Method and system for reconstructing missing tooth morphology based on hierarchical feature cascade fusion strategy
Through a method based on a cascade fusion strategy of layered features, combined with voxelization processing and three-dimensional crown reconstruction network model, the Marching Cubes algorithm is used to reconstruct the three-dimensional morphology of missing teeth, which solves the problem of insufficient reconstruction details in the existing technology, and achieves efficient and accurate tooth morphology reconstruction.
Patent Information
- Application Number
- CN202310439858.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-04-23
- Publication Date
- 2025-05-06
- Estimated Expiration
- 2043-04-23
AI Technical Summary
The prior art is difficult to efficiently and with high quality to reconstruct the occlusal morphology of missing teeth, especially in preserving the details of the tooth surface.
Using a method based on a cascade fusion strategy of hierarchical features, the three-dimensional morphology of missing teeth is reconstructed by combining voxelization and three-dimensional crown reconstruction network model. The method includes voxelization treatment, feature layer extraction and cascade fusion to ensure detailed features and functionality of the reconstruction results.
It realizes efficient and precise reconstruction of missing teeth, retains the detailed characteristics of the tooth surface, improves the design quality and efficiency of the functional morphology of the occlusal surface, and avoids the "unique" design problem in traditional methods.
Smart Images
Figure CN116452772B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of crown restoration reconstruction, and more specifically to a missing tooth morphology reconstruction method and system based on a hierarchical feature cascade fusion strategy. Background Art
[0002] In order to design and produce restorations suitable for patients' chewing functions, the existing digital restoration technology has the problems of low design efficiency and poor quality, and is heavily dependent on the design experience and operation techniques of experts, resulting in the "stereotyped" occlusal morphology of the designed full crown restorations; At present, the deep learning network-assisted crown restoration technology uses the crown depth image generated by the network model to reconstruct the three-dimensional morphology of the occlusal surface of the missing teeth. This method mainly includes pre-processing of the jaw model (calculation of the depth map of the tooth occlusal surface), construction of the depth map generation network model, and post-processing of the tooth morphology (three-dimensional reconstruction of the occlusal surface based on bidirectional reversible mapping of pixel-distance, and morphological design of the connector between the occlusal surface and the cervical margin line based on the digital geometry processing method). Among them, the pre- / post-processing process requires two bidirectional reversible mappings of "three-dimensional jaw data and two-dimensional depth image", which will inevitably lead to serious loss of the functional details of the crown surface.
[0003] Therefore, neither traditional digital restoration technology nor existing crown restoration technology based on deep learning has fully realized efficient, high-quality, and personalized intelligent design of the crown morphology of missing teeth. Summary of the invention
[0004] In view of the fact that the existing methods for designing the occlusal surface morphology of missing teeth cannot efficiently and high-quality reconstruct the detailed features of the maxillofacial morphology of teeth, the present invention provides a method and system for reconstructing the morphology of missing teeth based on a hierarchical feature cascade fusion strategy to restore the complete geometric morphology of missing teeth with sufficient surface detail features.
[0005] In order to achieve the above object, the present invention provides the following technical solutions:
[0006] A missing tooth morphology reconstruction method based on a hierarchical feature cascade fusion strategy comprises the following steps:
[0007] Step 1: voxelize the three-dimensional data of the tooth to be restored;
[0008] Step 2, inputting the voxelized three-dimensional data of the tooth to be restored into the three-dimensional crown reconstruction network model to obtain the occupancy rate of the missing tooth;
[0009] Step 3: Based on the occupancy rate of the missing teeth and the three-dimensional data of the teeth to be restored after voxelization, a Marching Cubes algorithm is used to reconstruct the three-dimensional shape of the teeth to be restored after restoration.
[0010] Optionally, the three-dimensional data of the tooth to be restored includes two adjacent teeth of the missing tooth.
