A jaw pad three-dimensional image generation system and generation method

Abstract

The present invention discloses a jaw pad three-dimensional image generation system and generation method, comprising a data acquisition module: collecting oral dynamic parameters through a wearable intraoral patch sensor, the oral dynamic parameters including oral image data and oral environment data; a micro feature module: optimizing micro sampling of oral dynamic parameters using an improved MobileNetV3 network based on oral image data to obtain a micro image of the tooth edge; wherein, the improved MobileNetV3 network can significantly improve the extraction accuracy of micro features of the tooth edge by optimizing the initial convolution layer and the depthwise separable convolution layer of the original MobileNetV3 network and adding a deformable convolution block after the last convolution layer of the MobileNetV3 network, thereby reducing the error between the jaw pad edge and the tooth anatomical structure and improving the fit and comfort.

Description

Technical Field

[0001] The present invention relates to the technical field of jaw pad three-dimensional image generation, and in particular to a jaw pad three-dimensional image generation system and generation method. Background Art

[0002] Jaw pads are widely used in the field of dentistry, such as treating bruxism and occlusion disorders. Traditional jaw pad production often relies on rough oral models and experience, lacking precision and personalization.

[0003] At present, in the traditional jaw pad three-dimensional image generation system, the jaw pad production relies on manual molding and empirical design, and has insufficient anatomical adaptability. For example, manual production based on plaster models is difficult to accurately capture the microscopic features of tooth edges (such as cusp inclination and pit and fissure morphology), resulting in poor fit between the jaw pad and the teeth, which can easily cause foreign body sensation or occlusal interference. In addition, there is a lack of multi-physical field coupling analysis, and the dynamic effects of chemical corrosion (saliva pH value), temperature changes (hot and cold stimulation of eating), and mechanical loads (bite force distribution) in the oral cavity are not comprehensively considered, resulting in the jaw pad being prone to corrosion wear, thermal deformation or mechanical failure. Therefore, a jaw pad three-dimensional image generation system and generation method are proposed herein. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned objectives, the present invention proposes the following technical solutions:

[0005] A jaw pad three-dimensional image generation system, comprising:

[0006] Data acquisition module: collects oral dynamic parameters through a wearable intraoral patch sensor, the oral dynamic parameters including oral image data and oral environment data;

[0007] Microscopic feature module: Based on oral image data, the improved MobileNetV3 network is used to optimize the microscopic sampling of oral dynamic parameters to obtain microscopic images of tooth edges;

[0008] The improved MobileNetV3 network is optimized by optimizing the initial convolution layer and the depth-separable convolution layer of the original MobileNetV3 network and adding a deformable convolution block after the last convolution layer of the MobileNetV3 network.

[0009] Physical Analysis Module: Analyzes the physical field in the oral cavity based on oral environment data, establishes a lightweight physical optimization model for multi-physical field coupling, and uses an iterative solution method to solve the multi-physical field coupling model to obtain the optimal physical characteristic parameters;

[0010] Three-dimensional construction module: construct the initial implicit surface model of the jaw pad based on the microscopic image of the tooth edge, optimize the initial implicit surface model of the jaw pad based on the optimal physical property parameters to obtain the optimized jaw pad three-dimensional model, and output the jaw pad three-dimensional image based on the optimized jaw pad three-dimensional model.

[0011] The oral dynamic parameters include oral image data and oral environment data;

[0012] The oral environment data includes saliva pH value data, temperature distribution data and bite force distribution data.

[0013] The initial convolutional layer optimization process is:

[0014] The initial convolutional layer of the original MobileNetV3 network is adjusted by using an asymmetric convolution kernel combination, replacing the 3×3 convolution kernel with 3×1 and 1×3 convolution kernels;

[0015] The optimization process of the depthwise separable convolutional layer is as follows:

[0016] For the depthwise separable convolutional layer, a channel attention mechanism is introduced to obtain the global information of the feature map of each channel based on the channel attention mechanism. Then, the channel weight is obtained through two fully connected layers. The channel weight is multiplied with the original feature map to complete the optimization of the depthwise separable convolutional layer.

[0017] The process of adding a deformable convolution block after the last convolution layer of the MobileNetV3 network is as follows:

[0018] The deformable convolution block learns the offset through an additional convolution layer, and optimizes the original convolution kernel sampling point coordinates by learning the optimal offset to obtain the new sampling point coordinates.

