An ai denture personalized design method based on multi-modal oral data fusion

Abstract

The application relates to an AI denture individual design method based on multi-modal oral cavity data fusion. The method comprises the following steps: acquiring multi-modal oral cavity data of a target patient, and performing time synchronization and space registration on the multi-modal oral cavity data to obtain registered oral cavity data; inputting the registered oral cavity data into a feature fusion model to perform feature extraction and feature fusion, and obtaining oral cavity feature data; inputting the oral cavity feature data into a denture shape generation model to perform feature coding, shape decoding and parameter mapping, and obtaining initial denture design data; based on the initial denture design data, performing shape matching degree calculation, occlusion parameter adjustment and aesthetic parameter optimization on the oral cavity feature data to obtain optimized denture design data; and performing three-dimensional reconstruction, occlusion simulation verification and design parameter correction on the optimized denture design data to obtain denture design data. The method can improve the individual adaptation degree and design consistency of the denture.

Description

Technical Field

[0001] This invention belongs to the field of oral medicine, and in particular relates to an AI-based personalized denture design method based on multimodal oral data fusion. Background Technology

[0002] With the integration of dental prosthodontics and digital healthcare, personalized denture fabrication methods based on 3D scanning and computer-aided design are gradually being applied. This technology uses an intraoral scanner to acquire the patient's dentition morphology data, matching and adjusting it against a pre-set standard tooth shape library to generate an initial prosthesis shape. Compared to traditional manual tooth arrangement, digital methods improve denture fabrication efficiency and morphological consistency to a certain extent.

[0003] In traditional techniques, denture design primarily relies on single three-dimensional oral cavity morphological data. Technicians then manually scale, rotate, and position standard tooth shapes using their experience, followed by static occlusal adjustments via an articulator. While some approaches incorporate two-dimensional imaging data as a reference, these two methods are spatially and temporally independent, resulting in limited information integration and an inability to fully reflect the patient's oral functional status and anatomical characteristics.

[0004] However, current denture design methods have significant shortcomings. Single-modal data cannot fully characterize the complex relationships between oral soft and hard tissue morphology, occlusal function, and dynamic movement characteristics, resulting in a low degree of matching between denture morphology and individual patient functional needs. Fragmented data sources lead to one-sided design basis, strong reliance on technician experience, poor consistency and low repeatability of design results. In addition, static morphological matching makes it difficult to predict denture interference and discomfort in actual occlusal movements, often requiring multiple clinical adjustments, prolonging the restoration period and reducing patient satisfaction. Therefore, how to more fully integrate multimodal oral data and improve the level of personalized denture design has become an urgent technical problem to be solved in this field. Summary of the Invention

[0005] Therefore, it is necessary to provide an AI-based personalized denture design method based on multimodal oral data fusion to address the aforementioned technical issues.

[0006] Firstly, this application provides an AI-based personalized denture design method based on multimodal oral data fusion, including:

[0007] S1. Acquire multimodal oral data of the target patient, and perform time synchronization and spatial registration on the multimodal oral data to obtain registered oral data;

[0008] S2. Input the registered oral cavity data into the feature fusion model for feature extraction and feature fusion to obtain oral cavity feature data;

[0009] S3. Input the oral cavity feature data into the denture morphology generation model for feature encoding, morphology decoding and parameter mapping to obtain the initial denture design data;

[0010] S4. Based on the initial denture design data, calculate the morphological matching degree, adjust the occlusal parameters and optimize the aesthetic parameters of the oral cavity feature data to obtain optimized denture design data;

[0011] S5. Perform three-dimensional reconstruction, occlusal simulation verification, and design parameter correction on the optimized denture design data to obtain denture design data.

[0012] Secondly, this application also provides an AI-based personalized denture design system based on multimodal oral data fusion, including:

[0013] The multimodal data acquisition and registration module is used to acquire multimodal oral data of the target patient and perform time synchronization and spatial registration on the multimodal oral data to obtain registered oral data.

[0014] The oral cavity feature extraction and fusion module is used to input registered oral cavity data into the feature fusion model for feature extraction and feature fusion to obtain oral cavity feature data;

[0015] The denture morphology generation module is used to input oral feature data into the denture morphology generation model for feature encoding, morphology decoding and parameter mapping to obtain initial denture design data.

[0016] The denture parameter optimization module is used to calculate the morphological matching degree, adjust the occlusal parameters, and optimize the aesthetic parameters of oral feature data based on the initial denture design data, so as to obtain optimized denture design data.

[0017] The denture simulation and verification module is used to perform three-dimensional reconstruction, occlusal simulation verification, and design parameter correction on optimized denture design data to obtain denture design data.

[0018] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0019] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0020] The aforementioned AI-based personalized denture design method, based on multimodal oral data fusion, effectively integrates 3D oral scanning data, 2D imaging data, and functional data through temporal synchronization and spatial registration of multimodal oral data. This allows for a comprehensive characterization of the complex relationships between oral soft and hard tissue morphology, occlusal function, and dynamic movement characteristics. Utilizing techniques such as local and global feature extraction and adaptive weighted fusion, it improves the matching degree between denture morphology and individual patient functional needs, enhancing the consistency and repeatability of design results. Furthermore, through 3D reconstruction, dynamic occlusal simulation verification, and parameter correction, it proactively avoids interference problems in actual occlusal movements, reducing the number of clinical adjustments, shortening the restoration cycle, and improving the level of personalized denture design. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating an AI-based personalized denture design method based on multimodal oral data fusion in one embodiment.

[0023] Figure 2 This is a schematic diagram of the structure of an AI-based personalized denture design system based on multimodal oral data fusion in one embodiment. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0025] refer to Figure 1 The document presents a flowchart illustrating an AI-based personalized denture design method based on multimodal oral data fusion, as provided in this application. The method includes the following steps:

[0026] S1. Acquire multimodal oral data of the target patient, and perform time synchronization and spatial registration on the multimodal oral data to obtain registered oral data.

[0027] For example, multimodal oral data can comprehensively characterize a patient's oral anatomy and physiological functions. It belongs to a multi-source heterogeneous data type and mainly includes three-dimensional oral scan data, two-dimensional oral imaging data, and oral functional data. Because different types of data are acquired by independent acquisition devices, there are inconsistencies in the acquisition sequence over time and independent coordinate systems in space. Directly performing subsequent feature analysis will lead to information misalignment and data coupling failure. A two-layer processing logic of temporal correction and spatial coordinate unification can be adopted to achieve the fusion and preprocessing of heterogeneous data.

[0028] Optionally, in terms of time synchronization, all data are aligned in time dimension based on a unified time series benchmark, eliminating time series deviations caused by differences in sampling rhythms of different acquisition devices, and enabling time-series binding of oral cavity morphology information, imaging information, and motor function information under the same physiological state. The core principle of its time synchronization can be expressed by the following formula. ,in, Indicates the first The timestamp of the data after synchronization correction; For the first The original timestamp of the data collection; Indicates the first The offset between each acquisition device and the reference time is used to unify the time reference of data collected by different devices through this formula.

[0029] Optionally, in the spatial registration stage, relying on the bidirectional extraction and matching mechanism of anatomical feature points, a mapping relationship between the two-dimensional image space and the three-dimensional scanning space is established. A global spatial transformation model is used to unify the coordinate system, enabling different modal data to spatially overlap based on the same anatomical reference. The mathematical model of spatial transformation can be expressed as follows: ,in, It is a three-dimensional coordinate vector after spatial registration; The coordinate vector in the original 3D scan data; It is a matrix containing rotations Translation vector The transformation matrix, i.e. This matrix operation enables spatial transformation of the original data, achieving spatial alignment of data from different modalities.

[0030] Optionally, after time synchronization and spatial registration, various multimodal oral data can form a complete and unified dataset, eliminating data fragmentation and information misalignment, and providing a standardized data input foundation for subsequent cross-modal feature joint extraction and deep fusion.

[0031] S2. Input the registered oral cavity data into the feature fusion model for feature extraction and feature fusion to obtain oral cavity feature data.

[0032] Optionally, the feature fusion model can adopt an overall architecture of dual-branch parallel extraction combined with adaptive weighted fusion, consisting of seven functional modules: local feature extraction branch, global feature extraction branch, first feature distance calculation unit, second feature distance calculation unit, fusion weight calculation unit, feature fusion unit, and feature mapping unit. Each module operates collaboratively in a fixed logical order.

[0033] Optionally, the registered oral data can be used to carry local anatomical details, global occlusal correlation information, and multimodal coupling information. Local anatomical details include microstructural features such as tooth morphology and alveolar bone structure, while global occlusal correlation information covers macroscopic functional features such as the maxillary-mandibular occlusal relationship and occlusal motion trajectory. Since a single feature extraction method cannot simultaneously capture the complete information of both microstructural and macroscopic functional features, this step employs a dual-branch network structure to achieve differentiated feature extraction: the local feature extraction branch focuses on local anatomical details, while the global feature extraction branch focuses on global occlusal correlation information.

[0034] Optionally, relying on the feature distance quantification evaluation mechanism, the distance between local features and global features is calculated by the first feature distance calculation unit and the second feature distance calculation unit, respectively, to determine the credibility and suitability of different types of features. The fusion weight calculation unit performs deep weighted fusion of multiple feature classes based on the feature distance calculation results and an adaptive weight allocation rule. The calculation formula can be... ,in, For the first The fusion weights of class features For the first Distance between class features This represents the total number of feature categories.

[0035] Optionally, the weighted fusion of composite feature data still suffers from dimensionality redundancy and information overlap. Therefore, the feature mapping module performs feature dimensionality reduction and key information extraction. This module uses dimensionality reduction algorithms such as Principal Component Analysis (PCA) to eliminate invalid and redundant features, retaining core fusion information strongly correlated with personalized denture design. The processed oral feature data can simultaneously cover multi-dimensional information such as soft and hard tissue anatomy, occlusal movement patterns, and oral aesthetic structure.