[0011] Optionally, the three-dimensional crown reconstruction network model includes a first enhanced residual module, a second enhanced residual module, a first convolution block, a second convolution block, a splicing layer and a third convolution block connected in sequence;
[0012] The first enhanced residual module and the second enhanced residual module are the same, both of which include a first channel and a second channel, the first channel includes a convolution Conv, a batch normalization BN, and a rectified linear unit ReLU connected in sequence, the second channel includes a hybrid attention mechanism module, a batch normalization BN, and a rectified linear unit ReLU connected in sequence, and the first channel and the second channel are connected with the maximum pooling Max Pool after the addition operation; the hybrid attention mechanism module includes a third channel and a fourth channel, the third channel includes a spatial attention network, the fourth channel includes a parallel channel attention mechanism PCAM, and the spatial attention network and the parallel channel attention mechanism PCAM are connected with the batch normalization BN of the second channel after multiplication operation; the parallel channel attention mechanism PCAM includes a fifth channel and a sixth channel, the fifth channel includes a maximum pooling Max Pool, a convolution Conv, a rectified linear unit ReLU, and a convolution Conv connected in sequence, the sixth channel includes an average pooling Avg Pool, a convolution Conv, a rectified linear unit ReLU, and a convolution Conv connected in sequence, and the fifth channel and the sixth channel are connected with the activation function Sigmod after the addition operation;
[0013] The first convolution block includes a convolution Conv, a rectified linear unit ReLU, a convolution Conv, a rectified linear unit ReLU, a batch normalization BN, and a maximum pooling Max Pool connected in sequence;
[0014] The second convolution block includes a convolution Conv, a rectified linear unit ReLU, a convolution Conv, a rectified linear unit ReLU, and a batch normalization BN connected in sequence;
[0015] The third convolution block includes convolution Conv, rectified linear unit ReLU, convolution Conv, rectified linear unit ReLU, convolution Conv, rectified linear unit ReLU, and convolution Conv, which are connected in sequence.
[0016] Optionally, in step 2, the training method of the three-dimensional crown reconstruction network model is:
[0017] Step 2.1, obtaining training samples, including a 3D data sample of a tooth to be restored and a corresponding 3D data sample of a target tooth, wherein the 3D data sample of the tooth to be restored includes two adjacent teeth of the missing tooth, and the 3D data sample of the target tooth includes the designed missing tooth and two adjacent teeth;
[0018] Step 2.2, sample preprocessing, voxel processing of the three-dimensional data sample of the tooth to be repaired; adding noise points to the point cloud data of the target tooth three-dimensional data sample, performing binary occupancy processing and implicit expression on the target tooth three-dimensional data sample, and extracting neighborhood features from the point cloud data after adding noise points to obtain the feature vector f 0 ;
[0019] Step 2.3: Compare the voxelized 3D data sample of the tooth to be restored with the feature vector f 0 The voxelized three-dimensional data sample of the tooth to be restored is input into the three-dimensional crown reconstruction network model, and passes through the first enhanced residual module, the second enhanced residual module, the first convolution block and the second convolution block in sequence, wherein the first enhanced residual module obtains the feature vector f 1 , the second enhanced residual module obtains the feature vector f 2 , the first convolution block obtains the feature vector f 3 , the second convolution block obtains the feature vector f 4 , the eigenvector f 0 Respectively with the eigenvector f 1 , eigenvector f 2 , eigenvector f 3 , eigenvector f 4 The fusion is performed to obtain four fused feature vectors, and the four fused feature vectors are all input into the splicing layer for splicing to obtain a total feature vector. The total feature vector passes through the third convolution block to obtain the occupancy rate of missing teeth.
[0020] Optionally, the implicit expression uses a potential vector Encode the target tooth 3D data sample:
[0021]
[0022] Where p∈R 3 is a given query point; V represents a vector set; R 3 represents a vector space domain.
[0023] Optionally, the neighborhood feature extraction method is a trilinear difference method.
[0024] Optionally, a binary cross entropy loss (BCE loss) is used to measure the difference between the occupancy rate and the implicit expression of the target tooth three-dimensional data sample.
