Remote oral rehabilitation diagnosis and treatment system and method based on 5G network

Abstract

The invention relates to the technical field of communication protocols, in particular to a remote oral rehabilitation diagnosis and treatment system and method based on a 5G network, and the system comprises an image collection module, a feature analysis module, a strategy generation module, a trajectory prediction module and a scheme feedback module. According to the method, the oral cavity three-dimensional scanning data stream is transmitted in real time, the Gaussian filtering algorithm is adopted to filter noise, the signal-to-noise ratio of original data is improved, environment interference and equipment errors are reduced, contour boundaries are extracted based on edge detection, edge acutance and curvature parameters are calculated, lesion geometric features are quantified, and the irregular defect recognition precision is improved; the convolutional neural network fuses defect area, depth value and damage index, generates repair precision coefficient, dynamically maps multi-dimensional index and repair strategy, improves scheme adaptability, predicts defect change trend by LSTM time sequence, pre-judges focus development trajectory, dynamically adjusts repair parameters, avoids restoration failure, forms a closed loop logic chain in each link, and improves restoration efficiency. And the data processing and decision-making efficiency is optimized.

Description

Technical Field

[0001] This invention relates to the field of communication protocol technology, and in particular to a remote dental restoration and treatment system and method based on a 5G network. Background Technology

[0002] The field of communication protocol technology encompasses a series of rules, standards, and control methods established in communication networks to achieve data transmission, resource sharing, and terminal interoperability. The core elements of communication protocols include data formats, signal exchange sequences, error detection and correction mechanisms, session control, synchronization mechanisms, and security measures. Overall, the field of communication protocol technology mainly covers protocol standards at each layer: physical layer, data link layer, network layer, transport layer, and application layer. It achieves end-to-end data transmission and information exchange functions through combinations of different protocol stacks, involving various communication environments including wireless communication, wired communication, satellite communication, and Internet protocols. It is widely used in various information systems and smart terminal devices, supporting seamless communication and collaborative operation between networks.

[0003] The remote oral restoration diagnosis and treatment system refers to a system based on the 5G communication protocol that connects oral restoration experts and patients through a remote medical platform, enabling interactive diagnosis and treatment information, real-time transmission of image data, and surgical operation guidance. The patent addresses the following technical aspects: utilizing the ultra-low latency and high reliability bandwidth of 5G to remotely upload three-dimensional imaging data of the patient's oral cavity in real time; the expert's end performs oral restoration diagnosis and treatment plan formulation based on the received data, and sends control commands to the local restoration equipment via a remote control terminal to guide the local equipment in precise restoration operations; the system employs a standardized communication protocol to ensure a unified format for diagnosis and treatment data, and the transmission process uses network encryption protocols to ensure data security, guaranteeing accurate and timely information transmission during remote diagnosis and treatment.

[0004] Current technologies rely on 5G to transmit raw scan data, but lack integrated preprocessing mechanisms. Noise and artifacts are not filtered out, leading to potential distortion in remotely received 3D images and impacting diagnostic accuracy. Existing systems use fixed thresholds to segment lesion areas, resulting in insufficient accuracy in recognizing complex morphological edges and prone to feature extraction bias, leading to discrepancies between the restoration plan and the actual lesion's geometric parameters. Traditional methods rely on expert experience to formulate static restoration strategies, lacking the ability to model and predict the dynamic development of lesions. Restoration parameters cannot be adjusted as the disease progresses, potentially causing a long-term decline in the compatibility between the restoration and dental tissue. Existing data encryption mechanisms only target the transport layer, failing to incorporate 3D image data features into the encryption algorithm design, posing a risk of partial data decryption. Insufficient cross-layer collaboration in existing protocol stacks leads to poor synchronization between data flow and control commands in network fluctuation scenarios, potentially causing device response delays or operational interruptions. For example, unoptimized protocol compatibility can result in packet loss during high-concurrency transmission, affecting the continuity of restoration operations. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing a remote oral restoration and treatment system and method based on a 5G network.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A remote oral restoration and treatment system based on a 5G network includes: The image acquisition module is used to receive oral cavity 3D scan data stream through 5G real-time transmission bandwidth, call Gaussian filtering algorithm to filter out noise, extract the coordinate set of lesion area, generate 3D reconstruction data, and transmit it to the feature analysis module. The feature analysis module is used to receive the three-dimensional reconstruction data, locate the contour boundary based on the edge detection algorithm, calculate the edge sharpness value and curvature parameter, generate lesion feature parameters, and transmit the lesion feature parameters and the edge sharpness value to the strategy generation module; The strategy generation module is used to receive the lesion feature parameters and the edge sharpness value, input them into the convolutional neural network model for lesion classification, output the defect area, depth value, and structural damage index, generate repair accuracy coefficients based on the product, map them to repair technology numbers, generate a dynamic repair strategy table, and pass the dynamic repair strategy table and the depth value to the trajectory prediction module. The trajectory prediction module is used to receive the dynamic repair strategy table and the depth value, call the LSTM time series model to model the depth change rate, combine the defect area growth rate to generate the evolution trajectory prediction vector, output the update strategy instruction, and pass it to the scheme feedback module.

[0007] As a further aspect of the present invention, the three-dimensional reconstruction data includes a set of spatial coordinates of the lesion area, surface point cloud data, and texture feature data after noise reduction. The lesion feature parameters include edge sharpness distribution, curvature change matrix, and morphological complexity index. The dynamic repair strategy table includes repair technology number, repair priority, and estimated repair cycle. The evolution trajectory prediction vector includes depth change rate sequence, area expansion rate curve, and evolution time node marker.

[0008] As a further aspect of the present invention, the image acquisition module includes: The 5G transmission submodule acquires real-time oral cavity 3D scan data stream, establishes a communication link through the 5G transmission protocol, divides the data stream into independent data packets and adds timestamps, and completes data reassembly according to the transmission sequence number to generate a transmission data stream; The filtering and noise reduction submodule calls the Gaussian filtering algorithm, constructs a three-dimensional spatial coordinate system based on the transmitted data stream, performs neighborhood weighted average calculation on each coordinate point, sets the standard deviation parameter to 0.8, eliminates outlier coordinate offsets, and generates filtered data. The lesion coordinate generation submodule performs gradient calculation based on the filtered data, performs differential operation on the gray values ​​of adjacent coordinate points, selects regions with gray value differences exceeding 0.12 as boundary judgment benchmarks, extracts the vertex coordinate set of continuous closed regions, and generates three-dimensional reconstruction data.

