combination of additive manufacturing processes to verify the technical effects of the biomimetic design mentioned above. Specifically, this includes: using selective laser sintering (SLS) technology to manufacture the denture base blank, faithfully replicating the gradient pore structure to ensure internal mechanical conduction and mucosal adhesion performance; using stereolithography (SLA) technology to manufacture the artificial tooth blank, accurately replicating the microstructure of natural tooth pits and fissures to restore chewing contact characteristics; and using laser cladding technology to form the connector, achieving a gradient transition from titanium alloy to PEEK to ensure the effective construction of a structure with a gradual change in elastic modulus. The three processes are carried out in parallel in different areas and assembled in one go, avoiding the errors of traditional splicing. After molding, multi-dimensional surface modification is performed: the denture base is plasma etched to introduce micro-nano bumps and hydrophilic groups to accelerate mucosal adhesion; the artificial tooth is coated with a nano-hydroxyapatite coating to form a tooth-like enamel mineralization layer, improving wear resistance and bioactivity; and the connector is electrochemically polished to cut fracture sources and reduce plaque adhesion.

[0050] After assembly, the denture was calibrated and fixed on a digital occlusal calibration platform. Performance testing showed that the denture base bending strength was ≥100MPa, the artificial tooth compressive strength was ≥300MPa, the wear resistance was ≤0.01mm³ / N·m, and it passed the ISO10993 biocompatibility evaluation. During the clinical fitting phase, intraoral scanners were used to obtain denture-mucosal occlusal surface data. If gaps exceeded the standard, they were precisely adjusted using a five-axis milling machine. Occlusal contact points were detected using T-Scan; if occlusal high points were found, the occlusal surface morphology was adjusted using CAD and then secondary curing was performed. Ultimately, after the trial fitting, the patient's chewing efficiency improved by ≥80% compared to before restoration, and the VAS comfort score was ≤2 points, completing the clinical fitting. Specific testing and experimental parameters are as follows: Mechanical property testing Three-point bending test of the base Instron 5969 Universal Testing Machine Span 20mm, loading speed 2mm / min, ambient temperature 25±2℃ Flexural strength ≥100MPa, elastic modulus ≥3GPa Mechanical property testing Artificial tooth compressive strength test Same as above The loading head has a diameter of 5mm and a loading speed of 1mm / min. Compressive strength ≥300MPa Mechanical property testing Artificial tooth wear resistance test CSM Friction and Wear Testing Machine Load 5N, rotation speed 100r / min, wear time 30min Wear amount ≤ 0.01 mm³ Mechanical property testing Shear strength test of connector Same as above Shear rate 1 mm / min, shear area 10 mm² Shear strength ≥ 50 MPa Biocompatibility testing Cytotoxicity test Thermo 3111 Cell Culture Incubator <![CDATA[L929 cells, seeding density 1×10 4 cells / well, cultured for 72 h]]> Cell viability ≥90% (MTT method, ISO10993-5) Biocompatibility testing Sensitization test GMP Standard Laboratory (Guinea Pig Maximization Test) Induction dose: 0.1 mL; stimulation dose: 0.05 mL; observation period: 14 days. No erythema / edema (score ≤0, ISO10993-10) Biocompatibility testing Mucosal irritation test Same as above (rabbit eye stimulation test) Add 0.1 mL of the material extract and observe for 72 hours. No corneal / conjunctival irritation (score ≤ 0) Clinical aptability testing Denture-mucosal fit assessment 3ShapeTrios5 intraoral scanner The scanning sampling density is 1000 points / mm², covering the entire area of ​​the missing tooth. Fitting gap ≤ 0.03mm Clinical aptability testing Dynamic occlusal function assessment T-Scan10 Dynamic Occlusal Recorder Sampling frequency 1000Hz, recording 3 types of bite states The bite contact points are uniform and free from interference (interference force value ≤50N). Clinical aptability testing Patient comfort assessment Visual Analogue Scale (VAS) Rating range: 0-10 (0 = no discomfort, 10 = severe pain) VAS score ≤ 2 points Working principle and usage of this invention: The working principle of this invention lies in constructing a fully digital prosthetic restoration system that considers diverse data parameters and incorporates biomimetic design and closed-loop verification. First, by simultaneously acquiring point clouds of the oral surface, CBCT voxels, and dynamic occlusal trajectories, a multi-stage registration algorithm is used to achieve spatiotemporal alignment and feature-level fusion of multi-source heterogeneous data, constructing a comprehensive 3D model that combines static geometric accuracy with dynamic functionality. Second, based on this model, the PointNet++ deep learning architecture is used to automatically identify anatomical landmarks and predict prosthetic segmentation boundaries, defining the precise placement space for the denture base, artificial teeth, and connectors. Finally, through biomimetic parametric modeling, a gradient porosity and elastic modulus-gradient structure are constructed within the solid structure, and finite element simulation and an automatic iterative correction mechanism are introduced to ensure that the restoration matches the natural dental and jaw tissue in terms of mechanical response, achieving a leap from morphological replication to functional biomimicry.