[0019] The optimal offset learning process is:

[0020] The offset is limited to a specific range, a dynamic adjustment function is obtained, the eigenvalue at the original convolution kernel sampling point coordinate is divided by the maximum value for normalization, and then input into the multi-layer perceptron, activated by the Sigmoid activation function, and the dynamic adjustment coefficient is obtained;

[0021] Obtain the offset through two-stage recursive learning;

[0022] The two-stage recursive learning includes a first stage of coarse migration and a second stage of fine-tuning network calculation;

[0023] The initial offset is generated based on the coarse offset in the first stage, and the correction value is obtained through the dilated gated convolution based on the fine-tuning network in the second stage, taking the initial offset and residual features as input;

[0024] The optimal offset is determined by dynamically adjusting parameters based on the initial offset and the corrected offset.

[0025] The process of obtaining the optimal physical property parameters is as follows:

[0026] The multi-physical fields include chemical fields, temperature fields and mechanical fields, and energy functionals of all fields in the multi-physical fields are obtained;

[0027] Construct the objective function of the lightweight physical optimization model based on the energy functional of all fields in the multi-physics field;

[0028] The objective function is optimized and solved using an iterative solution method to obtain the optimal physical characteristic parameters.

[0029] The process of constructing the initial implicit surface model of the jaw pad based on the microscopic image of the tooth edge is as follows:

[0030] Obtain the coordinates of the optimal control points based on the microscopic image of the tooth edge;

[0031] Based on the optimization algorithm and presetting a control point control amount, the control weight of the initial jaw pad implicit surface model is obtained;

[0032] The determined optimal control point coordinates and control weights are substituted into the radial basis function formula, and a scalar function is defined. When the scalar function is 0, the surface of the model is determined, and the radial basis function is used to construct the implicit surface to generate the initial jaw pad implicit surface model.

[0033] The process of optimizing the initial jaw pad implicit surface model based on the optimal physical characteristic parameters is as follows:

[0034] According to the porosity distribution information of different regions in the optimized physical property parameters, the internal scalar function of the implicit surface model is adjusted;

[0035] Define the porosity distribution matrix and determine an initial implicit surface scalar field. Use piecewise linear interpolation to adjust the initial implicit surface scalar field and set physical constraints to obtain the adjusted scalar.

[0036] The surface of the jaw pad three-dimensional model is determined by finding the isosurface with the adjusted scalar value equal to 0. The marching cubes algorithm is used to convert the scalar field data into a surface model represented by a triangular mesh to obtain the optimized jaw pad three-dimensional model.

[0037] The physical constraints include:

[0038] When the porosity distribution matrix is ​​greater than 0.7 and is in the high porosity area, the adjusted scalar increases by 0.2;

[0039] When the porosity distribution matrix is ​​less than 0.3 and is in the low porosity area, the adjusted scalar increases by 0.1.

[0040] A method for generating a three-dimensional image of a jaw pad, the method comprising the following steps:

[0041] S1: Collecting oral dynamic parameters through a wearable intraoral patch sensor, wherein the oral dynamic parameters include oral image data and oral environment data;

[0042] S2: Based on the oral image data, the improved MobileNetV3 network is used to optimize the micro-sampling of oral dynamic parameters to obtain microscopic images of tooth edges;

[0043] S3: Analyze the physical field in the oral cavity based on oral environment data, establish a lightweight physical optimization model for multi-physics field coupling, and use an iterative solution method to solve the multi-physics field coupling model to obtain the optimal physical characteristic parameters;

[0044] S4: Construct an initial implicit surface model of the jaw pad based on the microscopic image of the tooth edge, optimize the initial implicit surface model of the jaw pad based on the optimal physical property parameters to obtain an optimized jaw pad three-dimensional model, and output a jaw pad three-dimensional image based on the optimized jaw pad three-dimensional model.

[0045] The present invention has the following beneficial effects:

[0046] In this invention, firstly, by improving the MobileNetV3 network through asymmetric convolution kernels, channel attention mechanism and deformable convolution blocks, the extraction accuracy of tooth edge micro features (such as 0.2mm pit and fissure depth and 45° cusp inclination) is significantly improved, so that the error between the jaw pad edge and the tooth anatomical structure is reduced, and the fit and comfort are improved;

[0047] Secondly, through multi-physics field coupling optimization, a lightweight coupling model of the chemical field (pH value-corrosion rate), temperature field (thermal expansion-stress), and mechanical field (bite force-deformation) was constructed. By iteratively solving the objective function, regional optimization of the material porosity was achieved. This ensured that the jaw pad had better corrosion resistance in high-acid areas and enhanced mechanical properties in areas with concentrated stress. This comprehensively improved the overall performance of the jaw pad in complex oral environments and extended its service life.