[0036] S3. Input the oral cavity feature data into the denture morphology generation model for feature encoding, morphology decoding and parameter mapping to obtain the initial denture design data.

[0037] Optionally, the denture morphology generation model can be based on a hybrid architecture of Variational Autoencoder (VAE) and Conditional Generative Adversarial Network (cGAN), consisting of an end-to-end morphology generation chain comprised of a feature encoding layer, a morphology decoding unit, and a parameter mapping unit. This model uses condensed oral cavity feature data as its sole input and, through deep learning-based feature abstraction and geometric morphology reconstruction techniques, achieves the automatic transformation from abstract feature parameters to concrete denture morphology data.

[0038] Optionally, the feature encoding layer can employ a combination of a multilayer perceptron (MLP) and a convolutional neural network (CNN) to encode high-dimensional oral cavity feature data. Mapping to low-dimensional latent space representation The encoding process is achieved by optimizing the variational lower bound:

[0039]

[0040] in, As an inference network, it can be used to compute given oral cavity features. Hidden variables The posterior distribution of; To generate the network, latent variables are... Decoded into oral cavity feature data; The KL divergence is used to measure the posterior distribution. and prior distribution Differences; and These are the parameters for the generator network and the inference network, respectively.

[0041] Alternatively, the morphological decoding unit can use a three-dimensional convolutional neural network (3D-CNN) to decode the latent space vectors. Restored to continuous spatial geometric information By progressively upsampling through deconvolution, the low-dimensional vector is expanded into a three-dimensional voxel mesh or polygon mesh representation, the expression of which can be: ,in, For a 3D-CNN based decoder, Its network parameters; The three-dimensional geometric shape representing the denture. These represent the height, width, and depth of the voxel mesh, respectively.

[0042] Optionally, the parameter mapping unit can establish geometric data through a bidirectional long short-term memory network (Bi-LSTM). and standardized denture design parameters The mapping relationship. Forward LSTM learning of the encoding from geometric shape to parameters can be represented as... The reverse LSTM learning of decoding from parameters to geometry can be represented as: ,in, and These are time-series slices representing geometric shape and design parameters, respectively. and These are the hidden states of the forward and reverse LSTM. The reconstruction error is minimized. This enables a two-way binding between geometric shape and design parameters.

[0043] Optionally, the entire generation process is driven entirely by the patient's personalized oral feature data. Through the latent space modeling of VAE and the conditional generation capability of cGAN, combined with the structured processing of 3D-CNN and Bi-LSTM, initial denture data containing spatial geometric morphology information and standardized design parameters is automatically generated, providing an editable data foundation for subsequent multi-dimensional optimization.

[0044] S4. Based on the initial denture design data, calculate the morphological matching degree, adjust the occlusal parameters, and optimize the aesthetic parameters of the oral cavity feature data to obtain optimized denture design data.

[0045] For example, the initial denture design data is automatically generated by a 3D model generation algorithm based on generative adversarial networks (GANs), which can match the patient's basic oral anatomy features. However, it does not combine refined local morphological adaptation, dynamic occlusal coordination and oral aesthetic proportion constraints, resulting in potential problems such as insufficient local morphological fit, unreasonable occlusal contact relationship and uncoordinated appearance proportions.

[0046] Optionally, the Iterative Closest Point (ICP) algorithm can be used for global morphological matching quantification. By minimizing the sum of squared Euclidean distances between the source and target point clouds, the degree of fit between the initial denture shape and the patient's natural oral anatomy can be objectively evaluated, and the morphological deviation area can be accurately located. The calculation formula is as follows:

[0047]

[0048] in, This indicates the shape matching error. The total number of points in the point cloud data. The first point in the 3D point cloud representing the initial denture design data One point, It is the three-dimensional point cloud of the patient's oral cavity feature data and The nearest corresponding point. As a loop variable, it iterates through each point in the point cloud data, quantifying the morphological differences between the two by calculating the mean of the sum of squared Euclidean distances between corresponding points. The smaller the value, the higher the morphological matching degree.

[0049] Optionally, quantify the results using morphological matching. Based on this as the core adjustment basis, and combined with the occlusal function information carried by oral feature data, a support vector regression (SVR) model is used to perform global adaptation adjustments of occlusal parameters. By constructing a regression relationship between occlusal parameters and occlusal function indicators, static occlusal contact defects and potential dynamic occlusal interference are corrected. Assume the occlusal parameter vector is... The occlusal function index is The regression equation of the SVR model can be expressed as: ,in, It is a weight vector. It is a bias term. This is the allowable regression error. The number of dimensions representing the occlusal parameters is used to find the optimal value by training the SVR model. and This minimizes the error between the predicted occlusal function indicators and the actual requirements, thereby achieving optimized adjustment of occlusal parameters.

[0050] Optionally, while ensuring anatomical fit and occlusal function are adequate, and in accordance with the principles of oral natural aesthetic design, a deep learning-based image style transfer algorithm is employed to globally optimize and calibrate aesthetic parameters, coordinating aesthetic indicators such as denture contour proportions, arrangement, and edge lines. Through closed-loop optimization and correction across the three dimensions of morphology, function, and aesthetics, the detailed deficiencies in the initial denture design data are comprehensively compensated, allowing the denture design scheme to simultaneously meet anatomical fit requirements, physiological occlusal requirements, and aesthetic requirements, resulting in optimized denture design data with stronger comprehensive adaptability.

[0051] S5. Perform three-dimensional reconstruction, occlusal simulation verification, and design parameter correction on the optimized denture design data to obtain denture design data.

[0052] Optionally, the optimized denture design data is a composite data in the form of parametric and point cloud, which cannot be directly used for occlusal simulation and clinical processing applications. It needs to be transformed into a continuous and complete three-dimensional solid model through three-dimensional surface reconstruction. The final processing flow of simulation verification and closed-loop correction can be built based on NURBS (non-uniform rational B-spline) surface fitting and finite element analysis (FEA) technology.

[0053] Optionally, based on the optimized morphological data, NURBS surface fitting technology can be used to complete high-precision 3D model reconstruction. The NURBS surface equation can be expressed as:

[0054]

[0055] in: The coordinates of a point on the surface are with respect to the parameters. and The function; These are the coordinates of the control points, which define the shape of the surface. express The index of the direction control point ranges from 0 to... ; express The index of the direction control point, with a value range from arrive ; Corresponding control points The weights are used to adjust the distance between the surface and the control points; and They are direction and Direction Subsequent The B-spline basis functions, calculated recursively, determine the shape of the surface in the corresponding direction. Using the above formulas, a physical denture model suitable for mechanical and motion simulations is constructed.

[0056] Optionally, by combining dynamic occlusal motion feature data extracted from oral cavity feature data, finite element analysis (FEA) technology can be used to replicate the patient's actual mandibular movement trajectory and occlusal force pattern. In finite element analysis, stress... With strain The relationship can be expressed by Hooke's Law as follows: ,in, The stress tensor can be used to describe the stress state inside a material. The strain tensor reflects the degree of deformation of a material. The elasticity matrix contains parameters such as the material's elastic modulus and Poisson's ratio, which can be used to characterize the material's elastic properties. Based on this, a full-cycle dynamic engagement simulation test was conducted to comprehensively detect hidden problems such as engagement interference, stress concentration, and insufficient movement avoidance during dynamic motion.

[0057] Optionally, based on the full-domain results of the occlusal simulation, the least squares method is used to retrospectively trace the defect locations and parameter deviations in the denture design. The objective function of the least squares method can be expressed as:

[0058]

[0059] in, It is the objective function, which can be used to measure the error between the model's predicted value and the actual value; It is the sample size; It is the first The actual observed values ​​of each sample; It is the first Input features of each sample; It is the parameter vector to be optimized; Based on parameters and input The model's predicted value.

[0060] Optionally, by minimizing the above objective function, the design parameters can be specifically corrected and optimized in a closed loop, eliminating hidden defects left over from the early design stages. After reconstruction, simulation, and correction, the output final denture design data can fully adapt to the actual application scenarios of clinical dental restoration, possessing comprehensive characteristics of reasonable structure, kinematic adaptability, and feasible fabrication.

[0061] The aforementioned AI-based personalized denture design method based on multimodal oral data fusion achieves deep integration of three-dimensional oral scanning data, two-dimensional image data, and oral functional data through temporal synchronization and spatial registration of multimodal oral data. This comprehensively characterizes the complex relationship between oral soft and hard tissue morphology, occlusal function, and dynamic motion characteristics. By relying on differentiated feature mining and adaptive weighted fusion of local and global feature extraction branches, the matching degree between denture morphology and individual patient functional needs is improved, ensuring the consistency and repeatability of design results. Combined with dynamic occlusal motion simulation verification and parameter correction, interference and discomfort issues in real occlusal scenarios are avoided in advance, reducing the number of clinical adjustments and shortening the restoration cycle. The method fully leverages the advantages of multimodal data fusion to enhance the accuracy and practicality of personalized denture design.

[0062] In one embodiment, S1 includes:

[0063] S11. Perform timestamp alignment processing on the oral cavity 3D scan data, oral cavity 2D image data and oral cavity function data in the multimodal oral cavity data to obtain time-aligned data; wherein, the time-aligned data includes time-aligned oral cavity 3D scan data and time-aligned oral cavity 2D image data.

[0064] Various types of oral data operate independently using different acquisition terminals during the acquisition phase. Each terminal has its own independent timing system and sampling trigger logic, resulting in the inability to correlate the original timestamps of different modal data and thus failing to directly reflect the comprehensive oral information of a patient under the same chewing state and occlusal posture. This implementation method selects oral functional data as a global temporal benchmark. This data is dynamically and continuously acquired based on the physiological changes in mandibular movement and occlusion, resulting in stronger temporal continuity and higher physiological correlation, enabling precise matching of dynamic changes in the oral cavity.