[0009] As a further aspect of the present invention, the feature analysis module includes: The edge detection submodule calls the edge detection algorithm based on the 3D reconstruction data, calculates the gradient magnitude and direction for each coordinate point, sets high and low thresholds of 0.3 and 0.1 respectively, selects points with gradient magnitudes higher than the high threshold as boundary points, connects adjacent points to form a closed contour, and generates a boundary coordinate set; The sharpness curvature calculation submodule calls the boundary coordinate set, performs second derivative calculation on the coordinates in the neighborhood of each boundary point, extracts the maximum absolute value of curvature as the curvature parameter, normalizes the gradient magnitude, calculates the average value of the gradient direction change rate of adjacent points, and generates the edge sharpness value and curvature parameter. The feature parameter generation submodule performs a weighted fusion calculation on regions with sharpness values ​​exceeding 0.25 based on the edge sharpness value and the curvature parameter. The curvature weight coefficient is set to 0.6 and the sharpness weight coefficient is set to 0.4. The module integrates the point parameters within the region to generate the average value and standard deviation, thereby obtaining the lesion feature parameters.

[0010] As a further aspect of the present invention, the strategy generation module includes: The classification calculation submodule, based on the lesion feature parameters and the edge sharpness value, inputs the parameter matrix into the fully connected layer of the convolutional neural network, extracts multi-dimensional feature vectors, performs linear superposition operation of feature vectors and weight matrix, and outputs probability distribution through activation function to obtain the defect area, depth value and structural damage index; The coefficient generation submodule standardizes the defect area per unit area, converts the depth value into a relative depth ratio, performs a product operation on the three factors in conjunction with the structural damage index, introduces a weight correction factor to adjust the product result, and generates a repair accuracy coefficient. The strategy mapping submodule matches the corresponding technical number in a preset mapping table according to the numerical range of the repair accuracy coefficient, divides the depth value into levels with an accuracy of 0.1mm, integrates the number and depth level, and establishes a dynamic repair strategy table.

[0011] As a further aspect of the present invention, the trajectory prediction module includes: The deep modeling submodule calls the depth value sequence in the dynamic repair strategy table, inputs it into the LSTM time series model, calculates the time weight through the forget gate and the input gate, performs sliding window segmentation on the depth value, outputs the hidden layer state vector, and generates the depth change rate. Based on the depth change rate, the trajectory generation submodule calculates the difference between adjacent time points of the defect area, fits the area growth curve through linear regression, extracts the slope parameter as the growth rate, and concatenates the change rate and the growth rate into a matrix to generate an evolution trajectory prediction vector. The strategy update submodule performs a matching operation between the vector and the preset strategy rules based on the dimensional characteristics of the predicted vector of the evolution trajectory, filters candidate strategies that meet the error threshold, marks the optimal strategy as the priority, and outputs the update strategy instruction.

[0012] As a further aspect of the present invention, the system further includes: The solution feedback module is used to receive the update strategy instruction, match the repair material database, extract the compressive strength and adhesion coefficient, calculate the fit by combining real-time bite force monitoring data, generate a repair solution, and synchronize the repair solution to the wearable sensor at the patient end. The repair solution includes material compatibility parameters, compressive strength values, and adhesion performance indicators.

[0013] As a further aspect of the present invention, the solution feedback module includes: The material matching submodule calls the repair technology number in the update strategy instruction, traverses the primary key of the repair material database based on the number hash value, and after verifying the validity of the entry, extracts the compressive strength and adhesion coefficient fields to generate a material parameter set. The fit calculation submodule obtains the compressive strength of the material parameter set as the numerator and the real-time bite force monitoring data as the denominator to perform a division operation, extracts the product term of the adhesion coefficient and the rate of change of bite force, standardizes the two results and superimposes the weight coefficient to generate the fit coefficient. The solution generation submodule divides the fit level range according to the numerical distribution of the fit coefficient, selects material combinations that meet the range requirements for both compressive strength and adhesion coefficient, aligns the biting force data and material parameters according to the time series to generate an index matrix, and establishes a repair solution.

[0014] A remote dental restoration treatment method based on a 5G network, wherein the method is executed based on the aforementioned remote dental restoration treatment system based on a 5G network, and includes the following steps: S1: Receives real-time oral cavity 3D scan data stream via 5G communication protocol, calls Gaussian filtering algorithm to filter out noise from the data stream, extracts the coordinate set of the lesion area, and generates 3D reconstruction data; S2: Input the three-dimensional reconstruction data into the edge detection algorithm to locate the contour boundary, calculate the gradient magnitude of the boundary pixels to generate the edge sharpness value, and combine the curvature difference of adjacent pixels to generate lesion feature parameters. S3: Input the lesion feature parameters and the edge sharpness value into the convolutional neural network model for classification training, and output the defect area, depth value and structural damage index. Multiply the depth value and the structural damage index to generate a repair accuracy coefficient. Match the repair accuracy coefficient with a preset repair technology mapping table to generate a dynamic repair strategy table. S4: Based on the dynamic repair strategy table and the depth value, call the LSTM time series model to predict the rate of change of the depth value, combine the adjacent timestamp difference of the defect area to generate an evolution trajectory prediction vector, and output an update strategy instruction containing correction parameters. S5: According to the update strategy instruction, traverse the repair material database, match the candidate material set that meets the compressive strength and adhesion coefficient, and calculate the cosine similarity between the candidate material and the evolution trajectory prediction vector by combining the real-time collected biting force monitoring data, and generate a repair scheme.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a Gaussian filtering algorithm is used to filter out noise after real-time transmission of oral 3D scan data streams, reducing environmental interference and equipment errors during data preprocessing, improving the signal-to-noise ratio of the original data, and avoiding misjudgments caused by noise in subsequent analysis. An edge detection algorithm is used to extract contour boundaries and calculate edge sharpness and curvature parameters, quantifying the geometric features of the lesion area and enhancing the accuracy of identifying irregular defect morphologies. A convolutional neural network model is used to fuse defect area, depth, and structural damage index to generate repair accuracy coefficients, establishing a dynamic mapping relationship between multidimensional indicators and clinical repair strategies, improving the adaptability of the plan. An LSTM time series model is used to predict the evolution trend of defect depth change rate and area growth rate, predicting the lesion development trajectory and dynamically adjusting repair parameters, avoiding prosthesis failure due to uncontrollable development in traditional static plans. The technical means at each stage form a closed-loop logical chain, optimizing data processing efficiency and the scientific nature of decision-making. Attached Figure Description

[0016] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a flowchart of the image acquisition module of the present invention; Figure 3 This is a flowchart of the feature analysis module of the present invention; Figure 4 This is a flowchart of the strategy generation module of the present invention; Figure 5 This is a flowchart of the trajectory prediction module of the present invention; Figure 6 This is a flowchart of the feedback module of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0019] Please see Figure 1 This invention provides a technical solution: a remote dental restoration and treatment system based on a 5G network, comprising: The image acquisition module is used to receive oral cavity 3D scan data stream through 5G real-time transmission bandwidth, call Gaussian filtering algorithm to filter out noise, extract the coordinate set of lesion area, generate 3D reconstruction data, and transmit it to the feature analysis module. The feature analysis module is used to receive 3D reconstruction data, locate the contour boundary based on the edge detection algorithm, calculate the edge sharpness value and curvature parameter, generate lesion feature parameters, and pass the lesion feature parameters and edge sharpness value to the strategy generation module. Edge sharpness value: refers to the clarity of the edge of an image. It is usually calculated by the image gradient and measures the prominence of the edge and the steepness of the transition.