[0048] Finally, a customized three-dimensional model is generated based on the individual's oral dynamic parameters (such as the bite force distribution of unilateral chewing and the pH value of high acid secretion). The entire process from data acquisition (wearable sensors obtain parameters in real time) to three-dimensional modeling (implicit surface + MarchingCubes algorithm) is digitized, and the modeling time is reduced by reducing the number of trial wear and adjustment times. It integrates computer vision, computational physics, geometric modeling, etc. to achieve a highly integrated system. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a system block diagram of a jaw pad three-dimensional image generation system and generation method proposed in the present invention.

[0050] Figure 2 This is a method step diagram of a jaw pad three-dimensional image generation system and generation method proposed by the present invention. DETAILED DESCRIPTION

[0051] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0052] Example 1

[0053] like Figure 1 As shown, the present invention proposes a jaw pad three-dimensional image generation system, comprising:

[0054] Data acquisition module: collects oral dynamic parameters through a wearable intraoral patch-type multimodal sensor, wherein the oral dynamic parameters include oral image data F and oral environment data H;

[0055] Oral image data F acquisition process:

[0056] The wearable intraoral patch-type microsensor uses complementary metal oxide semiconductor technology. Its pixel unit structure is a combination of a photodiode and an active pixel circuit. When the photodiode receives photons, it generates photogenerated carriers (electron-hole pairs). The number of carriers is proportional to the light intensity. The active pixel circuit amplifies the photogenerated carriers and converts them into voltage signals. The analog voltage signal is then converted into oral image data through an analog-to-digital conversion circuit.

[0057] Oral environment data H collection process:

[0058] Oral environment data includes saliva pH value data, temperature distribution data, and bite force distribution data, among which:

[0059] Saliva pH data is based on the principle of ion-selective electrodes. The sensitive membrane is made of a specially synthesized polymer material that is highly selective for hydrogen ions. When hydrogen ions in saliva come into contact with the sensitive membrane, a stable electrochemical potential difference is quickly established on both sides of the membrane. The pH data measured at different locations is converted into distribution information within the oral space. By integrating pH data into the oral dynamic parameter system, it assists in analyzing the impact of the oral chemical environment on jaw pad design.

[0060] Temperature distribution data is collected through a micro-thermal sensor. The resistance of the thermistor changes with temperature in an accurate exponential relationship, which affects the thermal expansion of the jaw pad material. By obtaining temperature distribution data, the stress condition of the jaw pad is determined based on the thermal-bite force.

[0061] The bite force distribution data is collected using the piezoresistive principle, with a Wheatstone bridge structure composed of silicon-based piezoresistors. When subjected to bite force, the piezoresistor deforms, and its resistance value changes precisely according to the piezoresistive effect. Analysis of the bite force distribution data can simulate the patient's bite process and habits, and adjust the morphological structure of the jaw pad. For example, for patients with unilateral bite, targeted adjustments can be made to the jaw pad design to improve the bite function, enhance chewing efficiency, and enhance the structural coupling relationship of comfort.

[0062] Microscopic feature module: Based on oral image data, the improved MobileNetV3 network is used to optimize the microscopic sampling of oral dynamic parameters to obtain microscopic images of tooth edges;

[0063] Use the oral image data F as the input image data of the improved MobileNetV3 network;

[0064] The original MobileNetV3 network has limitations when processing microscopic images of tooth edges in oral images. We optimized the initial convolutional layer and depthwise separable convolutional layer of the original MobileNetV3 network and added a deformable convolution block after the last convolution layer of the MobileNetV3 network.

[0065] The initial convolutional layer of the original MobileNetV3 network is adjusted by using an asymmetric convolution kernel combination, replacing the 3×3 convolution kernel with 3×1 and 1×3 convolution kernels. This reduces the amount of computation while increasing the ability to capture image edge features.