[0025] Optionally, in step 3, based on the occupancy rate of the missing tooth and the three-dimensional data of the tooth to be restored after voxel processing, the method for reconstructing the three-dimensional shape of the tooth to be restored after restoration is:
[0026] First, create a 256×256×256 cubic grid as the volume space for generating teeth;
[0027] Based on the occupancy rate of the missing teeth and the three-dimensional data of the teeth to be restored after voxelization, the marching cubes algorithm is used to reconstruct the three-dimensional shape of the teeth to be restored after restoration.
[0028] A missing tooth morphology reconstruction system based on hierarchical feature cascade fusion strategy, comprising:
[0029] A voxel processing module, used for voxel processing of the three-dimensional data of the tooth to be restored;
[0030] A missing tooth occupancy rate determination module is used to input the voxelized three-dimensional data of the tooth to be restored into the three-dimensional crown reconstruction network model to obtain the occupancy rate of the missing tooth;
[0031] The tooth reconstruction module is used to reconstruct the three-dimensional shape of the tooth to be restored after restoration by using the Marching Cubes algorithm based on the occupancy rate of the missing tooth and the three-dimensional data of the tooth to be restored after voxelization.
[0032] It can be seen from the above technical solutions that the present invention provides a missing tooth morphology reconstruction method and system based on a hierarchical feature cascade fusion strategy, which has the following beneficial effects compared with the prior art:
[0033] (1) The present invention adopts the implicit encoding technology of tooth morphology based on the placeholder network for implicit expression, which can better describe the detailed features of the tooth surface and assist the three-dimensional crown reconstruction network model to reconstruct more detailed features of the occlusal surface function.
[0034] (2) The present invention adopts a hierarchical feature cascade fusion strategy to progressively fuse tooth features with different resolutions extracted from the feature layer to achieve accurate reconstruction of tooth functional detail features.
[0035] (3) The present invention proposes an enhanced residual module based on a hybrid attention mechanism module, which further improves the feature expression ability of the three-dimensional crown reconstruction network model.
[0036] (4) The present invention can reconstruct the overall shape of a personalized missing tooth restoration with more natural tooth functional anatomical features, avoiding the problem that digital restoration technology and existing tooth restorations designed based on deep learning technology are difficult to directly adapt to the patient's normal chewing function, reducing the occlusal surface adjustment work and intermediate connector design work, and improving the design quality and efficiency of the occlusal surface functional shape. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0038] Figure 1 It is a schematic diagram of the method principle of the present invention;
[0039] FIG2( a ) is a schematic diagram of a target tooth three-dimensional data sample;
[0040] FIG2( b ) is a schematic diagram of a three-dimensional data sample of a tooth to be restored;
[0041] FIG2(c) is a schematic diagram of the voxelization result of the three-dimensional data sample of the tooth to be restored;
[0042] FIG2(d) is a schematic diagram of the implicit expression result of the target tooth 3D data sample;
[0043] Figure 3(a) is a structural diagram of the enhanced residual module;
[0044] Figure 3(b) is a diagram of the parallel channel attention mechanism PCAM structure;
[0045] Figure 4(a) is a schematic diagram of a cubic grid;
[0046] FIG4( b ) is a schematic diagram of the three-dimensional morphology of the reconstructed tooth;
[0047] FIG4( c ) is a schematic diagram showing the magnified result of the three-dimensional morphological details of the reconstructed tooth;
[0048] FIG4( d ) is a schematic diagram of the crown reconstruction result;
[0049] FIG5( a ) is a schematic diagram of a target natural tooth;
[0050] Figure 5(b) is a schematic diagram of the reconstructed crown restoration. DETAILED DESCRIPTION
[0051] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 are within the scope of protection of the present invention.
[0052] The embodiment of the present invention discloses a method for reconstructing missing tooth morphology based on a hierarchical feature cascade fusion strategy, see Figure 1 , including the following steps:
[0053] Step 1: voxelize the three-dimensional data of the tooth to be restored; the three-dimensional data of the tooth to be restored includes two adjacent teeth of the missing tooth.