[0065] For example, temporal interpolation correction is performed on oral cavity 3D scan data with reference to a baseline temporal sequence system. Assume the original time series of the oral cavity 3D scan data is... The corresponding morphological data is The baseline time series of oral function data is When constructing a time-series fitting model of morphological changes, the Lagrange interpolation formula can be used:

[0066]

[0067] in, Indicates at the base time The following is the three-dimensional morphological data of the oral cavity after interpolation correction; The first of the original three-dimensional oral cavity scan data A timestamp; For the corresponding Three-dimensional morphological data of the oral cavity; In the original three-dimensional oral cavity scan data, excluding the first Other timestamps besides the one; The first in the benchmark time series of oral function data A timestamp. This formula is used to complete the transitional morphology information within the time interval, correcting the original acquisition time sequence offset problem.

[0068] For example, image frame temporal matching correction is performed on oral two-dimensional image data. Let the original image frame sequence of the oral two-dimensional image data be... The corresponding collection time series is Based on stable anatomical landmarks within the images as references, a temporal mapping relationship is established. It is assumed that a monotonically increasing mapping function exists. This makes the corrected time series satisfy ,in, For the original two-dimensional oral imaging data Frame acquisition time; For the corrected first Frame acquisition time; mapping function Must meet That is, through the Through optimization and adjustment, the temporal stretching and compression adaptation of the image sequence is completed, so that the image acquisition timing is fully in line with the reference timing system.

[0069] By employing the aforementioned differentiated temporal correction methods, the global timestamps of the three types of data are uniformly calibrated, and time-aligned data with complete temporal synchronization is constructed. This ensures that each set of three-dimensional morphological data can be bound one-to-one with the corresponding two-dimensional image data and oral function data, thereby realizing the temporal linkage expression of static anatomical structure and dynamic physiological function and effectively solving the core problem of temporal dimension fragmentation of multi-source data.

[0070] S12. Extract feature points from the time-aligned oral cavity 3D scan data to obtain the source data point set.

[0071] Optionally, time-aligned 3D oral scan data records the spatial morphological information of all oral soft and hard tissues, including the patient's dentition, gingiva, and alveolar tissue, in the form of discrete point clouds. This data is massive and contains a high proportion of redundant information; directly performing spatial registration would significantly increase the computational load. Furthermore, invalid point clouds in non-feature regions can interfere with registration accuracy. A 3D spatial feature detection algorithm can be used to filter and analyze the 3D point cloud data. Based on the morphological changes of oral anatomical tissues, the algorithm uses changes in surface curvature, spatial normal vector distribution characteristics, and anatomical contour edge changes as core criteria to identify highly recognizable stable anatomical landmark regions within the oral cavity.

[0072] Optionally, surface curvature The expression can be ,in, and It is the covariance matrix Two eigenvalues. Covariance matrix. It can be derived from the point set within the local neighborhood of the point cloud. The covariance matrix was calculated. The calculation formula can be ,in, It is the number of points in the local neighborhood. Represents the local neighborhood of the point cloud. The three-dimensional coordinates of the points These are the centroid coordinates of a point set within a local neighborhood, and their calculation formula can be... ,in, and Corresponding to the covariance matrix The maximum and minimum eigenvalues ​​can be used to measure the degree of curvature of a surface in different directions.

[0073] Space normal vector The calculation is usually based on the covariance matrix of the local neighborhood of the point cloud. Eigenvalue decomposition. Covariance matrix. The eigenvector corresponding to the smallest eigenvalue is the spatial normal vector of that point. It is used to describe the orientation information of the point cloud surface.

[0074] Optionally, based on the above formula, the changes in surface curvature, spatial normal vector distribution characteristics, and anatomical contour edge changes are used as core criteria to identify stable anatomical landmark regions with high recognizability within the oral cavity. These landmark regions are concentrated in the cusp structures, incisal edges, alveolar ridge contour edges, and proximal contact areas of teeth. These structures exhibit strong morphological stability, significant individual differences, and are not easily affected by soft tissue deformation or acquisition angle. The algorithm traverses all 3D point cloud samples, filters spatial points that meet the anatomical feature determination criteria, removes invalid and redundant point clouds from smooth and homogeneous regions, and integrates the selected effective feature points to construct a source data point set with uniform structural distribution and high anatomical recognizability. This data can serve as a benchmark data source for 3D spatial registration, balancing subsequent computational efficiency with spatial matching accuracy.

[0075] S13. Extract feature points from time-aligned two-dimensional oral imaging data to obtain the target data point set.

[0076] Optionally, the time-aligned two-dimensional oral imaging data includes two types of imaging information: tomographic images and intraoral surface images. The data presents the internal tissue structure of the oral cavity and the appearance of the surface teeth in the form of a pixel matrix. The images contain large areas of homogeneous grayscale and irrelevant background information, which cannot be directly used for cross-modal space matching operations. This implementation uses a two-dimensional image feature detection algorithm to perform region-by-region analysis of the entire image. Combining the grayscale distribution characteristics of the oral images, the gradient changes of the contour edges, and the differences in tissue texture, it identifies stable feature regions such as tooth contour boundaries, root anatomical landmarks, and gingival margin lines.

[0077] Optionally, differentiated feature enhancement processing can be applied to different types of 2D images. Tomographic images focus on enhancing grayscale differences at hard tissue boundaries, while intraoral images focus on enhancing color boundaries between teeth and soft tissues, thereby improving the recognition ability of weak feature regions. During feature enhancement, the degree of grayscale or color difference enhancement can be quantified using the following formula:

[0078]

[0079] in, Indicates the first in the image The enhanced feature intensity of each pixel; For the first The original grayscale or color value of each pixel; It is the average grayscale / color value of the entire or local area of ​​the image, used to measure the overall benchmark; This is the strengthening coefficient; adjusting this parameter controls the strengthening amplitude. This is a non-linear adjustment index used to adjust the shape of the enhancement curve. At the same time, it can enhance the difference in high-contrast areas. This focuses on preserving details in low-contrast areas.

[0080] Optionally, after feature region enhancement, the Harris corner detection algorithm can be used to extract stable and unique pixel feature points within the image. The expression for its core response function can be: ,in, Corner response value for each pixel, A larger value indicates that the point is more likely to be a corner point; It is the structure tensor matrix of pixels, which is calculated by examining the image in... and The gradient in the direction is obtained; Representation matrix The determinant of the matrix reflects the product of eigenvalues ​​of the autocorrelation matrix of the local image; Representation matrix The trace is the sum of the elements on the main diagonal of the matrix; This is an empirical constant, typically ranging from 0.04 to 0.06, used for balancing. and The method effectively avoids temporary interference features caused by image shooting angle and lighting conditions, ensuring the stability and uniqueness of extracted feature points by assigning weights to them.

[0081] Optionally, all the selected two-dimensional feature points can be structurally integrated to obtain a standardized target data point set. This data set maintains consistency with the source data point set in terms of anatomical correspondence, providing a reliable target reference basis for subsequent cross-dimensional feature matching.

[0082] S14. Perform feature point matching on the source data point set and the target data point set to obtain matching feature point pairs.

[0083] For example, the source data point set is a set of anatomical features in a three-dimensional spatial coordinate system, while the target data point set is a set of image features in a two-dimensional pixel coordinate system. Since the two have different dimensions and coordinate rules, it is necessary to establish a cross-dimensional feature association mapping relationship in order to realize the subsequent spatial registration operation.

[0084] Optionally, a perspective projection mapping model can be constructed based on the calibration parameters of the imaging equipment to convert the pixel coordinates of the two-dimensional image feature points into virtual three-dimensional projection coordinates, thereby completing the dimensional unification transformation of the two types of feature point sets and eliminating the matching barriers caused by dimensional differences. After dimensional unification, a spatial distance evaluation model is constructed to quantitatively calculate the spatial deviation between the source data point set and the target data point set after projection transformation, and preliminary candidate matching combinations are selected based on the spatial deviation value. To avoid the impact of abnormal matching points and incorrect mapping relationships on the overall matching effect, a random sampling consensus algorithm can be introduced to iteratively screen and correct the candidate matching combinations, identify and eliminate incorrect matching points caused by tissue deformation and imaging interference, and retain effective combinations with stable anatomical correspondences. After multiple rounds of screening and optimization verification, feature points with accurate spatial correspondences and mutually matching anatomical positions are bound in pairs to form one-to-one matching feature point pairs. This data provides the core constraints for solving the subsequent spatial transformation matrix, ensuring the rationality of the spatial registration operation.

[0085] S15. Based on the matching feature point pairs, the spatial registration transformation matrix is ​​calculated.

[0086] Alternatively, the expression for the spatial registration transformation matrix can be: In the formula, This represents the spatial registration transformation matrix obtained from the final solution; The overall coordinate matrix representing the source data point set; The overall coordinate matrix representing the target data point set; This represents a global spatial linear transformation function, which can be used to realize spatial coordinate transformation between the source data point set and the target data point set.

[0087] Optionally, the core optimization objective in the calculation process is to minimize the overall spatial error of the two sets of feature points. The matrix solution operation is completed by relying on the least squares optimization algorithm. The spatial linear transformation function includes two basic transformation logics: spatial rotation operation and spatial translation operation. It can fully cover the rigid spatial adjustment requirements required for oral data registration and conforms to the objective physiological characteristics of human oral anatomical tissues without elastic deformation.

