Construction method and device of automatic tooth arrangement model

Abstract

The invention provides a construction method and device of an automatic tooth arrangement model. The construction method comprises the following steps: firstly, obtaining multiple pieces of three-dimensional tooth scanning data and tooth arrangement sample data corresponding to each piece of three-dimensional tooth scanning data; then obtaining first width data corresponding to each piece of three-dimensional tooth scanning data; and according to all the three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data, training by using a preset deep learning algorithm to obtain an automatic tooth arrangement model. The input of the constructed automatic tooth arrangement model comprises the first width data, and the first width data can be used for representing the directed width between the teeth of the upper jaw and the directed width between the teeth of the lower jaw before tooth arrangement, so that compared with the prior art, a better tooth arrangement effect can be obtained while automatic tooth arrangement is realized. In addition, when the automatic tooth arrangement model is used, the first width data can be obtained according to the instruction input by the user, the controllable tooth arrangement effect can be achieved according to needs, and the user experience can be improved.

Description

Technical Field

[0001] The present application relates to the field of digital medical technology, and in particular to a method and device for constructing an automated tooth arrangement model. Background Art

[0002] Orthodontics is a discipline that corrects dental and maxillofacial deformities through tooth movement and jaw reconstruction. Traditional tooth arrangement technology mainly relies on the experience and skills of orthodontists, and the correction plan is formulated manually, but the efficiency is limited and is restricted by the judgment and operation accuracy of human doctors.

[0003] With the advancement of science and technology, computer-aided design and manufacturing technology (CAD / CAM) has been gradually introduced into the field of orthodontics, which has significantly improved the accuracy and efficiency of orthodontic treatment. However, the current digital auxiliary tooth arrangement system requires orthodontists to manually input initial parameters and make a large number of interactive adjustments, which has the disadvantages of high professional knowledge requirements and non-automation. With the development of artificial intelligence technology, in the field of orthodontics, a large amount of oral images and orthodontic case data have been analyzed based on machine learning algorithms, and the three-dimensional position information of teeth and jaws has been automatically extracted to generate personalized tooth arrangement plans. However, most of the existing tooth arrangement methods based on machine learning ignore the prior posture information of teeth and dentition itself, while the arch shape of dentition is an important basis for doctors to diagnose and formulate plans in clinical practice, and it is also an important criterion for evaluating the quality of tooth arrangement. Therefore, the tooth arrangement effect of the existing personalized tooth arrangement algorithm based on machine learning needs to be improved. Summary of the invention

[0004] In view of this, one of the technical problems solved by the present application is to provide a method and device for constructing an automated tooth arrangement model, so as to overcome the shortcomings of manual tooth arrangement in the prior art, such as high professional knowledge requirements and non-automation, and the problem of poor tooth arrangement effect of personalized tooth arrangement algorithms based on machine learning.

[0005] The first aspect of the embodiment of the present application discloses a method for constructing an automatic tooth arrangement model, the method comprising:

[0006] Obtaining a plurality of three-dimensional tooth scanning data and tooth arrangement sample data corresponding to each of the three-dimensional tooth scanning data; wherein the three-dimensional tooth scanning data is used to characterize the state of all teeth in the oral cavity before tooth arrangement; and the tooth arrangement sample data is used to characterize the state of all teeth in the oral cavity after tooth arrangement;

[0007] Obtaining first width data corresponding to each of the three-dimensional tooth scanning data; wherein the first width data is used to characterize the directional width between the teeth of the upper jaw and the directional width between the teeth of the lower jaw;

[0008] According to all the three-dimensional tooth scanning data and the corresponding first width data and the tooth arrangement sample data, an automated tooth arrangement model is obtained by training with a preset deep learning algorithm; wherein, the input of the automated tooth arrangement model is the three-dimensional tooth scanning data and the first width data, and the output of the automated tooth arrangement model is the tooth arrangement calculation data, and the tooth arrangement calculation data is used to characterize the state of all teeth in the oral cavity after tooth arrangement.

