Denture occlusal surface three-dimensional modeling optimization method and system based on dynamic pressure calibration

Abstract

The invention discloses a dynamic pressure calibration-based false tooth occlusal surface three-dimensional modeling optimization method and system, and particularly relates to the technical field of false tooth occlusal modeling, and the method comprises the steps: obtaining soft tissue viscoelastic parameters of a patient through ultrasonic elastography and palpation type pressure collection, and carrying out the classification marking; on the basis of viscoelastic parameters and typical pressure time sequence mapping association, a graph neural network fusion time sequence memory unit is trained, and online prediction from dynamic pressure to soft tissue deformation and denture displacement is achieved; performing three-dimensional interpolation and deformation multiplying power correction on the real-time pressure array by utilizing a prediction result, and eliminating soft tissue disturbance through geometric closed-loop feedback iteration; a key contact point cloud is extracted based on corrected equivalent pressure data and predicted displacement, and a high-precision occlusal surface three-dimensional grid is generated by combining weighted registration, reconstruction and local high-density resampling, so that the problem of pressure offset and geometry inconsistency caused by neglecting soft tissue deformation in static modeling is effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of denture occlusal modeling, and more particularly to a method and system for optimizing three-dimensional modeling of denture occlusal surfaces based on dynamic pressure calibration. Background Art

[0002] During clinical denture restoration, after the patient puts on the denture, chewing behavior causes dynamic contact and pressure distribution changes between the alveolar ridge mucosa, gums, and base. Three-dimensional modeling of the denture occlusal surface generally requires obtaining geometric information about the soft and hard tissues of the patient's oral cavity through intraoral scanning, and then using 3D modeling software to design the occlusal surface and base. During actual chewing, the viscoelastic deformation of the soft tissue, fluctuations in bite force, and slight displacements between the base and the alveolar ridge make it difficult for the geometric model obtained by static scanning to accurately reflect the pressure distribution under dynamic occlusion. Under soft tissue disturbances, the position and posture of the base also change, resulting in a shift in the contact point position and pressure center. This makes it impossible for traditional 3D modeling methods based on static geometry to truly reproduce the occlusal relationship during clinical chewing.

[0003] Existing technologies have significant shortcomings in addressing the coupling of dynamic occlusal pressure and soft tissue. First, most 3D modeling methods treat the denture and base as absolutely rigid bodies, calculating the distribution of occlusal contact points and pressure based solely on static geometric models obtained through intraoral optics or tomography, while ignoring the tiny perturbations caused by the viscoelastic deformation of soft tissue. When the soft tissue deformation reaches or exceeds the submillimeter level, the contact point distribution and pressure values ​​calculated by traditional static models will deviate significantly, resulting in the denture and base not being able to fit the real oral environment, causing base tenderness, decreased chewing efficiency, and difficulty for patients to adapt.

[0004] In summary, existing technologies cannot avoid the hidden disturbances caused by soft tissue deformation, thereby affecting the accuracy of denture design and patient comfort. This problem is particularly prominent under various chewing soft tissue disturbances. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a three-dimensional modeling optimization method and system for the denture occlusal surface based on dynamic pressure calibration. Through a multi-source dynamic mutual calibration and partition correction mechanism, the pressure measurement distortion caused by soft tissue deformation is eliminated, and combined with the closed-loop optimization driven by wear probability, the problems raised in the above-mentioned background technology are solved.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for optimizing three-dimensional modeling of denture occlusal surfaces based on dynamic pressure calibration, comprising: Step 1: Acquire ultrasonic shear wave elastic images of oral soft tissues (such as the alveolar ridge mucosa and gums) with the patient's mouth open. Obtain an oral array pressure matrix and base inertial measurement displacement through a palpation pressure test. Classify and annotate these images according to preset deformation and displacement thresholds to generate individualized soft tissue feature data, including viscoelastic modulus, loss modulus, and soft tissue disturbance measurement indicators, and store them in a soft tissue parameter library. Step 2: Construct an oral simulation geometric model based on individualized soft tissue feature data: Use the patient's oral imaging data to reconstruct an oral simulation geometric model including soft tissue, bones, and dentures. Based on the nonlinear viscoelastic constitutive model, obtain the soft tissue material properties and assign elastic-plastic material properties to the dentures. Perform soft tissue perturbation finite element simulation, output pressure mapping to data pairs of soft tissue deformation and base displacement, and construct an oral simulation prediction model by fusing a graph neural network with a temporal memory unit to achieve rapid prediction of soft and hard tissue coupling for arbitrary real-time pressure. Step 3: When the deformation or base displacement predicted by the real-time pressure array and the oral simulation prediction model exceeds the preset threshold, the real-time pressure array is mapped to the base coordinate system using three-dimensional interpolation. The real-time pressure array is initially corrected in combination with the deformation magnification factor. The deformation magnification factor is iteratively updated with the geometric closed-loop residual driving ratio until the geometric closed-loop residual converges. The equivalent pressure data is output to eliminate the implicit disturbance caused by soft tissue deformation and provide accurate pressure input for three-dimensional mesh reconstruction. Step 4: Extract key contact points based on equivalent pressure data and predicted displacement. First, perform weighted least squares registration to fuse the optical point cloud and then perform 3D reconstruction. When the curvature error or displacement threshold meets the conditions, drive local high-density resampling, re-perform local inverse correction, and generate a continuous 3D mesh that takes into account both the overall smoothness and local refinement of the occlusal surface. Explanation: Prior weighted least squares registration is a rigid point cloud fusion method that assigns weights to key contact points and minimizes the sum of squared weighted spatial distances between each pair of points during the registration process to ensure that high-weight areas are aligned first. The specific approach is: Weight assignment: assign a weight to each contact point. The weight can be calculated by combining the corrected rigid pressure, normal curvature, or base displacement confidence, indicating the importance of the point in the registration process. Objective function: Weighted least squares registration is defined as solving a rigid transformation (rotation matrix and translation vector) to minimize the sum of the squared Euclidean distances between weighted points and corresponding points. In other words, in each iteration, the weights of all paired points multiplied by the sum of their squared spatial distances are minimized, thereby giving higher priority to points with larger weights and larger errors in the global registration. Iterative process: First, the nearest neighbor pairing is selected for the initial point cloud according to the weighted rule, then the rigid transformation update parameters with the minimum current registration error are solved, and then the weighted correspondence is recalculated until the error converges or the number of iterations reaches the upper limit; Result output: After completing the weighted least squares registration, a point cloud in a fused consistent coordinate system is obtained, providing a geometric basis for subsequent Poisson reconstruction and constrained smoothing.

[0007] Step 5: After obtaining the local correction pressure and updating the point cloud, the frame-by-frame pressure energy and base displacement are accumulated to generate the contact energy index and combined with the local normal curvature to calculate the wear probability distribution. When the wear probability exceeds the preset threshold, the sampling density of the next cycle is automatically increased and the high-risk areas are marked in the final three-dimensional grid to achieve proactive prevention of denture micro-area wear.

[0008] Preferably, the specific operation process of step 1 includes the following steps: Step 101: attaching a shear wave elastography probe to the surface of oral soft tissue, and based on the shear wave velocity and soft tissue density, expressing the storage modulus as the product of the soft tissue density and the square of the shear wave velocity; using the same shear wave elastography probe to measure the shear wave velocity at two adjacent frequencies, respectively, expressing the loss modulus as the product of the shear wave velocity per unit frequency, the average frequency, and an empirical calibration coefficient, wherein the empirical calibration coefficient is used to compensate for the uniformity error of the shear wave elastography probe and the oral soft tissue; the storage modulus is used to characterize the elastic stiffness of the tissue in the vibration frequency band, and the loss modulus reflects the viscous dissipation of the oral soft tissue during shear vibration; Step 102: Press the calibrated pressure sensor into the oral soft tissue surface at a constant independent vertical pressure, and record the real-time pressure displacement and the real-time soft tissue deformation at the denture base-mucosal contact interface. , input the nonlinear least squares fitting viscoelastic model, and output the viscoelastic parameter set and pressure-soft tissue deformation time series curve; Step 103: An inertial measurement device is attached to the back of the denture base in the center to record three-dimensional linear acceleration and angular velocity in real time. After zero-bias calibration and complementary filtering algorithm, the output of the inertial measurement device is fused into the base displacement relative to the oral coordinate system. Step 104: A piezoresistive sensing unit is placed below the occlusal surface of the denture, with a synchronous sampling frequency set to 200 Hz, to output a pressure matrix. Each element of the pressure matrix represents the instantaneous pressure data at a specific moment and position coordinate. The oral pressure center is obtained by calculating the weighted average of all position coordinates, where the weight is the instantaneous pressure data measured at the specific moment, and the normalization factor is the sum of the pressures of the entire array. Step 105: Classify the soft tissue disturbance measurement indicators (oral soft tissue deformation, base displacement) into several levels. In the embodiment of the present invention, the levels are 3. Step 106: Match the outputs of steps 101 to 104 with the disturbance measurement indicators one by one to generate individualized soft tissue feature data. Based on the disturbance measurement indicators, the individualized soft tissue feature data are marked as three scenarios: normal soft tissue disturbance, secondary abnormal soft tissue disturbance, and high-level abnormal soft tissue disturbance; and stored in the soft tissue parameter library respectively.