[0088] Optionally, in the actual calculation process, the spatial centroid coordinates of the two sets of points are first solved by relying on matching feature point pairs. The computational interference caused by spatial position offset is eliminated by decentralization, reducing the influence of translation components on the solution of the rotation matrix. Then, the correlation covariance matrix of the two sets of points is constructed, and the matrix decomposition and orthogonal matrix reconstruction are completed by combining the singular value decomposition algorithm. The spatial rotation component and the spatial translation component are solved in turn, and the two types of components are integrated to form a complete homogeneous spatial registration transformation matrix.

[0089] Optionally, the entire solution process uses the global spatial error convergence condition as the termination criterion for the operation, and continuously iterates and optimizes the matrix parameters until the overall spatial deviation of the two sets of feature points reaches the optimal state. This ensures that the transformation matrix obtained can accurately realize the spatial coordinate unification of multimodal data, while strictly following the spatial distribution law of oral anatomical structure, and avoiding structural distortion and anatomical misalignment during the transformation process.

[0090] S16. Based on the spatial registration transformation matrix, perform spatial coordinate transformation on the source data point set to obtain the registered oral cavity data.

[0091] Optionally, the spatial registration transformation matrix integrates standardized spatial rotation parameters and spatial translation parameters, enabling a globally unified rigid coordinate transformation of all points within the source data point set. The transformation process does not change the morphology and relative positional relationship of the oral tissue itself, but only adjusts the spatial reference system to which the overall coordinates belong.

[0092] Optionally, for the three-dimensional coordinates of the source data point set It can be mapped to coordinates in a unified spatial system using the following transformation formula. :

[0093]

[0094] in, Rotation matrix , ; The cosine value represents the rotation along different coordinate axes, describing the rotational relationship in three-dimensional space; Translation vector , , , They represent in , , Translation along the axis is used to adjust the position of the coordinate origin.

[0095] Optionally, all three-dimensional coordinate parameters from the source data point set are uniformly substituted into the aforementioned spatial transformation operation model. Point-by-point coordinate conversion is then performed using the transformation matrix, allowing the originally independent coordinate system of the three-dimensional scan data to be completely mapped to the unified spatial system corresponding to the two-dimensional image data. After completing the global coordinate transformation, global registration accuracy is verified by calculating matching feature point pairs. and Euclidean distance between To assess registration error, the expression for Euclidean distance can be: ,in, and The first and second parts are respectively the first and second parts before and after the transformation. The coordinates of each feature point. The residual spatial deviation of multiple feature point pairs is compared through sampling. Take their average value To determine the overall registration effect, if If the value exceeds a reasonable range, the feature point matching combination and transformation matrix parameters are optimized in reverse iteration, and the coordinate transformation process is re-executed.

[0096] Optionally, adaptive spatial correction is performed simultaneously on the two-dimensional oral imaging data. The spatial reference datum of the image is adjusted by combining the mapping rules of the transformation matrix, thereby achieving bidirectional spatial alignment between the three-dimensional data and the two-dimensional imaging data. After complete coordinate transformation and accuracy verification, various multimodal oral data achieve synchronous unification in the temporal dimension and coordinate fusion in the spatial dimension, integrating and generating registered oral data with complete information and unified datum, eliminating the heterogeneity defects of multi-source data.

[0097] In one embodiment, the feature fusion model includes a local feature extraction branch, a global feature extraction branch, a first feature distance calculation unit, a second feature distance calculation unit, a fusion weight calculation unit, a feature fusion unit, and a feature mapping unit, wherein S2 includes:

[0098] S21. Input the registered oral cavity data into the local feature extraction branch to perform local feature extraction and obtain local anatomical feature data.

[0099] Optionally, the local feature extraction branch can be constructed using a multi-level convolutional network structure to meet the detailed feature mining needs of multimodal fusion data, relying on multi-scale convolution operations, feature activation processing, and hierarchical downsampling processing. After preprocessing, the registered oral cavity data is converted into a standardized multi-channel feature tensor, which can be directly input into the local feature extraction branch for computation.

[0100] Optionally, in multi-scale convolution operations, for the input feature map (in For feature map height, The width of the feature map. (number of channels), through convolution kernel ( , The first The height and width of each convolutional kernel, The convolution operation is performed on the number of output channels. The calculation process can be expressed as follows:

[0101]

[0102] in, For the first The output feature map after each convolutional kernel operation This represents the convolution operation. This corresponds to the bias term. Convolutional kernels of different scales are adjusted by setting different bias terms. and It can specifically capture the morphological differences in the subtle anatomical structures of the oral cavity.

[0103] Optionally, in the standardization and normalization process between network layers, taking batch normalization as an example, for the first... Layer input Its normalized output The calculation is as follows:

[0104]

[0105] in, This represents the mean of the data in this batch. This represents the variance of the batch of data. It is a very small constant (usually 1). (Order level), used to prevent the denominator from being zero. Through normalization, the ability to express subtle features can be enhanced, while the interference of irrelevant noise information can be weakened.

[0106] Optionally, within each branch, receptive fields of different scales are set to specifically capture morphological differences in subtle local anatomical structures of the oral cavity. During the computation, edge features, texture features, and local contour features are extracted and enhanced layer by layer, focusing on microscopic anatomical information such as the surface morphology of a single tooth, the gingival margin structure, the morphology of the proximal surfaces of teeth, and the contour of local alveolar tissue, to accurately capture the patient's personalized local oral structural differences.

[0107] Optionally, after multiple rounds of hierarchical feature extraction and multi-scale information fusion, the anatomical representation information of all local dimensions is integrated to generate structured local anatomical feature data. This data can accurately reflect the individual differences in oral microanatomy and provide detailed feature support for personalized design needs such as denture edge fitting, proximal surface fitting, and local morphological replication.

[0108] S22. Input the registered oral cavity data into the global feature extraction branch to perform global feature extraction and obtain global occlusal feature data.

[0109] Optionally, the global feature extraction branch can be constructed based on a global attention coding structure, which has the ability to capture long-distance feature associations and can overcome the limitations of local field of view to achieve a holistic analysis of the structural and functional association information of the entire oral cavity. This branch uses fully registered oral data as input, abandons the logic of local fragmented extraction, and carries out feature mining from macroscopic dimensions such as the overall distribution of the dentition, the relationship between the positions of the upper and lower jaws, and the linkage of occlusal movements.

[0110] Optionally, a global feature correlation matrix can be constructed based on a multi-head attention mechanism to quantitatively analyze the interaction relationships between different oral structures. The computational process of the multi-head attention mechanism can be represented as follows: ,in, This is a query vector, which can be used to retrieve relevant features from oral data; The key vector (Key) can be... Provide search criteria; The value vector can contain the actual oral cavity feature information to be extracted. This indicates the number of attention heads, each of which can capture oral cavity structural features from different angles. Indicates the first The calculation results of each attention point can be obtained through calculation. and The similarity, for We obtain the result by weighted summation. It is the output weight matrix, which can be used to perform a linear transformation on the multi-head attention calculation results to generate the final feature representation.

[0111] Optionally, during global feature extraction, the focus is on capturing the linkage between static occlusal alignment and dynamic mandibular movement. Simultaneously, the global distribution patterns after multimodal data fusion are integrated to achieve a correlated feature expression of anatomical structure and physiological function. Within each branch, a multi-layered coding structure can be used to abstract and upgrade global features, gradually filtering out superficial and ineffective global information and strengthening the weight of core global features such as occlusal function, tooth arrangement, and jaw position. The final output global occlusal feature data can comprehensively characterize the patient's overall oral cavity occlusal functional state and macroscopic structural distribution characteristics, complementing local anatomical feature data and compensating for the technical limitations of local features in reflecting the overall occlusal coordination.

[0112] S23. Input the local anatomical feature data into the first feature distance calculation unit to calculate the feature distance and obtain the local anatomical feature distance.

[0113] Optionally, the first feature distance calculation unit can be used to quantitatively evaluate the standardization and fit of local anatomical feature data, and perform calculations based on a multi-dimensional feature difference measurement model. A standardized oral anatomical feature template is used as a reference benchmark. This template is constructed based on the statistical regularities of features from a large range of clinical oral samples and possesses a universal anatomical structure evaluation standard. The extracted local anatomical feature data is compared with the standard anatomical feature template dimensionally, and a composite feature distance calculation model is constructed by combining multiple evaluation dimensions, including differences in feature spatial distribution, differences in feature numerical distribution, and differences in feature structural correlation.

[0114] Optionally, let the local anatomical feature data be... The standard anatomical feature template is ,in The number of feature dimensions, Indicates the first Dimensional local anatomical feature data, Indicates the first Standard anatomical feature data of dimensionality, distance of local anatomical features The expression can be ,in, For the first The weighting coefficients of the dimensional features are used to reflect the differences in importance of different dimensional features in the evaluation. This formula transforms the abstract differences in anatomical features into quantifiable feature distance parameters by summing the weighted square differences and then taking the square root. The magnitude of the feature distance directly corresponds to the degree of individualized differences in local anatomical features and the effectiveness of the features. Stable local anatomical structures that conform to physiological standards correspond to smaller feature distances, while areas with structural abnormalities or specialized tissue morphology will exhibit larger feature distances.

[0115] Optionally, by performing dimension-wise calculations across the entire domain and integrating the differential quantification results of all local features, a complete set of local anatomical feature distances can be obtained. This data provides a quantitative reference for the dynamic allocation of subsequent fusion weights, enabling differentiated control based on feature quality.

[0116] S24. Input the global occlusal feature data into the second feature distance calculation unit to calculate the feature distance and obtain the global occlusal feature distance.