[0051] Compared with existing technologies, this invention, through precise fusion of all-dimensional data and AI intelligent design, completely eliminates the occlusal interference and clinical revision risks caused by data fragmentation in traditional dentures, significantly improving the clinical fit rate of the prosthesis. By introducing gradient porosity and gradual elastic modulus design, it effectively solves the problems of denture base fracture caused by stress concentration in homogeneous material dentures and the compression and absorption of the remaining alveolar ridge, greatly improving the biomechanical compatibility and long-term stability of the prosthesis. In addition, by combining additive manufacturing and digital surface modification processes, it achieves non-destructive transformation from virtual design to physical entity, ensuring that the denture has excellent mechanical strength, wear resistance, and bioactivity, ultimately improving patients' chewing efficiency by more than 80% and reducing the VAS comfort score to below 2 points, demonstrating significant clinical application value and promising prospects for promotion.

[0052] The following are several embodiments from comparative experiments provided in this application. Example 1: Comparison Experiment of Static Geometric Accuracy Experimental objective: To verify the improvement effect of the method of the present invention on the static geometric accuracy of three-dimensional models of oral soft and hard tissues.

[0053] Experimental subjects: Ten patients with missing teeth (all with unilateral mandibular posterior tooth loss) were selected, and modeling was performed using both the traditional method and the method of this invention.

[0054] Conclusion: The method of this invention improves the static geometric accuracy compared with traditional methods by using a deep learning fine registration algorithm based on PointNet++ (registration error ≤0.005mm) and combining it with the accuracy evaluation of 1000 uniform sampling points. It significantly reduces the fitting error between the model and oral tissues, laying the foundation for the precise design of dentures.

[0055] Example 2: Comparative Experiment on Dynamic Occlusal Function Adaptability Experimental objective: To evaluate the advantages of the method of the present invention in terms of dynamic bite trajectory and force value adaptability.

[0056] Experimental subjects: 5 patients requiring complete denture restoration, and data on three movement states—centric occlusion, protruding occlusion, and lateral occlusion—were collected simultaneously.

[0057] Occlusal trajectory overlap (R) 75%–82% (mean 78%) 98%–99.5% (mean 98.8%) <![CDATA[Force value distribution uniformity (C v )]]> 0.45-0.60 (mean 0.52) 0.20-0.28 (mean 0.24) Occlusal peak incidence 20% (1 / 5 of cases) 0% (0 / 5 cases) Conclusion: The method of this invention significantly improves the dynamic occlusal functional adaptability of dentures through dynamic occlusal data mapping technology. Compared with traditional methods without dynamic data fusion, the occlusal trajectory overlap of this invention is increased from 78% to 98.8%, and the force distribution uniformity coefficient (C) is also improved. v The occlusal velocity decreased from 0.52 to 0.24, and the occlusal high point was completely eliminated (occurrence rate decreased from 20% to 0%). Based on the fusion technology of dynamic data and static model with timestamps, combined with the multi-stage registration algorithm, the accurate simulation of occlusal trajectory and force value was achieved.

[0058] Example 3: Clinical Application Effect Comparison Experiment Experimental objective: To compare the differences between the two methods in terms of clinical rework rate, patient satisfaction, and manufacturing cycle.

[0059] Subjects: 20 patients with removable partial dentures were randomly divided into a traditional group (10 cases) and an invention group (10 cases).