[0009] A second aspect of an embodiment of the present application discloses a device for constructing an automated tooth arrangement model, the device comprising:

[0010] A data acquisition module, used to obtain a plurality of three-dimensional tooth scanning data and tooth arrangement sample data corresponding to each of the three-dimensional tooth scanning data; wherein the three-dimensional tooth scanning data is used to characterize the state of all teeth in the oral cavity before tooth arrangement; and the tooth arrangement sample data is used to characterize the state of all teeth in the oral cavity after tooth arrangement;

[0011] A width acquisition module, used to obtain first width data corresponding to each of the three-dimensional tooth scanning data; wherein the first width data is used to represent the directional width between the teeth of the upper jaw and the directional width between the teeth of the lower jaw;

[0012] A model training module is used to obtain an automated tooth arrangement model by training using a preset deep learning algorithm based on all the three-dimensional tooth scanning data and the corresponding first width data and the tooth arrangement sample data; wherein the input of the automated tooth arrangement model is the three-dimensional tooth scanning data and the first width data, and the output of the automated tooth arrangement model is tooth arrangement calculation data, and the tooth arrangement calculation data is used to characterize the state of all teeth in the oral cavity after tooth arrangement.

[0013] It can be seen from the invention content of the present application that the method and device for constructing an automated tooth arrangement model disclosed in the present application first obtain multiple three-dimensional tooth scanning data, and tooth arrangement sample data corresponding to each three-dimensional tooth scanning data; then obtain the first width data corresponding to each three-dimensional tooth scanning data; and then obtain the automated tooth arrangement model by training using a preset deep learning algorithm based on all the three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data. Since the input of the automated tooth arrangement model constructed in this embodiment includes the first width data, which can be used to characterize the directional width between the teeth of the maxilla and the directional width between the teeth of the mandible, compared with the prior art, while realizing automated tooth arrangement, a better tooth arrangement effect can also be obtained. In addition, since the first width data can also be obtained according to the instructions manually input by the user when using the automated tooth arrangement model, a controllable tooth arrangement effect can be achieved on demand, which is conducive to improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 It is a flow chart of a method for constructing an automatic tooth arrangement model disclosed in Example 1 of the present application;

[0016] Figure 2 It is a schematic diagram of the numbering of the teeth in the upper and lower jaws;

[0017] Figure 3 It is a schematic diagram of three-dimensional dental scan data;

[0018] Figure 4 It is a schematic diagram of the result of automatically arranging teeth using the automatic tooth arrangement model directly on the three-dimensional tooth scanning data;

[0019] Figure 5 It is a schematic diagram of the result of automatic tooth arrangement of three-dimensional tooth scanning data using an automatic tooth arrangement model under a network retraction command;

[0020] Figure 6 It is a schematic diagram of the result of automatic tooth arrangement of three-dimensional tooth scanning data using an automatic tooth arrangement model under network expansion instructions;

[0021] Figure 7 It is a flow chart of a method for constructing an automated tooth arrangement model disclosed in Embodiment 2 of the present application;

[0022] Figure 8 It is a schematic block diagram of the structure of a device for constructing an automatic tooth arrangement model disclosed in Example 3 of the present application. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0024] It should be noted that the terms "first", "second", "third" and "fourth" in the specification and claims of the present application are used to distinguish different objects rather than to describe a specific order. The terms "including" and "having" in the embodiments of the present application and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment including a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0025] Embodiment 1

[0026] like Figure 1 As shown, Figure 1 This is a schematic flow chart of a method for constructing an automated tooth arrangement model disclosed in Embodiment 1 of the present application. The method for constructing an automated tooth arrangement model includes:

[0027] Step S101, obtaining a plurality of three-dimensional tooth scanning data and tooth arrangement sample data corresponding to each three-dimensional tooth scanning data.