[0020] Curvature parameter: describes the degree of curvature of a curve at a certain point, and is usually used to judge changes in surface shape.

[0021] The strategy generation module receives lesion feature parameters and edge sharpness values, inputs them into a convolutional neural network model for lesion classification, and outputs defect area, depth value, and structural damage index. Based on the product, it generates repair accuracy coefficients, maps them to repair technology numbers, generates a dynamic repair strategy table, and passes the dynamic repair strategy table and depth values ​​to the trajectory prediction module. Defect area: Represents the surface area of ​​the lesion region in three-dimensional space, which is usually obtained by geometric calculation of the point cloud data of the lesion region.

[0022] Depth value: The depth of the lesion area from the surface to the substrate, which is usually measured by spatial coordinates in three-dimensional reconstruction data.

[0023] Structural damage index: A numerical value that assesses the structural integrity of the damaged area, calculated based on characteristic parameters and comparisons with adjacent areas.

[0024] The trajectory prediction module receives the dynamic repair strategy table and depth value, calls the LSTM time series model to model the depth change rate, combines the defect area growth rate to generate the evolution trajectory prediction vector, outputs the update strategy instruction, and passes it to the scheme feedback module. The solution feedback module is used to receive update strategy instructions, match the repair material database, extract compressive strength and adhesion coefficient, calculate the fit by combining real-time bite force monitoring data, generate a repair plan, and synchronize the repair plan to the wearable sensor at the patient's end.

[0025] Compressive strength: The ability of a material to resist external pressure. It is commonly measured by testing methods such as compression tests. The higher the value, the more durable the material.

[0026] Adhesion coefficient: describes the adhesion of materials to the bonding surface, and is usually an inherent value obtained through shear or tensile tests.

[0027] The 3D reconstruction data includes the spatial coordinate set of the lesion area, surface point cloud data, and texture feature data after noise reduction. The lesion feature parameters include edge sharpness distribution, curvature change matrix, and morphological complexity index. The dynamic repair strategy table includes repair technology number, repair priority, and estimated repair cycle. The evolution trajectory prediction vector includes depth change rate sequence, area expansion rate curve, and evolution time node marker. The repair scheme includes material compatibility parameters, compressive strength value, and adhesion performance index.

[0028] Edge sharpness distribution: This indicates the distribution of edge sharpness in different regions of the image, providing detailed information about the boundaries of the lesion.

[0029] Curvature variation matrix: By analyzing the point cloud of the lesion area surface, the curvature variation of different points is obtained, reflecting the complexity of the surface.

[0030] Morphological complexity index: used to quantitatively assess the complexity of lesion shape, usually calculated in combination with parameters such as boundary length and area.

[0031] Please see Figure 2 The image acquisition module includes: The 5G transmission submodule acquires real-time oral cavity 3D scan data stream, establishes a communication link through the 5G transmission protocol, divides the data stream into independent data packets and adds timestamps, and completes data reassembly according to the transmission sequence number to generate a transmission data stream; The 5G transmission submodule acquires real-time oral cavity 3D scanning data streams. First, a handheld intraoral scanner, using structured light or confocal laser technology, scans the inside of the patient's mouth (e.g., tooth number 26 to be repaired, with distal occlusal caries). The probe moves across the tooth surface, acquiring a series of point cloud data in real time, containing spatial coordinates (x, y, z) and possible color or grayscale information (c), forming raw data points. For example, point P1(10.1, 20.3, 5.5, 0.7) is acquired in the first second. Tens of thousands of points, such as P2(10.2,20.4,5.6,0.75), P3(10.1,20.5,5.5,0.72), constitute the initial data stream. Subsequently, using the 5G communication chip integrated in the scanning device or connected computing unit, a 5G communication protocol stack is initiated with the remote server or edge computing node to establish a communication link based on TCP / IP or UDP protocol, which is more suitable for real-time streaming. This data stream is dynamically divided into appropriately sized data packets; for example, each data packet is configured to carry... For 1000 data points, the first 1000 points constitute data packet #1, the next 1000 points constitute data packet #2, and so on. During packetization, a timestamp accurate to milliseconds is appended to each data packet; for example, the timestamp for data packet #1 is 2025-04-28 10:30:01.123, and the timestamp for data packet #2 is 2025-04-28 10:30:01.256. Simultaneously, each data packet is assigned a strictly increasing transmission sequence number; for example, the sequence number for data packet #1 is SN. =1, the sequence number of data packet #2 is SN=2. At the receiving end, due to network fluctuations, the data packets may arrive out of order (e.g., SN=2 is received first, then SN=1). In this case, the received data packets are sorted and buffered according to the sequence number in the data packets, and the data is recombined in the order of SN=1, SN=2, SN=3... to form a continuous sequence of data points. For example, the data points P1 to P1000 of SN=1 are connected in sequence with the data points P1001 to P2000 of SN=2 to generate a transmission data stream.

[0032] The filtering and noise reduction submodule calls the Gaussian filtering algorithm, constructs a three-dimensional spatial coordinate system based on the transmitted data stream, performs neighborhood weighted average calculation for each coordinate point, sets the standard deviation parameter to 0.8, eliminates outlier coordinate offsets, and generates filtered data. The filtering and noise reduction submodule receives the transmitted data stream, which is a sequence of three-dimensional coordinate points, such as P1(10.1,20.3,5.5), P2(10.2,20.4,5.6), ..., Pn(x_n,y_n,z_n). First, a three-dimensional spatial coordinate system is established based on the coordinate range of these points. Then, each coordinate point Pi(xi,yi,zi) in the data stream is processed, and a neighborhood is defined around it, for example, a spherical region with a radius of 1.5mm centered on Pi. All neighboring points Pj, Pk, ... within this region are found, and the Euclidean distance between Pi and each neighboring point Pk is calculated. Based on the set standard deviation parameter mm, calculate Gaussian weights ,this The values ​​are empirical values ​​derived from the analysis of noise characteristics of a large amount of oral scan data, aiming to balance denoising effect and detail preservation. For example, if point Pj is 1.0mm away from Pi, then its weight is... If the distance between point Pk and Pi is 0.5mm, then its weight... For Pi itself, the distance is 0, and the weight is... Next, calculate the new coordinates of Pi. Its calculation method is the weighted average of the coordinates of all points in the neighborhood (including Pi itself). , , For example, if the neighborhood of Pi(10.0,20.0,5.0) contains only Pj(11.0,20.2,5.1) and Pk(9.8,19.9,4.9), and the calculated distance and weights are as in the example above, then... Calculate the same By comparing the offsets of the original coordinates Pi and the new coordinates P'i The system identifies and processes points whose offsets are much greater than the average offset of the neighborhood. These points are considered outliers, and their coordinates are smoothly adjusted to eliminate drastic deviations in the coordinates of isolated points caused by scanning noise or artifacts, thereby generating filtered data.

