Dental restoration and model training method, electronic equipment, storage medium and product
By generating a simulated dataset of missing teeth based on preset rules and training a tooth restoration model using the PointAttn model, the problem of insufficient accuracy and applicability of tooth restoration model training in existing technologies is solved, and a more comprehensive tooth restoration effect is achieved.
Patent Information
- Application Number
- CN202411053814.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-01
- Publication Date
- 2026-02-03
AI Technical Summary
Existing technologies for training dental restoration models lack accuracy and applicability, especially in the restoration of teeth after caries and attachment removal. The artificially restored teeth are not exactly the same as the original teeth, leading to inaccurate restoration results.
By processing individual complete teeth in user historical data based on preset removal rules, simulated missing teeth are obtained. The data information of individual complete teeth and simulated missing teeth are used as datasets to train a point cloud restoration model. The PointAttn model is used to train the tooth restoration model. By combining preset classification standards and binary classification models, the consistency and accuracy of input and output data are ensured.
It enables more comprehensive restoration of missing teeth, the restoration results are closer to the actual situation, and the accuracy and applicability of the repair are improved.
Smart Images
Figure CN121458931A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The embodiment of the present application relates to the technical field of tooth orthodontics, in particular to a tooth repair and model training method, an electronic device, a storage medium and a product. BACKGROUND
[0002] In the production of the appliance, the missing part in the tooth model needs to be repaired, and there are usually two situations that need to be repaired, one is to repair the caries, and the other is to repair the original accessory part after removing the accessory from the digital oral scanning data. The existing tooth repair for missing teeth is usually manually repaired in software, so it is time-consuming and laborious.
[0003] In order to solve the above problems, the prior art will first complete the missing tooth by using software, and then train the model according to the tooth information after completion and the missing tooth information. Subsequently, the missing tooth can be directly generated after repair according to the trained model. However, the inventors found that in the prior art, the model can only be trained according to the actual obtained missing tooth information, and it is difficult to obtain more comprehensive different types of missing information. In addition, the completed tooth as the input of the model is also completed by artificial, and it cannot be guaranteed that the completed tooth is the same as the original tooth, so the accuracy of the trained model cannot be guaranteed. Therefore, the scope and accuracy of the repair method in the prior art cannot be guaranteed. SUMMARY
[0004] The purpose of the embodiment of the present application is to provide a tooth repair and model training method, an electronic device, a storage medium and a product, which can repair more comprehensive missing types and the result of the missing tooth repair is also closer to the real tooth.
[0005] To solve the above technical problems, the embodiment of the present application provides a tooth repair model training method, comprising:
[0006] Based on the preset removal rule, a single complete tooth in the user historical data is processed to obtain a simulated missing tooth, wherein the simulated missing tooth is a tooth with part of the crown removed according to the preset removal rule; and the data information of the single complete tooth and the data information of the simulated missing tooth are used as a data set to train a point cloud repair model to obtain the tooth repair model training method.
[0007] The embodiment of the present application further provides a tooth repair method, comprising: obtaining a tooth to be repaired; and repairing the tooth to be repaired by using the point cloud repair model trained by the tooth repair model training method as described above to obtain a corresponding completed tooth.
[0008] The embodiment of the present application also provides an electronic device, comprising: at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the tooth restoration model training method or the tooth restoration method.
[0009] The embodiment of the present application also provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the tooth restoration model training method or the tooth restoration method.
[0010] The embodiment of the present application also provides a computer, which comprises computer instructions, and the computer instructions are executed by a processor to implement the tooth restoration model training method or the tooth restoration method.
[0011] In the embodiment of the present application, a single complete tooth in user historical data is processed based on a preset removal rule to obtain a simulated missing tooth, wherein the simulated missing tooth is a tooth with a part of a tooth crown removed according to the preset removal rule; and data information of the single complete tooth and data information of the simulated missing tooth are used as a data set training point cloud restoration model to obtain the tooth restoration model training method. The data set for training is constructed by removing the tooth crown part of the complete tooth in various ways to obtain the missing tooth, and various missing teeth do not need to be actually obtained, so more comprehensive missing types can be included in the point cloud restoration model. In addition, in the prior art, the complete tooth after artificial completion is used for training, and the completed tooth may not be completely the same as the original tooth due to various reasons. The present application trains the complete tooth in the historical data, so that the result of the missing tooth restoration is closer to the real situation.
