Full crown repair preparation body edge line extraction method and system based on deep learning
Through the full crown repair preparatory edge line extraction method based on deep learning, using three-dimensional mesh semantic segmentation and smoothing processing technology, the problems of low edge line extraction complexity and accuracy in the existing technology are solved, and the rapid and accurate Edge line extraction reduces subjective errors and improves the efficiency of crown design and production.
Patent Information
- Application Number
- CN202411505994.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-28
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-10-28
AI Technical Summary
The existing methods of edge line extraction of full crown preparatory bodies rely on the experience of doctors or technicians, resulting in complex operation, low accuracy, time-consuming and prone to labeling errors caused by subjective factors.
Adopt the edge line extraction method of full crown repair preparatory body based on deep learning. By obtaining three-dimensional mesh data, using the three-dimensional mesh semantic segmentation network model for segmentation and preprocessing, combining smoothing and projection processing, fully automated edge line extraction is achieved. .
Fast and accurate extraction of the edge line of the entire crown preparatory body is achieved, reducing labeling errors caused by subjective factors, saving restoration design and production time, and ensuring the complete alignment of the crown and the edge of the preparatory body.
Smart Images

Figure CN119027443A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of full crown restoration, and in particular to a method and system for extracting edge lines of a full crown restoration preparation based on deep learning. Background Art
[0002] Full crown restoration is a dental restoration method to restore the tissue morphology and chewing function of the prepared body, and is the preferred treatment option for fixed restoration of tooth defects. With the rapid development of digital technology, optical impressions and computer-aided design / computer-aided manufacturing (CAD / CAM) have become key links in the design and production of restorations. Although the digital technology of restoration-assisted design has improved the efficiency of restoration diagnosis and treatment to a certain extent, in the wave of full-process digital development, especially in the face of challenges such as differences in design capabilities between dentists and dental technicians and the rapid growth of oral scan data, it is still necessary to seek more efficient and stable solutions.
[0003] The existing method for extracting the edge line of the full crown preparation is usually based on the clinical experience of doctors or technicians. However, this method has the problems of high experience requirements for doctors and technicians, different subjective judgments of different doctors and technicians, and more manual interaction processes, which wastes a lot of time, is complicated to operate, and has low accuracy.
[0004] Therefore, there is an urgent need for a full-crown restoration preparation margin line extraction method and system based on deep learning, which can assist in marking the preparation margin line, save restoration design and production time, and reduce problems such as marking errors caused by subjective factors. Summary of the invention
[0005] In order to solve the above technical problems, the present invention provides a method and system for extracting the edge line of a full-crown restoration preparation based on deep learning, which can assist in marking the edge line of the preparation, save time in restoration design and production, and reduce problems such as marking errors caused by subjective factors.
[0006] The present invention provides a method for extracting edge lines of a full crown restoration preparation based on deep learning, comprising the following steps: S1. Obtaining three-dimensional mesh data of the prepared body of the tester; S2, inputting the three-dimensional mesh data of the prepared body of the tester into the three-dimensional mesh semantic segmentation network model to obtain the three-dimensional mesh segmentation result of the prepared body of the tester; S3, preprocessing the three-dimensional mesh segmentation result of the prepared body, and smoothing the preprocessed three-dimensional mesh segmentation result to obtain an accurate three-dimensional mesh segmentation result; S4, calculating the segmentation edge according to the accurate three-dimensional mesh segmentation result, and obtaining the rough edge line of the preparation body; S5. Perform smoothing and projection processing on the rough edge line of the prepared body to obtain a smooth edge line of the prepared body on the three-dimensional grid.
[0007] Furthermore, in S1, obtaining the three-dimensional mesh data of the prepared body of the test subject includes: The intraoral scanning 3D mesh data of the single abutment tooth preparation of the tester is obtained.
[0008] Furthermore, in S2, the training process of the 3D mesh semantic segmentation network model is as follows: S2.1. Construct a 3D mesh semantic segmentation sample data set; wherein the sample data set includes 3D mesh data of several non-test subject preparations and segmentation labels corresponding to the 3D mesh data of several non-test subject preparations; S2.2, build a neural network model; S2.3. Use the three-dimensional mesh semantic segmentation sample data set to train the neural network model until the loss value of the neural network model is less than the preset loss value and the segmentation accuracy of the neural network model is greater than the preset accuracy, thereby obtaining a trained three-dimensional mesh semantic segmentation neural network model.