[0117] Optionally, the second feature distance calculation unit uses a standardized occlusal function feature template as a reference system to conduct a global differential quantitative analysis of global occlusal feature data, focusing on the feature adaptability evaluation of global dimensions such as oral occlusal function patterns, jaw position correlation rules, and dynamic movement characteristics. Internally, the second feature distance calculation unit can employ a correlation degree measurement algorithm adapted to global high-dimensional features, comprehensively considering the distribution deviation of global feature vectors, the degree of linear correlation, and the degree of functional pattern matching to construct a feature distance calculation model adapted to occlusal function evaluation.

[0118] Optionally, let the global occlusal feature data be... The standardized occlusal function feature template is ,in, This represents the number of dimensions for the global bite feature. Indicates the first 3D global occlusal feature data, Indicates the first The standard occlusal functional feature data of the dimension. Then the global occlusal feature distance. It can be calculated using the following formula:

[0119]

[0120] In the above formula, by calculating the mean relative deviation between the global occlusal feature data and the data of each dimension of the standard template, the degree of difference between the patient's personalized occlusal features and the normal physiological occlusal features can be quantitatively determined, and conventional occlusal features, special pathological features and personalized occlusal features can be accurately distinguished.

[0121] Optionally, global occlusal feature distances can objectively reflect the stability and structural coordination of occlusal function. Feature data with coordinated occlusal relationships and movement patterns conforming to physiological standards correspond to smaller distance values, while special conditions such as occlusal interference and jaw abnormalities will increase the feature distance values. The final output set of global occlusal feature distances can form a bidirectional reference with local anatomical feature distances, jointly supporting the accurate calculation of fusion weights.

[0122] S25. Input the local anatomical feature distance and the global occlusal feature distance into the fusion weight calculation unit to calculate the fusion weight.

[0123] Alternatively, the expression for the fusion weights can be:

[0124]

[0125] In the formula, Representing the The fusion weight coefficients corresponding to the features; Representing the Distance of local anatomical features; Representing the Global bite feature distance; This represents a pre-set scale adjustment parameter, which can be used to adjust the magnitude of the influence of feature distance on the weight value; This represents the natural exponent operation, which can be used to achieve a negative correlation between feature distance and weight value.

[0126] Optionally, the fusion weight calculation unit relies on the normalized exponential function to complete the adaptive weight allocation calculation. The scale adjustment parameter can be used to adjust the influence of feature distance on the weight value, balance the weight discrimination of different feature distance intervals, and ensure the smoothness and rationality of weight allocation.

[0127] Optionally, during the calculation, all local anatomical feature distances and global occlusal feature distances are first integrated to construct a unified set of feature distances. Each independent feature distance is then substituted into an exponential decay calculation logic, using a negative exponential function to achieve a negative correlation between the feature distance and the weight value. The smaller the feature distance value, the higher the exponential calculation result, and the larger the proportion of the allocated fusion weight, thereby achieving weight enhancement for highly adaptable and reliable features.

[0128] Optionally, the exponential calculation results of all features are globally summed, and the sum is used as the normalized denominator to proportionally convert the exponential calculation results of individual features, ensuring the global normalization constraint of all fusion weight coefficients. This weight calculation logic can dynamically adjust the fusion ratio of the two types of features according to the individual differences of the patient's oral cavity features, automatically strengthening the fusion ratio of high-quality and effective features, and weakening ineffective features with large deviations and low fit, thus achieving adaptive allocation of multiple types of features.

[0129] S25. Input the fusion weights, local anatomical feature data and global occlusal feature data into the feature fusion unit for weighted fusion to obtain fused feature data.

[0130] Optionally, the feature fusion unit first completes the data dimension unification adaptation processing of the two types of heterogeneous features, eliminates the differences in dimensional structure and data form between local anatomical feature data and global occlusal feature data, constructs a feature system of the same dimension that can be directly computed, and ensures the smooth execution of weighted fusion operation.

[0131] Optionally, local anatomical feature data can be ,in, Represents the first in the local anatomical feature vector Each feature dimension This represents the total number of dimensions for local anatomical features; global occlusal feature data can be... ,in Represents the first element in the global bite feature vector. Each feature dimension This represents the total number of dimensions of the global bite features. The adaptive fusion weight coefficient vector can be... ,in After unifying the dimensions through methods such as zero padding, the local anatomical feature data and the global occlusal feature data are weighted and superimposed channel-by-channel and dimension-by-dimensional based on the adaptive fusion weight coefficients obtained from the solution. The expression for the fused feature data can be: ,in, The th element of the fused feature vector One dimension, For the first The fusion weights corresponding to each dimension have a range of values. This formula allocates the information proportion of the two types of features in the composite feature according to the weight ratio. For local features with small anatomical deviations, it automatically increases the fusion proportion, and for global features with stronger occlusal function coordination, it strengthens the information output, thereby achieving personalized feature fusion ratio.

[0132] Optionally, the fusion process retains the advantages of detailed representation of local feature data and the functional correlation advantages of global feature data, achieving deep coupling between microscopic anatomical information and macroscopic functional information, and avoiding the information loss problem caused by single feature fusion. Simultaneously, feature correlation constraints can be introduced to ensure that the internal structural logic of the fused feature data conforms to the physiological correlation between oral anatomy and occlusion, preventing information conflicts in cross-feature fusion. After global weighted fusion and logical verification, fused feature data with complete information coupling and reasonable weight allocation is generated, completing the integrated integration of multimodal deep features.

[0133] S26. Input the fused feature data into the feature mapping unit for feature dimensionality reduction to obtain oral cavity feature data.

[0134] Optionally, the feature mapping unit can adopt a non-linear autoencoder mapping structure design. Its core function is to compress the redundant dimensions of the fused feature data, condense the core effective information, reduce the complexity of subsequent model operations, and at the same time retain the key correlations within the fused feature data.

[0135] Optionally, the encoding stage progressively compresses feature dimensions through multi-layer nonlinear transformations, filtering out repetitive, noisy, and redundant information generated during the fusion process, focusing on the core anatomical morphology, occlusal relationships, and oral aesthetic structural features required for denture design. The specific encoding process can be represented as follows: ,in, Indicates the first Hidden layer output of the layer; Indicates the previous layer (the first layer) The hidden layer output of the layer; It is the first The layer weight matrix is ​​used to control the influence of the previous layer's output on the current layer; It is the first The layer's bias vector is used to adjust the position of the activation function; Non-linear activation functions, such as ReLU, can be used to introduce non-linear relationships, enabling the model to learn complex feature maps.

[0136] Optionally, feature distribution reconstruction and correction are performed simultaneously during the decoding stage to ensure that dimensionality reduction does not destroy the correlation logic between features and individualized differences, thus maintaining the integrity of multimodal fusion information. The overall mapping process relies on unsupervised feature learning logic to optimize parameters, adaptively adapting to the distribution patterns of fusion feature data from different patients, and ensuring the generalization ability of the dimensionality-reduced data. For example, assume the input data is... Latent variables are obtained through encoding. The decoding process can be represented as ,in, It is the data reconstructed after decoding; These are latent variables obtained during the encoding stage; and These are the weight matrix and bias vector during the decoding process, respectively. As an activation function, it can be used to map latent variables back to the original data space.

[0137] Optionally, after feature mapping dimensionality reduction and key information condensation, the output is oral feature data with simplified dimensions and higher information density. This data serves as the core intermediate data connecting the feature fusion stage and the denture morphology generation stage, and can fully carry all the personalized key information of the patient's oral cavity.

[0138] In one embodiment, the denture morphology generation model includes a feature encoding layer, a morphology decoding unit, and a parameter mapping unit, wherein S3 includes:

[0139] S31. Input the oral cavity feature data into the feature encoding layer for feature encoding to obtain the latent space feature vector.

[0140] Optionally, the feature encoding layer can be composed of multiple stacked deep encoding networks, focusing on high-order abstraction of oral feature data and mining the hidden nonlinear correlations and morphological mapping patterns within the feature data. After receiving standardized oral feature data, the encoding layer gradually compresses the surface expression information of the features through multi-layer fully connected transformation and nonlinear activation processing, extracting deep hidden features that are strongly correlated with tooth morphology, tooth arrangement, and occlusal surface structure.

[0141] Optionally, the expression for the encoding operation can be ,in, Indicates the number of layers in the coding network. For the first The hidden layer feature vector output by the layer; It is the previous layer (i.e., the first layer) The output of the layer; For the first The layer weight matrix can be used to perform linear transformations on the input features; It is the first The layer bias vector can be used to adjust the result after linear transformation; Nonlinear activation functions, such as ReLU, introduce nonlinearity to enhance the network's expressive power. The expression can be .

[0142] Optionally, during the encoding process, a hierarchical parameter constraint mechanism is used to enhance the preservation of individual oral cavity features and filter out generic, indifferent features, ensuring that the encoded latent vectors fully carry the patient's unique oral cavity structural information. Through multi-layered progressive encoding, the dimensionally simplified oral cavity feature data is mapped to a high-dimensional latent space system, generating latent space feature vectors with unified dimensions and stronger feature representation capabilities. The latent space feature vectors represent abstract morphological semantics and do not directly present geometric structural information, but they completely store all the core constraints required for denture morphology generation. They are a crucial intermediate representation connecting oral cavity feature data and physical geometry, ensuring the personalized generation effect of subsequent morphological decoding.

[0143] S32. Input the latent space feature vector into the morphological decoding unit for morphological decoding to obtain denture morphological point cloud data.

[0144] Optionally, the morphological decoding unit can be constructed based on a 3D generative decoding network, possessing the core capability of restoring 3D spatial geometric data from abstract latent vectors. It uses latent space feature vectors as the decoding driving source, performing upsampling and spatial morphological restoration operations layer by layer. The decoding process follows the following mathematical logic: first, the latent space feature vector is defined as... ,in Given the latent vector dimension, its mapping relationship through the morphological decoding unit can be represented as follows: ,in, The output denture morphology point cloud data is a three-dimensional point set. , This represents the number of points in the point cloud data; The mapping function represents the 3D generative decoding network, which consists of multiple layers of 3D transposed convolutions.