[0009] Preferably, the method includes a multi-source data mutual calibration step, and the multi-source data mutual calibration is performed based on step 101 and step 102, including: comparing the measured storage modulus with the storage modulus fitted by viscoelastic parameter inversion, and calculating the elastic deviation ratio point by point; when the local area deviation ratio exceeds a threshold, triggering viscoelastic parameter re-optimization, including: At each position point of the three-dimensional ultrasound imaging grid, calling the measured storage modulus of step 101; Based on the viscoelastic parameter set output in step 102, the viscoelastic parameter inversion fitting storage modulus is output through the following formula : ; Where ω represents the angular frequency, 、E1、E2、 、 All are derived from the fitting results of step 102; Compare the measured storage modulus with the storage modulus obtained by inversion fitting of viscoelastic parameters. If the relative error between the two exceeds a preset threshold (e.g., 10%) at the same or multiple frequency points, the viscoelastic parameter set re-optimization is triggered. The initial values ​​of the nonlinear least-squares viscoelastic model are readjusted or weights are optimized until the inverse-fitted storage modulus of the viscoelastic parameters is consistent with the measured storage modulus. This mutual calibration mechanism effectively integrates the two data sources of shear wave imaging and palpation fitting, improving the reliability and consistency of viscoelastic parameters.

[0010] Preferably, the viscoelastic parameter set is , the nonlinear least squares fitting viscoelastic model is used to solve the viscoelastic parameter set, and the viscoelastic model satisfies the following formula: ; in, is the long-term elastic modulus (the long-term elastic modulus refers to the stable elastic stiffness of the soft tissue after a sufficiently long period of stress relaxation after the external pressure loading is completed. It is regarded as the ratio of the equilibrium stress to the corresponding strain in the relaxation stage of the pressure-soft tissue deformation time series curve. It is obtained by recording the stress decay curve over time in the constant speed indentation test. After the stress tends to a steady state, the ratio of the steady-state stress to the corresponding indentation depth is taken and converted in combination with the shear wave elastography results). is the indentation depth varying with time, represents the instantaneous rate of displacement over time; the integral upper limit t accumulates from 0 to the current moment to achieve the accumulation of historical effects, and s is the historical time variable in the integral; the value of i is 1 or 2; E1 and E2 are the elastic moduli corresponding to the first and second Maxwell viscoelastic elements, respectively, which are used to characterize the stiffness of the soft tissue under short-term and medium- and low-frequency loading. and are the relaxation times of the first and second Maxwell units, respectively, which are used to describe the rate at which stress relaxes to the corresponding time scale; in the palpation constant speed indentation test of step 1, E1 and It mainly reflects the rapid elastic-viscous response of the mucosa or gums under high-frequency and small-amplitude loading, E2 and It reflects the hysteresis deformation characteristics under medium and low frequencies or large deformation conditions.

[0011] Preferably, the method further comprises the step of intelligently removing motion artifacts: The output of the inertial measurement device and the pressure center timing mapping are mutually verified. By comparing and synchronizing the base displacement output by the inertial measurement device with the pressure center vector timing calculated by the pressure sensor array, the normalized amplitude of the vector cross product of the base displacement and the pressure center vector is used to construct a timing consistency criterion. When the normalized amplitude exceeds the threshold, the base displacement and pressure center vector data at the corresponding moment are automatically eliminated, thereby dynamically eliminating oral movement artifacts and improving the accuracy of the base displacement data.

[0012] Preferably, the specific operation process of step 2 includes the following steps: Step 201: Based on the soft tissue parameter library, the optical intraoral scan is used to perform three-dimensional geometric registration with the patient's oral data to reconstruct an oral simulation geometric model, including soft tissue, bone, denture base, and occlusal surface meshes; Step 202: Multi-condition simulation: Call pressure time series data under three scenarios from the soft tissue parameter library and apply them to the occlusal mesh nodes of the finite element model according to the node distribution. The application time of each scenario is 3 to 5 seconds. The initial soft tissue deformation vector and base displacement are 0. Use the explicit time integration method to simulate under the Courant stability condition, and record the soft tissue deformation vector, base displacement and time lag parameter at each time step. The soft tissue deformation vector represents the maximum displacement of the node under the occlusal surface in the normal direction. The base displacement is the mean of the displacement vectors of all base nodes. The time lag parameter refers to the difference between the time corresponding to the maximum soft tissue deformation and the time of peak pressure. The Courant stability condition ensures that the numerical method can correctly capture the propagation process of physical quantities by constraining the ratio of the time step to the space step. Step 203: Prediction model design and training: construct an oral simulation prediction model based on the GNN-LSTM hybrid network to predict the soft tissue deformation vector, base displacement, and time lag parameters; Step 204: Online update and optimization: When the residual error of the soft tissue deformation predicted by the oral simulation prediction model exceeds a threshold, incremental online fine-tuning is automatically triggered to ensure that the oral simulation prediction model is continuously optimized under the new scenario.

[0013] Preferably, the loss function of the oral simulation prediction model consists of three parts: first, the deformation prediction error term measures the absolute difference between the soft tissue deformation output by the network and the finite element simulation annotation value, which is used to ensure that the model accurately reproduces the soft tissue deformation numerically; second, the displacement and time lag joint error term simultaneously calculates the difference between the mean base displacement and the response time lag prediction value and the simulation value, to ensure that the model can capture the overall micro-displacement trend and reflect the soft tissue delay effect; finally, the physical consistency error term measures the deviation between the network prediction and the sum of elastic energy storage and viscous dissipation obtained by simulation at the energy level. This term emphasizes not only minimizing numerical errors, but also being consistent with the real physical process in terms of energy conservation and viscous dissipation. The overall loss is the sum of these three errors. By taking into account the deformation, displacement time lag and energy consistency at the same time, the network is driven to converge jointly in both numerical and physical senses.

[0014] Preferably, the operation process of step 3 includes the following steps: Step 301: Read the real-time pressure array data and the soft tissue deformation and base displacement predicted by the oral simulation geometric model at the same time, and call the deformation magnification coefficient. The deformation magnification coefficient is a pre-set dimensionless adjustment parameter used to map and associate the predicted soft tissue deformation with the pressure array data. Step 302: Based on linear approximate mapping, the position information of the real-time pressure array is added to the base displacement predicted by the oral simulation geometric model. The corresponding real-time pressure array value is obtained at the mapped coordinate using a three-dimensional interpolation technique. The pressure value is then combined with the soft tissue deformation using a deformation multiplication factor to obtain preliminary equivalent pressure data. The preliminary equivalent pressure data represents the pressure distribution when the base is assumed to be rigid and soft tissue disturbances are ignored, and is used for subsequent geometric verification. Step 303: Map the pressure point positions corresponding to the preliminary equivalent pressure to the three-dimensional point cloud obtained by the oral optical scan. Calculate the minimum geometric distance from each pressure mapping point to the point cloud using the nearest neighbor distance method. Aggregate the distances to obtain a distance set, and use the maximum value of the distance set as the geometric residual. If the geometric residual does not exceed the preset geometric tolerance limit, the correction is determined to have converged and the iteration is terminated. Otherwise, proceed to the next step. Step 304: When the geometric residual is greater than the preset value, the difference between the current geometric residual and the tolerance limit is used as the basis for adjustment. The deformation magnification coefficient is slightly increased or decreased to ensure that the change does not exceed the predetermined threshold. Then, the second step is called again and the next equivalent pressure is calculated in the same manner. The process of steps 302 and 303 is repeated until the geometric residual meets the preset standard or the maximum number of iterations is reached. If the convergence condition is not met after exceeding the maximum number of iterations, the corresponding moment is marked as an abnormal disturbance and the original data is retained for subsequent analysis. Step 305: After the iteration is completed, the final converged equivalent pressure data and the corresponding geometric residual are output. The equivalent pressure data is used to guide the subsequent occlusal surface geometry reconstruction and dynamic correction.

[0015] Preferably, the preliminary equivalent pressure data satisfies the following formula: ; in, represents preliminary equivalent pressure data, represents mapping the position of the i-th pressure sensing unit from the soft tissue perturbation back to the rigid reference coordinates; is the Euclidean norm for predicting soft tissue deformation, represents the deformation magnification factor, is the dimensionless adjustment coefficient, express The pressure values ​​at the corresponding coordinates have not undergone any geometric mapping or soft tissue / base correction, and reflect the actual uncorrected occlusal pressure distribution collected.