[0009] Furthermore, in S2.1, the process of constructing the segmentation labels corresponding to the three-dimensional mesh data of the prepared bodies of the non-test subjects includes: S2.11. Connect the ordered points on the edge line of any preparation of a non-test subject end to end to form a closed curve; S2.12. According to the closed curve, the three-dimensional mesh of the prepared body of the non-test subject is divided into the prepared area and the non-prepared area, and a label is assigned to each triangular mesh in the prepared area and the non-prepared area, respectively, to obtain the segmentation label corresponding to the three-dimensional mesh of the prepared body of the non-test subject; wherein the prepared area refers to the area containing only the tooth body, and the non-prepared area refers to the area containing other tissues in addition to the tooth body.
[0010] Further, in S3, the three-dimensional mesh segmentation result of the prepared body is preprocessed, wherein the preprocessing includes: outlier removal processing and hole filling processing; The clustering algorithm is used to remove outliers and fill holes in the three-dimensional mesh segmentation results of the preparation.
[0011] Furthermore, in S3, the preprocessed three-dimensional mesh segmentation result is smoothed to obtain an accurate three-dimensional mesh segmentation result including: According to the preprocessed 3D mesh segmentation result, smoothing is performed through a fully connected conditional random field algorithm to obtain an accurate 3D mesh segmentation result.
[0012] Furthermore, in S4, based on the accurate three-dimensional mesh segmentation result, the segmentation edge is calculated to obtain the rough edge line of the prepared body, including: S41, according to the accurate three-dimensional mesh segmentation result, obtain all segmentation edge triangle meshes to form a triangle set; S42. Calculate the centroid of each segmented edge triangle mesh according to the vertices of the segmented edge triangle mesh in the triangle set, use the centroid of the segmented edge triangle mesh as a rough edge point to form a rough edge line scatter point set, and use the edge line formed by the rough edge points as the rough edge line of the preparation.
[0013] Further, in S5, the rough edge line of the prepared body is smoothed and projected to obtain a smooth edge line of the prepared body on the three-dimensional grid, including: S51, downsampling the rough edge line scattered point set into an unordered point set; S52, sorting the unordered point set to obtain an ordered point set that is connected end to end to form a closed curve; S53, performing smoothing processing by a polynomial regression algorithm according to the ordered point set to obtain a smoothed preparation edge line point cloud; S54, projecting the smoothed edge line point cloud of the preparation onto the three-dimensional grid of the preparation to obtain the smooth edge line of the preparation on the three-dimensional grid.
[0014] The present invention also provides a full crown restoration preparation edge line extraction system based on deep learning, which is used to execute any of the full crown restoration preparation edge line extraction methods based on deep learning described above, and the system includes the following modules: A data acquisition module, used to acquire three-dimensional mesh data of the prepared body of the tester; A three-dimensional mesh data segmentation module, connected to the data acquisition module, is used to input the three-dimensional mesh data of the prepared body of the tester into the three-dimensional mesh semantic segmentation network model to obtain the three-dimensional mesh segmentation result of the prepared body of the tester; The segmentation result precision module is connected to the 3D mesh data segmentation module, and is used to pre-process the 3D mesh segmentation result of the preparation, and to smooth the pre-processed 3D mesh segmentation result to obtain a precise 3D mesh segmentation result; A rough edge line calculation module is connected to the segmentation result refinement module and is used to calculate the segmentation edge according to the precise three-dimensional grid segmentation result to obtain the rough edge line of the preparation body; The smooth edge line extraction module is connected to the rough edge line calculation module and is used to perform smoothing and projection processing on the rough edge line of the prepared body to obtain the smooth edge line of the prepared body on the three-dimensional grid.
[0015] The embodiments of the present invention have the following technical effects: The present invention uses a 3D mesh semantic segmentation network model to accurately segment the 3D mesh of the preparation based on the intraoral scanned 3D mesh data, and performs preprocessing, smoothing, and projection processing on the segmentation results to achieve accurate, fast, and fully automated extraction of the smooth edge line of the preparation. The subsequent crown is designed and manufactured based on the smooth edge line extracted by this solution to ensure that the manufactured crown and the edge of the preparation are completely consistent. The present invention can assist in marking the edge line of the preparation, save time in the design and manufacture of the restoration, and reduce problems such as marking errors caused by subjective factors. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0017] Figure 1 It is a flow chart of a method for extracting margin lines of a full crown restoration preparation based on deep learning provided by an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a three-dimensional grid semantic segmentation network model provided by an embodiment of the present invention; Figure 3 is a schematic diagram of a rough edge line of a preparation provided by an embodiment of the present invention; Figure 4 is a schematic diagram of a set of unordered points after downsampling provided by an embodiment of the present invention; Figure 5 is a schematic diagram of a smooth edge line of a preparation provided by an embodiment of the present invention; Figure 6 is a schematic diagram of a smooth edge line of a preparation on a three-dimensional grid provided by an embodiment of the present invention; Figure 7 It is a structural schematic diagram of a full crown restoration preparation margin line extraction system based on deep learning provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0018] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described clearly and completely below. 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 creative work belong to the scope of protection of the present invention.