[0066] For the depth-wise separable convolutional layer, a channel attention mechanism is introduced to obtain the global information of the feature map of each channel based on the channel attention mechanism;

[0067] Obtain the global information of the feature map of each channel through the global average pooling of the channel attention mechanism;

[0068] Then, two fully connected layers are used to obtain the channel weights. Finally, the channel weights are multiplied by the original feature map (oral image data F) to enhance the focus on important channel features and complete the optimization of the depthwise separable convolution layer. Its mathematical expression is:

[0069]

[0070] Among them, F is the input oral image data, GAP is the global average pooling, MLP is the multi-layer perceptron (consisting of two fully connected layers), is the Sigmoid activation function, is the channel attention weight;

[0071] Add a deformable convolution block after the last convolution layer of the MobileNetV3 network. The deformable convolution block learns the offset through an additional convolution layer. Suppose the coordinates of the original convolution kernel sampling point are (n, m). By learning the optimal offset The original convolution kernel sampling point coordinates are optimized, and the new sampling point coordinates become ;

[0072] Among them, the optimal offset The learning process is:

[0073] The offset Restricted to a specific range, such as ;

[0074] Get a dynamic adjustment function , the original convolution kernel sampling point coordinates The eigenvalue at is divided by the maximum value for normalization, then input into the multi-layer perceptron, and finally activated by the Sigmoid activation function to obtain the dynamic adjustment coefficient, which is expressed as follows:

[0075]

[0076] in, Represents the coordinates in the input feature map The eigenvalue at (feature information obtained after previous convolution operations), is the maximum value in the feature map G;

[0077] Specifically, dynamically adjust the function It will play a regulatory role in subsequent calculations, dynamically adjusting the offset calculation process according to the characteristics of different positions in the feature map;

[0078] The offset is obtained through a two-stage recursive learning process. The specific process is:

[0079] The first stage is the rough offset generation, which uses 3×3 depth-separable convolution to make a preliminary prediction of the offset to obtain the initial offset. The depth-wise separable convolution decomposes the standard convolution into depth-wise convolution and point-wise convolution, while reducing the amount of calculation, and performs preliminary feature extraction and processing on the input feature map, thereby preliminarily predicting the initial offset. ,This step obtains a relatively rough offset estimate;

[0080] The second stage fine-tunes the network calculation, with the initial offset And residual features as input, the correction offset is calculated through the dilated gated convolution (DilatedGatedConv) ;

[0081] The receptive field of the convolution kernel can be expanded without increasing the number of parameters through the dilated convolution, and the gating mechanism can filter and control the information. By combining the two, the network can And the residual features of the input feature map (that is, the difference from the standard features), further mining the finer structural information in the feature map, and calculating the corrected offset used to correct the preliminary offset ;

[0082] Finally based on the initial offset and corrected offset By dynamically adjusting parameters Determining the optimal offset , the formula is:

[0083]

[0084] in, Function is used to correct the amount Perform nonlinear transformation, compress its value to the (-1,1) interval, and dynamically adjust the parameters Used to control the degree of influence of the correction amount on the final offset;

[0085] As the training progresses, the influence of the correction amount is gradually reduced to make the adjustment of the offset more stable, and the offset used to adjust the convolution kernel sampling position is finally obtained. , to achieve more accurate capture of microscopic images of tooth edges (such as cusp inclination, pit and fissure morphology, etc.);

[0086] Specifically, the deformable convolution block dynamically learns the offset through an additional convolution layer and can adaptively adjust the convolution kernel sampling position. In this process, the deformable convolution block is based on the input feature map F and adjusts the sampling position according to the learned optimal offset. The output feature map is calculated, allowing the convolution kernel to break through the limitations of traditional fixed sampling positions and actively obtain the key positions of microscopic features of tooth edges, such as cusp inclination, and:

[0087] The convolution kernel can move to the edge of the tooth cusp according to the learned offset, accurately collecting feature information related to the tooth cusp inclination. For pit and fissure morphology, it can penetrate deep into the pit and fissure to capture microscopic feature information such as the direction and depth of the pit and fissure.

[0088] Finally, an improved MobileNetV3 network was obtained by optimizing the initial convolutional layer and depthwise separable convolutional layer of the original MobileNetV3 network and adding a deformable convolution block after the last convolution layer of the MobileNetV3 network. Oral image data was input into the improved MobileNetV3 network to obtain microscopic images of tooth edges.

[0089] For the input oral image data F, the output of the improved MobileNetV3 network is expressed as:

[0090]

[0091] in, is the output feature map coordinate, is the convolution kernel weight, To improve the output of the MobileNetV3 network, i.e. the microscopic image of the tooth edge;

[0092] Specifically, the role of overall network optimization and network infrastructure optimization (using a hybrid asymmetric convolution kernel structure and introducing dilated convolution technology) has been achieved. The input image has been subjected to multi-level and multi-dimensional feature extraction and enhancement, so that the feature map F contains rich information related to the tooth edge. On this basis, the deformable convolution block further focuses and extracts the microscopic features of the tooth edge by dynamically adjusting the offset, and filters and enhances the information related to the microscopic features of the tooth edge from the feature map F. The final output feature map This also presents the best microscopic image characteristics of the tooth edge.