[0016] Preferably, in the real-time pressure array correction, the base is not set to be absolutely rigid, but is given elastic parameters matching the actual base material in the finite element model, thereby allowing the base to produce slight elastic deformation during the iterative mapping process.

[0017] Preferably, the occlusal surface is divided into several fixed sub-areas, the geometric residual is calculated for each sub-area and the respective deformation magnification coefficient is updated, and the correction results are merged as a whole after the sub-area correction is completed.

[0018] Preferably, the specific operation process of step 4 includes the following steps: Step 401: Extracting the initial contact point cloud: In a virtual occlusal environment, based on the equivalent pressure data distribution output in step 3, an initial contact point cloud distribution is generated in the centric occlusal posture. Several pressure measurement points with pressure values ​​in the top 20% are selected as key contact points, and the contact point clouds corresponding to the key contact points are collected to represent the pressure concentration areas during occlusion. Step 402: Global translation and normal fine-tuning: Each point in the contact point cloud is translated and corrected along the local displacement vector predicted by the oral simulation geometric model. Then, a small displacement is added to the local normal direction of the translated point based on the soft tissue deformation value predicted by the oral simulation geometric model to reflect the change in contact surface curvature caused by soft tissue deformation, thereby obtaining a corrected contact point cloud. Step 403: Point cloud fusion and preliminary mesh reconstruction: The corrected contact point cloud is fused with the oral optical 3D point cloud using a weighted least squares registration algorithm. A 3D mesh generation method based on Poisson reconstruction is applied to the fused point cloud. Laplace smoothing is performed on the reconstructed preliminary mesh to ensure that the mesh maintains both overall smoothness and the geometric characteristics of the local contact points. At the same time, the local curvature error is calculated at each contact point by comparing the surface curvature of the original point cloud to provide a quantitative basis for subsequent resampling. Step 404: Local high-density resampling trigger and recalibration: When the base displacement corresponding to the corrected point cloud exceeds a preset displacement threshold or the curvature error of any contact point exceeds a preset curvature threshold, a spherical area with a fixed radius is defined with the contact point that meets the threshold condition as the center. High-resolution optical point cloud and pressure data are recollected within the spherical area, and steps 401 to 403 are repeated. Local inverse mapping and geometric closed-loop correction are performed to obtain refined local equivalent pressure data and point cloud data. Step 405: Final 3D mesh construction and residual output: The locally resampled and corrected point cloud is merged with the original fused point cloud, and the merged point cloud is again subjected to Poisson reconstruction and Laplace smoothing while retaining the corrected contact point position constraints to generate a final continuous 3D mesh model. The overall geometric residual is determined by calculating the minimum distance from each contact point to the final mesh, and the output includes the 3D mesh model and its corresponding geometric residual result.

[0019] Preferably, the operation process of step 5 includes the following steps: Step 501: Accumulate frame-by-frame pressure energy and base displacement The local corrected rigid pressure at each sampling moment and the corresponding predicted displacement of the oral simulation geometric model are collected, and the cumulative pressure energy and displacement energy of each contact point in the time series are calculated to quantify the loading history of the contact point during continuous chewing. Step 502: Calculate the wear probability based on the local normal curvature For each contact point, the normal curvature of its corresponding position in the optical point cloud is obtained, and the accumulated pressure energy is combined with the normal curvature to output the wear probability of the contact point; Step 503: Automatically increase sampling density when wear probability exceeds threshold When the wear probability of any contact point exceeds the preset threshold, the sampling area within a fixed radius of the contact point is automatically switched to high-density sampling mode to re-collect optical point cloud and pressure data at a higher resolution; Step 504: Mark high-risk areas on the final 3D grid In the three-dimensional mesh model, all contact points where the wear probability exceeds the threshold are visually marked as high-risk areas, providing a reference for denture material optimization and subsequent design.

[0020] Preferably, in the construction of the wear probability prediction model: the cumulative pressure energy, displacement energy and local normal curvature characteristics of each contact point are obtained through laboratory cyclic loading and clinical repair sample collection and normalized preprocessing is performed; a multi-layer perceptron and lightweight gradient boosting tree hybrid classifier is constructed, and cross-validation and early stopping strategies are used to optimize the model parameters; the trained model is deployed on an online interface to calculate the wear probability of each contact point in real time.

[0021] To achieve the above objectives, the present invention provides the following technical solution: a three-dimensional modeling and optimization system for denture occlusal surfaces based on dynamic pressure calibration, comprising: The dynamic acquisition module simultaneously acquires ultrasonic elastic images, array pressure, and base displacement data while the patient's mouth is open. After verification through palpation, it generates individualized soft tissue feature data based on the deformation / displacement dual threshold. The simulation modeling module reconstructs the geometric model of oral soft and hard tissues based on individualized soft tissue feature data, assigns nonlinear viscoelastic properties to the soft tissues and elastoplastic properties to the dentures, generates pressure-deformation data pairs through finite element simulation, and constructs an oral simulation prediction model. The output data pairs map pressure to soft tissue deformation and base displacement. When the real-time pressure array and the predicted deformation or base displacement of the oral simulation prediction model exceed the preset threshold, the pressure correction module uses three-dimensional interpolation to map the real-time pressure array to the base coordinate system. The real-time pressure array is initially corrected in combination with the deformation magnification factor. The deformation magnification factor is updated iteratively using the geometric closed-loop residual to drive the proportional update until the geometric closed-loop residual converges. The equivalent pressure data is then output to eliminate the implicit disturbance caused by soft tissue deformation. The reconstruction optimization module fuses equivalent pressure data with optical point cloud data and extracts key contact points through weighted least squares registration. When the curvature error exceeds the threshold, local resampling is triggered. After inverse correction, a smooth 3D mesh is generated and output to the wear warning module. The wear warning module accumulates frame-by-frame pressure energy and displacement to generate a contact energy index, combines local curvature to calculate the wear probability distribution, automatically increases the sampling density in areas where the probability exceeds the limit, marks high-risk coordinates in the grid, and synchronously feeds back the sampling instructions to the reconstruction module to form a closed-loop optimization.

[0022] Technical effects and advantages of the present invention: The present invention acquires the viscoelastic parameters of the patient's soft tissue through ultrasonic elastography and palpation-type pressure acquisition and classifies and labels them; based on the association between viscoelastic parameters and typical pressure time series mapping, and training of graph neural networks fused with time series memory units, online prediction of dynamic pressure to soft tissue deformation and base displacement is achieved; the prediction results are used to perform three-dimensional interpolation and deformation magnification correction on the real-time pressure array, and soft tissue disturbances are eliminated through geometric closed-loop feedback iteration; based on the corrected equivalent pressure data and predicted displacement, the key contact point cloud is extracted, and a high-precision three-dimensional mesh of the occlusal surface is generated by combining weighted alignment, reconstruction and local high-density resampling, effectively solving the problems of pressure offset and geometric inconsistency caused by static modeling ignoring soft tissue deformation. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a simplified flow chart of the denture occlusal surface three-dimensional modeling optimization method of the present invention. DETAILED DESCRIPTION

[0024] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0025] At the same time, it should be understood that for the convenience of description, the sizes of the various parts shown in the drawings are not drawn according to the actual proportional relationship.

[0026] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the present disclosure, its application, or uses.

[0027] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered part of the specification.