[0019] The present invention proposes a method for extracting the margin line of a full crown restoration preparation based on deep learning. Figure 1 is a flowchart of a method for extracting margin lines of a full crown restoration preparation based on deep learning provided by an embodiment of the present invention, see Figure 1 , including: S1. Obtain three-dimensional mesh data of the prepared body of the tester.
[0020] Specifically, the three-dimensional mesh data of the tester's preparation includes the intraoral scanned three-dimensional mesh data of the tester's single abutment tooth preparation; wherein, the single abutment tooth preparation includes a single tooth from the gums to the tooth body; the three-dimensional mesh data is a three-dimensional model file, including information such as the numbers and coordinates of all vertices contained in the triangular meshes.
[0021] S2. Input the three-dimensional mesh data of the prepared body of the tester into the three-dimensional mesh semantic segmentation network model to obtain the three-dimensional mesh segmentation result of the prepared body of the tester.
[0022] Specifically, Figure 2 is a schematic diagram of the structure of a three-dimensional grid semantic segmentation network model provided by an embodiment of the present invention, see Figure 2 ,The training process of the 3D mesh semantic segmentation network model is as follows: S2.1. Construct a 3D mesh semantic segmentation sample data set; wherein the sample data set includes 3D mesh data of several non-test subject preparations and segmentation labels corresponding to the 3D mesh data of several non-test subject preparations.
[0023] Furthermore, intraoral scanned 3D mesh data of single abutment tooth preparations of several non-test subjects and corresponding manually marked preparation edge lines are obtained.
[0024] S2.11. Connect the ordered points on the edge line of any preparation of a non-test subject end to end to form a closed curve.
[0025] S2.12. According to the closed curve, the three-dimensional mesh of the preparation body of the non-test subject is divided into a preparation area and a non-preparation area, and a label is assigned to each triangular mesh in the preparation area and the non-preparation area, respectively, to obtain a segmentation label corresponding to the three-dimensional mesh of the preparation body of the non-test subject.
[0026] The prepared area refers to an area containing only tooth body, and the unprepared area refers to an area containing other tissues (such as gums) in addition to tooth body. Exemplarily, the triangle mesh label in the prepared area is set to 1, and the triangle mesh label in the unprepared area is set to 0.
[0027] Furthermore, data enhancement is performed on the constructed three-dimensional mesh semantic segmentation sample dataset; wherein the data enhancement includes random flipping, random scaling, random rotation, random jittering and random discarding.
[0028] S2.2. Build a neural network model.
[0029] Specifically, Picasso++ is used as the neural network model in this embodiment. The neural network includes an encoder and a decoder. The output channels of the encoder are 32, 64, 96, 128, 192, and 256, respectively, and the output channels of the decoder are 128, 128, 96, 96, and 96, respectively.
[0030] The encoder feature extraction mainly includes the initial convolution module, the grid encoder module and the dual encoder module. The decoder module is mainly composed of upsampling, feature splicing and 1×1 convolution. The maximum pooling is used on the encoder side for 5 downsamplings with a pooling step of 4, 3, 3, 2, 2. The decoder side performs 5 upsamplings to restore the size, and uses jump connections at each stage to splice the corresponding encoding layer features with the upsampled decoding layer features.
[0031] Furthermore, the specific operations involved in the Picasso++ neural network model include: ① Use spherical harmonics to represent the convolution filter as a continuous function of azimuth and elevation angles.
[0032] The continuous function expression is as follows: ; ; Among them, F represents the convolution filter function, ! represents the factorial, θ∈[0,2π) is the azimuth angle, ∈[0,π) is the pitch angle, spherical harmonic function is the basis function, the integers l and m represent the degree and order of the spherical harmonic function respectively. are linear coefficients, where Represents the vector space of complex numbers, representing a lm The range of is a set of complex values; represents the associated Legendre polynomials without the Condon–Shortley phase.
[0033] ②Following the expression of spherical harmonic functions, the features of the vertices after the face-to-point convolution module are obtained.