[0054] Step 2: Input the voxelized three-dimensional data of the tooth to be restored into the three-dimensional crown reconstruction network model to obtain the occupancy rate of the missing tooth.
[0055] Step 3: First, establish a 256×256×256 cubic grid (see Figure 4(a)) as the volume space for generating teeth; then, based on the occupancy rate of the missing teeth and the three-dimensional data of the teeth to be restored after voxelization, use the Marching Cubes algorithm to reconstruct the three-dimensional shape of the teeth to be restored after restoration, see Figure 4(b), Figure 4(c), and Figure 4(d).
[0056] The three-dimensional crown reconstruction network model proposed in this invention combines the hierarchical feature cascade fusion strategy to reconstruct the functional anatomical morphology of the missing tooth surface. The multi-branch structure of the network model can make full use of the hierarchical features of each feature extraction module to achieve more accurate reconstruction of tooth surface features. Figure 1 , including a first enhanced residual module, a second enhanced residual module, a first convolution block, a second convolution block, a splicing layer and a third convolution block connected in sequence;
[0057] The first enhanced residual module is the same as the second enhanced residual module, which introduces a hybrid attention mechanism module into the ordinary residual network to construct an enhanced residual module to further improve the feature expression ability of the network. The enhanced residual module includes a first channel and a second channel, see Figure 3 (a), the first channel includes a convolution Conv, a batch normalization BN, and a rectified linear unit ReLU connected in sequence, the second channel includes a hybrid attention mechanism module, a batch normalization BN, and a rectified linear unit ReLU connected in sequence, the first channel and the second channel are added and connected to the maximum pooling Max Pool; the hybrid attention mechanism module includes a third channel and a fourth channel, the third channel includes a spatial attention network, the fourth channel includes a parallel channel attention mechanism PCAM, the spatial attention network and the parallel channel attention mechanism PCAM are multiplied and connected to the batch normalization BN of the second channel; the parallel channel attention mechanism PCAM includes a fifth channel and a sixth channel, see Figure 3 (b), the fifth channel includes a maximum pooling Max Pool, a convolution Conv, a rectified linear unit ReLU, and a convolution Conv connected in sequence, the sixth channel includes an average pooling Avg Pool, a convolution Conv, a rectified linear unit ReLU, and a convolution Conv connected in sequence, the fifth channel and the sixth channel are added and connected to the activation function Sigmod;
[0058] The first convolution block includes a convolution Conv, a rectified linear unit ReLU, a convolution Conv, a rectified linear unit ReLU, a batch normalization BN, and a maximum pooling Max Pool connected in sequence;
[0059] The second convolution block includes a convolution Conv, a rectified linear unit ReLU, a convolution Conv, a rectified linear unit ReLU, and a batch normalization BN connected in sequence;
[0060] The third convolution block includes convolution Conv, rectified linear unit ReLU, convolution Conv, rectified linear unit ReLU, convolution Conv, rectified linear unit ReLU, and convolution Conv, which are connected in sequence.
[0061] Optionally, in step 2, the training method of the three-dimensional crown reconstruction network model is:
[0062] Step 2.1, obtaining training samples, including a 3D data sample of a tooth to be restored and a corresponding 3D data sample of a target tooth, see FIG2(a) and FIG2(b), wherein the 3D data sample of the tooth to be restored includes two adjacent teeth of the missing tooth, and the 3D data sample of the target tooth includes the designed missing tooth and two adjacent teeth;
[0063] Step 2.2, sample preprocessing: voxelize the three-dimensional data sample of the tooth to be repaired, see Figure 2(c), voxel resolution: 128×128×128; perform two processes on the target tooth three-dimensional data sample: (1) first add noise points to its point cloud data, the purpose is to enhance the generalization ability of the network model, and then perform binary occupancy processing and implicit expression on the target tooth three-dimensional data sample. The implicit expression uses a ray tracing method to classify the points in the tooth point cloud data into surface internal points (marked as 1, see the red points in Figure 2(d)) and surface external points (marked as 0, see the green points in Figure 2(d)), so as to better describe the detailed features of the tooth surface; (2) perform neighborhood feature extraction on the point cloud data after adding noise points, mainly using trilinear interpolation to extract tooth features, and obtain the feature vector f 0 , allowing voxel features (neighboring teeth) to fit the target tooth point cloud features, and then establish the jaw relationship between the missing teeth and the adjacent teeth. In addition, it is used to assist the tooth features extracted from different feature layers to be more easily combined.