[0033] The lesion coordinate generation submodule performs gradient calculation based on the filtered data, performs differential operation on the gray values ​​of adjacent coordinate points, selects regions with gray value differences exceeding 0.12 as boundary judgment benchmarks, extracts the vertex coordinate set of continuous closed regions, and generates three-dimensional reconstruction data.

[0034] The lesion coordinate generation submodule receives the filtered data, which is a smoother 3D point cloud. For each point in the point cloud, if the scanned data contains grayscale information (e.g., reflecting surface reflectivity or material density), the grayscale value is used directly. If not, a pseudo-grayscale value can be estimated based on local point density or surface normal changes. Then, grayscale difference operations are performed on adjacent coordinate points in space. For example, considering point Pi(xi,yi,zi) and its immediate neighbors in the x, y, and z directions, Px+(xi+1,yi,zi), Py+(xi,yi+1,zi), and Pz+(xi,yi,zi+1), the grayscale difference is calculated. , , , thus obtaining the gradient vector The gradient amplitude approximates the intensity of grayscale changes. This gradient amplitude is compared with a preset boundary judgment benchmark of 0.12. This benchmark value of 0.12 is set based on statistical analysis of grayscale / density differences between a large number of known healthy tooth tissues and carious tissues (such as enamel demineralization or dentin caries) under specific scanning modes to distinguish between normal structural changes and pathological boundaries. For example, if the grayscale value of point Pa is 0.85 and the grayscale value of its adjacent point Pb is 0.70, the difference is 0.15, which is greater than 0.12. Then the boundary between Pa and Pb is marked as part of the potential lesion boundary. If the difference between another pair of adjacent points Pc (0.80) and Pd (0.75) is 0.05, which is less than 0.12, then it is not marked as a boundary. The boundary segments formed by all adjacent point pairs with a gray-level difference exceeding 0.12 are selected. Then, region growing or contour tracing is used to connect these continuous boundary segments to find the set of boundary points that can form a closed loop. For example, starting from a boundary point that meets the conditions, the search continues to connect its neighboring boundary points that also meet the conditions until it returns to the starting point or can no longer be connected, forming a closed region. The three-dimensional coordinate set of all vertices (i.e., boundary points) that constitute these continuous closed regions is extracted, such as the set {(x1,y1,z1),(x2,y2,z2),…,(xm,ym,zm)}. These coordinate points together delineate the contour of the potential lesion area and generate three-dimensional reconstruction data.

[0035] Please see Figure 3 The feature analysis module includes: The edge detection submodule calls the edge detection algorithm based on the 3D reconstruction data, calculates the gradient magnitude and direction for each coordinate point, sets high and low thresholds of 0.3 and 0.1 respectively, selects points with gradient magnitudes higher than the high threshold as boundary points, connects adjacent points to form a closed contour, and generates a boundary coordinate set; The edge detection submodule receives the 3D reconstructed data, which is a set of vertex coordinates {(x1,y1,z1),…,(xm,ym,zm)} that defines the boundaries of potential lesions. For each coordinate point Pi in the set, the gradient information in its local neighborhood is calculated again, this time focusing more on the changes in spatial coordinates or the gradient related to gray level / density in the original filtered data, and the gradient magnitude is calculated. and gradient direction For example, by applying the 3D Sobel operator to the coordinates or gray values ​​around Pi, a gradient vector can be obtained. ,but , Set a high threshold for the direction of this vector. With low threshold These two thresholds were determined through histogram analysis of the gradient magnitude distribution at known lesion boundary points and non-boundary points. For example, thresholds were selected such that the gradient magnitude of most strong edges (such as the edge of a carious cavity) was higher than that of the boundary points. While most weak edges and noise are below Filter out all values. Points that are strong boundaries are marked as such. For example, if point Pk has a strong boundary boundary... Then Pk is a strong boundary point, if point Pl Then it is a weak boundary point (between...) and (between), if Pq If the boundary condition is not met, it is ignored. Starting from the strong boundary point, a connection is made to its neighboring points that also satisfy the boundary conditions (i.e., ...). Here, "proximity" not only refers to close spatial distance, but may also require similar gradient directions, such as an angle difference of less than 30 degrees. Continue to connect until a closed contour is formed by strong boundary points and weak boundary points connected by strong boundary points. Discard those contours that are completely composed of weak boundary points or are not closed, and generate a boundary coordinate set.

[0036] The sharpness curvature calculation submodule calls the boundary coordinate set, performs second derivative calculation on the coordinates in the neighborhood of each boundary point, extracts the maximum absolute value of curvature as the curvature parameter, normalizes the gradient magnitude, calculates the average rate of change of gradient direction of adjacent points, and generates edge sharpness value and curvature parameter. The sharpness curvature calculation submodule receives the boundary coordinate set, which is an ordered set of points P1, P2, ..., Pk forming a closed contour. For each boundary point Pi, its immediate neighbors Pi-1 and Pi+1 on the contour are examined (for the first and last points P1 and Pk, a loop connection is performed, i.e., the point before P1 is Pk, and the point after Pk is P1). The second derivative is calculated using the coordinates of these three points to approximate the local curvature. One calculation method is... Where ||…|| represents the vector norm (length), the curvature value of point Pi is calculated. Then, within a small neighborhood of Pi (e.g., including Pi-2, Pi-1, Pi, Pi+1, Pi+2), the maximum absolute value of these curvature values ​​is found, and this maximum value is used as the curvature parameter of Pi. For example, in the forjneari algorithm, if the calculated curvatures of points near Pi are -1.2, -1.8, 1.5, and 0.9 mm, then the curvature parameter of Pi is... Simultaneously, retrieve the gradient magnitude of each boundary point Pi calculated in the previous step. For all boundary points The values ​​are normalized, for example, linearly normalized to the [0,1] interval: ,in and The minimum and maximum values ​​of the gradient magnitudes at all boundary points are given. Next, the rate of change of the gradient direction between adjacent points is calculated. For point Pi, its gradient direction vector is calculated. Compared to the previous point And the next point The angle difference between and The average of these two angle differences is taken as the edge sharpness value of point Pi. For example, if , , The calculated angle differences are respectively and Then the sharpness value (Or use radians) to generate edge sharpness values ​​and curvature parameters for each boundary point.