[0012] In addition, in one example, the tooth restoration model training method processes a single complete tooth in user historical data based on a preset removal rule, comprising: when the obtained simulated missing tooth is a first missing tooth, forming holes corresponding to the shapes of different accessories on the tooth crown surface of the complete tooth according to the shapes of the removed different accessories to obtain missing tooth crowns of different shapes; wherein the first missing tooth is a tooth missing due to removal of accessories. For the tooth model missing due to removal of accessories, holes corresponding to the shapes can be formed on the tooth crown to simulate the missing.
[0013] In addition, in one example, the dental restoration model training method further includes, when the simulated missing tooth is a second missing tooth, cutting the tooth crown of the complete tooth from different angles using a cutting surface in a three-dimensional space to obtain different forms of missing tooth crowns, wherein the cutting surface is a plane or a curved surface of different shapes, and the second missing tooth is a tooth missing due to non-removable attachments. For tooth models missing due to non-removable attachments, a plane or a curved surface can be used to simulate the missing tooth. The cutting method is diversified and can be applied to more missing types.
[0014] In addition, in one example, the dental restoration model training method further includes, when the simulated missing tooth is a second missing tooth, cutting the tooth crown of the complete tooth from different angles using a cutting surface in a three-dimensional space to obtain different forms of missing tooth crowns, wherein the cutting surface is a plane or a curved surface of different shapes, and the second missing tooth is a tooth missing due to non-removable attachments. For tooth models missing due to non-removable attachments, a plane or a curved surface can be used to simulate the missing tooth. The cutting method is diversified and can be applied to more missing types.
[0015] In addition, in one example, the dental restoration model training method further includes, when the simulated missing tooth is a second missing tooth, cutting the tooth crown of the complete tooth from different angles using a cutting surface in a three-dimensional space to obtain different forms of missing tooth crowns, wherein the cutting surface is a plane or a curved surface of different shapes, and the second missing tooth is a tooth missing due to non-removable attachments. For tooth models missing due to non-removable attachments, a plane or a curved surface can be used to simulate the missing tooth. The cutting method is diversified and can be applied to more missing types.
[0016] In addition, in one example, the dental restoration model training method further includes, when the simulated missing tooth is a second missing tooth, cutting the tooth crown of the complete tooth from different angles using a cutting surface in a three-dimensional space to obtain different forms of missing tooth crowns, wherein the cutting surface is a plane or a curved surface of different shapes, and the second missing tooth is a tooth missing due to non-removable attachments. For tooth models missing due to non-removable attachments, a plane or a curved surface can be used to simulate the missing tooth. The cutting method is diversified and can be applied to more missing types.
[0017] In addition, in one example, the tooth restoration method adopts the point cloud restoration model trained by the tooth restoration model training method to restore the tooth to be restored, including: when the tooth to be restored is a defective tooth; points in the tooth to be restored point cloud are divided into third type points and fourth type points by a pre-trained second binary classification model; the third type points include vertices of the defective surface in the point cloud, and the fourth type points include the remaining points in the point cloud; and the point cloud from which the third type points are removed is input into the restoration model after training. When the tooth is restored, which is not defective due to an accessory, the vertices of the defective surface are removed in advance by the binary classification model to ensure that the input information and the data distribution input when the model is trained are consistent, thereby improving the restoration accuracy.
[0018] In addition, in one example, the tooth restoration method adopts the point cloud restoration model trained by the tooth restoration model training method to restore the tooth to be restored, including: when the tooth to be restored is a defective tooth; points in the tooth to be restored point cloud are divided into third type points and fourth type points by a pre-trained second binary classification model; the third type points include vertices of the defective surface in the point cloud, and the fourth type points include the remaining points in the point cloud; and the point cloud from which the third type points are removed is input into the restoration model after training. When the tooth is restored, which is not defective due to an accessory, the vertices of the defective surface are removed in advance by the binary classification model to ensure that the input information and the data distribution input when the model is trained are consistent, thereby improving the restoration accuracy.