[0145] Optionally, in specific operations, taking the three-dimensional transpose convolution operation as an example, its calculation process can be represented as follows:

[0146]

[0147] in, Is the output feature map in Location feature values; Input feature map; , , It is the stride parameter, used to control the sliding stride of the convolution kernel in each dimension of three-dimensional space; , , These are the dimensions of the convolution kernel in the three dimensions of three-dimensional space; Indicates that the convolution kernel is in Position weight; This is the bias term. Through this operation, the decoding network can expand the spatial dimension and construct a continuously distributed set of discrete spatial points.

[0148] Optionally, the decoding network follows oral anatomy and physiology constraints, automatically limiting the range of changes in tooth physiological structure during morphological generation to avoid deformities that violate common sense about oral physiology. Simultaneously, based on the occlusal and anatomical information carried by the latent space feature vectors, it adaptively adjusts morphological features such as the overall tooth contour, surface curvature, and structural proportions. The final decoded denture morphology point cloud data consists of a massive number of spatially discrete points, completely covering the entire external structure of the denture. Each point possesses independent three-dimensional spatial coordinate information, accurately representing the overall shape and local surface details of the initial denture, providing raw geometric data support for subsequent 3D model reconstruction and parameter analysis.

[0149] S33. Input the denture morphology point cloud data into the parameter mapping unit to extract vertex coordinates and obtain the denture vertex coordinate set. Perform parameter mapping on the denture morphology point cloud data through the parameter mapping unit to obtain denture morphology parameter data. Integrate the denture morphology parameter data and the denture vertex coordinate set through the parameter mapping unit to obtain the initial denture design data.

[0150] Optionally, the parameter mapping unit integrates both geometric analysis and parametric modeling functions. First, it performs a global traversal analysis of the denture morphology point cloud data, extracting the three-dimensional coordinate information of all spatial points in batches. Assuming the point cloud data contains... Each vertex has three-dimensional coordinates, which can be represented as a set of three-dimensional vectors. ,in, Let the set of vertex coordinates of the denture be... Indicates the first The three-dimensional coordinate vector of each vertex. , , These are the coordinates of the vertex in the Cartesian coordinate system. , , axis coordinate values, The value range is from 1 to It fully preserves the original geometric morphology information generated by decoding.

[0151] Optionally, based on a pre-defined system of analytical rules for oral restoration parameters, the surface structure, contour dimensions, angular relationships, and contact structures of the point cloud data are subjected to full-domain quantitative analysis. Taking the calculation of surface curvature parameters as an example, for points on the surface... Its Gaussian curvature It can be calculated using the following formula:

[0152]

[0153] in, and Points The first and second basic form matrices at the location. This represents the determinant operation of a matrix. The formula quantizes the points by the ratio of the determinants of two basic matrix forms. The curvature of the surface is determined, which in turn forms standardized denture morphological parameter data. The parameters cover multiple design dimensions such as geometric dimensions, surface curvature, occlusal structure, and proximal fit.

[0154] Optionally, after independently parsing the coordinate set and morphological parameters, the concrete set of denture vertex coordinates and the quantified denture morphological parameter data are integrated according to data association and binding rules to establish a one-to-one correspondence between morphological structure and design parameters. The integrated initial denture design data combines the visualization capabilities of a geometric model with the adjustability of design parameters. It can be directly used for 3D model preview and also supports subsequent quantitative adjustments to occlusal and aesthetic parameters, enabling automated, personalized, and rapid generation of the denture's basic morphology.

[0155] In one embodiment, S4 includes:

[0156] S41. Extract the denture vertex coordinates from the initial denture design data and extract the reference vertex coordinates from the oral cavity feature data.

[0157] Optionally, from the set of denture vertex coordinates built into the initial denture design data, the spatial vertex coordinates of all denture structures can be completely extracted to form a complete denture vertex coordinate system, the expression of which can be: ,in, Represents the coordinate system of the denture vertices. The first vertex in the denture coordinate system The three-dimensional coordinates of each vertex. This represents the total number of vertices in the denture vertex coordinate system. The coordinate information fully covers all external surfaces and contact surfaces of the denture, ensuring the globality of morphological evaluation.

[0158] Optionally, relying on the anatomical benchmark information such as the patient's natural dentition, alveolar tissue, and adjacent tooth structures stored within the oral cavity feature data, the standard anatomical benchmark region inside the oral cavity used for denture fitting evaluation is located. Continuous spatial point information of the benchmark region is extracted, and a reference vertex coordinate system is constructed, the expression of which can be: ,in, Indicates the reference vertex coordinate system. Represents the first vertex in the reference vertex coordinate system The three-dimensional coordinates of each vertex. The total number of vertices in the reference vertex coordinate system.

[0159] Optionally, the two types of vertex coordinate systems maintain the same sampling distribution rules and point density distribution patterns, which can be achieved by setting a uniform sampling interval. With density function This ensures that subsequent matching calculations can achieve point-to-point correspondence across the entire domain. The reference vertex coordinates are taken entirely from the patient's actual oral anatomy features, possessing unique individual differences, and can serve as an objective benchmark for evaluating denture morphology fit, ensuring that the morphology matching evaluation fully conforms to the patient's own oral conditions.

[0160] S42. Calculate the morphological matching degree between the denture vertex coordinates and the reference vertex coordinates to obtain the morphological matching degree.

[0161] Optionally, the expression for morphological matching degree can be: In the formula, The overall morphological fit value is a dimensionless quantitative indicator used to characterize the degree of fit between the denture shape and the patient's oral anatomy. The lower the value, the better the fit between the denture shape and the oral anatomy. It is the first of the coordinate vectors of the denture vertex. One portion, Indicates the first in the sampling point sequence One point, . It is the first vertex coordinate vector of the oral cavity reference. Each component. The total number of sampling points representing the coordinates of the denture apex is averaged by averaging the sum of squared deviations of all points to eliminate the influence of the number of sampling points on the results, making the matching degree values ​​more universal and comparable. The square operation, representing the absolute value, can be used to quantify the coordinates of a single set of corresponding vertices (i.e., the first set of vertices). The vertex of the denture and the first Spatial deviation (at each reference vertex). By squaring, the error proportion in high-deviation areas can be amplified, highlighting areas of local morphological abnormalities.

[0162] Optionally, taking the calculation of the matching degree between two simple geometric shapes (such as triangles) on a two-dimensional plane as an example, assuming the coordinates of the vertex of the denture triangle are... , , The coordinates of the vertex of the reference triangle are , , ,at this time The formula for calculating morphological matching degree is then expanded as follows:

[0163]

[0164] The formula for calculating morphological fit can employ global mean square error calculation logic. By calculating the spatial deviation between the coordinates of the denture vertices and the coordinates of the reference anatomical vertices point by point, it quantifies the degree of fit between the overall morphology of the denture and the patient's natural oral structure. In actual 3D oral model calculations, the coordinates of each vertex are represented as a 3D vector. The operational logic is the same as the two-dimensional example, except that it adds... Dimensional deviation calculation. Based on this quantitative calculation method, the morphological adaptation effect can be digitally and objectively evaluated, eliminating the uncertainty of subjective human judgment and providing quantitative data support for subsequent precise parameter adjustments.

[0165] S43. Based on the morphological matching degree, adjust the occlusal parameters of the initial denture design data to obtain the adjusted denture design data.

[0166] Optionally, by combining the overall numerical value of the global morphological matching degree with the distribution pattern of deviations in local areas, the adjustment priorities for different occlusal structural regions can be determined, with a focus on directional parameter correction for occlusal surfaces, proximal surfaces, and cusp contact structures where matching degree deviations are concentrated. To quantify the regional adjustment priorities, a priority coefficient calculation formula can be introduced:

[0167]

[0168] in, Indicates the first The adjustment priority coefficient for each occlusal structure region indicates that the region needs to be adjusted first; For the first The morphological deviation of a region can be used to reflect the degree of difference between that region and the ideal morphology. Representing all The sum of morphological deviations of each occlusal structural region is used for normalization. For the first The weighting coefficients for each region are set manually based on the functional importance of the region (such as the higher weighting of key areas like the occlusal surface and cusps).

[0169] Optionally, relying on the global occlusal feature information carried in oral cavity feature data and combined with the physiological movement law of the human mandible, the core occlusal parameters such as denture occlusal height, cusp angle, occlusal surface curvature, and adjacent tooth contact relationship can be adaptively fine-tuned. The adjustment process follows the principle of occlusal physiological balance, balancing the static occlusal contact distribution with the dynamic movement avoidance requirements, eliminating functional defects such as occlusal high points, occlusal interference, and excessively tight or large interproximal contacts. An occlusal contact force balance formula can be introduced. The equilibrium state of the resultant force at each contact point in the horizontal direction during static engagement is quantified; among which, Representing the The magnitude of the contact force at each biting contact point can be used to reflect the first biting contact point. The biting force borne by each biting contact point; For the first The angle between the force direction at each contact point and the horizontal direction; This represents the total number of occlusal contact points. When the formula result approaches 0, it indicates that the static occlusal contact distribution has reached a state of equilibrium.

[0170] Optionally, the adjustment range of adaptive control parameters can be adjusted according to the degree of morphological deviation in different areas. This ensures that while correcting local morphology, the integrity and morphological coordination of the overall denture structure are maintained, avoiding new structural imbalances caused by excessive local adjustments. After completing the iterative adjustment of the occlusal parameters across the entire domain, the local morphological matching effect and occlusal contact relationship are re-verified to ensure that the adjusted denture structure can adapt to the patient's natural occlusal movement pattern, generating more reasonable denture design data for occlusal function.

[0171] S44. Optimize the aesthetic parameters of the adjusted denture design data to obtain optimized denture design data.