[0037] The feature parameter generation submodule performs weighted fusion calculations on regions with sharpness values ​​exceeding 0.25 based on edge sharpness values ​​and curvature parameters. The curvature weight coefficient is set to 0.6 and the sharpness weight coefficient is set to 0.4. The module integrates the point parameters within the region to generate the average value and standard deviation, thus obtaining the lesion feature parameters.

[0038] The feature parameter generation submodule receives the edge sharpness value. With the curvature parameter (For all boundary points i=1tok), first filter out regions whose sharpness values ​​exceed a specific threshold, and set the sharpness threshold to 1. (The unit here must be consistent with the sharpness calculation. If sharpness is expressed in angles, it is, for example, 15 degrees; if a normalized value is used, it is 0.25.) This threshold is based on the distribution analysis of sharpness values ​​for different dental lesions (such as sharp cavity edges vs. relatively gentle wear), selecting values ​​that can better distinguish sharp features and identifying all those that meet the criteria. The continuous boundary point segments constitute high-sharpness regions. For each point Pi located within these high-sharpness regions, a weighted fusion calculation is performed, and a curvature weight coefficient is set. and sharpness weighting coefficient These two weighting coefficients ( This reflects the relative importance of curvature (morphological complexity) and sharpness (edge ​​clarity / accentuation) when assessing lesion severity or characteristics. Based on clinical experience or data analysis, it is believed that morphological complexity (reflected by curvature) may be more important than edge sharpness in certain diagnostic scenarios. Therefore, curvature is given a higher weight of 0.6. This weight setting can be determined by optimizing on the training set to maximize the correlation between the generated feature parameters and clinical diagnostic results. A fusion value is calculated for each point Pi within the region. For example, a highly sharp region contains points P1 and P2, with parameter P1( ) and P2( ),but , Then, the fusion parameters of all points within the region are integrated. Calculate these average value and standard deviation Where N is the number of points in the region, continuing the previous example, , The characteristic parameters of the lesion were obtained.

[0039] Please see Figure 4 The strategy generation module includes: The classification calculation submodule is based on lesion feature parameters and edge sharpness values. It inputs the parameter matrix into the fully connected layer of the convolutional neural network, extracts multi-dimensional feature vectors, performs linear superposition of feature vectors and weight matrix, and outputs probability distribution through activation function to obtain the defect area, depth value and structural damage index. The classification calculation submodule receives the lesion feature parameters (average fusion value). and standard deviation And may also include global edge sharpness statistics (e.g., the average sharpness of the entire boundary). These parameters are combined into a feature vector. The vector is input into the input layer of a pre-trained convolutional neural network (referring to the fully connected layer in its structure). Through the forward propagation process of the network, the feature vector first passes through the fully connected layer, where it interacts with the weight matrix of that layer. The linear superposition operation, plus the bias term. ,Right now For example, if a fully connected layer has three output nodes, corresponding to preliminary estimates of the defect area, depth value, and structural damage index, respectively, the weight matrix... It is The matrix, bias It is The vector, the calculation process is as follows , , ,in and These are the parameters obtained from network training, and are calculated. The vector is then passed through an activation function; if the output is a continuous value (such as area or depth), a ReLU or linear activation function may be used. or If it is a classification probability (such as a damage index classification), then the Softmax function is used. Assuming Using linear activation, preliminary estimates are obtained for the corresponding area and depth. For example, the area is estimated to be 14.5 (mm²) and the depth is estimated to be 2.8 (mm). The relevant outputs are used for the damage index (divided into three levels: mild, moderate, and severe). The probability distribution of each level is output through Softmax. For example, if P(mild) = 0.1, P(moderate) = 0.6, and P(severe) = 0.3, then the moderate level with the highest probability is selected as the structural damage index, and the defect area (14.5 mm²), depth value (2.8 mm), and structural damage index (moderate) are obtained.

[0040] The coefficient generation submodule standardizes the defect area per unit area, converts the depth value into a relative depth ratio, performs a product operation on the three factors in combination with the structural damage index, introduces a weight correction factor to adjust the product result, and generates a repair accuracy coefficient. The coefficient generation submodule receives the defect area obtained in the previous step. mm², depth value mm, and the structural damage index $Index=$moderate, firstly, the defect area is standardized by unit area, for example, if 1 mm² is used as the standard unit area. mm², then the standardized area (Dimensionless), then the depth value is converted into a relative depth ratio, which requires a reference maximum depth. This can be determined based on tooth type and anatomical structure. For example, for premolars, the maximum depth that caries can typically reach (before reaching the pulp chamber) might be estimated at 4.5 mm, thus the relative depth... Then, the structural damage index is quantified, with values ​​set as follows: mild = 1, moderate = 2, severe = 3. Multiply these three quantitative indicators: Introduce a weighting correction factor To adjust the product result, this correction factor The settings may be based on the patient's age and the location of the affected tooth (e.g., posterior teeth bear greater force). A comprehensive assessment should be made, taking into account factors such as the possibility of slightly higher tooth height and the presence of complications. For example, for a middle-aged patient's posterior teeth (such as tooth number 26). It may be set to 1.15. This value is determined by performing regression analysis on a historical case database to find an adjustment value that allows the final coefficient to better predict repair complexity and prognosis. This means that for this type of situation, the basic risk assessment value needs to be increased by 15% to calculate the final repair accuracy coefficient. Generate repair accuracy coefficients.

[0041] The strategy mapping submodule matches the corresponding technical number in the preset mapping table according to the numerical range of the repair accuracy coefficient, divides the depth value into levels with an accuracy of 0.1mm, integrates the number and depth level, and establishes a dynamic repair strategy table.

[0042] The strategy mapping submodule receives the repair accuracy coefficient. Based on this value, the corresponding range is found in a preset technical mapping table. This mapping table (see example table 1 below) establishes the correspondence between the repair accuracy coefficient range and the specific repair technology number.

[0043] Table 1 Repair Strategy Mapping Table ; As shown in Table 1, the repair accuracy coefficient The value falls within the range of 10.01-25, therefore the matched technology number is T02, corresponding to the technology type of direct composite resin filling. Simultaneously, the previously calculated depth value... Depth levels are classified in mm with an accuracy of 0.1 mm and calculated accordingly. Integrate the technology number T02 and depth level 28, and add or update the pair of information {technology number: T02, depth level: 28} as an entry to the dynamic repair strategy table for this case. This table records the condition assessment and recommended strategies that change over time. For example, an entry in the table may contain {timestamp: 2025-04-28 11:00:00, case ID: P123, technology number: T02, depth level: 28} to establish the dynamic repair strategy table.