[0019] In addition, in one example, the tooth restoration method further includes, before registration: determining the dental axis of the tooth to be restored and the complete tooth, respectively; and aligning the dental axes of the tooth to be restored and the complete tooth. By pre-aligning the dental axes, the approximate pose can be determined, and the subsequent registration is more efficient. BRIEF DESCRIPTION OF DRAWINGS
[0020] One or more embodiments are illustrated by way of example in the figures that are part of this disclosure and which are illustrative, but not restrictive, of the embodiments, wherein elements having the same reference numbers designate corresponding elements and wherein the notation “some” or “one or more” or “at least one” shall not be construed as excluding the presence of zero of the element in question. The figures in the drawings are not to scale.
[0021] Figure 1 is a flowchart of a tooth restoration model training method according to an embodiment of the present application;
[0022] Figure 2 is a structural schematic diagram of a PointAttn model according to an embodiment of the present application;
[0023] Figure 3 is a structural schematic diagram of a geometric detail perception machine (GDP) module according to an embodiment of the present application;
[0024] Figure 4 is a structural schematic diagram of a self-feature enhancement module according to an embodiment of the present application;
[0025] Figure 5is a flow chart of a tooth restoration method according to an embodiment of the present application;
[0026] Figure 6 is a schematic diagram of an electronic device structure according to an embodiment of the present application. DETAILED DESCRIPTION
[0027] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below with reference to the drawings. However, those skilled in the art can understand that, in the embodiments of the present application, many technical details are presented in order to make the reader better understand the present application. However, the technical solutions claimed by the present application can be implemented even without these technical details and based on various changes and modifications of the following embodiments. The division of the following embodiments is for the convenience of description, and should not constitute any limitation on the specific embodiments of the present application, and the embodiments can be combined and referenced with each other without contradiction.
[0028] In existing appliance production, a hot pressing film process is usually used, and the hot pressing film work needs to obtain a physical model of teeth in advance and then make the appliance on the physical model. When the physical model is incomplete, the appliance will be concave. In order to avoid the concave during the appliance production by the hot pressing film forming process, the appliance cannot be finally worn into the patient's mouth, and the incomplete physical model needs to be repaired in the appliance production.
[0029] In the present application, the physical model is obtained according to a three-dimensional tooth digital model manufactured by scanning the mouth or a plaster model. The three-dimensional tooth digital model can be directly repaired, wherein the incomplete teeth mainly include caries, the hollow tooth crown left on the tooth surface in the three-dimensional model after removing the virtual accessory, and some patients have part of the tooth loss due to other factors.
[0030] One embodiment of the present invention relates to a method for training a tooth restoration model, which can be applied to terminals capable of model training, such as computers. In this embodiment, a single complete tooth in a user's historical data is processed based on preset removal rules to obtain a simulated missing tooth. The simulated missing tooth is a tooth whose crown has been partially removed according to the preset removal rules. The point cloud restoration model is trained using the data information of the single complete tooth and the data information of the simulated missing tooth as a dataset to obtain the tooth restoration model training method. By using a complete tooth to remove the crown portion in various ways to obtain the missing tooth dataset for training, it is not necessary to actually obtain various types of missing teeth, thus allowing for a wider and more comprehensive inclusion of different types of missing teeth in the point cloud restoration model. Furthermore, in existing technologies, training is performed using artificially restored complete teeth, which may result in the restored tooth not being entirely identical to the original tooth due to various reasons. The present invention trains using complete teeth from historical data, making the restored result closer to reality. The implementation details of the tooth restoration model training method of this embodiment are described below. The following details are provided for ease of understanding and are not essential for implementing this solution.
[0031] like Figure 1 As shown, in step 101, a single complete tooth in the user's historical data is processed based on a preset removal rule to obtain a simulated missing tooth. The simulated missing tooth is a tooth whose crown has been partially removed according to the preset removal rule. The user's historical data consists of virtual 3D models of the complete tooth morphology of various patients collected.