[0064] Implicit expression: This embodiment proposes a tooth morphology implicit encoding technology based on a placeholder network to obtain a continuous expression of tooth shape and improve the ability of the network model to effectively capture the fine-grained features of complex tooth surfaces. Encode the target tooth 3D data sample:
[0065]
[0066] Where p∈R 3 is a given query point; V represents a vector set; R 3 Represents the vector space domain. The points are classified as inside the surface (classified as 1) or outside the surface (classified as 0). Therefore, the implicit expression of the surface can be represented as a point on the decision boundary, {p∈R 3 |f(v,p)=t}, and the threshold parameter is experimentally verified to be t=0.5. Therefore, once f(·) is learned by the network, it can be queried at continuous point locations without being restricted by the resolution of a typical voxel grid. This representation can be used to express models with complex structures and preserve the detailed features in the model.
[0067] Step 2.3: Compare the voxelized 3D data sample of the tooth to be restored with the feature vector f 0 The voxelized three-dimensional data sample of the tooth to be restored is input into the three-dimensional crown reconstruction network model, and passes through the first enhanced residual module, the second enhanced residual module, the first convolution block and the second convolution block in sequence, wherein the first enhanced residual module obtains the feature vector f 1 , the second enhanced residual module obtains the feature vector f 2, the first convolution block obtains the feature vector f 3 , the second convolution block obtains the feature vector f 4 , the eigenvector f 0 Respectively with the eigenvector f 1 , eigenvector f 2 , eigenvector f 3 , eigenvector f 4 The fusion is performed to obtain four fusion feature vectors, and the feature vector f of the voxelized three-dimensional data sample of the tooth to be repaired is constructed. 1 , eigenvector f 2 , eigenvector f 3 , eigenvector f 4 and the target tooth 3D data sample feature vector f 0 The connection between them is concatenated into a total feature vector containing low-level and high-level feature information through the concatenation layer to enhance the geometric and semantic feature information in the final feature map. The total feature vector then passes through the third convolution block to obtain the occupancy rate of missing teeth.
[0068] The above results use binary cross entropy loss BCE loss to measure the difference between the occupancy rate and the implicit expression of the target tooth 3D data sample.
[0069] The occlusal grooves and cusp distribution on the crown surface obtained by the present invention are very close to those of the target tooth, see Figure 5(a) and Figure 5(b), and the present invention can also reconstruct the buccal side, lingual side and mesiodistal side of the crown together, and the four side surfaces of the crown obtained by the present invention are more in line with the physiological needs of the mouth and jaw than the side surfaces designed by the CAD method.
[0070] A missing tooth morphology reconstruction system based on hierarchical feature cascade fusion strategy, comprising:
[0071] A voxel processing module, used for voxel processing of the three-dimensional data of the tooth to be restored;
[0072] A missing tooth occupancy rate determination module is used to input the voxelized three-dimensional data of the tooth to be restored into the three-dimensional crown reconstruction network model to obtain the occupancy rate of the missing tooth;
[0073] The tooth reconstruction module is used to reconstruct the three-dimensional shape of the tooth to be restored after restoration by using the Marching Cubes algorithm based on the occupancy rate of the missing tooth and the three-dimensional data of the tooth to be restored after voxelization.
[0074] As for the system module disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.