[0060] Dental repair rate 30% (3 / 10 cases) 5% (1 in 20 cases, due to material compatibility issues) Patient comfort score (VAS) 6.2 ± 1.5 points (out of 10) 9.1 ± 0.8 points (out of 10) Manufacturing cycle 7-10 days 3-4 days Basement mucosal fit satisfaction 65% (6 / 10 cases) 95% (19 / 20 cases) Conclusion: The method of this invention, through gradient porosity base design, biomimetic elastic modulus gradient structure and surface modification treatment, reduces the clinical rework rate by 83.3%, improves patient comfort by 46.8%, and shortens the manufacturing cycle by 57.1%, with the overall clinical effect being significantly better than the traditional method.

[0061] In summary, the method of this invention achieves breakthrough improvements in static accuracy, dynamic function, and clinical application through multi-source data AI fusion (static + dynamic data), biomimetic structural design (gradient materials and pores), and mechanical closed-loop optimization (finite element iterative correction), providing a high-precision and personalized solution for digital design of dental prostheses.

[0062] The above description is only used to illustrate the technical solutions of the present invention and is not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Any equivalent structural or procedural transformations made using the content of the present invention specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A digital design method for dental prostheses based on AI intelligence, characterized in that, include: S1: Simultaneously collect the three-dimensional point cloud data of the patient's oral cavity surface, the three-dimensional data of the bone tissue structure, and the dynamic occlusal motion trajectory and force data to obtain the basic design parameters data of the denture, and perform data preprocessing on the basic design parameters data of the denture. S2: The preprocessed denture basic design parameter data are fused using a deep learning-based multi-stage registration algorithm to obtain a spatiotemporally aligned oral soft and hard tissue-dynamic occlusion dataset, and a comprehensive three-dimensional model with static geometric accuracy and dynamic occlusion function is reconstructed based on the dataset. S3: Based on the comprehensive 3D model, key point identification, automatic coordinate system alignment and key feature extraction of point cloud are performed on the patient's 3D dental and jaw tissue, and the point cloud classification prediction results are output to form the denture segmentation boundary prediction. S4: Based on the segmentation boundary prediction results, perform three-dimensional intelligent modeling and filling of the denture to generate an initial three-dimensional denture model. Optimize the initial three-dimensional denture model through a segmentation refinement algorithm. Combined with biomimetic structural parametric modeling, generate a three-dimensional denture model with biomimetic structural features.

2. The AI-based digital design method for dental prostheses according to claim 1, characterized in that: The accuracy of the integrated 3D model is evaluated using two dimensions: geometric deviation and dynamic accuracy. Geometric deviation is calculated by measuring the average distance error between the integrated 3D model and the high-precision oral reference point cloud data acquired preoperatively. An assessment was conducted, in which... , where N is the number of sampling points ≥ 1000, evenly distributed in the edentulous area and adjacent teeth; , , These are the three-dimensional coordinates of the sampling points in the model; The coordinates of the actual oral cavity sampling points obtained by a coordinate measuring machine are given, with an accuracy of 0.001 mm. 'i' is the index, representing the coordinates of the i-th sampling point, 'm' represents the model value, and 't' represents the actual measured value. , , () represents the three-dimensional coordinates of the i-th sampling point in the model, and the corresponding... The superscript 't' in the text represents the actual measured value, that is, the coordinates of the actual oral sampling point obtained by the coordinate measuring machine. It is required that... ≤0.005mm; Dynamic accuracy is evaluated by calculating the degree of overlap, R, between the bite trajectory simulated by the integrated three-dimensional model and the dynamic bite movement trajectory actually collected from the patient. ,in: The overlap length between the simulated bite trajectory of the 3D model and the actual collected trajectory; The total length of the actual bite trajectory is required to be R≥98%.

3. The AI-based digital design method for dental prostheses according to claim 1, characterized in that: The three-dimensional point cloud data of the oral cavity surface was acquired by an intraoral scanner, and the spatial resolution of the scan was 0.01 mm. The three-dimensional data of the skeletal tissue structure was acquired by cone-beam CT scanning. The three-dimensional data of the skeletal tissue structure includes alveolar bone height, bone density data and nerve canal location data. The interslice spacing of the tomographic image reconstructed after cone-beam CT scanning is 0.1 mm. The dynamic occlusal motion trajectory and force data are collected by an intraoral dynamic occlusal recorder with a sampling frequency of 1000Hz, covering three motion states: centric occlusion, protruding occlusion, and lateral occlusion.