[0044] Please see Figure 5 The trajectory prediction module includes: The deep modeling submodule calls the depth value sequence in the dynamic repair strategy table, inputs it into the LSTM time series model, calculates the time weight through the forget gate and the input gate, performs sliding window segmentation on the depth value, outputs the hidden layer state vector, and generates the depth change rate. The depth modeling submodule calls a series of depth values ​​recorded in the dynamic repair strategy table for a specific lesion (e.g., tooth 26 in case P123). If the lesion has been monitored multiple times, a time series is formed, for example, the depth values ​​(in mm) of the past three examinations are respectively... The current value is 2.8mm. Sequence this depth value... The input is processed into a Long Short-Term Memory (LSTM) unit structure. Inside the LSTM unit, for each time point in the sequence... depth value It will be in the same hidden state as the previous moment. and unit state Together they participate in the calculation, first passing through the forget gate. The calculation determines the state of the cell. How much information is discarded, and what is the weight? and bias These are learning parameters. It's the sigmoid function, then calculated through the input gate. Decide which new information to update, and the status of candidate units. Prepare the information to add, then update the cell status. ( (representing element-wise multiplication), and finally passed through the output gate. Calculate and determine the hidden state of the output This process is performed stepwise on the sequence [2.1, 2.4, 2.8], ultimately yielding the hidden layer state vector. (T being the end of the sequence) contains dynamic information about the sequence, and Through a linear layer (or directly using) (Partial dimensions) are used to predict the depth change at the next time step, i.e., the depth change rate. For example, based on the growth trend of the sequence [2.1, 2.4, 2.8] (intervals of 0.3 mm and 0.4 mm respectively), the model may predict the rate of change in depth for the next time unit (e.g., the next 3 months). mm / unit time, generating depth change rate.

[0045] The trajectory generation submodule calculates the difference between adjacent time points of the defect area based on the depth change rate, fits the area growth curve through linear regression, extracts the slope parameter as the growth rate, and concatenates the change rate and the growth rate into a matrix to generate the evolution trajectory prediction vector. The trajectory generation submodule receives the depth change rate. mm / unit time, and obtain the defect area sequence corresponding to the depth sequence, for example mm², calculate the difference between adjacent time points in the defect area sequence to obtain the area change. mm², mm², in order to fit the area growth trend, the area value A linear regression analysis was performed with the corresponding time points (or sequence indices 1, 2, 3) to establish a model. ,in It is a time point or index, and the slope is calculated using the least squares method. For example, performing linear regression on points (1, 12.5), (2, 13.8), and (3, 14.5) yields the calculated slope. (i.e., area growth rate) ) may be mm² / unit time, the calculated depth change rate With area growth rate Perform matrix concatenation (or vector concatenation in this case) to form a two-dimensional evolution trajectory prediction vector. This vector represents the expected changes in depth and area of ​​the lesion within a future time unit, generating an evolution trajectory prediction vector.

[0046] The strategy update submodule performs a matching operation between the vector and the preset strategy rules based on the dimensionality characteristics of the predicted vector of the evolution trajectory, filters candidate strategies that meet the error threshold, marks the optimal strategy as the priority, and outputs the update strategy instruction.

[0047] The strategy update submodule receives the evolution trajectory prediction vector. Based on the dimensional characteristics of the vector (i.e., the rate of change of depth) and area growth rate The vector is matched against a set of predefined policy adjustment rules. These rules define policy adjustment suggestions for different predicted trajectory intervals. For example, the rule set might include: Rule A: If and The candidate strategy is {shorten the follow-up visit period to 3 months and consider upgrading the repair material level}; Rule B: If or The candidate strategy is {immediate follow-up visit to assess whether more invasive treatment, such as root canal preparation, is needed, and the restoration plan is upgraded to an inlay / high inlay}; Rule C: If and The candidate strategy is then {maintain the current observation / repair plan, and the follow-up period can be appropriately extended}, and the current prediction vector is... Matching these rules, it was found that it satisfies the condition of rule A. and Therefore, the selected candidate strategy set is {shorten the follow-up visit cycle to 3 months, consider upgrading the repair material grade}. Next, the candidate strategies need to be evaluated to select the optimal strategy. This may involve evaluating factors such as the expected treatment effect, risk, and cost under different strategies, and setting an error threshold or risk scoring standard. For example, if "upgrading the material grade" can reduce the predicted long-term failure rate by more than 10% (below the error threshold), it is considered better than "shortening the follow-up visit cycle" alone. Assuming that the evaluation results show that "upgrading the repair material grade" combined with "shortening the follow-up visit cycle" is the optimal choice, the combined strategy containing these two actions is marked as the highest priority, and an update strategy instruction containing specific instructions is generated, such as {instruction type: strategy update, case ID: P123, suggested operation: [shorten the follow-up visit cycle to 3 months, upgrade the repair material by one grade]}, and the update strategy instruction is output.

[0048] Please see Figure 6 The solution feedback module includes: The material matching submodule calls the repair technology number in the update strategy instruction, traverses the primary key of the repair material database based on the number hash value, and after verifying the validity of the entry, extracts the compressive strength and adhesion coefficient fields to generate a material parameter set. The material matching submodule receives the update strategy instruction, such as a requirement to upgrade the repair material by one grade. Combined with the currently recommended repair technology number (assuming it remains T02, but a higher-performance composite resin is required), this instruction triggers a query to the repair material database. First, based on the technology number T02 (direct composite resin filling) and the new requirement (e.g., previously using A2 grade material, now needing to be upgraded to A1 or a higher strength grade), the query conditions are determined. Using the technology number T02 or its hash value, and the material grade or performance requirements (such as compressive strength range) as the primary key or index, the database is traversed to find matching repair material entries. For example, if the database contains material records M1{ID:RCM005,Name:BrandXFlowableA2,Type:T02,Strength:150MPa,Adhesion:0.7} and M2{ID:RCM010,Name:BrandYUniversalA1,Type:T02,Strength:300... The program verifies the validity of these entries (e.g., whether they are within the validity period and comply with regional regulations), and then extracts key performance parameter fields, mainly compressive strength and adhesion coefficient (this coefficient may represent the bond strength or an indicator related to the bonding ability with tooth tissue). Assuming that M2 and M3 are both valid and meet the upgrade requirements (strength higher than M1), the extracted parameter set is: [{Material:M2,Strength:300MPa,Adhesion:0.85},{Material:M3,Strength:380MPa,Adhesion:0.90}], generating the material parameter set.