[0034] The formula for calculating the eigenvalue of a vertex is: ; Among them, g v represents the characteristics of vertex v, hf Represents the related features of adjacent faces, Represents the set of adjacent faces (i.e., triangular faces) of vertex v. The relevant features of these adjacent faces are , where f represents any adjacent face, represents the azimuth of any adjacent face, Represents the pitch angle of any adjacent face.
[0035] ③Following the expression of spherical harmonic functions, the features of the triangular surface are obtained after the point-to-surface convolution module.
[0036] The formula for calculating the eigenvalue of a triangular face is: ; Among them, g f represents the characteristics of the triangular face f, and {h1,h2,h3} represents the characteristics of the three vertices {v1,v2,v3} of the triangular face.
[0037] ④Following the expression of spherical harmonics, the features of the triangular surface are obtained after the face-to-face convolution module.
[0038] The formula for calculating the eigenvalue of a triangular face is: ; Among them, k represents any point in the triangle surface, K represents the texture resolution, , , A min and A max Represent the minimum and maximum areas of the triangular mesh surface, A represents the area of the current triangular mesh surface, α and β are hyperparameters, , represents the set of non-negative integers. θ k represents the azimuth of any point in the triangle surface, represents the pitch angle of any point in the triangle surface, Represents the color vector set of all points on the triangle face f, Represents the RGB color value range.
[0039] ⑤Calculate the geometric features of the surface.
[0040] For any triangular face in a three-dimensional mesh, its geometric features are expressed as [l, θ, n], where l = (l1, l2, l3) represents the length of the three sides of the triangular face, n represents the normal of the triangular face, and θ = (θ1, θ2, θ3) represents the internal angle of the triangular face. The calculation formula is: ; ; ; Among them, x1, x2, x3 represent the three-dimensional coordinate vectors of the three vertices of the triangular face, and <·> represents the inner product operation.
[0041] S2.3. Use the three-dimensional mesh semantic segmentation sample data set to train the neural network model until the loss value of the neural network model is less than the preset loss value and the segmentation accuracy of the neural network model is greater than the preset accuracy, thereby obtaining a trained three-dimensional mesh semantic segmentation neural network model.
[0042] Specifically, the training strategy of the Picasso++ neural network model includes: ①The loss function of Picasso++ neural network is the cross entropy loss function.
[0043] The loss function formula is: ; Among them, L CE represents the cross entropy loss function, p is the true value, q is the predicted value, C is the number of detection categories, and j represents the jth detection category. In this embodiment, the detection category C=2.
[0044] ②During training, the optimizer is Adam, the initial learning rate is set to 0.001, and the weight decay is set to 1×e -5 , the learning rate decays as the number of training rounds increases, and the learning rate is dynamically adjusted according to the following formula: ; Among them, lr represents the learning rate after each round of training, lr init represents the initialization learning rate, t represents the current number of training rounds, is the attenuation coefficient.
[0045] Preferably, lr init =0.001, =0.98.
[0046] Furthermore, the training process of the Picasso++ neural network model includes: A. Obtain a series of data from the 3D mesh semantic segmentation sample dataset as input to the neural network model; the data includes the coordinates of the 3D mesh vertices, the vertex index of each triangle in the 3D mesh, the number of vertices, and the number of triangles.
[0047] B. Use the series of data obtained in step A to perform feature calculation and spherical harmonic function coefficient calculation; wherein the feature calculation mainly includes extracting the geometric features of the surface and calculating the normal vector of the surface.
[0048] C. The input geometric features are preliminarily processed through the initial convolutional layer to extract low-level features and map them to a high-dimensional feature space.
[0049] D. The output features of the initial convolutional layer are input into the encoder, which extracts features layer by layer through a series of convolutional layers and downsamples through maximum pooling to increase the receptive field.
[0050] E. Input the bottom-level output features of the encoder into the decoder and decode them through the decoder module. The image size is restored through upsampling operation, and the output features of the corresponding layers of the encoder are fused using jump connections to fuse the shallow information with the deep information.
[0051] F. Input the decoder output features into the prediction layer to obtain the label of each triangular face of the predicted 3D mesh data as the final prediction result.
[0052] G. Calculate the loss value between the prediction result and the true label, back-propagate to update the network weights, and repeat the training for 500 rounds to obtain the trained 3D mesh semantic segmentation neural network model.
[0053] S3. Preprocessing the three-dimensional mesh segmentation result of the prepared body, and smoothing the preprocessed three-dimensional mesh segmentation result to obtain an accurate three-dimensional mesh segmentation result.