[0075] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0076] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A missing tooth morphology reconstruction method based on hierarchical feature cascade fusion strategy, characterized in that: The following steps are involved: Step 1: voxelize the three-dimensional data of the tooth to be restored; Step 2, inputting the voxelized three-dimensional data of the tooth to be restored into the three-dimensional crown reconstruction network model to obtain the occupancy rate of the missing tooth; In step 2, the training method of the three-dimensional crown reconstruction network model is: Step 2.1, obtaining training samples, including a 3D data sample of a tooth to be restored and a corresponding 3D data sample of a target tooth, wherein the 3D data sample of the tooth to be restored includes two adjacent teeth of the missing tooth, and the 3D data sample of the target tooth includes the designed missing tooth and two adjacent teeth; Step 2.2, sample preprocessing, voxelizing the three-dimensional data sample of the tooth to be repaired; Adding noise points to the point cloud data of the target tooth three-dimensional data sample, performing binary occupancy processing and implicit expression on the target tooth three-dimensional data sample, and performing neighborhood feature extraction on the point cloud data after adding the noise points to obtain a feature vector f0; Step 2.3, input the voxelized three-dimensional data sample of the tooth to be repaired and the feature vector f0 into the three-dimensional crown reconstruction network model, the voxelized three-dimensional data sample of the tooth to be repaired passes through the first enhanced residual module, the second enhanced residual module, the first convolution block and the second convolution block in sequence, wherein the first enhanced residual module obtains the feature vector f1, the second enhanced residual module obtains the feature vector f2, the first convolution block obtains the feature vector f3, and the second convolution block obtains the feature vector f4, the feature vector f0 is fused with the feature vector f1, the feature vector f2, the feature vector f3, and the feature vector f4 respectively to obtain four fused feature vectors, the four fused feature vectors are all input into the splicing layer for splicing to obtain the total feature vector, the total feature vector passes through the third convolution block to obtain the occupancy rate of the missing teeth; Step 3: Based on the occupancy rate of the missing teeth and the three-dimensional data of the teeth to be restored after voxelization, a Marching Cubes algorithm is used to reconstruct the three-dimensional shape of the teeth to be restored after restoration.
2. A missing tooth morphology reconstruction method based on hierarchical feature cascade fusion strategy according to claim 1, characterized in that: The three-dimensional data of the tooth to be restored includes two adjacent teeth of the missing tooth.
3. The missing tooth morphology reconstruction method based on hierarchical feature cascade fusion strategy according to claim 1, characterized in that: The three-dimensional crown reconstruction network model includes a first enhanced residual module, a second enhanced residual module, a first convolution block, a second convolution block, a splicing layer and a third convolution block connected in sequence; The first enhanced residual module and the second enhanced residual module are the same, both of which include a first channel and a second channel, the first channel includes a convolution Conv, a batch normalization BN, and a rectified linear unit ReLU connected in sequence, the second channel includes a hybrid attention mechanism module, a batch normalization BN, and a rectified linear unit ReLU connected in sequence, and the first channel and the second channel are connected with the maximum pooling Max Pool after the addition operation; the hybrid attention mechanism module includes a third channel and a fourth channel, the third channel includes a spatial attention network, the fourth channel includes a parallel channel attention mechanism PCAM, and the spatial attention network and the parallel channel attention mechanism PCAM are connected with the batch normalization BN of the second channel after multiplication operation; the parallel channel attention mechanism PCAM includes a fifth channel and a sixth channel, the fifth channel includes a maximum pooling Max Pool, a convolution Conv, a rectified linear unit ReLU, and a convolution Conv connected in sequence, the sixth channel includes an average pooling Avg Pool, a convolution Conv, a rectified linear unit ReLU, and a convolution Conv connected in sequence, and the fifth channel and the sixth channel are connected with the activation function Sigmod after the addition operation; The first convolution block includes a convolution Conv, a rectified linear unit ReLU, a convolution Conv, a rectified linear unit ReLU, a batch normalization BN, and a maximum pooling Max Pool connected in sequence; The second convolution block includes a convolution Conv, a rectified linear unit ReLU, a convolution Conv, a rectified linear unit ReLU, and a batch normalization BN connected in sequence; The third convolution block includes convolution Conv, rectified linear unit ReLU, convolution Conv, rectified linear unit ReLU, convolution Conv, rectified linear unit ReLU, and convolution Conv, which are connected in sequence.