[0032] In one example, missing teeth can be mainly divided into two categories. One category consists of missing teeth models where virtual attachments are removed from the 3D model, resulting in gaps on the tooth surface. This is because the need for attachments varies at different stages of orthodontic treatment; sometimes, the original attachments need to be removed. In the 3D intraoral digital model, this is equivalent to removing the mesh of that attachment, leaving voids and causing tooth defects in the 3D model (but the actual patient's teeth will not be missing). The other category consists of missing teeth without removing attachments. This mainly includes cavities and teeth partially missing due to other factors. The patient's actual teeth are missing, and the resulting 3D tooth model will also be incomplete. During orthodontic treatment, these missing teeth need to be restored to a complete tooth.
[0033] Specifically, when the obtained simulation missing tooth is the first missing tooth, the first missing tooth is a tooth missing due to removal of an attachment, and according to the shape of the removed different attachment, a hole similar to a rectangle, a semicircle or a crescent can be randomly dug on the buccal side (or lingual side) surface of the complete tooth crown part to simulate the shape of the base (the contact surface of the attachment and the tooth crown) of the commonly used attachment on the tooth crown surface. Here, the complete tooth is obtained from the current or historical tooth data of other users, which is beneficial to make the result after repairing the missing tooth closer to the real situation. Alternatively, when the obtained simulation missing tooth is the second missing tooth, the second missing tooth is a tooth missing due to non-removal of an attachment, such as a missing tooth similar to a caries. Considering that the missing tooth has various shapes, such as missing an angle, or missing more than half of the tooth, or having a hole in the middle, etc., a plane or a curved surface of different shapes can be used to cut the complete tooth crown from multiple different angles in three-dimensional space, so that the generated tooth crown presents different shapes to simulate the actual missing tooth crown. The same tooth crown can generate multiple different missing teeth. In addition, similar methods are used for different types of tooth crowns, such as incisors, canines, premolars and molars.
[0034] After obtaining the simulation missing tooth, the tooth repair model training method is obtained by training the point cloud repair model with the data information of the single complete tooth and the data information of the simulation missing tooth as the data set training points.
[0035] In one example, after obtaining the data information of the complete tooth and the simulation missing tooth, since the input and output scales of the neural network model are fixed, the input data and labels in the generated data set also need to be sampled. There are many specific sampling methods, such as uniform sampling, random sampling, and sampling based on geometric features. For example, for incisors and canines, we can choose a sampling method for sampling. For premolars and molars, since the occlusal surfaces of these two types of teeth have many details, such as multiple dental grooves in the first permanent tooth, and the sampling based on geometric features can better preserve the details of the point cloud data, so the sampling based on geometric features is used for these two types of teeth. In addition, in order to better preserve the topological structure of the missing tooth, it is necessary to ensure that some or all of the sampling points on the missing tooth and the sampling points on the complete tooth are the same. The following method can be used to select the sampling points: first, sample M points on the tooth crown of the complete tooth, and then select the points corresponding to the structure of the missing tooth from these sampling points. If the number of selected missing tooth points is not enough, continue to sample in the missing tooth point cloud until the number of points corresponding to the size of the model input is met.