[0093] Physical analysis module: Analyzes the multi-physical fields in the oral cavity based on oral environment data, establishes a multi-physical field coupling objective function model, and uses an iterative solution method to solve the objective function model to obtain the optimal physical characteristic parameters;

[0094] Among them, chemical fields, temperature fields, and mechanical fields are generally collectively referred to as multi-physics fields. Chemical fields involve physical chemistry such as chemical reaction kinetics and chemical thermodynamics. Temperature fields are based on physical theories such as heat transfer. Mechanical fields are based on physical knowledge such as elasticity and solid mechanics. They all belong to the fields of different branches of the field of physics.

[0095] By inputting saliva pH value data, temperature distribution data and bite force distribution data in the oral environment data as input physical parameters, and constructing a multi-physical field based on the oral environment data, the multi-physical field in the oral cavity is analyzed;

[0096] The process of analyzing the multi-physics field in the oral cavity is as follows:

[0097] The physical fields in the oral cavity include chemical fields (based on saliva pH data), temperature fields (based on temperature distribution data), and mechanical fields (based on bite force distribution data). The energy functionals of all fields in the multi-physics field are obtained. ;

[0098] Specifically, the process of obtaining the chemical field energy functional is as follows:

[0099] The pH value of saliva reflects the concentration of acidic substances in the mouth, which directly affects the degree of corrosion of teeth and jaw pad materials. Using the three-dimensional anatomical model of the mouth as the basic framework, the pH value data measured at different positions are converted into continuous distribution information in the oral space through a spatial interpolation algorithm to obtain the chemical field energy functional. ;

[0100] The process of determining the temperature field energy functional is:

[0101] The temperature data collected at different time points are sorted out according to the temperature data to form a time-space temperature data set, and a dynamic curve of temperature change over time is obtained. The thermal expansion coefficient of the jaw pad material is combined with the temperature dynamic curve to obtain the thermal expansion amount of the jaw pad material at different temperatures. By analyzing the distribution of thermal expansion in different parts of the jaw pad, the effect of temperature on the thermal expansion of the jaw pad material is evaluated, and the energy functional of the temperature field is determined. ;

[0102] The process of determining the energy functional of the mechanical field is:

[0103] Based on the bite contact point position, bite force magnitude and direction determined by the bite force distribution data, and combined with the oral anatomical structure (such as the geometry and mechanical properties of the teeth and jaws), the finite element analysis method is used to establish an oral mechanical model, simulate the transmission and distribution of bite force in the oral cavity, and obtain the mechanical field energy functional. ;

[0104] The objective function of the lightweight physical optimization model is constructed based on the energy functional of multi-physics fields. The objective function is expressed as:

[0105]

[0106] in, 、 、 is the weight coefficient, which is determined according to the degree of influence of different physical fields on the performance of the jaw pad in reality;

[0107] The objective function is optimized and solved by adopting an iterative solution method;

[0108] In the solution process, the current physical parameters are substituted into the energy functional of each physical field, and the parameter weight coefficients in the model are adjusted to minimize the objective function J. The physical parameters corresponding to the minimum objective function J are the optimal solution under the multi-physical field coupling condition, that is, the optimal physical characteristic parameters, which are expressed as ,in, represents the physical parameters corresponding to the chemical field, represents the physical parameters corresponding to the temperature field, Represents the physical parameters corresponding to the mechanical field;

[0109] Optimal physical property parameters It reflects the parameter range in which the jaw pad has the best physical properties under the interaction of chemical, temperature and mechanical fields.