[0049] The fit calculation submodule obtains the compressive strength of the material parameter set as the numerator and the real-time bite force monitoring data as the denominator to perform a division operation, extracts the product term of the adhesion coefficient and the rate of change of bite force, standardizes the two results and superimposes the weight coefficient to generate the fit coefficient. The fit calculation submodule receives the material parameter set [{M2,300MPa,0.85},{M3,380MPa,0.90}] generated in the previous step, and acquires the patient's real-time or recent monitored occlusal force data, for example, the maximum chewing force on the side of the patient undergoing repair measured by an occlusal force sensor. N, the average rate of change of biting force during chewing. N / s, firstly, the occlusal force needs to be converted into stress, which requires estimating the stress-bearing area of ​​the restoration. Assuming that for a T02 restoration, the average contact area is approximately mm², then the maximum bite stress MPa, calculate two parameters for each candidate material: 1. Strength ratio = Material compressive strength / Maximum interlocking stress - for M2: -For M3: 2. Adhesion-kinetic term = Material adhesion coefficient Change rate of bite force (assuming adhesion coefficient is in the range of 0-1, change rate in N / s) - for M2: -For M3: Next, we standardize these two indicators. Assuming that, based on extensive data statistics, the typical range for the strength ratio is [1.5, 5.0], and the typical range for the adhesion-dynamic term is [30, 100], we perform linear standardization to the [0, 1] interval: -M2: -M3: - -M2: -M3: Set weighting coefficients; for example, for posterior tooth restorations that need to withstand large occlusal forces, compressive strength may be more important, so assign a weight to compressive strength. Adhesion-related factors weights These two weights are determined by clinical experts based on the importance of the repair type and location, and Calculate the final fitness coefficient. :-M2: -M3: Generate fitness coefficients.

[0050] The solution generation submodule divides the fit level range according to the numerical distribution of the fit coefficient, selects material combinations that meet the range requirements for both compressive strength and adhesion coefficient, aligns the interlocking force data and material parameters according to the time series to generate an index matrix, and establishes a repair solution.

[0051] The scheme generation submodule receives the fit coefficients of each candidate material, M2: M3: Based on the numerical distribution of these coefficients, a suitability level range is defined, for example: [0-0.3): Not recommended; [0.3-0.5): Acceptable; [0.5-0.7]: Good; [0.7-1.0]: Excellent. Accordingly, material M2 (0.426) belongs to the "acceptable" level, and material M3 (0.592) belongs to the "good" level. Materials that meet the preset minimum suitability level (e.g., "acceptable" or higher) are selected, and their original compressive strength and adhesion coefficient must also meet the basic requirements of this repair technology (T02, direct composite resin filling). Assuming the basic requirements of T02... If Strength > 250 MPa and Adhesion > 0.8, and both M2 (300 MPa, 0.85) and M3 (380 MPa, 0.90) meet these conditions, then both are candidate materials. However, M3 has a higher fit. When generating the final solution, the material combination with the highest fit is usually recommended first (if multiple materials are needed). In this example, material M3 (BrandZHighStrengthBody) is recommended first. Then, the selected material information (such as M3's IDRCM015), the timestamp or features of the bite force data used to calculate the fit (such as...) are used to... N, The key parameters of the material (Strength=380MPa, Adhesion=0.90) and the fit score (0.592) are aligned and recorded. An index matrix or structured record containing this information is created to finally determine and record the detailed restoration plan. For example, the plan is recorded as {Case ID:P123, Tooth position:26, Recommended technique:T02, Recommended material:M3(RCM015), Fit:0.592 (good), Key parameters:{Strength:380MPa, Adhesion:0.90}, Reference occlusal force:{Max:600N, Rate:70N / s}}, thus establishing the restoration plan.

[0052] A remote dental restoration treatment method based on a 5G network, which is executed based on the aforementioned remote dental restoration treatment system based on a 5G network, includes the following steps: S1: Receives real-time oral cavity 3D scan data stream via 5G communication protocol, calls Gaussian filtering algorithm to filter out noise from the data stream, extracts the coordinate set of the lesion area, and generates 3D reconstruction data; S2: Input the 3D reconstruction data into the edge detection algorithm to locate the contour boundary, calculate the gradient magnitude of the boundary pixels to generate the edge sharpness value, and combine the curvature difference of adjacent pixels to generate lesion feature parameters; S3: Input the lesion feature parameters and edge sharpness values ​​into the convolutional neural network model for classification training, and output the defect area, depth value and structural damage index. Multiply the depth value and structural damage index to generate the repair accuracy coefficient. Match the repair accuracy coefficient with the preset repair technology mapping table to generate a dynamic repair strategy table. S4: Based on the dynamic repair strategy table and depth value, call the LSTM time series model to predict the rate of change of depth value, combine the adjacent timestamp difference of the defect area to generate the evolution trajectory prediction vector, and output the update strategy instruction containing the correction parameters. S5: According to the update strategy instructions, traverse the repair material database, match the candidate material set that meets the compressive strength and adhesion coefficient, and calculate the cosine similarity between the candidate material and the evolution trajectory prediction vector by combining the real-time collected biting force monitoring data, and generate a repair scheme.

[0053] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A remote dental restoration and treatment system based on a 5G network, characterized in that, The system includes: The image acquisition module is used to receive oral cavity 3D scan data stream through 5G real-time transmission bandwidth, call Gaussian filtering algorithm to filter out noise, extract the coordinate set of lesion area, generate 3D reconstruction data, and transmit it to the feature analysis module. The feature analysis module is used to receive the three-dimensional reconstruction data, locate the contour boundary based on the edge detection algorithm, calculate the edge sharpness value and curvature parameter, generate lesion feature parameters, and transmit the lesion feature parameters and the edge sharpness value to the strategy generation module; The strategy generation module is used to receive the lesion feature parameters and the edge sharpness value, input them into the convolutional neural network model for lesion classification, output the defect area, depth value, and structural damage index, generate repair accuracy coefficients based on the product, map them to repair technology numbers, generate a dynamic repair strategy table, and pass the dynamic repair strategy table and the depth value to the trajectory prediction module. The trajectory prediction module is used to receive the dynamic repair strategy table and the depth value, call the LSTM time series model to model the depth change rate, combine the defect area growth rate to generate the evolution trajectory prediction vector, output the update strategy instruction, and pass it to the scheme feedback module.

2. The remote oral restoration and treatment system based on a 5G network according to claim 1, characterized in that, The three-dimensional reconstruction data includes a set of spatial coordinates of the lesion area, surface point cloud data, and texture feature data after noise reduction. The lesion feature parameters include edge sharpness distribution, curvature change matrix, and morphological complexity index. The dynamic repair strategy table includes repair technology number, repair priority, and estimated repair cycle. The evolution trajectory prediction vector includes depth change rate sequence, area expansion rate curve, and evolution time node marker.