[0054] Among them, the preprocessing includes: outlier removal processing and hole filling processing.
[0055] The clustering algorithm is used to remove outliers and fill holes in the three-dimensional mesh segmentation results of the preparation.
[0056] Specifically, according to the three-dimensional mesh segmentation result of the prepared body, a clustering algorithm based on the triangular mesh topological structure is executed to remove outliers and fill holes, wherein a clustering algorithm based on the triangular mesh topological structure is an algorithm suitable for processing triangular mesh data to find connected clusters, and the algorithm is used to cluster connected triangles in the triangular mesh, wherein connected triangles refer to triangles connected by edges.
[0057] Furthermore, the steps of performing outlier removal and hole filling processing include: S31. According to the three-dimensional mesh segmentation result of the prepared body, the original triangular mesh is mapped back, and the triangular mesh with the predicted segmentation label of 1 is screened out to form a new triangular mesh data.
[0058] S32. Execute a clustering algorithm for the new triangular mesh data in S31 to find connected triangle clusters, select the largest triangle cluster, which is the three-dimensional mesh after outliers are removed, map the index of the three-dimensional mesh back to the original three-dimensional mesh data, set the triangle label values corresponding to these indexes to 1, and set the remaining label values to 0, to obtain the predicted segmentation labels after outliers are removed.
[0059] S33, mapping back to the original triangular mesh according to the predicted segmentation label after the outliers are removed, filtering out the triangular mesh with the predicted label of 0, and forming a new triangular mesh data.
[0060] S34. Execute a clustering algorithm for the new triangular mesh data in S33 to find connected triangle clusters, select the largest triangle cluster, which is the three-dimensional mesh after the holes are filled, map the index of the three-dimensional mesh back to the original three-dimensional mesh data, set the triangle label values corresponding to these indexes to 0, and set the remaining label values to 1, to obtain the predicted segmentation labels after the holes are filled.
[0061] S35. According to the preprocessed three-dimensional mesh segmentation result, a smoothing process is performed using a fully connected conditional random field algorithm to obtain an accurate three-dimensional mesh segmentation result.
[0062] Specifically, the core idea of the fully connected conditional random field algorithm is to optimize label assignment by maximizing the conditional probability P(X|I), where X is the label set and I represents the input grid data. The fully connected conditional random field model describes the cost of label configuration through the energy function E(X), and the goal is to find the label configuration that minimizes the energy. The energy function E(X) usually contains a unary potential energy function and a binary potential energy function: ; in, Represents a univariate potential energy function, which means that when the observed value of the index i of a triangular face is y i When , the triangle belongs to the category label x i The probability of The index i of the triangle face is assigned to the label x i The probability of , I represents the input mesh data, i, j represent the i-th index and j-th index of the triangle face respectively, the value range of i, j is from 1 to N, where N is the total number of triangle faces in the input mesh data (i.e. the total number of samples); represents the binary potential energy function, x i , x j Representing two different triangular faces, the specific formula of the binary potential energy function is: ; in, represents the label compatibility term, which constrains the conduction conditions between adjacent triangle faces; f i and f j They represent the features of point i and point j respectively, K G represents the Gaussian kernel function, is the characteristic function, Represents the weight function, m represents different feature channels or feature types, and each m represents an independent Gaussian kernel function.
[0063] Furthermore, the post-processing steps of the fully connected conditional random field algorithm are: The input is the segmentation label after outlier removal and hole filling, that is, the input is a one-dimensional vector. A fully connected conditional random field model is constructed, in which the label confidence of the univariate energy term is set to 0.95, and the compatibility parameter of the binary energy term is set to 1. After 50 iterations, the optimized label is output, and the optimized label is a one-dimensional vector.
[0064] S4. According to the accurate three-dimensional mesh segmentation result, the segmentation edge is calculated to obtain the rough edge line of the preparation body.
[0065] S41. According to the accurate three-dimensional mesh segmentation result, all segmentation edge triangle meshes are obtained to form a triangle set.
[0066] Specifically, the corresponding labels of the mesh vertices are obtained through the triangular mesh segmentation labels. Assume that the original preparation mesh is O. For any triangle T, its label is u, and the value of u is 0 or 1. The label u of triangle T is corresponded to the three vertices v1, v2 and v3 of the triangle, that is, the vertex label is set to be the same as the label of the triangle. At this time, the label of each vertex of the original preparation 3D mesh data is obtained; traverse each triangle in the preparation 3D mesh data. If the three vertex labels of the triangle are different, it is considered to be a segmentation edge triangle. All segmentation edge triangle meshes are obtained to form a triangle set C. edge .