[0110] 3D construction module: constructs an initial implicit surface model of the jaw pad based on the microscopic image of the tooth edge, optimizes the initial implicit surface model of the jaw pad based on the optimal physical characteristic parameters to obtain an optimized 3D model of the jaw pad, and outputs a 3D image of the jaw pad based on the optimized 3D model of the jaw pad;

[0111] The process of constructing the initial implicit surface model of the jaw pad based on the microscopic image of the tooth edge is as follows:

[0112] The optimal control point coordinates are obtained based on the microscopic image of the tooth edge, and the control weights of the initial jaw pad implicit surface model are obtained based on the optimization algorithm;

[0113] Optimal control point coordinates This is achieved by extracting key feature points from microscopic images of tooth edges;

[0114] For example, the apex of the tooth tip, the starting and ending points of the pit and fissure, etc. These points can accurately reflect the morphological characteristics of the tooth edge;

[0115] Obtain the control weight of the initial jaw pad implicit surface model through optimization algorithm , so that the implicit surface approaches the actual shape of the tooth edge. In the calculation, a control point control amount b is preset to optimize the shape and position of the surface, such as:

[0116] After calculation, the coordinates of the three best control points mentioned above are 、 、 The control weights are 、 、 For the control point control amount b, set b = 0 initially. If the generated implicit surface is found to be biased to the upper left, it needs to be translated to the lower right to better fit the teeth. This can be achieved by adjusting b. For example, b can be adjusted to (1, -1) (here, it represents the control point control amount in the horizontal and vertical coordinate directions in two dimensions), so that the surface is translated 1 unit to the right and 1 unit downward to achieve the best fit with the actual shape of the tooth edge.

[0117] Substitute the determined optimal control point coordinates and control weights into the radial basis function formula, define a scalar function V, determine the surface of the model when the scalar function is 0 (V = 0), and use the radial basis function to construct the implicit surface to generate the initial jaw pad implicit surface model. The formula is expressed as:

[0118]

[0119] in, are the coordinates of the spatial point, is the control weight, is the radial basis function, is the coordinate of the optimal control point, b is the control amount of the control point, is the index of the optimal control point coordinate quantity;

[0120] The process of optimizing the initial jaw pad implicit surface model based on the optimal physical characteristic parameters is as follows:

[0121] According to the porosity distribution information of different regions in the optimized physical property parameters, the internal scalar function of the implicit surface model is adjusted;

[0122] Specifically, in the process of constructing the implicit surface model of the jaw pad, the porosity distribution information of different regions in the physical property parameters is optimized. The porosity distribution reflects the density of the jaw pad in different parts and has a significant impact on its physical properties (chemical, thermal, and mechanical).

[0123] Define the porosity distribution matrix ,in, Represents the coordinates of a spatial point, with a size of (100×100×100) (representing voxel resolution), and a value range of ,in Indicates dense area, Indicates high porosity area;

[0124] At the same time, determine the initial implicit surface scalar field , whose domain is , negative values ​​indicate the inside of the surface, positive values ​​indicate the outside of the surface, initial isosurface , represents the target surface boundary;

[0125] Use piecewise linear interpolation to adjust the initial implicit surface scalar field And set physical constraints to obtain the adjusted scalar;

[0126] Set physical constraints: When , in the high porosity area, let Increase 0.2 to expand the influence of high porosity area; when , in the low porosity (dense) area, let Reduce by 0.1 to achieve shrinkage adjustment of dense areas;

[0127] Constraint calculation formula:

[0128]

[0129] in, is the base porosity, , is the adjustment coefficient, is the adjusted scalar;

[0130] Through constrained calculation, the adjusted scalar field is obtained ;

[0131] By finding the adjusted scalar to 0 The surface of the jaw pad 3D model is determined by the isosurface of the voxel. The marching cubes algorithm is used to convert the scalar field data into a surface model represented by a triangular mesh. The algorithm analyzes each voxel unit and determines its intersection with the isosurface based on the scalar value of the voxel vertex, thereby generating corresponding triangular facets to form a complete surface, that is, to optimize the jaw pad 3D model.

[0132] Specifically, the optimized jaw pad 3D model is essentially a mesh structure composed of a large number of triangular facets. Each facet is defined by three vertex coordinates and a normal vector. These facets are interconnected to form a jaw pad shape that reflects the adjustment of physical parameters such as porosity. For example, high-porosity areas correspond to locally looser structural features of the model, while low-porosity areas are denser.

[0133] Convert the generated triangular mesh model data into a common 3D file format (such as STL format). The STL file stores the vertex coordinates and normal vectors of all triangular facets in text or binary form. Finally, in the modeling software or programming environment, use the export function to save the model as an STL or other format file to output the jaw pad 3D image.