[0028] In this embodiment, the three-dimensional tooth scanning data is used to characterize the state of all teeth in the oral cavity before tooth arrangement, and is usually obtained by scanning the human oral cavity with professional oral scanning equipment. The specific method of characterizing the state of all teeth in the oral cavity before tooth arrangement using the three-dimensional tooth scanning data is not limited, and can be reasonably selected according to actual application requirements. For example, the state of all teeth in the oral cavity before tooth arrangement can be characterized in the form of an image, a grid, or a point cloud.

[0029] In addition, the number of three-dimensional tooth scanning data obtained in step S101 is not limited and can be reasonably selected according to actual application requirements. For example, in order to achieve a better tooth arrangement effect of the final automatic tooth arrangement model, the number of three-dimensional tooth scanning data obtained in step S101 can preferably exceed 100.

[0030] In this embodiment, the tooth arrangement sample data is used to characterize the state after all teeth in the oral cavity are arranged. There is no limit to the way to obtain the tooth arrangement sample data, and a reasonable choice can be made according to actual application requirements. For example, it can be obtained by an orthodontist using professional software to adjust the tooth position and angle of the three-dimensional tooth scanning data; it can also be obtained by adjusting the tooth position and angle of the three-dimensional tooth scanning data using other algorithms with better automatic tooth arrangement effects.

[0031] Optionally, considering that each tooth needs to be analyzed and processed when arranging teeth, and considering that some people may have incomplete teeth, in order to achieve a better tooth arrangement effect of the final automated tooth arrangement model, the teeth in the digital model obtained by directly scanning the oral scanning device can be completed and segmented. Specifically, step S101 includes the following sub-steps S101a-S101c:

[0032] Sub-step S101a, using an oral scanning device to acquire a plurality of original three-dimensional scanning data.

[0033] The original three-dimensional scanning data is a digital model directly generated by an oral scanning device after scanning the oral cavity. The specific type of oral scanning device is not limited and can be reasonably selected according to actual application requirements.

[0034] Sub-step S101b, performing crown segmentation processing on each original three-dimensional scanning data to obtain digital mesh data corresponding to each tooth.

[0035] The specific method of performing crown segmentation processing on the original three-dimensional scanning data is not limited and can be reasonably selected according to actual application requirements. For example, it can be processed manually using specific software operations or by a specific automatic crown segmentation algorithm.

[0036] Sub-step S101c, completing the digitized mesh data corresponding to all teeth to obtain a plurality of three-dimensional tooth scanning data.

[0037] The specific method for completing the digital mesh data corresponding to the teeth is not limited and can be reasonably selected according to actual application requirements. For example, it can be processed manually using specific software operations or by a specific automatic crown segmentation algorithm.

[0038] Step S102, obtaining first width data corresponding to each three-dimensional tooth scanning data.

[0039] In this embodiment, the first width data is used to characterize the directional width between the teeth of the upper jaw and the directional width between the teeth of the lower jaw. The reason for obtaining the directional width between the teeth of the upper jaw and the directional width between the teeth of the lower jaw is to characterize the state of the dental arch, that is, to characterize the arch shape of the upper and lower jaw teeth continuously arranged in the alveolar bone to form a parabola. Among them, the specific characterization method of the first width data is not limited, and can be reasonably selected according to actual application requirements.

[0040] In addition, there is no limit to the way of obtaining the first width data, and a reasonable choice can be made according to actual application requirements. For example, it can be obtained by measuring the tooth size represented by the three-dimensional tooth scanning data using a specific software function module; or it can be obtained according to a manual instruction input by a user. Among them, when the first width data is obtained according to a manual instruction input by a user, the order of executing capture S101 and step S102 is not limited.

[0041] Optionally, the first width data may accurately characterize the dental arch by the directional width between all adjacent teeth.

[0042] Optionally, in order to more accurately characterize the dental arch while reducing the difficulty and complexity of subsequent training of the automated tooth arrangement model, the first width data may also accurately characterize the dental arch by measuring the directional width between some teeth at specific positions.