[0028] Example 1, see Figure 1 The present invention provides a simplified flow chart of the three-dimensional modeling optimization method for the denture occlusal surface. Figure 1 A method for optimizing the three-dimensional modeling of the denture occlusal surface based on dynamic pressure calibration is shown, comprising: Step 1: Acquire ultrasonic shear wave elastic images of oral soft tissues (such as the alveolar ridge mucosa and gums) with the patient's mouth open. Obtain an oral array pressure matrix and base inertial measurement displacement through a palpation pressure test. Classify and annotate these images according to preset deformation and displacement thresholds to generate individualized soft tissue feature data, including viscoelastic modulus, loss modulus, and soft tissue disturbance measurement indicators, and store them in a soft tissue parameter library. The specific implementation is as follows: Step 101: attaching a shear wave elastography probe to the surface of oral soft tissue. Based on the shear wave velocity and soft tissue density, the storage modulus is expressed as the product of the soft tissue density and the square of the shear wave velocity. The shear wave velocity is measured at two adjacent frequencies using the same shear wave elastography probe. The loss modulus is expressed as the product of the shear wave velocity per unit frequency, the average frequency, and an empirical calibration coefficient. The empirical calibration coefficient is used to compensate for the uniformity error of the shear wave elastography probe and oral soft tissue and has a value range of 0.8-1.2. The storage modulus is used to characterize the elastic stiffness of the tissue in the vibration frequency band, and the loss modulus reflects the viscous dissipation of the oral soft tissue during shear vibration. Explanation: The ultrasound probe was kept perpendicular to the tissue surface when in contact with soft tissue, clamped with a pre-calibrated spring clamp, and the scanning depth was controlled by fixing the probe holder at the same depth. The intraoral temperature was maintained at 36 degrees Celsius ± 0.5 degrees Celsius using an inflatable silicone pad to ensure consistency in the storage modulus and loss modulus collected by different implementers. Step 102: Press the calibrated pressure sensor into the oral soft tissue surface at a constant independent vertical pressure, and record the real-time pressure displacement and the real-time soft tissue deformation at the denture base-mucosal contact interface. , input the nonlinear least squares fitting viscoelastic model, and output the viscoelastic parameter set and pressure-soft tissue deformation time series curve; Explanation: In the implementation of step 102, the upper limit of the pressing depth is taken as the safety limit, and the upper limit of the pressing depth is set to 2.0 mm. The test time range is from 0 to 2 seconds. The upper limit of the time is obtained by dividing the upper limit of the depth by the pressing speed. Explanation: The viscoelastic parameter set is , the nonlinear least squares fitting viscoelastic model is used to solve the viscoelastic parameter set, and the viscoelastic model satisfies the following formula: ; in, is the long-term elastic modulus (the long-term elastic modulus refers to the stable elastic stiffness of the soft tissue after a sufficiently long period of stress relaxation after the external pressure loading is completed. It is regarded as the ratio of the equilibrium stress to the corresponding strain in the relaxation stage of the pressure-soft tissue deformation time series curve. It is obtained by recording the stress decay curve over time in the constant speed indentation test. After the stress tends to a steady state, the ratio of the steady-state stress to the corresponding indentation depth is taken and converted in combination with the shear wave elastography results). is the indentation depth varying with time, represents the instantaneous rate of displacement over time; the integral upper limit t accumulates from 0 to the current moment to achieve the accumulation of historical effects, and s is the historical time variable in the integral; the value of i is 1 or 2; E1 and E2 are the elastic moduli corresponding to the first and second Maxwell viscoelastic elements, respectively, which are used to characterize the stiffness of the soft tissue under short-term and medium- and low-frequency loading. and are the relaxation times of the first and second Maxwell units, respectively, which are used to describe the rate at which stress relaxes to the corresponding time scale; in the palpation constant speed indentation test of step 1, E1 and It mainly reflects the rapid elastic-viscous response of the mucosa or gums under high-frequency and small-amplitude loading, E2 and It reflects the hysteresis deformation characteristics under medium and low frequency or large deformation conditions; Furthermore, based on steps 101 and 102, multi-source data intercalibration is performed, including: comparing the measured storage modulus with the storage modulus obtained by inversion fitting of viscoelastic parameters, and calculating the elastic deviation ratio point by point; when the deviation ratio of a local area exceeds a threshold, triggering viscoelastic parameter re-optimization, including: At each position point of the three-dimensional ultrasound imaging grid, calling the measured storage modulus of step 101; Based on the viscoelastic parameter set output in step 102, the viscoelastic parameter inversion fitting storage modulus is output through the following formula : ; Where ω represents the angular frequency, 、E1、E2、 、 All are derived from the fitting results of step 102; Compare the measured storage modulus with the storage modulus obtained by inversion fitting of viscoelastic parameters. If the relative error between the two exceeds a preset threshold (e.g., 10%) at the same or multiple frequency points, the viscoelastic parameter set re-optimization is triggered. The initial values ​​of the nonlinear least-squares viscoelastic model are readjusted or weights are optimized until the inverse-fitted storage modulus of the viscoelastic parameters is consistent with the measured storage modulus. This mutual calibration mechanism effectively integrates the two data sources of shear wave imaging and palpation fitting, improving the reliability and consistency of viscoelastic parameters.

[0029] Step 103: An inertial measurement device is attached to the back of the denture base in the center to record three-dimensional linear acceleration and angular velocity in real time. After zero-bias calibration and complementary filtering algorithm, the output of the inertial measurement device is fused into the base displacement relative to the oral coordinate system. Step 104: Place a (5×5 micro) piezoresistive sensor unit (unit size 0.5mm) below the occlusal surface of the denture. 2), the synchronous sampling frequency is set to 200 Hz, and the pressure matrix is ​​output; each pressure matrix element represents the instantaneous pressure data at a specific time and position coordinate. The oral pressure center is obtained by calculating the weighted average of all position coordinates. The weight is the instantaneous pressure data measured at a specific time, and the normalization factor is the sum of the pressure of the entire array, in millimeters; Furthermore, the method further includes a step of intelligently removing motion artifacts: The output of the inertial measurement device and the pressure center timing mapping are mutually verified. By comparing and synchronizing the base displacement output by the inertial measurement device with the pressure center vector timing calculated by the pressure sensor array, the normalized amplitude of the vector cross product of the base displacement and the pressure center vector is used to construct a timing consistency criterion. When the normalized amplitude exceeds the threshold (i.e., when the angle between the displacement and the pressure direction is abnormal), the base displacement and pressure center vector data at the corresponding moment are automatically eliminated, thereby dynamically eliminating oral movement artifacts and improving the accuracy of the base displacement data; Explanation: Motion artifacts are a common problem in fields like medical imaging. Involuntary patient movements, such as coughing and shaking, can be recorded by imaging devices, creating abnormal images that don't reflect the actual situation. This artifact, known as motion artifact, blurs the image and interferes with diagnosis. This step, based on vector cross product technology, accurately identifies and removes these artifacts, improving image quality and providing doctors with more accurate diagnostic evidence.

[0030] Summary: Steps 101-104 above perform ultrasonic shear wave elastography and constant-speed palpation indentation tests on oral soft tissues (such as alveolar ridge mucosa and gums) with the patient's mouth open, and deploy array pressure sensing units and inertial measurement units on the back of the denture base to collect soft tissue deformation, base displacement, and instantaneous pressure data under multiple soft tissue perturbations and chewing in real time, thereby obtaining storage modulus, loss modulus, pressure-soft tissue deformation time series curves, and dynamic pressure and displacement time series data. After comparative analysis, the long-term elastic modulus, second-order Maxwell viscoelastic parameters, and maximum deformation under each soft tissue perturbation are extracted, and the data are labeled as soft tissue perturbations based on deformation and displacement thresholds to generate a soft tissue parameter library containing oral soft tissue deformation data; the oral soft tissue deformation data at least includes: saturated storage modulus, viscosity attenuation coefficient, and multiple soft tissue perturbation time series pressure; Step 105: Classify the soft tissue disturbance measurement indicators (oral soft tissue deformation, base displacement) into several levels. In the embodiment of the present invention, the levels are 3. Step 106: Match the outputs of steps 101 to 104 with the disturbance measurement indicators one by one to generate individualized soft tissue feature data. Based on the disturbance measurement indicators, the individualized soft tissue feature data are marked as three scenarios: normal soft tissue disturbance, secondary abnormal soft tissue disturbance, and high-level abnormal soft tissue disturbance; and stored in the soft tissue parameter library respectively.

[0031] Step 2: Construct an oral simulation geometric model based on individualized soft tissue feature data: Use the patient's oral imaging data to reconstruct an oral simulation geometric model including soft tissue, bones, and dentures. Based on the nonlinear viscoelastic constitutive model, obtain the soft tissue material properties and assign elastic-plastic material properties to the dentures. Perform soft tissue perturbation finite element simulation, output pressure mapping to data pairs of soft tissue deformation and base displacement, and construct an oral simulation prediction model by fusing a graph neural network with a temporal memory unit to achieve rapid prediction of soft and hard tissue coupling for arbitrary real-time pressure. Explanation: Soft tissue perturbation finite element simulation refers to the numerical process of calculating the normal deformation of soft tissue and base displacement and its time-delay response under different dynamic pressure time series using the explicit time integration method in the soft tissue, elastic-plastic denture base and rigid bone coupling model described by the secondary Maxwell viscoelastic constitutive model.

[0032] Explanation: Based on the soft tissue viscoelastic parameters obtained in step 1 and the pressure time series curve of the typical chewing scene, the individualized soft tissue feature data is used to reconstruct the soft tissue, bone and denture geometric model (i.e., the oral simulation geometric model), and the denture base and occlusal surface are given elastic-plastic or elastic material properties, and the bone is set as a rigid boundary condition. Multi-condition finite element simulation is carried out to obtain the pressure P(t) and soft tissue deformation. , base displacement Data pairs; Based on the simulation results, a hybrid deep network with graph neural network (GNN) as the backbone and fused with time series memory unit (LSTM) is designed to achieve real-time prediction of the corresponding dynamic pressure input P(t) 、 and hysteresis In the training process, a cross-scene enhancement strategy (combining different forces and sideways to generate virtual pressure sequences) and multi-task learning (simultaneously predicting Δs, Δx and To expand the generalization capability of the network; The specific implementation is as follows: Step 201: Based on the soft tissue parameter library generated in step 1, the optical intraoral scan is used to perform 3D geometric registration with the patient's oral data to reconstruct an oral simulation geometric model, including soft tissue, bone, denture base, and occlusal surface meshes. The specific process is as follows: Geometric registration: First, three pairs of anatomical landmarks, such as the maxillary anterior tooth margin and the buccal zygomatic foramen, are selected for initial rigid registration, and then the ICP algorithm is used for iterative alignment. Meshing: Tetrahedron elements were used for bones with a maximum side length of 1.5 mm; the maximum side length of the soft tissue area was 1.0 mm, and the base and occlusal surfaces were locally refined to a maximum side length of 0.5 mm. Material property assignment: Soft tissue: Map the viscoelastic parameters obtained in step 1 to the corresponding fields in the material library; different layers (mucosal layer, submucosal layer, deep layer) use the same constitutive model but with independent unit partitions; Base and occlusal surface: If resin material is used, set the Young's modulus according to the tensile test results ( ) and Poisson's ratio ( ); If a metal alloy is used, the Young's modulus is obtained experimentally ( ), Poisson's ratio ( ).