[0134] For example:

[0135] An initial implicit surface model is obtained, and its radial basis function parameters are:

[0136] Optimal control point coordinates: cusp (5,5,0), pit and fissure starting point (3,2,0), and end point (7,3,0);

[0137] Control weight: , , ;

[0138] Control point control amount b=(1,−1,0) (two-dimensional adjustment example);

[0139] Gaussian function width σ=1;

[0140] Generate the surface equation:

[0141] According to the porosity distribution information of different regions in the optimized physical property parameters, the porosity distribution matrix is:

[0142] Size: 100×100×100 voxels;

[0143] voxel, molar region (dense area) θ = 0.2;

[0144] Anterior tooth area (high porosity area) θ = 0.7;

[0145] Scalar field adjustment:

[0146] Base porosity =0.3, adjustment coefficient α=0.3;

[0147] Constraints:

[0148] High porosity region (θ>0.7): ;

[0149] Dense region (θ<0.3): ;

[0150] 3D model construction:

[0151] Use the MarchingCubes algorithm to generate a triangular mesh containing 5000 facets and export it to STL format;

[0152] Output of jaw pad 3D image;

[0153] Visualization results: The STL file was imported into Blender and rendered to display the three-dimensional image of the jaw pad. The molar area (dense area) showed a solid structure, while the anterior area (high porosity area) showed honeycomb pores.

[0154] Key dimensions (examples):

[0155] Overall dimensions of the jaw pad: 30mm×25mm×3mm, thickness 2.5mm in the molar area, 1.5mm in the anterior area, pore diameter 0.5mm (high porosity area).

[0156] Example 2

[0157] like Figure 2 As shown, a method for generating a three-dimensional image of a jaw pad includes the following steps:

[0158] S1: Collecting oral dynamic parameters through a wearable intraoral patch sensor, wherein the oral dynamic parameters include oral image data and oral environment data;

[0159] S2: Based on the oral image data, the improved MobileNetV3 network is used to optimize the micro-sampling of oral dynamic parameters to obtain microscopic images of tooth edges;

[0160] S3: Analyze the physical field in the oral cavity based on oral environment data, establish a lightweight physical optimization model for multi-physics field coupling, and use an iterative solution method to solve the multi-physics field coupling model to obtain the optimal physical characteristic parameters;

[0161] S4: Construct an initial implicit surface model of the jaw pad based on the microscopic image of the tooth edge, optimize the initial implicit surface model of the jaw pad based on the optimal physical property parameters to obtain an optimized jaw pad three-dimensional model, and output a jaw pad three-dimensional image based on the optimized jaw pad three-dimensional model.

[0162] In the application, several formulas involved are calculated by taking their numerical values ​​after removing the dimensions, and the formulas are established by collecting a large amount of data and performing software simulation to obtain a formula for the most recent real situation. Some coefficients or weights in the formulas are set by technical personnel in this field according to actual conditions, so they will not be elaborated here.

[0163] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0164] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A jaw pad three-dimensional image generation system, characterized in that: include: Data acquisition module: collects oral dynamic parameters through a wearable intraoral patch sensor, the oral dynamic parameters including oral image data and oral environment data; Microscopic feature module: Based on oral image data, the improved MobileNetV3 network is used to optimize the microscopic sampling of oral dynamic parameters to obtain microscopic images of tooth edges; The improved MobileNetV3 network is optimized by optimizing the initial convolution layer and the depth-separable convolution layer of the original MobileNetV3 network and adding a deformable convolution block after the last convolution layer of the MobileNetV3 network. Physical Analysis Module: Analyzes the physical field in the oral cavity based on oral environment data, establishes a lightweight physical optimization model for multi-physical field coupling, and uses an iterative solution method to solve the multi-physical field coupling model to obtain the optimal physical characteristic parameters; The process of obtaining the optimal physical property parameters is as follows: The multi-physical fields include chemical fields, temperature fields and mechanical fields, and energy functionals of all fields in the multi-physical fields are obtained; Construct the objective function of the lightweight physical optimization model based on the energy functional of all fields in the multi-physics field; The objective function is optimized and solved by an iterative solution method to obtain the optimal physical characteristic parameters; Three-dimensional construction module: construct the initial implicit surface model of the jaw pad based on the microscopic image of the tooth edge, optimize the initial implicit surface model of the jaw pad based on the optimal physical property parameters to obtain the optimized jaw pad three-dimensional model, and output the jaw pad three-dimensional image based on the optimized jaw pad three-dimensional model.

2. The jaw pad three-dimensional image generation system according to claim 1, characterized in that: The oral dynamic parameters include oral image data and oral environment data; The oral environment data includes saliva pH value data, temperature distribution data and bite force distribution data.