4. The AI-based digital design method for dental prostheses according to claim 1, characterized in that, Data preprocessing is performed on the aforementioned denture basic design parameter data, including the following steps: A global reference coordinate system is established based on the three-dimensional point cloud data of the oral cavity surface. Rigid body transformation registration is performed on the cone-beam CT data, and the dynamic occlusal motion trajectory and force data are resampled according to the sampling timestamp to unify the spatial scale and temporal dimension of multi-source heterogeneous data. For the three-dimensional point cloud data of the oral cavity surface, normal consistency filtering is used to remove flying points, and Gaussian filtering is used for smoothing and noise reduction. For the three-dimensional data of bone tissue structure, median filtering is performed based on the HU threshold range to suppress ring artifacts. For the dynamic occlusal motion trajectory and force data, sliding window moving average filtering is used to eliminate instantaneous interference force values. Based on prior information about the anatomical structure of the edentulous area, non-target areas in the maxillary sinus, mandibular canal, and distal free soft tissue are automatically eliminated, retaining only the effective design domain data within the edentulous area and the range of the three adjacent teeth on each side.

5. The AI-based digital design method for dental prostheses according to claim 1, characterized in that, In step S2, the multi-stage registration algorithm is used to achieve multi-source data fusion, and its processing flow includes: The point cloud density of the preprocessed oral cavity surface point cloud data was reduced to 500 points / mm². The voxel data obtained after the cone-beam CT device performs three-dimensional X-ray scanning of the patient's maxillofacial region is segmented by HU value thresholding. First, the bone voxels are picked out using the HU threshold, and then the MarchingCubes algorithm is used to directly generate triangular meshes on the bone surface, that is, to extract the three-dimensional model of the jawbone. Using the occlusal surface feature points of adjacent teeth as a reference, the preprocessed three-dimensional point cloud data of the oral cavity surface and the generated three-dimensional model of the jawbone are coarsely registered using the ICP algorithm, with an error ≤0.05mm; A deep learning model based on PointNet++ is used to perform fine registration of the three-dimensional point cloud data of the oral cavity surface after coarse registration with the three-dimensional model of the jawbone. The 128-dimensional feature vector is fused, and the key areas within the effective design domain are focused through the attention mechanism to achieve a registration error of ≤0.005mm. The preprocessed dynamic occlusal motion trajectory and force data are mapped to the finely registered 3D model through the timestamp to obtain a spatiotemporally aligned oral soft and hard tissue-dynamic occlusal dataset.

6. The AI-based digital design method for dental prostheses according to claim 1, characterized in that: Step S3 includes the following steps: Based on the comprehensive three-dimensional model, anatomical landmarks of the cusps, central fossa and alveolar crest of the three-dimensional dental and maxillary tissues are identified, and a local coordinate system is constructed using the anatomical landmarks to automatically orient the three-dimensional dental and maxillary tissues in the coordinate system. On the three-dimensional dental and jaw tissue point cloud of the patient automatically aligned by the coordinate system, the dental arch curve and the key curve of the cervical margin are fitted, and the curvature, normal vector and neighborhood density are calculated point by point to extract 128-dimensional key feature vectors. The 128-dimensional key feature vector is input into a deep learning model based on the PointNet++ architecture, and the output is a point-by-point classification label, which includes at least the basement region, the artificial tooth region, and the connector region. Based on the point-by-point classification labels, the boundaries at the intersections of different category regions are extracted to generate denture segmentation boundary prediction results.

7. The AI-based digital design method for dental prostheses according to claim 6, characterized in that, Step S4 includes: Based on the initial denture segmentation boundary and the denture segmentation boundary prediction result, the three-dimensional intelligent modeling of the denture is filled to generate the initial three-dimensional denture model. Based on the initial three-dimensional model of the denture, the outer surface of the initial three-dimensional model of the denture is smoothed by the graph cut algorithm, which is the segmentation and refinement algorithm, to eliminate jagged edges. The point cloud classification labels are mapped to weighted graph nodes, and an energy function containing boundary smoothing and region consistency terms is defined. The optimal boundary is solved by the minimum cut / maximum flow algorithm to achieve high-precision fitting between the segmentation boundary and the natural tooth jaw anatomical interface. Based on the initial three-dimensional model of the denture after high-precision fitting, and combined with biomimetic structural parametric modeling, a three-dimensional model of the denture with biomimetic structural features is generated.