[0036] Specifically, after the sampling is completed and the data set is prepared, the data set can be used for training of the point cloud repair model. There are many models for point cloud repair and completion, such as PointAttn (attention point cloud model) proposed in recent years, FBNet (lightweight convolutional neural network architecture), SPCNet (spatial pyramid clustering network), AdaPointTr (adaptive point cloud transformation network), and CRA-PCN (cross-resolution point cloud completion model), etc. In the present embodiment, the PointAttn model is taken as an example for illustration, and a structural diagram of the PointAttn model is shown in Figure 2 The overall structure of the model is divided into encoding and decoding structures. The encoding structure is composed of a feature extractor, and the decoding structure is divided into two parts, one part being a seed generator (Seed generator) and the other part being an up-sampled point cloud generator (Point generator). The feature extractor is mainly composed of a geometric detail perception (GDP) module and a self-feature augmentation module, and also contains some multi-layer perception (MLP) layers. The structure of the geometric detail perception (GDP) module is generally similar to the encoder in the Transformer structure, as shown in Figure 3 The module mainly accumulates the information of unordered point clouds through cross-attention mechanism. For the geometric detail perception (GDP) module, the input X is sampled at a down-sampling rate d to obtain the input Y, then X and Y are input into the multi-head self-attention block, the output and q are spliced to obtain the feature F, F is input into the feed forward network (FFN), and after passing through the residual connection structure, it is spliced with the input q again to obtain the final output of the GDP module. The mathematical formula is as follows:
[0037] F=Norm(q+MultiHead(q,K,V))
[0038] q=YW q ,K=XW KV ,V=XW KV
[0039] GDP(X,d)=Concat(F+FFN(F),q)
[0040] The structure of the self-feature augmentation module is as shown in Figure 4The input of this module is feature X, where u represents the up-sampling rate. The feature input is further fused by the multi-head self-attention module for different points, and the feature F is obtained through the form of residual connection, and then the output feature of this module is obtained by passing through the feed forward network (FFN) and fusing the feature F again through the residual connection. The mathematical formula is as follows:
[0041] F = Norm(q + MultiHead(q, K, V))
[0042] q = XW qu K = XW KVu V = XW KVu
[0043] The input of the entire model can not only be the three-dimensional space coordinates (the dimension of the feature vector is 3) of each point in the point cloud, but also the splicing of the three-dimensional space coordinates and the corresponding normal features (the dimension of the feature vector is 6) of each point. However, the output of the model is always the three-dimensional space coordinates of each point in the final complete point cloud. In the model training stage, the loss function is composed of three parts, respectively from the seed generator and two cascaded point cloud generators, and the corresponding points are P0, P1 and P2, and the corresponding labels are S0, S1 and S2. The loss function is calculated by using the chamfer distance (CD) of the point cloud. See the following two formulas:
[0044]
[0045] λi in the above formula is a hyperparameter for adjusting the proportion of the loss values calculated in the three different stages.
[0046] In this embodiment, the complete single tooth point cloud data can be down-sampled twice. The complete crown point cloud data is equivalent to S2, and after one down-sampling, the point cloud is S1, and after one down-sampling of S1, S0 is obtained, S0 is the point cloud label generated by the seed generator, S1 is the point cloud label generated by the first level point cloud generator, and S2 is the point cloud label generated by the second level point cloud generator.
[0047] The input of the whole model is the point cloud corresponding to the crown of the missing tooth. In order to ensure the consistency of the model input, the missing crown is sampled, and the number of sampling points is set to M. The sampling number of all crown data used to train the model is M, which ensures the consistency of the dimension of the model input data. For each point in the crown point cloud input to the model, the feature of each point can be the three-dimensional spatial coordinates (x, y, z) of each point, so that the input dimension of the model is M x 3; or the three-dimensional spatial coordinates of each point are spliced with the three normal vectors (x, y, z, nx, ny, nz) of each point, which has a total of six dimensions, so that the input dimension of the model is M x 6.
[0048] The training data is first input to the feature extractor for encoding to obtain the encoded features (shape code), and then the encoded features and the model input features are input to the seed generator for decoding to generate a sparse complete crown point cloud P0. The point cloud is used as the input of the next first-level point cloud generator, and the encoded features are also input to the first-level point cloud generator. The output of the first-level point cloud generator is a more dense crown point cloud P1, which is used as the input of the second-level point cloud generator, and the encoded features (shape code) are also input to the second-level point cloud generator. The output of the second-level point cloud generator is used as the final model-predicted complete crown point cloud P2.
[0049] In the training process, the loss function is calculated by P0, P1, P2 and labels S0, S1, S2 using the above formula.