3. The jaw pad three-dimensional image generation system according to claim 2, characterized in that: The initial convolutional layer optimization process is: The initial convolutional layer of the original MobileNetV3 network is adjusted by using an asymmetric convolution kernel combination, replacing the 3×3 convolution kernel with 3×1 and 1×3 convolution kernels; The optimization process of the depthwise separable convolutional layer is as follows: For the depthwise separable convolutional layer, a channel attention mechanism is introduced to obtain the global information of the feature map of each channel based on the channel attention mechanism. Then, the channel weight is obtained through two fully connected layers. The channel weight is multiplied with the original feature map to complete the optimization of the depthwise separable convolutional layer.

4. The jaw pad three-dimensional image generation system according to claim 3, characterized in that: The process of adding a deformable convolution block after the last convolution layer of the MobileNetV3 network is as follows: The deformable convolution block learns the offset through an additional convolution layer, and optimizes the original convolution kernel sampling point coordinates by learning the optimal offset to obtain the new sampling point coordinates.

5. The jaw pad three-dimensional image generation system according to claim 4, characterized in that: The optimal offset learning process is: The offset is limited to a preset range, a dynamic adjustment function is obtained, the eigenvalue at the original convolution kernel sampling point coordinate is divided by the maximum value for normalization, and then input into the multi-layer perceptron, activated by the Sigmoid activation function, and the dynamic adjustment coefficient is obtained; Obtain the offset through two-stage recursive learning; The two-stage recursive learning includes a first stage of coarse migration and a second stage of fine-tuning network calculation; The initial offset is generated based on the coarse offset in the first stage, and the correction value is obtained through the dilated gated convolution based on the fine-tuning network in the second stage, taking the initial offset and residual features as input; The optimal offset is determined by dynamically adjusting parameters based on the initial offset and the corrected offset.

6. The jaw pad three-dimensional image generation system according to claim 1, characterized in that: The process of constructing the initial implicit surface model of the jaw pad based on the microscopic image of the tooth edge is as follows: Obtain the coordinates of the optimal control points based on the microscopic image of the tooth edge; Based on the optimization algorithm and presetting a control point control amount, the control weight of the initial jaw pad implicit surface model is obtained; The determined optimal control point coordinates and control weights are substituted into the radial basis function formula, and a scalar function is defined. When the scalar function is 0, the surface of the model is determined, and the radial basis function is used to construct the implicit surface to generate the initial jaw pad implicit surface model.

7. The jaw pad three-dimensional image generation system according to claim 6, characterized in that: The process of optimizing the initial jaw pad implicit surface model based on the optimal physical characteristic parameters is as follows: According to the porosity distribution information of different regions in the optimized physical property parameters, the internal scalar function of the implicit surface model is adjusted; Define the porosity distribution matrix and determine an initial implicit surface scalar field. Use piecewise linear interpolation to adjust the initial implicit surface scalar field and set physical constraints to obtain the adjusted scalar. The surface of the jaw pad three-dimensional model is determined by finding the isosurface with the adjusted scalar value equal to 0. The marching cubes algorithm is used to convert the scalar field data into a surface model represented by a triangular mesh to obtain the optimized jaw pad three-dimensional model.

8. The jaw pad three-dimensional image generation system according to claim 7, characterized in that: The physical constraints include: When the porosity distribution matrix is ​​greater than 0.7 and is in the high porosity area, the adjusted scalar increases by 0.2; When the porosity distribution matrix is ​​less than 0.3 and is in the low porosity area, the adjusted scalar increases by 0.

1.

9. A method for generating a three-dimensional image of a jaw pad, which is implemented by the generation system according to any one of claims 1 to 8, characterized in that: The method steps include: S1: Collecting oral dynamic parameters through a wearable intraoral patch sensor, wherein the oral dynamic parameters include oral image data and oral environment data; S2: Based on the oral image data, the improved MobileNetV3 network is used to optimize the micro-sampling of oral dynamic parameters to obtain microscopic images of tooth edges; S3: Analyze the physical field in the oral cavity based on oral environment data, establish a lightweight physical optimization model for multi-physics field coupling, and use an iterative solution method to solve the multi-physics field coupling model to obtain the optimal physical characteristic parameters; S4: Construct an initial implicit surface model of the jaw pad based on the microscopic image of the tooth edge, optimize the initial implicit surface model of the jaw pad based on the optimal physical property parameters to obtain an optimized jaw pad three-dimensional model, and output a jaw pad three-dimensional image based on the optimized jaw pad three-dimensional model.

Images