[0067] S42, calculating the centroid of each segmented edge triangle mesh according to the vertices of the segmented edge triangle mesh in the triangle set, and constructing a rough edge line scattered point set according to the centroid of the segmented edge triangle mesh as the rough edge line of the prepared body.
[0068] Specifically, Figure 3 is a schematic diagram of the rough edge line of the preparation provided by an embodiment of the present invention, see Figure 3 , for the triangle set C edge For any cut edge triangle mesh in , assuming its three vertices are v1, v2 and v3, according to the formula Calculate the centroid of the segmented edge triangle mesh, and use the centroid of the segmented edge triangle mesh as the rough edge point to form a rough edge line scattered point set C centor , and the edge line formed by the rough edge points is used as the rough edge line of the prepared body.
[0069] S5. Perform smoothing and projection processing on the rough edge line of the prepared body to obtain a smooth edge line of the prepared body on the three-dimensional grid.
[0070] S51, downsampling the rough edge line scattered point set into an unordered point set.
[0071] Specifically, Figure 4 is a schematic diagram of a set of unordered points after downsampling provided by an embodiment of the present invention, see Figure 4 , the rough edge line scattered point set C centor Put it into a large voxel grid, each voxel is a cube, that is, the entire space of scattered points is divided into several voxels. centor (x, y, z) calculates its voxel index according to the size of the given parameter voxel_size. In each voxel, calculate the center of all points contained in the voxel, which is the sampled point v sample , all v sample The downsampled point set is composed of the unordered point set C. sample .
[0072] S52. Sort the unordered point set to obtain an ordered point set that is connected end to end to form a closed curve.
[0073] Specifically, the unordered point set C sample The scattered points in the set are disordered, which is not conducive to the smoothing of the scattered points. The overall shape of the scattered points constitutes the outline of the edge line of the preparation, which is a closed curve. 01 As the initial point, use the k-dimensional tree algorithm to search for its nearest neighbor v 02 , also use the k-dimensional tree algorithm to search for v 02 Except v 01 The nearest point v outside 03 As v 03 , and to prevent v 03 and v 01 On the same side, calculate the vector (v 01 -v 02 ) and vector (v 03 -v 02 ), if the angle is greater than 90°, then v 03 Add the point set, otherwise continue to look for points other than v 01 、v 03The nearest point outside the , until the angle is greater than 90°, repeat the above steps to find v 03 , and so on, until the nth point v is found 0n The nearest neighbor point to v 01 distance, and obtain an ordered set of points that can form a closed curve by connecting end to end.
[0074] S53, performing smoothing processing by a polynomial regression algorithm according to the ordered point set to obtain a smoothed preparation body edge line point cloud.
[0075] Specifically, the ordered three-dimensional point set includes v 01 、v 02 ,...,v 0n There are n points in total. The Euclidean distance between two points is calculated and accumulated to obtain a distance set: {d0=0,d1=|v 02 -v 01 |,d2=d1+|v 03 -v 02 |,...,d n-1 =d n-2 +|v 0n -v 0n-1 |} has a total of n elements, and the sorted point set also has n elements, each element has three coordinates x, y, and z. Let {x1,x2,...,x n}, {y1,y2,...,y n} and {z1,z2,...,z n} respectively with {d0,d1,...,d n-1 Perform twelve-order polynomial regression to obtain the curve expression after regression, and interpolate the curve after regression to obtain a smooth preparation edge line point cloud P.
[0076] S54, projecting the smoothed edge line point cloud of the preparation onto the three-dimensional grid of the preparation to obtain the smooth edge line of the preparation on the three-dimensional grid.
[0077] Specifically, Figure 5 is a schematic diagram of a smooth edge line of a preparation provided by an embodiment of the present invention, Figure 6 is a schematic diagram of a smooth edge line of a preparation on a three-dimensional grid provided by an embodiment of the present invention, see Figure 5 and Figure 6 (in, Figure 6 The 3D mesh of the preparation in the figure is not rendered). According to the smoothed preparation edge line point cloud P, traverse the points in the point cloud and for any point v p Find the triangle face T in the original preparation body 3D mesh with the shortest distance to it near , and for point v p To T nearDraw a perpendicular line, the intersection point is the projection point v, and traverse all points v in the point cloud P p , the set of projection points v is the smooth edge line of the final preparation on the three-dimensional grid.