[0043] Further, based on the summary of clinical experience and balancing the difficulty and complexity of subsequent training of the automated tooth arrangement model, the first width data may be preferably used to characterize the directional width between the two groups of teeth No. 1, 3, and 6 in the maxilla, and the directional width between the two groups of teeth No. 1, 3, and 6 in the mandible. Figure 2 The position of teeth in the oral cavity can be divided into four areas: upper right, lower right, upper left and lower left. The teeth in each area are numbered 1-8 from the middle to the edge. The two groups of teeth No. 1, 3, and 6 in the upper jaw correspond to 11, 13, 16, 21, 23, and 26 in the figure, and the two groups of teeth No. 1, 3, and 6 in the lower jaw correspond to 31, 33, 36, 41, 43, and 46 in the figure.

[0044] Step S103, based on all the three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data, a preset deep learning algorithm is used for training to obtain an automated tooth arrangement model.

[0045] In this embodiment, the input of the automatic tooth arrangement model is the three-dimensional tooth scanning data and the first width data, and the output of the automatic tooth arrangement model is the tooth arrangement calculation data, which is used to characterize the state of all teeth in the oral cavity after the tooth arrangement. In the subsequent use of the automatic tooth arrangement model, the input first width data can be manually input by the user, or can be obtained by measuring the tooth size represented by the input three-dimensional tooth scanning data using a specific software function module, which is not limited in this embodiment.

[0046] For example, see Figure 3-Figure 6 , Figure 3is a schematic diagram of 3D tooth scanning data, which shows the state of all teeth in the oral cavity before tooth arrangement. If the tooth size represented by the input 3D tooth scanning data is measured using a specific software function module to obtain the first width data, the result after subsequent processing using the automatic tooth arrangement model is as follows Figure 4 As shown; if the user inputs a network retraction command for instructing the shortening of the dental arch, after obtaining the first width data according to the network retraction command, the result after further processing using the automated tooth arrangement model is as follows Figure 5 As shown; if the user inputs a network expansion command for instructing the dental arch to lengthen, after obtaining the first width data according to the network expansion command, the result after further processing using the automated tooth arrangement model is as follows Figure 6 shown.

[0047] In this embodiment, the preset deep learning algorithm may be composed of multiple functional modules. The specific type of algorithm adopted by each functional module is not limited and can be reasonably selected according to actual application requirements.

[0048] In this embodiment, the specific method of using the preset deep learning algorithm to train the model is not limited and can be reasonably selected according to actual application requirements. For example, all three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data can be used as a training set; or all three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data can be divided into three parts, the first part as a training set, the second part as a verification set, and the third part as a test set.

[0049] From the above description of the present embodiment, it can be seen that the present embodiment first obtains a plurality of three-dimensional tooth scanning data, and the tooth arrangement sample data corresponding to each three-dimensional tooth scanning data; then obtains the first width data corresponding to each three-dimensional tooth scanning data; and then, based on all the three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data, the preset deep learning algorithm is used to train and obtain the automatic tooth arrangement model. Since the input of the automatic tooth arrangement model constructed in the present embodiment includes the first width data, which can be used to characterize the directional width between the teeth of the maxilla and the directional width between the teeth of the mandible, compared with the prior art, a better tooth arrangement effect can be obtained while realizing automatic tooth arrangement. In addition, since the first width data can also be obtained according to the instructions manually input by the user when using the automatic tooth arrangement model, a controllable tooth arrangement effect can be achieved on demand, which is conducive to improving the user experience.

[0050] Embodiment 2

[0051] like Figure 7 As shown, Figure 7 This is a schematic flow chart of a method for constructing an automated tooth arrangement model disclosed in Embodiment 2 of the present application. The method for constructing an automated tooth arrangement model includes:

[0052] Step S201, obtaining a plurality of three-dimensional tooth scanning data and tooth arrangement sample data corresponding to each three-dimensional tooth scanning data.

[0053] In this embodiment, step S201 is substantially the same as or similar to step S101 in the aforementioned first embodiment, and will not be described in detail herein.

[0054] Step S202, performing farthest point sampling processing on the teeth represented by each three-dimensional tooth scanning data to obtain point cloud data corresponding to each three-dimensional tooth scanning data.