[0050] In one example, in order to improve the effect of model training, the teeth can be divided into 3 groups using the above prepared data. For each group, a model is trained using the same network structure as described above. Based on the similarity of tooth morphology, the specific grouping is as follows: upper and lower incisors and canines, upper and lower first and second premolars, and upper and lower first and second molars. Based on the characteristics of the teeth, the teeth are classified, and the training of the tooth restoration model is performed based on different types of teeth, which can make the result of the trained tooth restoration model more accurate. In this embodiment, the teeth are divided into three categories for training of the tooth restoration model, but in actual application, different categories can be divided according to needs.
[0051] In the embodiment, a single complete tooth in user historical data is processed based on a preset removal rule to obtain a simulated missing tooth, wherein the simulated missing tooth is a tooth with a part of a tooth crown removed according to the preset removal rule; and data information of the single complete tooth and data information of the simulated missing tooth are used as a data set training point cloud repair model to obtain the tooth repair model training method. The data set used for training is constructed by removing the tooth crown part of the complete tooth in various ways to obtain a missing tooth, and various missing teeth do not need to be actually obtained, so that more comprehensive missing types can be included in the point cloud repair model; in addition, the complete tooth after artificial completion in the prior art is used for training, and the completed tooth may not be completely the same as the original tooth due to various reasons, and the present application is trained according to the complete tooth in the historical data, so that the result of the missing tooth repair is closer to the real situation.
[0052] In another embodiment of the present application, the flow of the tooth repair method is as shown in Figure 5 Step 501 obtains a tooth to be repaired, specifically, a user's oral cavity and tooth model can be obtained by oral scanning, and the tooth to be repaired is identified.
[0053] After obtaining the tooth to be repaired, the tooth to be repaired is repaired by the point cloud repair model obtained by the tooth repair model training method as described above in step 502, to obtain a corresponding completed tooth.
[0054] In one example, in order to ensure that the distribution of the input data remains consistent when training the model and that the prediction result does not deviate, and to obtain more accurate repair data, a pre-trained binary classification model is used to classify the point cloud on the tooth to be repaired. After classification, the points on the tooth contour or the points on the tooth defect surface are removed. In this case, the removed part of the point cloud of the tooth to be repaired is more consistent with the previous training data in the training model. Specifically, the tooth repair method uses a tooth repair model training method to train a point cloud repair model to repair the tooth to be repaired, including: when the tooth to be repaired is a tooth with an attachment; by using a pre-trained first binary classification model, the points in the point cloud of the tooth to be repaired are divided into first class points and second class points; wherein the first class points include the vertices of the attachment in the point cloud, and the second class points include the remaining points in the point cloud; and the point cloud without the first class points is input into the repaired model after training. The first binary classification model can be trained using existing point cloud classification models, such as MeshSegNet (tooth classification network), iMeshSegNet (three-dimensional tooth classification network), TSGCNet (double-flow graph segmentation network), PointNet++ (point cloud segmentation network), etc. When repairing the tooth with the attachment removed, removing the vertices of the attachment in advance by the binary classification model can ensure that the distribution of the input information remains consistent when training the model, and improve the repair accuracy. Alternatively, when the tooth to be repaired is a defective tooth; by using a pre-trained second binary classification model, the points in the point cloud of the tooth to be repaired are divided into third class points and fourth class points; wherein the third class points include the vertices of the defect surface in the point cloud, and the fourth class points include the remaining points in the point cloud; and the point cloud without the third class points is input into the repaired model after training. The second binary classification model can also be trained using existing point cloud classification models, such as MeshSegNet (tooth classification network), iMeshSegNet (three-dimensional tooth classification network), TSGCNet (double-flow graph segmentation network), PointNet++ (point cloud segmentation network), etc.