[0078] The present invention uses a 3D mesh semantic segmentation network model to accurately segment the 3D mesh of the preparation based on the intraoral scanned 3D mesh data, and performs preprocessing, smoothing, and projection processing on the segmentation results to achieve accurate, fast, and fully automated extraction of the smooth edge line of the preparation. The subsequent crown is designed and manufactured based on the smooth edge line extracted by this solution to ensure that the manufactured crown and the edge of the preparation are completely consistent. The present invention can assist in marking the edge line of the preparation, save time in the design and manufacture of the restoration, and reduce problems such as marking errors caused by subjective factors.
[0079] Figure 7 is a structural schematic diagram of a full crown restoration preparation edge line extraction system based on deep learning provided in an embodiment of the present invention, and the system is used to execute the full crown restoration preparation edge line extraction method based on deep learning described in the above embodiment, such as Figure 7 As shown, the system includes the following modules: A data acquisition module, used to acquire three-dimensional mesh data of the prepared body of the tester; A three-dimensional mesh data segmentation module, connected to the data acquisition module, is used to input the three-dimensional mesh data of the prepared body of the tester into the three-dimensional mesh semantic segmentation network model to obtain the three-dimensional mesh segmentation result of the prepared body of the tester; The segmentation result precision module is connected to the 3D mesh data segmentation module, and is used to pre-process the 3D mesh segmentation result of the preparation, and to smooth the pre-processed 3D mesh segmentation result to obtain a precise 3D mesh segmentation result; A rough edge line calculation module is connected to the segmentation result refinement module and is used to calculate the segmentation edge according to the precise three-dimensional grid segmentation result to obtain the rough edge line of the preparation body; The smooth edge line extraction module is connected to the rough edge line calculation module and is used to perform smoothing and projection processing on the rough edge line of the prepared body to obtain the smooth edge line of the prepared body on the three-dimensional grid.
[0080] It should be noted that the terms used in the present invention are only for describing specific embodiments, rather than limiting the scope of the present application. As shown in the present specification, unless the context clearly indicates an exception, the words "one", "a", "a kind of" and / or "the" do not specifically refer to the singular, but may also include the plural. The terms "include", "comprise" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method or device. In the absence of more restrictions, the elements defined by the sentence "include one..." do not exclude the presence of other identical elements in the process, method or device including the elements.
[0081] It should also be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. Unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", etc. should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the technical solutions of the embodiments of the present invention.
Claims
1. A method for extracting margin lines of full crown restoration preparations based on deep learning, characterized in that: The steps include: S1. Obtaining three-dimensional mesh data of the prepared body of the tester; S2, inputting the three-dimensional mesh data of the prepared body of the test subject into a three-dimensional mesh semantic segmentation network model to obtain a three-dimensional mesh segmentation result of the prepared body of the test subject; S3, preprocessing the three-dimensional mesh segmentation result of the prepared body, and smoothing the preprocessed three-dimensional mesh segmentation result to obtain an accurate three-dimensional mesh segmentation result; S4, calculating the segmentation edge according to the accurate three-dimensional grid segmentation result to obtain the rough edge line of the preparation; S5. Performing smoothing and projection processing on the rough edge line of the prepared body to obtain a smooth edge line of the prepared body on the three-dimensional grid.
2. The method for extracting margin lines of full crown restoration preparations based on deep learning according to claim 1, characterized in that: In S1, obtaining the three-dimensional mesh data of the prepared body of the tester includes: The intraoral scanning 3D mesh data of the single abutment tooth preparation of the tester is obtained.
3. The method for extracting margin lines of full crown restoration preparations based on deep learning according to claim 2, characterized in that: In S2, the training process of the three-dimensional grid semantic segmentation network model is as follows: S2.
1. Construct a 3D mesh semantic segmentation sample data set; wherein the sample data set includes 3D mesh data of several non-test subject preparations and segmentation labels corresponding to the 3D mesh data of several non-test subject preparations; S2.2, build a neural network model; S2.
3. Use the three-dimensional mesh semantic segmentation sample data set to train the neural network model until the loss value of the neural network model is less than the preset loss value and the segmentation accuracy of the neural network model is greater than the preset accuracy, thereby obtaining a trained three-dimensional mesh semantic segmentation neural network model.
4. The method for extracting margin lines of full crown restoration preparations based on deep learning according to claim 3, characterized in that: In S2.1, the process of constructing segmentation labels corresponding to the three-dimensional mesh data of the prepared bodies of the non-test subjects includes: S2.11, connecting the ordered points in the edge line of any of the prepared bodies of the non-test subject end to end to form a closed curve; S2.