[0172] Optionally, based on the principles of natural oral restoration aesthetic design, and considering the morphological proportions, alignment curvature, and edge characteristics of the patient's natural dentition, a comprehensive aesthetic parameter optimization of the prosthesis design data can be performed. The core optimization process is explained below using mathematical formulas:

[0173] Optionally, optimizing the proportions of the tooth profile requires ensuring that the dimensions of all parts of the tooth are coordinated, which can be described by the following formula:

[0174]

[0175] in, The optimized tooth contour proportion index; Indicates the crown height, Indicates the length of the tooth root; Mesial width of the teeth Distal width of the tooth; and The weighting coefficient is adjusted individually based on the characteristics of the natural dentition to balance the influence of different dimensions on the overall proportion.

[0176] Optionally, to ensure that the curvature of the denture dentition matches that of the natural dentition, the least squares method can be used to fit the dentition curve, as shown in the following formula:

[0177]

[0178] In the formula, The number of teeth in the dentition; Indicates the first The coordinates of the position of each tooth in the oral cavity coordinate system; and Given the curve parameters to be solved, by minimizing the above error function, the denture arrangement curve that best matches the curvature of the natural dentition can be obtained.

[0179] The smoothness of the neckline shape can be measured by the rate of change of curvature. The optimization goal is to make the curvature change of the neckline curve continuous and natural, and the rate of change of curvature... The expression can be:

[0180]

[0181] in, For the curvature of the curve, This refers to the arc length parameter of the curve. and These are curve coordinates. and For arc length The first derivative, and This is the second derivative. During the optimization process, control... The range of fluctuations is adjusted to achieve a smooth transition of the neckline.

[0182] Optionally, the uniformity of the surface transition can be quantified by calculating the standard deviation of the angle between the normal vectors of adjacent surfaces. The expression can be ,in, This represents the number of adjacent surfaces. It is the first The angle between the normal vectors of adjacent surfaces, This is the average of these included angles. By reducing... The value of can improve the uniformity of the transition of the curved surface of the denture.

[0183] Optionally, the optimization covers core aesthetic indicators such as tooth contour proportions, dentition curvature, cervical margin morphology, and surface transition uniformity. It strictly matches and calibrates to the aesthetic characteristics of the patient's natural teeth, ensuring the prosthesis's appearance is harmonious and integrated with the natural teeth. Based on global proportional constraints, it corrects aesthetic defects such as localized contour imbalances, abrupt dentition curvature, and harsh edge transitions, maintaining the overall symmetry and smooth lines of the dentition. Aesthetic optimization is achieved entirely through quantitative parameter control, refining the appearance details without compromising anatomical fit accuracy and occlusal stability, balancing the clinical practicality and aesthetics of the restoration. After multi-dimensional aesthetic parameter synergistic optimization, the overall appearance of the prosthesis is highly consistent with the patient's natural oral aesthetic characteristics, ultimately obtaining optimized prosthesis design data that balances anatomical fit, occlusal stability, and aesthetics.

[0184] In one embodiment, S5 includes:

[0185] S51. Perform three-dimensional mesh reconstruction on the optimized denture design data to obtain three-dimensional denture model data.

[0186] Optionally, using the set of denture vertex coordinates within the optimized denture design data as the basic data source, a three-dimensional surface mesh reconstruction algorithm can be employed to continuously fit the discretized denture point data onto a surface, constructing a closed and complete triangular mesh model. During the three-dimensional surface mesh reconstruction process, surface fitting formulas can be used. The discrete data points are processed, among which, This represents a point on the fitted 3D surface. These are the parametric coordinates on the curved surface; For control points, where The value range is from 0 to , The value range is from 0 to These control points determine the shape of the surface; and They are direction and Direction Subsequent Secondary basis functions are used to control the shape changes of a surface in the corresponding direction.

[0187] Optionally, during reconstruction, the uniformity of mesh patch distribution and the smoothness of surface transitions are optimized, and the mesh density is adaptively adjusted according to the structural detail requirements of different areas of the denture. Mesh density adjustment can be achieved by controlling the area of ​​each triangular patch; the formula for calculating the area of ​​a triangular patch is as follows: ,in, Represents the area of ​​the triangular facet; , , These are the three-dimensional coordinate vectors of the three vertices of the triangle; This represents the vector cross product operation. In fine structural regions such as the occlusal surface, neck margin, and adjacent surfaces, the area calculated by the above formula is reduced. Refine the mesh; increase the area in the smooth base surface region. To simplify the mesh structure and balance model accuracy with computational efficiency.

[0188] Optionally, mesh topology defect detection and repair can be carried out simultaneously to eliminate modeling defects such as isolated meshes, overlapping patches, and broken boundaries, ensuring the closure and structural integrity of the denture 3D model data. The reconstructed denture 3D model data has a standard and universal 3D model format, which is compatible with various simulation analysis software and digital processing equipment.

[0189] S52. Extract occlusal motion feature data from oral cavity feature data.

[0190] Optionally, the oral feature data integrates multimodal oral functional information, internally preserving the patient's personalized mandibular movement patterns and occlusal physiological characteristics. It selectively filters movement-related feature information from the overall oral feature data, separating and extracting trajectory features and movement constraint parameters for various physiological movements such as open-and-close jaw movements, lateral occlusion movements, and protruding occlusion movements. Occlusal movement feature data can include key information such as the overall movement trajectory change patterns, the limit range of jaw movement, and the synergistic constraint relationships of occlusal movements, enabling a complete replication of the patient's unique oral dynamic movement patterns, distinct from generalized standardized movement templates.

[0191] For example, the law governing the change of the global motion trajectory can be expressed as follows: ,in, Represents time The three-dimensional spatial position vector of the mandibular movement at any given moment can be used to describe the trajectory of the mandible within the oral cavity; The number of fundamental functions constituting the trajectory can be used to reflect the complexity of trajectory decomposition. It is the first Basic functions The weighting coefficients can be used to adjust the contribution of each basic function to the overall trajectory; For the first A time The changing basic functions, such as trigonometric functions and polynomial functions, are combined to simulate the movement trajectory of the mandible; It is a three-dimensional offset vector that can be used to adjust the starting position and overall offset of the trajectory in space.

[0192] For example, the cooperative constraint relationship of biting motion can be expressed as: ,in, Indicates the first The motion component and the first A measure of the collaborative constraint relationship between individual motion components; and The first The and the first The position vector of each motion component in three-dimensional space; express Compared to The partial derivative matrix can be used to describe the relationship between changes in two motion components; It is a unit vector used to measure the direction of association between two motion components; It is a set constraint threshold that can be used to determine whether the coordination relationship between two motion components meets the physiological requirements of motion. Exceeding this threshold indicates an abnormal motion relationship.

[0193] Optionally, using occlusal motion feature data as motion-driven constraints for occlusal simulation can ensure that the simulation process is more in line with the patient's real physiological movement habits, avoid the simulation result deviation caused by using a general motion model, make occlusal simulation verification more in line with actual clinical use scenarios, and improve the comprehensiveness and accuracy of defect detection.

[0194] S53. Based on the three-dimensional model data of the denture and the occlusal motion characteristic data, perform occlusal motion simulation to obtain occlusal simulation result data.

[0195] Optionally, the three-dimensional model data of the denture and the occlusal motion feature data are imported into the dynamic simulation calculation system. The system uses the patient's actual mandibular movement trajectory as the driving condition to reproduce the full range of natural occlusal motion processes and simulate various oral motion states such as daily chewing, occlusal closure, and lateral sliding.

[0196] Optionally, during simulation, the interaction between the denture and surrounding tissues can be described by dynamic equations, such as the following dynamic equilibrium equations:

[0197]

[0198] in, Indicates the number of external forces acting on the denture. For the first An external force vector, including chewing force, friction force, etc.; This indicates the number of external torques acting on the denture. For the first An external torque vector. By solving this equation, the force balance state of the denture during the occlusal process can be analyzed, and the stress distribution on the denture surface can be determined.

[0199] Optionally, a distance function can be introduced when analyzing spatial interferometry. To quantify the spatial distances between the denture and opposing teeth, adjacent teeth, and gingival soft tissue, the expression for the distance function can be: ,in, and These are the three-dimensional coordinate vectors of points on the surfaces of the two objects. Representing three-dimensional space , , Three dimensions. When When the value is less than the safety threshold, spatial interference is determined to exist.

[0200] Optionally, during the simulation, the spatial positional changes between the denture and opposing teeth, adjacent teeth, and gingival soft tissue are monitored in real time, and comprehensive information such as contact relationships, spatial interference, and structural compression states are continuously collected during the movement. Simultaneously, combined with the occlusal force distribution pattern, the stress distribution and load transmission characteristics of the denture surface are analyzed to comprehensively identify various potential design flaws in dynamic scenarios. The simulation records the structural fit state at different movement stages throughout the entire process, integrates monitoring information from all movement cycles, and generates quantitative and visualized occlusal simulation results data. Abnormal locations such as occlusal interference areas, stress concentration areas, and edge compression areas are accurately marked, providing precise problem location basis for subsequent parameter correction.

[0201] S54. Based on the occlusal simulation results, the parameters of the three-dimensional model data of the prosthesis are corrected to obtain the prosthesis design data.

[0202] Optionally, the distribution and degree of deviation of various anomalies in the occlusal simulation results data are analyzed in depth, and a targeted parameter correction scheme is formulated in combination with oral anatomical constraints and denture design specifications. For example, for occlusal interference areas, a formula is used... The adjustment amount of the quantified surface shape and spatial position, among which, Indicates the first Within the first interlocking interference region, the first The position adjustment amount of each control point; This is a correction factor, and its value can be determined by the first... Within the first interlocking interference region, the first The severity of interference at each control point is determined by both the anatomical limitations and the overall interference. For the first The first region The magnitude of the interference force at each control point This represents the sum of interference forces across all interference regions, and normalization ensures that the adjustment amount is proportional to the interference force. These are the original position coordinates of the control points.