8. The AI-based digital design method for dental prostheses according to claim 1, characterized in that: The aforementioned three-dimensional intelligent modeling of the denture includes the denture base, the artificial tooth portion, and the connector portion. Based on the optimized denture basic model, biomimetic structural modeling is performed to obtain a three-dimensional denture model with biomimetic structural features, specifically including: The denture base structure was modeled, and a gradient pore structure design was adopted, defining the pore diameter as 50-100μm near the mucosa, 200-300μm in the middle layer, and 300-500μm on the outer layer. Multi-layer biomimetic modeling was performed on the artificial tooth portion, and the pit and fissure structure of natural teeth was replicated on the crown surface. The dentin layer was designed with a gradient in elastic modulus, decreasing from 80 GPa on the enamel side to 20 GPa on the dentin side. The gradient formula is as follows: ,in It is a gradual function of the elastic modulus of the dentin layer of artificial teeth, representing the elastic modulus at a distance z from the enamel side. It is the distance from the enamel side to the dentin side. The thickness of the dentin layer. A biomimetic transition model was created for the connection between the denture base and the artificial tooth, employing a biomimetic tendon-bone connection structure. The elastic modulus gradually changes from 30 GPa at the denture base end to 60 GPa at the artificial tooth end, with a transition length of 3-5 mm. The transition formula is as follows: Where E(t) is the gradual function of the elastic modulus of the connected body, The distance from the base of the denture to the end of the artificial tooth. The length of the transition section. Based on the above parameters, a three-dimensional model of the denture with biomimetic structural features is generated.

9. The AI-based digital design method for dental prostheses according to claim 8, characterized in that: The three-dimensional model of the denture was imported into the finite element simulation platform in STL format with a precision of 0.001mm, and the material properties of each part of the denture were configured: The base portion is made of PEEK-HA gradient composite material, with the hydroxyapatite (HA) mass fraction decreasing from 30% on the mucosal side to 10% on the outer side; the enamel layer of the artificial tooth portion is made of zirconia ceramic, and the dentin layer is made of PMMA-nano SiO2-HA composite resin; the connector portion is made of titanium alloy-PEEK composite structure, with the titanium alloy surface roughness set to Ra1.0-2.0μm; Three typical clinical dynamic loading conditions—centric occlusion, protruding occlusion, and lateral occlusion—were applied, and finite element mechanical calculations were performed in parallel to verify that the following three indicators were simultaneously satisfied: (a) Maximum equivalent stress σ of the Keetor max ≤80MPa; (b) Stress uniformity coefficient C of artificial tooth occlusal surface v ≤0.3, where C v =σ std / σ mean , σ mean σ represents the average stress at the interlocking joint surfaces. std (c) The stress concentration factor α of the connection body is ≤1.5, where α = σ peak / σ nom , σ peak σ is the maximum local stress of the connector. nom For the nominal stress in the far field; If any one of the indicators fails to meet the requirements, the parameter iteration correction process will be automatically triggered: adjust at least one of the following: the hierarchical pore diameter of the base plate gradient pore, the gradient slope of the elastic modulus of the artificial tooth, or the length of the transition section of the connector; regenerate the three-dimensional model of the denture and repeat the above simulation verification steps until all three indicators meet the requirements; use the finally qualified three-dimensional model of the denture as the direct digital data source for selective laser sintering, photopolymerization stereolithography, and laser cladding additive manufacturing.

10. The AI-based digital design method for dental prostheses according to claim 7, characterized in that: The process of generating a three-dimensional denture model with biomimetic structural features by combining biomimetic structural parametric modeling specifically includes: Based on the three-dimensional data of the skeletal tissue structure in the comprehensive three-dimensional model, the density distribution characteristics of alveolar bone in the edentulous area are obtained. Based on the density distribution characteristics, the gradient porosity parameters of the denture base, the elastic modulus gradient parameters of the artificial tooth, and the transition stiffness parameters of the connector are determined. Based on the gradient porosity parameters, elastic modulus gradient parameters, and transition stiffness parameters, the internal structure of the initial three-dimensional denture model after high-precision fitting is reconstructed to generate a three-dimensional denture model with biomimetic structural features.

Images