4. The missing tooth morphology reconstruction method based on hierarchical feature cascade fusion strategy according to claim 1, characterized in that: The implicit representation uses a latent vector Encode the target tooth 3D data sample: Where p∈R 3 is a given query point; V represents a vector set; R 3 represents a vector space domain.
5. The missing tooth morphology reconstruction method based on hierarchical feature cascade fusion strategy according to claim 1, characterized in that: The method for extracting neighborhood features is a trilinear difference method.
6. The missing tooth morphology reconstruction method based on hierarchical feature cascade fusion strategy according to claim 1, characterized in that: The binary cross entropy loss (BCEloss) is used to measure the difference between the occupancy rate and the implicit expression of the target tooth 3D data sample.
7. The missing tooth morphology reconstruction method based on hierarchical feature cascade fusion strategy according to claim 1, characterized in that: In step 3, based on the occupancy rate of the missing teeth and the three-dimensional data of the teeth to be restored after voxel processing, the method for reconstructing the three-dimensional shape of the teeth to be restored after restoration is: First, create a 256×256×256 cubic grid as the volume space for generating teeth; Based on the occupancy rate of the missing teeth and the three-dimensional data of the teeth to be restored after voxelization, the marching cubes algorithm is used to reconstruct the three-dimensional shape of the teeth to be restored after restoration.
8. A missing tooth morphology reconstruction system based on hierarchical feature cascade fusion strategy, characterized in that: include: A voxel processing module, used for voxel processing of the three-dimensional data of the tooth to be restored; A missing tooth occupancy rate determination module is used to input the voxelized three-dimensional data of the tooth to be restored into the three-dimensional crown reconstruction network model to obtain the occupancy rate of the missing tooth; The training method of the 3D crown reconstruction network model is: Step 2.1, obtaining training samples, including a 3D data sample of a tooth to be restored and a corresponding 3D data sample of a target tooth, wherein the 3D data sample of the tooth to be restored includes two adjacent teeth of the missing tooth, and the 3D data sample of the target tooth includes the designed missing tooth and two adjacent teeth; Step 2.2, sample preprocessing, voxelizing the three-dimensional data sample of the tooth to be repaired; Adding noise points to the point cloud data of the target tooth three-dimensional data sample, performing binary occupancy processing and implicit expression on the target tooth three-dimensional data sample, and performing neighborhood feature extraction on the point cloud data after adding the noise points to obtain a feature vector f0; Step 2.3, input the voxelized three-dimensional data sample of the tooth to be repaired and the feature vector f0 into the three-dimensional crown reconstruction network model, the voxelized three-dimensional data sample of the tooth to be repaired passes through the first enhanced residual module, the second enhanced residual module, the first convolution block and the second convolution block in sequence, wherein the first enhanced residual module obtains the feature vector f1, the second enhanced residual module obtains the feature vector f2, the first convolution block obtains the feature vector f3, and the second convolution block obtains the feature vector f4, the feature vector f0 is fused with the feature vector f1, the feature vector f2, the feature vector f3, and the feature vector f4 respectively to obtain four fused feature vectors, the four fused feature vectors are all input into the splicing layer for splicing to obtain the total feature vector, the total feature vector passes through the third convolution block to obtain the occupancy rate of the missing teeth; The tooth reconstruction module is used to reconstruct the three-dimensional shape of the tooth to be restored after restoration by using the Marching Cubes algorithm based on the occupancy rate of the missing tooth and the three-dimensional data of the tooth to be restored after voxelization.