[0033] Boundary conditions: The skeleton is not set to be strictly rigid, but the main constraint of the shell element with extremely high elastic stiffness is used to simulate micro-motion. The Young's modulus is set to 15 GPa and the Poisson's ratio is set to 0.30 to preserve the slight elastic deformation of the jaw during severe unilateral chewing. The soft tissue and the bone are connected by binding contact; the soft tissue and the base use an automatic surface-to-surface contact algorithm with a friction coefficient of μ = 0.30 and a contact tolerance of 0.05 mm; Step 202: Multi-condition simulation: Call pressure time series data under three scenarios from the soft tissue parameter library and apply them to the occlusal mesh nodes of the finite element model according to the node distribution. The application time of each scenario is 3 to 5 seconds. The initial soft tissue deformation vector and base displacement are 0. Use the explicit time integration method to simulate under the Courant stability condition, and record the soft tissue deformation vector, base displacement and time lag parameter at each time step. The soft tissue deformation vector represents the maximum displacement of the node under the occlusal surface in the normal direction. The base displacement is the mean of the displacement vectors of all base nodes. The time lag parameter refers to the difference between the time corresponding to the maximum soft tissue deformation and the time of peak pressure. The Courant stability condition ensures that the numerical method can correctly capture the propagation process of physical quantities by constraining the ratio of the time step to the space step. Explanation: In the finite element solver, the time step is automatically determined by the ratio of the minimum element size in the mesh to the elastic wave propagation velocity in soft tissue, ensuring that each calculation step captures wave details. For example, if the minimum element side length is approximately 0.5 mm and the shear wave propagation velocity in soft tissue is approximately 1,500 meters per second, the time interval between each simulation calculation is approximately 330 microseconds. The entire simulation process lasts three seconds, and the system records the simulation results every five milliseconds.

[0034] Step 203: Prediction model design and training: Construct an oral simulation prediction model based on the GNN-LSTM hybrid network to predict the soft tissue deformation vector, base displacement, and time lag parameters. The oral simulation prediction model includes: Graph neural network backbone: The pressure matrix input shape is 5×5, and two layers of Chebyshev graph convolution are first applied, each layer outputs 64-dimensional features, and the activation function is ReLU to extract spatial coupling information; Temporal memory unit: The GNN output features are input into two layers of LSTM according to the time series length T = 50 frames, with 128 hidden units in each layer and a dropout rate of 0.1 to extract temporal dependencies; Multi-task output branch: The output of the last layer of LSTM is connected to three fully connected sub-networks to predict 、 、 ; Training details: Cross-scenario enhancement: The mild median and severe lateral pressure sequences are spliced ​​together to generate virtual sequences. 200 virtual data points are generated for each scenario to supplement the scarce extreme scenarios. Optimizer and parameters: Adam, initial learning rate 0.001, batch size 32, training epochs 2000, validation set weight 20%, early stopping strategy (stop if validation set loss does not decrease for 50 consecutive epochs); The loss function consists of three parts: First, the deformation prediction error term measures the absolute difference between the soft tissue deformation output by the network and the annotated value from the finite element simulation, ensuring that the model accurately reproduces soft tissue deformation numerically. Second, the displacement and time-delay joint error term simultaneously calculates the difference between the mean base displacement and the response time-delay prediction and simulation values, ensuring that the model captures both the overall micro-displacement trend and the soft tissue delay effect. Finally, the physical consistency error term measures the energy-level deviation of the network predictions and the sum of elastic energy storage and viscous dissipation obtained from simulations. This term emphasizes not only minimizing numerical errors but also maintaining consistency with the real physical process in terms of energy conservation and viscous dissipation. The overall loss is the sum of these three errors. By simultaneously taking into account the deformation, displacement time-delay, and energy consistency, the network is driven to converge both numerically and physically.

[0035] Explanation: With the pressure time series data, soft tissue deformation and base displacement and time lag data pairs generated in sub-step 202 as annotations, a graph neural network backbone is designed to extract the spatial coupling characteristics of the pressure matrix, long short-term memory units are input to capture the dynamic change trend, and multi-task output branches are constructed to predict the maximum soft tissue deformation, the overall base displacement and the response time lag respectively; during the training process, the cross-scene enhancement strategy (combining mild occlusion with severe lateralization to generate a virtual pressure sequence) and the loss function (making the three predictions of soft tissue deformation, base displacement and time lag coordinated and optimized in parallel) are combined, and iterations are repeated until each prediction output converges on the verification set, and an oral simulation prediction model that can be inferred in real time is obtained.

[0036] Step 204: Online update and optimization: When the residual error of the soft tissue deformation predicted by the oral simulation prediction model exceeds a threshold, incremental online fine-tuning is automatically triggered to ensure that the oral simulation prediction model is continuously optimized under the new scenario.

[0037] Automatically triggered incremental online fine-tuning includes: Incremental sample selection: extract 10 frames of data before and after the trigger moment to form an incremental training set; retain the latest 50 frames of data as the incremental sample pool; Fine-tuning strategy: Use Adam optimizer with a learning rate of 1×10 -5 , fine-tune for 10 rounds each time; while continuing to calculate multi-task losses on incremental samples; Validation and rollback: After every 100 fine-tuning steps, the average residual is evaluated on the retained validation set. If the validation residual increases by more than 10%, the model is rolled back to the previous best model weight.

[0038] Step 3: When the deformation or base displacement predicted by the real-time pressure array and the oral simulation prediction model exceeds the preset threshold, the real-time pressure array is mapped to the base coordinate system using three-dimensional interpolation. The real-time pressure array is initially corrected in combination with the deformation magnification factor. The deformation magnification factor is iteratively updated with the geometric closed-loop residual driving ratio until the geometric closed-loop residual converges. The equivalent pressure data is output to eliminate the implicit disturbance caused by soft tissue deformation and provide accurate pressure input for three-dimensional mesh reconstruction. Explanation: The real-time pressure array refers to the force value directly measured by each pressure sensing unit arranged on the occlusal surface of the denture before any correction is performed; it includes the spatial offset and distribution error caused by the viscoelastic deformation of the soft tissue and the movement of the base, and is therefore not equivalent to the true occlusal pressure when the base is assumed to remain rigid; eliminating the implicit disturbance caused by soft tissue deformation here means deducting the measurement point position offset and pressure dispersion caused by the slight deformation of the mucosa or gums during chewing from the original pressure data, so that the final equivalent pressure data only reflects the assumption that the base is completely rigid and the soft tissue is not deformed. The actual occlusal load distribution is obtained by first adding the predicted overall displacement of the base to the position of each pressure measuring point, and then using three-dimensional interpolation to extract the force value corresponding to the translated coordinate from the original array. The force value is then multiplied by a deformation factor proportional to the soft tissue deformation amplitude to restore it to the equivalent pressure data as if there was no deformation. Next, the corrected equivalent pressure data is mapped back to the original geometry, and the geometric residual is measured by calculating the maximum projection distance between these points and the oral optical point cloud. The deformation factor is adjusted proportionally according to the residual size, and the above correction operation is repeated until the geometric residual converges. In this way, the spatial deviation introduced by the soft tissue deformation can be removed after several consecutive iterations. The output equivalent pressure data is the actual pressure distribution under the condition of assuming a rigid base and undeformed soft tissue, which is used for subsequent three-dimensional mesh reconstruction.