3. The remote oral restoration and treatment system based on a 5G network according to claim 2, characterized in that, The image acquisition module includes: The 5G transmission submodule acquires real-time oral cavity 3D scan data stream, establishes a communication link through the 5G transmission protocol, divides the data stream into independent data packets and adds timestamps, and completes data reassembly according to the transmission sequence number to generate a transmission data stream; The filtering and noise reduction submodule calls the Gaussian filtering algorithm, constructs a three-dimensional spatial coordinate system based on the transmitted data stream, performs neighborhood weighted average calculation on each coordinate point, sets the standard deviation parameter to 0.8, eliminates outlier coordinate offsets, and generates filtered data. The lesion coordinate generation submodule performs gradient calculation based on the filtered data, performs differential operation on the gray values ​​of adjacent coordinate points, selects regions with gray value differences exceeding 0.12 as boundary judgment benchmarks, extracts the vertex coordinate set of continuous closed regions, and generates three-dimensional reconstruction data.

4. The remote oral restoration and treatment system based on a 5G network according to claim 3, characterized in that, The feature analysis module includes: The edge detection submodule calls the edge detection algorithm based on the 3D reconstruction data, calculates the gradient magnitude and direction for each coordinate point, sets high and low thresholds of 0.3 and 0.1 respectively, selects points with gradient magnitudes higher than the high threshold as boundary points, connects adjacent points to form a closed contour, and generates a boundary coordinate set; The sharpness curvature calculation submodule calls the boundary coordinate set, performs second derivative calculation on the coordinates in the neighborhood of each boundary point, extracts the maximum absolute value of curvature as the curvature parameter, normalizes the gradient magnitude, calculates the average value of the gradient direction change rate of adjacent points, and generates the edge sharpness value and curvature parameter. The feature parameter generation submodule performs a weighted fusion calculation on regions with sharpness values ​​exceeding 0.25 based on the edge sharpness value and the curvature parameter. The curvature weight coefficient is set to 0.6 and the sharpness weight coefficient is set to 0.

4. The module integrates the point parameters within the region to generate the average value and standard deviation, thereby obtaining the lesion feature parameters.

5. The remote oral restoration and treatment system based on a 5G network according to claim 4, characterized in that, The strategy generation module includes: The classification calculation submodule, based on the lesion feature parameters and the edge sharpness value, inputs the parameter matrix into the fully connected layer of the convolutional neural network, extracts multi-dimensional feature vectors, performs linear superposition operation of feature vectors and weight matrix, and outputs probability distribution through activation function to obtain the defect area, depth value and structural damage index; The coefficient generation submodule standardizes the defect area per unit area, converts the depth value into a relative depth ratio, performs a product operation on the three factors in conjunction with the structural damage index, introduces a weight correction factor to adjust the product result, and generates a repair accuracy coefficient. The strategy mapping submodule matches the corresponding technical number in a preset mapping table according to the numerical range of the repair accuracy coefficient, divides the depth value into levels with an accuracy of 0.1mm, integrates the number and depth level, and establishes a dynamic repair strategy table.

6. The remote dental restoration and treatment system based on a 5G network according to claim 5, characterized in that, The trajectory prediction module includes: The deep modeling submodule calls the depth value sequence in the dynamic repair strategy table, inputs it into the LSTM time series model, calculates the time weight through the forget gate and the input gate, performs sliding window segmentation on the depth value, outputs the hidden layer state vector, and generates the depth change rate. Based on the depth change rate, the trajectory generation submodule calculates the difference between adjacent time points of the defect area, fits the area growth curve through linear regression, extracts the slope parameter as the growth rate, and concatenates the change rate and the growth rate into a matrix to generate an evolution trajectory prediction vector. The strategy update submodule performs a matching operation between the vector and the preset strategy rules based on the dimensional characteristics of the predicted vector of the evolution trajectory, filters candidate strategies that meet the error threshold, marks the optimal strategy as the priority, and outputs the update strategy instruction.

7. The remote oral restoration and treatment system based on a 5G network according to claim 6, characterized in that, The system also includes: The solution feedback module is used to receive the update strategy instruction, match the repair material database, extract the compressive strength and adhesion coefficient, calculate the fit by combining real-time bite force monitoring data, generate a repair solution, and synchronize the repair solution to the wearable sensor at the patient end. The repair solution includes material compatibility parameters, compressive strength values, and adhesion performance indicators.

8. The remote oral restoration and treatment system based on a 5G network according to claim 7, characterized in that, The solution feedback module includes: The material matching submodule calls the repair technology number in the update strategy instruction, traverses the primary key of the repair material database based on the number hash value, and after verifying the validity of the entry, extracts the compressive strength and adhesion coefficient fields to generate a material parameter set. The fit calculation submodule obtains the compressive strength of the material parameter set as the numerator and the real-time bite force monitoring data as the denominator to perform a division operation, extracts the product term of the adhesion coefficient and the rate of change of bite force, standardizes the two results and superimposes the weight coefficient to generate the fit coefficient. The solution generation submodule divides the fit level range according to the numerical distribution of the fit coefficient, selects material combinations that meet the range requirements for both compressive strength and adhesion coefficient, aligns the biting force data and material parameters according to the time series to generate an index matrix, and establishes a repair solution.

9. A remote dental restoration treatment method based on a 5G network, characterized in that, The method is used to implement the remote oral restoration and treatment system based on a 5G network as described in any one of claims 1-8, and includes the following steps: S1: Receives real-time oral cavity 3D scan data stream via 5G communication protocol, calls Gaussian filtering algorithm to filter out noise from the data stream, extracts the coordinate set of the lesion area, and generates 3D reconstruction data; S2: Input the three-dimensional reconstruction data into the edge detection algorithm to locate the contour boundary, calculate the gradient magnitude of the boundary pixels to generate the edge sharpness value, and combine the curvature difference of adjacent pixels to generate lesion feature parameters. S3: Input the lesion feature parameters and the edge sharpness value into the convolutional neural network model for classification training, and output the defect area, depth value and structural damage index. Multiply the depth value and the structural damage index to generate a repair accuracy coefficient. Match the repair accuracy coefficient with a preset repair technology mapping table to generate a dynamic repair strategy table. S4: Based on the dynamic repair strategy table and the depth value, call the LSTM time series model to predict the rate of change of the depth value, combine the adjacent timestamp difference of the defect area to generate an evolution trajectory prediction vector, and output an update strategy instruction containing correction parameters. S5: According to the update strategy instruction, traverse the repair material database, match the candidate material set that meets the compressive strength and adhesion coefficient, and calculate the cosine similarity between the candidate material and the evolution trajectory prediction vector by combining the real-time collected biting force monitoring data, and generate a repair scheme.

Images