[0055] In one example, the corresponding complete tooth obtained in the tooth restoration method comprises: registering the point cloud of the tooth to be restored with the point cloud of the complete tooth; extracting the points corresponding to the missing part of the tooth to be restored from the registered point cloud of the complete tooth; and adding the points corresponding to the missing part to the corresponding positions of the point cloud of the tooth to be restored. Specifically, for the tooth with an accessory, according to the subsequent medical needs, sometimes only the point cloud corresponding to the hole left after the accessory is removed in the generated complete point cloud data is needed, and the other part still needs to use the original point cloud. At this time, the input point cloud data and the generated point cloud data need to be registered by the ICP method, and the point cloud corresponding to the hole is extracted and added to the input point cloud data. In order to ensure the high accuracy of the point cloud alignment, when sampling points on the point cloud, the data in the hole filling place needs to be avoided. Similarly, for the tooth missing without removing the accessory, only the point cloud corresponding to the missing part in the generated complete point cloud data can be used, and the other part still needs to use the original point cloud.
[0056] In the IPC registration, in the embodiment, the hole position left after the accessory is removed or the missing part position of the caries can be determined according to the dental axis. For each dental crown, the dental axis has been determined in advance, and the complete dental crown still uses the original dental axis. Therefore, according to the direction of the dental axis, the general position of the hole left after the accessory is removed or the missing part position of the caries can be determined, and the random selected points can be avoided when registration is performed. After registration is completed, the corresponding points in the generated complete point cloud are taken out as the repair of the hole in the input point cloud according to the position of the hole.
[0057] The step division of the above method is only for clear description, and can be combined into one step or split into multiple steps in implementation, as long as the same logical relationship is included, and all are within the protection scope of the patent; adding irrelevant modifications or introducing irrelevant designs in the algorithm or flow, but not changing the core design of the algorithm and flow are within the protection scope of the patent.
[0058] Another embodiment of the present application relates to an electronic device, such as Figure 6 As shown in the figure, the electronic device comprises at least one processor; and a memory connected with the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the tooth restoration model training method or the tooth restoration method.
[0059] The memory and the processor are connected by a bus, which can include any number of interconnecting buses and bridges, and the bus connects the various circuits of the one or more processors and the memory together. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art, and therefore, will not be described any further. A bus interface provides an interface between the bus and a transceiver. The transceiver, which can be a single element or a plurality of elements, such as a plurality of receivers and transmitters, provides a communication path for communicating with various other devices over a transmission medium. Processed data is transmitted over a wireless medium via an antenna, and further, the antenna also receives data and transfers the data to the processor.
[0060] The processor is responsible for managing the bus and general processing, and can also provide various functions including timing, peripheral interfaces, voltage regulation, power management, and other control functions. The memory can be used for storing data used by the processor while executing operations.
[0061] Another embodiment of the present application relates to a computer readable storage medium, which stores a computer program. The computer program is executed by a processor to implement the method embodiments described above.
[0062] That is, those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by programs instructing relevant hardware, the programs are stored in a storage medium, and include a plurality of instructions for causing a device (which can be a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of the method described in various embodiments of the present application. The foregoing storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.
[0063] Another embodiment of the present application provides a computer program product, which includes a computer program, and the computer program is executed by a processor to implement the tooth restoration model training method described above or the tooth restoration method described above.
[0064] Those skilled in the art can understand that the above-mentioned embodiments are specific embodiments for implementing the present application, and in actual applications, various changes can be made in form and details without departing from the spirit and scope of the present application.
Claims
1. A dental restoration model training method, characterized in that, The method comprises the following steps: processing a single complete tooth in user historical data based on a preset removal rule to obtain a simulated missing tooth, wherein the simulated missing tooth is a tooth with a part of the tooth crown removed according to the preset removal rule; training a point cloud repair model using the data information of the single complete tooth and the data information of the simulated missing tooth as a data set, and obtaining a tooth repair model training method.
2. The dental restoration model training method of claim 1, wherein, The processing of the single complete tooth in the user historical data based on the preset removal rule comprises the following steps: when the obtained simulated missing tooth is a first missing tooth, forming a hole corresponding to the shape of the different accessories on the tooth crown surface of the complete tooth according to the shape of the removed different accessories, to obtain a missing tooth crown with different shapes; wherein the first missing tooth is a tooth missing due to removal of accessories.