[0039] The specific implementation is as follows: Step 301: Read the real-time pressure array data and the soft tissue deformation and base displacement predicted by the oral simulation geometric model at the same time, and call the deformation magnification coefficient, which is a pre-set dimensionless adjustment parameter used to map and associate the predicted soft tissue deformation with the pressure array data. (In the initial iteration, the deformation magnification coefficient is obtained from empirical values ​​or previous calibration). The real-time pressure array is derived from the multi-point pressure sensing units arranged on the occlusal surface of the denture, and the soft tissue deformation and base displacement are output by the trained oral simulation geometric model. The deformation magnification coefficient is a pre-set dimensionless adjustment parameter. Explanation: The deformation multiplication factor is a dimensionless factor that converts the predicted soft tissue deformation into a pressure correction amplitude proportionally. It is used to adjust the pressure value according to the deformation magnitude during the initial mapping. The deformation multiplication factor is modified in each subsequent iterative update, thereby gradually converging to the final equivalent pressure data. Step 302: Based on linear approximate mapping, the position information of the real-time pressure array is added to the base displacement predicted by the oral simulation geometric model. The corresponding real-time pressure array value is obtained at the mapped coordinate using a three-dimensional interpolation technique. The pressure value is then combined with the soft tissue deformation using a deformation multiplication factor to obtain preliminary equivalent pressure data. The preliminary equivalent pressure data represents the pressure distribution when the base is assumed to be rigid and soft tissue disturbances are ignored, and is used for subsequent geometric verification. In a possible embodiment, the preliminary equivalent pressure data satisfies the following formula: ; in, represents preliminary equivalent pressure data, represents mapping the position of the i-th pressure sensing unit from the soft tissue perturbation back to the rigid reference coordinates; is the Euclidean norm for predicting soft tissue deformation, represents the deformation magnification factor, is the dimensionless adjustment coefficient, express The pressure values ​​at the corresponding coordinates have not been subjected to any geometric mapping or soft tissue / base correction, and reflect the uncorrected occlusal pressure distribution actually collected; Step 303: Map the pressure point positions to which the preliminary equivalent pressure belongs (representing the pressure distribution at each point on the denture occlusal surface, including the pressure values ​​and their specific positions on the denture occlusal surface. The position information is obtained through the layout and calibration of the pressure sensors, and it can be understood that each pressure value corresponds to a coordinate point in three-dimensional space) to the three-dimensional point cloud obtained by the oral optical scan (a three-dimensional model of the internal structure of the oral cavity obtained through optical scanning technology). Calculate the minimum geometric distance from each pressure mapping point to the point cloud using the nearest neighbor distance method, summarize the distances to obtain a set, and use the maximum value of the distance set as the geometric residual. If the geometric residual does not exceed the preset geometric tolerance limit, the correction is determined to have converged and the iteration is terminated. Otherwise, proceed to the next step. Step 304: When the geometric residual is greater than the preset value, the difference between the current geometric residual and the tolerance limit is used as the basis for adjustment. The deformation magnification coefficient is slightly increased or decreased to ensure that the change does not exceed the predetermined threshold. Then, the second step is called again and the next equivalent pressure is calculated in the same manner. The process of steps 302 and 303 is repeated until the geometric residual meets the preset standard or the maximum number of iterations is reached. If the convergence condition is not met after exceeding the maximum number of iterations, the corresponding moment is marked as an abnormal disturbance and the original data is retained for subsequent analysis. Step 305: After the iteration is completed, the final converged equivalent pressure data and the corresponding geometric residual are output. The equivalent pressure data is used to guide the subsequent occlusal surface geometry reconstruction and dynamic correction.

[0040] Furthermore, in the real-time pressure array correction, the base is not set to be absolutely rigid. Instead, elastic parameters matching the actual base material are assigned in the finite element model. The initial correction pressure is calculated by combining the soft tissue deformation and the base elastic deformation. The geometric closed-loop residual drives the iteration of the shared magnification factor of the soft tissue and base until the geometric residual converges. The elastic equivalent pressure data is output, thereby eliminating the implicit disturbance caused by the elastic deformation of the soft tissue and base, providing accurate pressure input for 3D mesh reconstruction. The difference from the case where the base is assumed to be rigid is that: When calculating the pressure mapping position, the reference coordinates of the original pressure measurement point in space are superimposed with the overall rigid displacement component of the base, the elastic deformation component of the base, and the block deformation component of the soft tissue output by the simulation model. This generates a mapping coordinate that includes the elastic deflection of the base and the average deformation of the soft tissue, instead of only considering the translational offset caused by the rigid movement of the base. In the initial equivalent pressure correction, the original pressure value obtained by interpolation is first corrected by multiplying the predicted soft tissue normal deformation by the soft tissue deformation multiplication factor, and then multiplied by the base elasticity correction factor corresponding to the base elastic deformation to compensate for the influence of the base material deflection on the pressure distribution, thereby approximately restoring the pressure value under the assumption of base rigidity; In the geometric closed-loop iterative correction, the coordinates of the above-mentioned mapping points are regarded as the positions after the elastic deformation of the current base (i.e., the elastic base coordinates), and the coordinates are nearest-neighbor aligned with the optical point cloud to calculate the geometric residual. If the residual does not converge, the soft tissue deformation magnification factor and the base elastic correction factor are updated simultaneously according to the size of the residual to achieve dual iterative correction of the elastic disturbance of the soft tissue and the base.

[0041] The final output equivalent pressure data reflects the rigid approximate pressure distribution under the premise of considering the elastic deformation of the base.

[0042] Furthermore, the degree of soft tissue deformation and the amount of base movement in different occlusal areas may be uneven, making it difficult for the overall correction to take into account all areas; multi-area parallelism can improve local accuracy and avoid over-reliance on a single deformation multiplication coefficient; the occlusal surface is divided into several fixed sub-areas, and the geometric residual is calculated for each sub-area and the respective deformation multiplication coefficients are updated. After completing the sub-area correction, the correction results are merged as a whole; for example, the occlusal surface grid is divided into 3×3 nine-square areas, and the mapping and iteration in sub-steps 302–303–304 are performed on each area, and finally the equivalent pressure data of all areas are integrated.

[0043] Step 4: Extract key contact points based on equivalent pressure data and predicted displacement. First, perform weighted least squares registration to fuse the optical point cloud and then perform 3D reconstruction. When the curvature error or displacement threshold meets the conditions, drive local high-density resampling, re-perform local inverse correction, and generate a continuous 3D mesh that takes into account both the overall smoothness and local refinement of the occlusal surface. The specific implementation is as follows: Step 401: Extracting the initial contact point cloud: In a virtual occlusal environment, based on the equivalent pressure data distribution output in step 3, an initial contact point cloud distribution is generated in the centric occlusal posture. Several pressure measurement points with pressure values ​​in the top 20% are selected as key contact points, and the contact point clouds corresponding to the key contact points are collected to represent the pressure concentration areas during occlusion. Step 402: Global translation and normal fine-tuning: Each point in the contact point cloud is translated and corrected along the local displacement vector predicted by the oral simulation geometric model. Then, a small displacement is added to the local normal direction of the translated point based on the soft tissue deformation value predicted by the oral simulation geometric model to reflect the change in contact surface curvature caused by soft tissue deformation, thereby obtaining a corrected contact point cloud. Step 403: Point cloud fusion and preliminary mesh reconstruction: The corrected contact point cloud is fused with the oral optical 3D point cloud using a weighted least squares registration algorithm. A 3D mesh generation method based on Poisson reconstruction is applied to the fused point cloud. Laplace smoothing is performed on the reconstructed preliminary mesh to ensure that the mesh maintains both overall smoothness and the geometric characteristics of the local contact points. At the same time, the local curvature error is calculated at each contact point by comparing the surface curvature of the original point cloud to provide a quantitative basis for subsequent resampling. Step 404: Local high-density resampling trigger and recalibration: When the base displacement corresponding to the corrected point cloud exceeds a preset displacement threshold or the curvature error of any contact point exceeds a preset curvature threshold, a spherical area with a fixed radius is defined with the contact point that meets the threshold condition as the center. High-resolution optical point cloud and pressure data are recollected within the spherical area, and steps 401 to 403 are repeated. Local inverse mapping and geometric closed-loop correction are performed to obtain refined local equivalent pressure data and point cloud data. Step 405: Final 3D mesh construction and residual output: The locally resampled and corrected point cloud is merged with the original fused point cloud, and the merged point cloud is again subjected to Poisson reconstruction and Laplace smoothing while retaining the corrected contact point position constraints to generate a final continuous 3D mesh model. The overall geometric residual is determined by calculating the minimum distance from each contact point to the final mesh, and the output includes the 3D mesh model and its corresponding geometric residual result.