12. According to the closed curve, the three-dimensional mesh of the prepared body of the non-tester is divided into a prepared area and a non-prepared area, and a label is assigned to each triangular mesh in the prepared area and the non-prepared area, respectively, to obtain the segmentation label corresponding to the three-dimensional mesh of the prepared body of the non-tester; wherein the prepared area refers to the area containing only the tooth body, and the non-prepared area refers to the area containing other tissues in addition to the tooth body.
5. The method for extracting margin lines of full crown restoration preparations based on deep learning according to claim 1, characterized in that: In S3, the three-dimensional mesh segmentation result of the prepared body is preprocessed, wherein the preprocessing includes: outlier removal processing and hole filling processing; The three-dimensional mesh segmentation result of the prepared body is processed by clustering algorithm to remove outliers and fill holes.
6. The method for extracting margin lines of full crown restoration preparations based on deep learning according to claim 5, characterized in that: In S3, the preprocessed three-dimensional mesh segmentation result is smoothed to obtain an accurate three-dimensional mesh segmentation result, including: According to the preprocessed 3D mesh segmentation result, smoothing is performed through a fully connected conditional random field algorithm to obtain an accurate 3D mesh segmentation result.
7. The method for extracting margin lines of full crown restoration preparations based on deep learning according to claim 6, characterized in that: In S4, calculating the segmentation edge according to the accurate three-dimensional grid segmentation result to obtain the rough edge line of the prepared body includes: S41, according to the accurate three-dimensional mesh segmentation result, obtaining all segmentation edge triangle meshes to form a triangle set; S42. Calculate the centroid of each segmented edge triangle mesh according to the vertices of the segmented edge triangle mesh in the triangle set, use the centroid of the segmented edge triangle mesh as a rough edge point to form a rough edge line scatter point set, and use the edge line formed by the rough edge points as the rough edge line of the preparation.
8. The method for extracting margin lines of full crown restoration preparations based on deep learning according to claim 7, characterized in that: In S5, performing smoothing and projection processing on the rough edge line of the prepared body to obtain a smooth edge line of the prepared body on the three-dimensional grid includes: S51, downsampling the rough edge line scattered point set into an unordered point set; S52, sorting the unordered point set to obtain an ordered point set connected end to end to form a closed curve; S53, performing smoothing processing by a polynomial regression algorithm according to the ordered point set to obtain a smoothed preparation edge line point cloud; S54, projecting the smoothed edge line point cloud of the prepared body onto the three-dimensional grid of the prepared body to obtain the smooth edge line of the prepared body on the three-dimensional grid.
9. A system for extracting margin lines of full crown restoration preparations based on deep learning, used to execute the method for extracting margin lines of full crown restoration preparations based on deep learning as described in any one of claims 1 to 8, characterized in that: The system includes the following modules: A data acquisition module, used to acquire three-dimensional mesh data of the prepared body of the tester; A three-dimensional mesh data segmentation module, connected to the data acquisition module, for inputting the three-dimensional mesh data of the prepared body of the test subject into a three-dimensional mesh semantic segmentation network model to obtain a three-dimensional mesh segmentation result of the prepared body of the test subject; a segmentation result precision module, connected to the three-dimensional mesh data segmentation module, for preprocessing the three-dimensional mesh segmentation result of the prepared body, and performing smoothing processing on the preprocessed three-dimensional mesh segmentation result to obtain a precise three-dimensional mesh segmentation result; A rough edge line calculation module is connected to the segmentation result refinement module and is used to calculate the segmentation edge according to the precise three-dimensional grid segmentation result to obtain the rough edge line of the preparation; The smooth edge line extraction module is connected to the rough edge line calculation module and is used to perform smoothing and projection processing on the rough edge line of the prepared body to obtain the smooth edge line of the prepared body on the three-dimensional grid.
Citation Information
Patent Citations
Automatic denture neck edge line extraction method based on DBSCAN clustering
CN107679536A
Digital generation type preparation method based on three-dimensional printing technology
CN111920537A
Automatic detection, generation and / or correction of tooth features in digital models
CN114730466A
Automatic generation of dental restorations using machine learning
US20230277283A1
Method for segmenting and denoising triangle mesh
US20230334627A1
Cited By
False tooth edge line determination method and device, electronic equipment and storage medium
CN119339029A
A method, device, electronic device and storage medium for determining the edge line of a denture
CN119339029B
Deep-learning-based method and system for extracting finish line of full-crown-restoration preparation
WO2026092363A1