[0203] Optionally, for stress concentration areas, a stress distribution optimization formula can be used. ,in, The optimized surface contact area, This represents the original contact area; This is an area adjustment factor, set according to material properties and clinical standards; It is the maximum stress value in the stress concentration region. This is the average stress value in the region. This formula is used to reduce the stress concentration factor and optimize the curved surface transition structure.

[0204] Optionally, for soft tissue compression defects, the formula can be used. Correct the edge contour and fitting gap parameters, where, The adjusted fit gap, This is the initial design gap; The gap adjustment coefficient can be determined by the elastic modulus of soft tissue and clinical experience; This represents the amount of soft tissue deformation under pressure.

[0205] Optionally, parameter correction can adopt a local fine-tuning mode, which only targets the defective areas detected by simulation for targeted optimization, preserving the overall anatomical fit and aesthetic design of the denture, and avoiding large-scale adjustments that could destroy the previous multi-dimensional optimization results.

[0206] Optionally, after parameter correction, the geometric structure and corresponding design parameters of the three-dimensional denture model data are updated simultaneously to form a closed-loop optimization mechanism. The denture design data after final simulation verification and parameter correction completely eliminates the dynamic usage defects that cannot be predicted by static design. All design indicators can meet the long-term use requirements of oral physiological functions and can be directly applied to subsequent digital processing and clinical restoration of dentures.

[0207] In the aforementioned AI-based personalized denture design method based on multimodal oral data fusion, deep integration of three-dimensional oral scanning data, two-dimensional image data, and oral functional data is achieved through temporal synchronization and spatial registration technology of multimodal oral data. Relying on technologies such as bi-branch feature extraction and adaptive weighted fusion, the method fully explores the local anatomical details and global occlusal correlation features of the oral cavity. Combined with dynamic occlusal simulation verification and parameter optimization, the denture design not only conforms to the patient's individual anatomical structure but also adapts to the real occlusal movement law, improving the matching degree of denture morphology and function. At the same time, standardized data processing and automated design process can improve the consistency and repeatability of design results, effectively avoid the occlusal interference risks caused by static design, reduce the number of clinical adjustments, shorten the restoration cycle, and improve the accuracy and reliability of personalized denture design.

[0208] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0209] Based on the same inventive concept, this application also provides an AI denture personalized design system based on multimodal oral data fusion for implementing the aforementioned AI denture personalized design method based on multimodal oral data fusion. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the AI ​​denture personalized design system based on multimodal oral data fusion provided below can be found in the limitations of the AI ​​denture personalized design method based on multimodal oral data fusion described above, and will not be repeated here.

[0210] In one exemplary embodiment, such as Figure 2 As shown, a schematic diagram of the structure of an AI-powered personalized denture design system 10 based on multimodal oral data fusion is provided, including:

[0211] The multimodal data acquisition and registration module 11 is used to acquire multimodal oral data of the target patient and perform time synchronization and spatial registration on the multimodal oral data to obtain registered oral data.

[0212] The oral cavity feature extraction and fusion module 12 is used to input the registered oral cavity data into the feature fusion model for feature extraction and feature fusion to obtain oral cavity feature data;

[0213] The denture morphology generation module 13 is used to input oral feature data into the denture morphology generation model for feature encoding, morphology decoding and parameter mapping to obtain initial denture design data.

[0214] The denture parameter optimization module 14 is used to calculate the morphological matching degree, adjust the occlusal parameters and optimize the aesthetic parameters of the oral feature data based on the initial denture design data, so as to obtain optimized denture design data.

[0215] The denture simulation verification module 15 is used to perform three-dimensional reconstruction, occlusal simulation verification, and design parameter correction on the optimized denture design data to obtain denture design data.

[0216] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the AI-based personalized denture design method based on multimodal oral data fusion as described above.

[0217] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0218] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0219] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for personalized AI-based denture design based on multimodal oral data fusion, characterized in that, The method includes: S1. Acquire multimodal oral data of the target patient, and perform time synchronization and spatial registration on the multimodal oral data to obtain registered oral data; S2. Input the registered oral cavity data into the feature fusion model for feature extraction and feature fusion to obtain oral cavity feature data; S3. Input the oral cavity feature data into the denture morphology generation model for feature encoding, morphology decoding and parameter mapping to obtain the initial denture design data; S4. Based on the initial denture design data, perform morphological matching calculation, occlusal parameter adjustment and aesthetic parameter optimization on the oral cavity feature data to obtain optimized denture design data; S5. Perform three-dimensional reconstruction, occlusal simulation verification, and design parameter correction on the optimized denture design data to obtain denture design data.

2. The method according to claim 1, characterized in that, S1 includes: S11. Perform timestamp alignment processing on the oral cavity three-dimensional scan data, oral cavity two-dimensional image data and oral cavity function data in the multimodal oral cavity data to obtain time-aligned data; wherein, the time-aligned data includes time-aligned oral cavity three-dimensional scan data and time-aligned oral cavity two-dimensional image data; S12. Extract feature points from the time-aligned oral cavity three-dimensional scan data to obtain the source data point set; S13. Extract feature points from the time-aligned two-dimensional oral cavity image data to obtain the target data point set; S14. Perform feature point matching on the source data point set and the target data point set to obtain matching feature point pairs; S15. Based on the matched feature point pairs, calculate the spatial registration transformation matrix; wherein, the expression of the spatial registration transformation matrix is: In the formula, To register the transformation matrix, For the source data point set, For the target data point set, It is a transformation function; S16. Based on the spatial registration transformation matrix, perform spatial coordinate transformation on the source data point set to obtain the registered oral cavity data.

3. The method according to claim 1, characterized in that, The feature fusion model includes a local feature extraction branch, a global feature extraction branch, a first feature distance calculation unit, a second feature distance calculation unit, a fusion weight calculation unit, a feature fusion unit, and a feature mapping unit. S2 includes: S21. Input the registered oral cavity data into the local feature extraction branch to perform local feature extraction and obtain local anatomical feature data; S22. Input the registered oral cavity data into the global feature extraction branch to perform global feature extraction and obtain global occlusal feature data; S23. Input the local anatomical feature data into the first feature distance calculation unit to calculate the feature distance and obtain the local anatomical feature distance; S24. Input the global occlusal feature data into the second feature distance calculation unit to calculate the feature distance and obtain the global occlusal feature distance; S25. The local anatomical feature distance and the global occlusal feature distance are input into the fusion weight calculation unit to calculate the fusion weight; wherein, the expression for the fusion weight is: In the formula, For the first Each fusion weight, For the first Distance of local anatomical features For the first A global bite feature distance, These are preset scale parameters; S25. Input the fusion weights, the local anatomical feature data, and the global occlusal feature data into the feature fusion unit for weighted fusion to obtain fused feature data; S26. Input the fused feature data into the feature mapping unit for feature dimensionality reduction to obtain the oral cavity feature data.

4. The method according to claim 1, characterized in that, The denture morphology generation model includes a feature encoding layer, a morphology decoding unit, and a parameter mapping unit. S3 includes: S31. Input the oral cavity feature data into the feature encoding layer for feature encoding to obtain the latent space feature vector; S32. Input the latent space feature vector into the morphological decoding unit to perform morphological decoding and obtain denture morphological point cloud data; S33. Input the denture morphology point cloud data into the parameter mapping unit to extract vertex coordinates and obtain a denture vertex coordinate set. Perform parameter mapping on the denture morphology point cloud data through the parameter mapping unit to obtain denture morphology parameter data. Integrate the denture morphology parameter data and the denture vertex coordinate set through the parameter mapping unit to obtain the initial denture design data.

5. The method according to claim 1, characterized in that, S4 includes: S41. Extract the denture vertex coordinates from the initial denture design data, and extract the reference vertex coordinates from the oral cavity feature data; S42. Calculate the morphological matching degree between the vertex coordinates of the denture and the reference vertex coordinates to obtain the morphological matching degree; S43. Based on the morphological matching degree, adjust the occlusal parameters of the initial denture design data to obtain the adjusted denture design data; S44. Optimize the aesthetic parameters of the adjusted denture design data to obtain the optimized denture design data.

6. The method according to claim 1, characterized in that, S5 includes: S51. Perform three-dimensional mesh reconstruction on the optimized denture design data to obtain three-dimensional denture model data; S52. Extract the biting motion feature data from the oral cavity feature data; S53. Perform bite motion simulation based on the three-dimensional model data of the denture and the bite motion feature data to obtain bite simulation result data; S54. Based on the occlusion simulation results data, the parameters of the three-dimensional model data of the prosthesis are corrected to obtain the prosthesis design data.

7. An AI-powered personalized denture design system based on multimodal oral data fusion, characterized in that, The system includes: A multimodal data acquisition and registration module is used to acquire multimodal oral data of the target patient and perform time synchronization and spatial registration on the multimodal oral data to obtain registered oral data. The oral cavity feature extraction and fusion module is used to input the registered oral cavity data into the feature fusion model for feature extraction and feature fusion to obtain oral cavity feature data; The denture morphology generation module is used to input the oral cavity feature data into the denture morphology generation model for feature encoding, morphology decoding and parameter mapping to obtain initial denture design data. The denture parameter optimization module is used to calculate the morphological matching degree, adjust the occlusal parameters and optimize the aesthetic parameters of the oral feature data based on the initial denture design data, so as to obtain optimized denture design data. The denture simulation verification module is used to perform three-dimensional reconstruction, occlusal simulation verification, and design parameter correction on the optimized denture design data to obtain denture design data.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 6.

Images