[0044] Step 5: After obtaining the local correction pressure and updating the point cloud, the contact energy index is generated by accumulating the frame-by-frame pressure energy and base displacement. The wear probability distribution is then calculated in combination with the local normal curvature. When the wear probability exceeds a preset threshold, the sampling density for the next cycle is automatically increased, and high-risk areas are marked in the final 3D mesh, achieving proactive prevention of denture micro-area wear. The specific implementation is as follows: Step 501: Accumulate frame-by-frame pressure energy and base displacement The local corrected rigid pressure at each sampling moment and the corresponding predicted displacement of the oral simulation geometric model are collected, and the cumulative pressure energy and displacement energy of each contact point in the time series are calculated to quantify the loading history of the contact point during continuous chewing. Step 502: Calculate the wear probability based on the local normal curvature For each contact point, the normal curvature of its corresponding position in the optical point cloud is obtained. The accumulated pressure energy is combined with the normal curvature and input into a pre-trained wear probability prediction model to output the wear probability of the contact point. Step 503: Automatically increase sampling density when wear probability exceeds threshold When the wear probability of any contact point exceeds the preset threshold, the sampling area within a fixed radius of the contact point is automatically switched to high-density sampling mode to re-collect optical point cloud and pressure data at a higher resolution; Step 504: Mark high-risk areas on the final 3D grid In the three-dimensional mesh model, all contact points where the wear probability exceeds the threshold are visually marked as high-risk areas, providing a reference for denture material optimization and subsequent design.

[0045] Furthermore, in constructing the wear probability prediction model, the following steps were performed: The cumulative pressure energy, displacement energy, and local normal curvature characteristics of each contact point were obtained through laboratory cyclic loading and clinical repair sample collection, and normalized preprocessing was performed; a multi-layer perceptron and lightweight gradient boosting tree hybrid classifier was constructed, and model parameters were optimized using cross-validation and early stopping strategies; the trained model was deployed in an online interface to calculate the wear probability of each contact point in real time. The construction process of the wear probability prediction model includes the following steps: Step 521: Collect and label experimental and clinical wear samples. First, perform repeated masticatory cyclic loading tests on typical denture materials in a laboratory environment. Record the local pressure time series, cumulative pressure energy, and load duration corresponding to each contact point under each loading cycle. After the test, measure the wear depth and weight loss using a microscope or profilometer to classify the sample labels into high-wear and low-wear categories. Simultaneously, select repaired denture samples that have been worn for several months in a clinical environment. Extract the actual wear area and depth through optical scanning. Combined with the cumulative pressure energy, base displacement, and local curvature information output by the patient's oral simulation geometric model, map the actual wear degree of each contact point to a multi-level or binary classification label to ensure that the data sources cover different materials and different oral regions, and that diversity and balance are guaranteed. Step 522: Feature engineering and data preprocessing. Numerical features such as the cumulative pressure energy, displacement energy, local normal curvature, and cumulative number of chewing cycles corresponding to each contact point in the wear samples collected in the laboratory and clinical settings are normalized to eliminate the effects of different dimensions. Missing or abnormal data are corrected using nearest neighbor interpolation or median filling strategies, and obvious outliers are removed. If the labels are multi-level, one-hot encoding is used to convert the wear levels into classification vectors. Finally, a feature matrix and corresponding labels are generated for training to ensure stable and reliable input to the subsequent model. Step 523: Model architecture selection and principle design; Based on the feature dimension, a multilayer perceptron neural network is selected as the primary classifier. The number of nodes in the network input layer is consistent with the feature dimension. The hidden layer is set to two to three layers, with the number of nodes in each layer decreasing step by step and using the ReLU activation function. The output layer uses Sigmoid activation to obtain the binary classification wear probability. At the same time, for comparison purposes, a lightweight gradient boosting tree model (LightGBM) can be designed as a benchmark. The tree depth and number of leaf nodes are adjusted within a reasonable range based on the features and sample size. The loss function is binary cross entropy, which is used to measure the difference between the predicted probability and the true label. Key parameters include the learning rate, number of hidden units, batch size, and tree model hyperparameters. Step 524: Model training and validation strategy. The preprocessed dataset was split into a training set and a validation set at 80% / 20%, and a 50% cross-validation system was used to evaluate the generalization performance of the model during the training process. The neural network used the Adam optimizer, with an initial learning rate of 0.001 and a batch size of 32. The early stopping mechanism was initiated after the validation set loss did not decrease for 20 consecutive epochs. For the tree model, the default parameters of LightGBM were used for preliminary training, and key hyperparameters were tuned using grid search or Bayesian optimization. During the training process, the performance of different models was compared by calculating the AUC, precision, recall, F1 score, and confusion matrix, and the model architecture with the best performance was ultimately selected. Step 525: Model deployment and online update mechanism; export the trained best model into a portable format (such as ONNX or TorchScript) and deploy it to an edge or cloud server, receive real-time input of features such as cumulative pressure energy, displacement energy, and local curvature through a RESTful API interface, and output the wear probability of each contact point in real time; set a wear probability threshold (for example, 0.7) to trigger high-density sampling, and automatically increase the sampling accuracy when it is detected that the probability of a certain contact point exceeds the threshold; after deployment, regularly (such as quarterly) incorporate newly collected real wear and pressure-curvature data into the incremental training set, and update the model parameters through transfer learning or incremental training to maintain the model's adaptability to different patients and material environments.

[0046] The present invention also provides a denture occlusal surface three-dimensional modeling optimization system based on dynamic pressure calibration, comprising: The dynamic acquisition module simultaneously acquires ultrasonic elastic images, array pressure, and base displacement data while the patient's mouth is open. After verification through palpation, it generates individualized soft tissue feature data based on the deformation / displacement dual threshold. The simulation modeling module reconstructs the geometric model of oral soft and hard tissues based on individualized soft tissue feature data, assigns nonlinear viscoelastic properties to the soft tissues and elastoplastic properties to the dentures, generates pressure-deformation data pairs through finite element simulation, and constructs an oral simulation prediction model. The output data pairs map pressure to soft tissue deformation and base displacement. When the real-time pressure array and the predicted deformation or base displacement of the oral simulation prediction model exceed the preset threshold, the pressure correction module uses three-dimensional interpolation to map the real-time pressure array to the base coordinate system. The real-time pressure array is initially corrected in combination with the deformation magnification factor. The deformation magnification factor is updated iteratively using the geometric closed-loop residual to drive the proportional update until the geometric closed-loop residual converges. The equivalent pressure data is then output to eliminate the implicit disturbance caused by soft tissue deformation. The reconstruction optimization module fuses equivalent pressure data with optical point cloud data and extracts key contact points through weighted least squares registration. When the curvature error exceeds the threshold, local resampling is triggered. After inverse correction, a smooth 3D mesh is generated and output to the wear warning module. The wear warning module accumulates frame-by-frame pressure energy and displacement to generate a contact energy index, combines local curvature to calculate the wear probability distribution, automatically increases the sampling density in areas where the probability exceeds the limit, marks high-risk coordinates in the grid, and synchronously feeds back the sampling instructions to the reconstruction module to form a closed-loop optimization.

[0047] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A three-dimensional modeling and optimization method for denture occlusal surface based on dynamic pressure calibration, characterized in that: include: Step 1: Acquire ultrasonic shear wave elastic images of the patient's oral soft tissue. Obtain oral array pressure matrices and base inertial measurement displacements through palpation pressure testing. Classify and annotate them according to preset deformation and displacement thresholds to generate individualized soft tissue feature data. Step 2: Use the patient's oral imaging data to reconstruct an oral simulation geometric model including soft tissue, bone, and dentures. Based on the nonlinear viscoelastic constitutive model, the soft tissue material properties are obtained, and the dentures are given elastic-plastic material properties. A soft tissue perturbation finite element simulation is performed, and the output pressure is mapped to a data pair of soft tissue deformation and base displacement. A graph neural network is then integrated with a temporal memory unit to construct an oral simulation prediction model. Step 3: When the deformation or base displacement predicted by the real-time pressure array and the oral simulation prediction model exceeds the preset threshold, three-dimensional interpolation is used to map the real-time pressure array to the base coordinate system, and the real-time pressure array is initially corrected in combination with the deformation magnification factor. The deformation magnification factor is updated iteratively with the geometric closed-loop residual driving ratio until the geometric closed-loop residual converges, and the equivalent pressure data is output to eliminate the implicit disturbance caused by soft tissue deformation.

2. The method for three-dimensional modeling and optimization of denture occlusal surface based on dynamic pressure calibration according to claim 1, characterized in that: The process of acquiring individualized soft tissue feature data includes the following steps: Step 101: attaching a shear wave elastography probe to the surface of oral soft tissue, and based on the shear wave velocity and soft tissue density, expressing the storage modulus as the product of the soft tissue density and the square of the shear wave velocity; using the same shear wave elastography probe to measure the shear wave velocity at two adjacent frequencies, expressing the loss modulus as the product of the shear wave velocity per unit frequency, the average frequency, and an empirical calibration coefficient; Step 102: Press the pressure sensor vertically into the oral soft tissue surface and record the real-time displacement and the real-time soft tissue deformation at the denture base-mucosal contact interface. , input the nonlinear least squares fitting viscoelastic model, and output the viscoelastic parameter set and pressure-soft tissue deformation time series curve; Step 103: Affix an inertial measurement device to the back of the center of the denture base to record three-dimensional linear acceleration and angular velocity in real time; After zero bias calibration and complementary filtering algorithm, the output of the inertial measurement device is fused into the base displacement relative to the oral coordinate system; Step 104: Deploy a piezoresistive sensing unit below the occlusal surface of the denture to output a pressure matrix; each element of the pressure matrix represents instantaneous pressure data at a specific moment and position coordinate. The oral pressure center is obtained by calculating the weighted average of all position coordinates, where the weight is the instantaneous pressure data measured at the specific moment, and the normalization factor is the sum of the pressures of the entire array. Step 105: classifying the soft tissue disturbance measurement index into several levels; Step 106: Match the outputs of steps 101 to 104 with the disturbance measurement indicators one by one to generate individualized soft tissue feature data, which are stored in a soft tissue parameter library.