3. The dental restoration model training method of claim 1, wherein, The processing of the single complete tooth in the user historical data based on the preset removal rule further comprises the following steps: when the obtained simulated missing tooth is a second missing tooth, cutting the tooth crown of the complete tooth from different angles using a section in a three-dimensional space, to obtain a missing tooth crown with different shapes; wherein the section is a plane or a curved surface with different shapes; wherein the second missing tooth is a tooth missing due to non-removal of accessories.
4. The dental restoration model training method of claim 1, wherein, The use of the data information of the complete tooth and the simulated missing tooth as a data set comprises the following steps: after obtaining the data information of the complete tooth and the simulated missing tooth, sampling the data information of the complete tooth and the simulated missing tooth respectively, wherein the sampling method is any one of uniform sampling, random sampling, and sampling based on geometric features; using the sampled data information of the complete tooth and the simulated missing tooth as a data set.
5. The dental restoration model training method of claim 1, wherein, The training of the point cloud repair model using the data information of the single complete tooth and the data information of the simulated missing tooth as a data set comprises the following steps: dividing the single complete tooth and its corresponding simulated missing tooth into N different categories of data sets according to a preset classification standard; training a repair model according to the N different categories of data sets, wherein N is a natural number greater than or equal to 1.
6. The dental restoration model training method of claim 5, wherein, The preset classification standard includes the shape feature of the tooth, and the single complete tooth and its corresponding simulated missing tooth are divided into three different categories of data sets, and a repair model is trained according to the three different categories of data sets; the three different categories include upper and lower incisors and canines, upper and lower first and second premolars, and upper and lower first and second molars.
7. A dental restoration method, characterized by, The method comprises the following steps: obtaining a tooth to be repaired; using the point cloud repair model trained by the tooth repair model training method according to any one of claims 1 to 6 to repair the tooth to be repaired, to obtain a corresponding complete tooth.
8. The dental restoration method of claim 7, wherein, The repair of the tooth to be repaired using the point cloud repair model trained by the tooth repair model training method according to any one of claims 1 to 6 comprises the following steps: when the tooth to be repaired is a tooth with accessories; dividing the points in the tooth to be repaired point cloud into first-class points and second-class points by using a pre-trained first two-classification model; The first type of points includes vertices of the accessory in the point cloud, and the second type of points includes the remaining points in the point cloud. The point cloud from which the first type of points is removed is input into the repaired model after training.
9. The dental restoration method of claim 7, wherein, The point cloud repaired model trained by the tooth repair model training method according to any one of claims 1 to 6 is used to repair the tooth to be repaired, including: When the tooth to be repaired is a defective tooth; The points in the point cloud of the tooth to be repaired are divided into third type of points and fourth type of points by the pre-trained second binary classification model. The third type of points includes vertices of the defective surface in the point cloud, and the fourth type of points includes the remaining points in the point cloud. The point cloud from which the third type of points is removed is input into the repaired model after training.
10. The dental restoration method of claim 7, wherein, The corresponding complete tooth includes: The point cloud of the tooth to be repaired is registered with the point cloud of the complete tooth. The points corresponding to the defective part of the tooth to be repaired are extracted from the registered point cloud of the complete tooth. The points corresponding to the defective part are added to the corresponding positions of the point cloud of the tooth to be repaired.
11. The dental restoration method of claim 10, wherein, Before the registration, further including: The dental axes of the tooth to be repaired and the complete tooth are determined respectively. The dental axes of the tooth to be repaired and the complete tooth are aligned.
12. An electronic device, comprising: Including: At least one processor; And The memory is in communication connection with the at least one processor; wherein The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the tooth repair model training method according to any one of claims 1 to 6 or the tooth repair method according to any one of claims 7 to 11.
13. A computer readable storage medium storing a computer program, wherein the computer program comprises program instructions configured to cause a processor to perform the method according to any one of claims 1 to 12. The computer program is executed by the processor to implement the tooth repair model training method according to any one of claims 1 to 6 or the tooth repair method according to any one of claims 7 to 11.
14. A computer program product, characterised in that, The computer program is executed by the processor to implement the tooth repair model training method according to any one of claims 1 to 6 or the tooth repair method according to any one of claims 7 to 11.