3. The method for three-dimensional modeling and optimization of denture occlusal surface based on dynamic pressure calibration according to claim 2, characterized in that: The viscoelastic parameter set is , the nonlinear least squares fitting viscoelastic model is used to solve the viscoelastic parameter set, and the viscoelastic model satisfies the following formula: ； in, is the long-term elastic modulus, is the indentation depth that varies with time, represents the instantaneous rate of displacement over time; the integral upper limit t accumulates from 0 to the current moment to achieve the accumulation of historical effects, and s is the historical time variable in the integral; the value of i is 1 or 2; E1 and E2 are the elastic moduli corresponding to the first and second Maxwell viscoelastic elements, respectively; and are the relaxation times of the first and second Maxwell units, respectively.

4. The method for three-dimensional modeling and optimization of denture occlusal surface based on dynamic pressure calibration according to claim 3, characterized in that: The method includes a multi-source data mutual calibration step, including: comparing the measured storage modulus with the storage modulus obtained by inversion fitting of viscoelastic parameters, and calculating the elastic deviation ratio point by point; when the deviation ratio exceeds a threshold, triggering viscoelastic parameter reoptimization, re-adjusting the initial value of the nonlinear least squares fitting viscoelastic model or re-optimizing the weight until the storage modulus obtained by inversion fitting of the viscoelastic parameters is consistent with the measured storage modulus.

5. The method for optimizing the three-dimensional modeling of the denture occlusal surface based on dynamic pressure calibration according to claim 1, characterized in that: The method further comprises the step of intelligently removing motion artifacts: The output of the inertial measurement device and the pressure center timing mapping are mutually verified. By comparing and synchronizing the base displacement output by the inertial measurement device with the pressure center vector timing calculated by the pressure sensor array, the normalized amplitude of the vector cross product of the base displacement and the pressure center vector is used to construct a timing consistency criterion. When the normalized amplitude exceeds the threshold, the base displacement and pressure center vector data at the corresponding moment are automatically eliminated.

6. The method for three-dimensional modeling and optimization of denture occlusal surface based on dynamic pressure calibration according to claim 1, characterized in that: The oral simulation prediction model: Using the pressure time series data, soft tissue deformation, and base displacement and time lag data pairs generated in sub-step 202 as annotations, a graph neural network backbone is designed to extract the spatial coupling features of the pressure matrix; Long short-term memory units are input to capture dynamic change trends, and multi-task output branches are constructed to predict the maximum deformation of soft tissue, the overall displacement of the base, and the response lag respectively. During the training process, the cross-scene enhancement strategy and loss function are combined, and iterations are repeated until each prediction output converges on the validation set to obtain the oral simulation prediction model; the loss function consists of a deformation prediction error term, a displacement and time lag joint error term, and a physical consistency error term; The deformation prediction error term measures the absolute difference between the soft tissue deformation output by the network and the annotated value of the finite element simulation; the displacement and time lag joint error term simultaneously calculates the difference between the mean base displacement and the response time lag prediction value and the simulated value; the physical consistency error term is calculated by comparing the sum of elastic energy storage and viscous dissipation obtained by network prediction and simulation.

7. The method for optimizing the three-dimensional modeling of the denture occlusal surface based on dynamic pressure calibration according to claim 1, characterized in that: The operation process of step 3 includes the following steps: Step 301: Read the real-time pressure array data and the soft tissue deformation and base displacement predicted by the oral simulation geometric model at the same time, and call the deformation magnification coefficient. The deformation magnification coefficient is a pre-set dimensionless adjustment parameter used to map and associate the predicted soft tissue deformation with the pressure array data. Step 302: Based on linear approximate mapping, the position information of the real-time pressure array is added to the base displacement predicted by the oral simulation geometric model. The corresponding real-time pressure array value is obtained at the mapped coordinate using a three-dimensional interpolation technique. The pressure value is then combined with the soft tissue deformation using a deformation multiplication factor to obtain preliminary equivalent pressure data. The preliminary equivalent pressure data represents the pressure distribution when the base is assumed to be rigid and soft tissue disturbances are ignored, and is used for subsequent geometric verification. Step 303: Map the pressure point positions of the preliminary equivalent pressures to the three-dimensional point cloud obtained by the oral optical scan. Calculate the minimum geometric distance from each pressure mapping point to the point cloud using the nearest neighbor distance method. Summarize the distances to obtain a set, and use the maximum value of the distance set as the geometric residual. When the geometric residual does not exceed the preset geometric tolerance limit, the correction is determined to be converged and the iteration is terminated, otherwise it proceeds to the next step; Step 304: When the geometric residual is greater than the preset value, the difference between the current geometric residual and the tolerance limit is used as the basis for adjustment. The deformation magnification coefficient is slightly increased or decreased to ensure that the change does not exceed the predetermined threshold. Then, the second step is called again and the next equivalent pressure is calculated in the same manner. The process of steps 302 and 303 is repeated until the geometric residual meets the preset standard or the maximum number of iterations is reached. If the convergence condition is not met after exceeding the maximum number of iterations, the corresponding moment is marked as an abnormal disturbance and the original data is retained for subsequent analysis. Step 305: After the iteration is completed, the final converged equivalent pressure data and the corresponding geometric residual are output. The equivalent pressure data is used to guide the subsequent occlusal surface geometry reconstruction and dynamic correction.

8. The method for optimizing the three-dimensional modeling of the denture occlusal surface based on dynamic pressure calibration according to claim 7, characterized in that: The occlusal surface is divided into several fixed sub-areas. The geometric residual is calculated for each sub-area and the deformation magnification coefficient is updated. After the sub-area correction is completed, the correction results are merged as a whole.

9. The method for optimizing the three-dimensional modeling of the denture occlusal surface based on dynamic pressure calibration according to any one of claim 1, characterized in that: The method further comprises: Step 4: Extract key contact points based on equivalent pressure data and predicted displacement, perform 3D reconstruction after first fusing the optical point cloud with weighted least squares registration, and drive local high-density resampling when the curvature error or displacement threshold meets the conditions; Step 5: After obtaining the local correction pressure and updating the point cloud, the frame-by-frame pressure energy and base displacement are accumulated to generate the contact energy index and combined with the local normal curvature to calculate the wear probability distribution. When the wear probability exceeds the preset threshold, the sampling density of the next cycle is automatically increased and the high-risk areas are marked in the final three-dimensional grid to achieve proactive prevention of denture micro-area wear.

10. A three-dimensional modeling and optimization system for denture occlusal surfaces based on dynamic pressure calibration, used to implement the method of claim 9, characterized in that: include: The dynamic acquisition module simultaneously acquires ultrasonic elastic images, array pressure, and base displacement data while the patient's mouth is open. After verification through palpation, it generates individualized soft tissue feature data based on the deformation / displacement dual threshold. The simulation modeling module reconstructs the geometric model of oral soft and hard tissues based on individualized soft tissue feature data, assigns nonlinear viscoelastic properties to the soft tissues and elastoplastic properties to the dentures, generates pressure-deformation data pairs through finite element simulation, and constructs an oral simulation prediction model. The output data pairs map pressure to soft tissue deformation and base displacement. When the real-time pressure array and the predicted deformation or base displacement of the oral simulation prediction model exceed the preset threshold, the pressure correction module uses three-dimensional interpolation to map the real-time pressure array to the base coordinate system. The real-time pressure array is initially corrected in combination with the deformation magnification factor. The deformation magnification factor is updated iteratively using the geometric closed-loop residual to drive the proportional update until the geometric closed-loop residual converges. The equivalent pressure data is then output to eliminate the implicit disturbance caused by soft tissue deformation. The reconstruction optimization module fuses equivalent pressure data with optical point cloud data and extracts key contact points through weighted least squares registration. When the curvature error exceeds the threshold, local resampling is triggered. After inverse correction, a smooth 3D mesh is generated and output to the wear warning module. The wear warning module accumulates frame-by-frame pressure energy and displacement to generate a contact energy index, combines local curvature to calculate the wear probability distribution, automatically increases the sampling density in areas where the probability exceeds the limit, marks high-risk coordinates in the grid, and synchronously feeds back the sampling instructions to the reconstruction module to form a closed-loop optimization.

Images