[0055] In this embodiment, the point cloud data is used to represent the state of all teeth in the oral cavity before tooth arrangement in the form of point cloud.

[0056] Optionally, in order to more accurately characterize the state of all teeth in the oral cavity before tooth arrangement and balance the complexity of subsequent data processing, it is preferred that the point cloud data use 512 points to characterize each tooth.

[0057] Step S203: measuring and obtaining first width data corresponding to each three-dimensional tooth scanning data according to the point cloud data corresponding to each three-dimensional tooth scanning data.

[0058] In this embodiment, the specific method of using point cloud data to measure and obtain the first width data corresponding to each three-dimensional tooth scan data is not limited, and can be reasonably selected according to actual application requirements. For example, a unified reference plane or reference coordinate system can be determined first, and then the first width data corresponding to each three-dimensional tooth scan data can be measured and obtained.

[0059] Optionally, in order to more accurately and reasonably obtain the first width data corresponding to all the three-dimensional tooth scanning data, step S203 may include the following sub-steps S203a and S203b:

[0060] Sub-step S203a, fitting the median plane and occlusal plane of the dentition corresponding to the three-dimensional tooth scanning data according to the point cloud data corresponding to the three-dimensional tooth scanning data.

[0061] Among them, the median plane and occlusal plane of the dentition can be fitted through parameters such as the centroid of the point cloud.

[0062] Sub-step S203b, taking the median plane and the occlusal plane of the dentition corresponding to the three-dimensional tooth scanning data as a reference, measuring and obtaining the first width data corresponding to each three-dimensional tooth scanning data.

[0063] Among them, the coordinate system can be adjusted to the upward direction of the human body posture as the z direction, the forward direction of the human body posture as the y direction, and the x direction can be determined by the right-hand system, so as to further measure and obtain the first width data corresponding to each three-dimensional tooth scanning data in the determined coordinate system.

[0064] Step S204, based on all the three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data, a preset deep learning algorithm is used for training to obtain an automated tooth arrangement model.

[0065] In this embodiment, the input of the automatic tooth arrangement model is three-dimensional tooth scanning data and first width data, and the output of the automatic tooth arrangement model is tooth arrangement calculation data, which is used to characterize the state of all teeth in the oral cavity after tooth arrangement.

[0066] Optionally, in order to obtain an automated tooth arrangement model with a better tooth arrangement effect, step S204 may include the following sub-steps S204a and S204b:

[0067] Sub-step S204a, constructing an initial tooth arrangement model.

[0068] Among them, the functional modules included in the initial tooth arrangement model and the automatic tooth arrangement model are basically the same, and the difference mainly lies in the specific values ​​of some or all of the network parameters.

[0069] Sub-step S204b, training the initial tooth arrangement model for a preset number of rounds according to all the three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data, to obtain an automated tooth arrangement model.

[0070] In sub-step S204b, a round of training is performed on the initial tooth arrangement model according to all three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data, including the following steps AH:

[0071] Step A, based on the point cloud data corresponding to the three-dimensional tooth scanning data, the initial tooth arrangement model is used to calculate and obtain the centroid coordinate data, the first local feature data and the first global feature data corresponding to the three-dimensional tooth scanning data.

[0072] Among them, the centroid coordinate data is used to characterize the centroid coordinate values ​​of all teeth before tooth arrangement. The first local feature data is used to characterize the features of several dimensions of each tooth before tooth arrangement; the first global feature data is used to characterize the features of several dimensions of all teeth before tooth arrangement. The initial tooth arrangement model may include functional modules for extracting local features and overall features, as well as functional modules for extracting the centroid coordinates of each tooth. The specific algorithms adopted by these functional modules are not limited and can be reasonably selected according to actual application requirements.

[0073] For example, the point cloud data of each tooth corresponding to the three-dimensional dental scanning data can be spliced ​​together and then sent to a point cloud feature encoder to obtain the first global feature data corresponding to the three-dimensional dental scanning data; the point cloud data of each tooth corresponding to the three-dimensional dental scanning data can be sent to the same point cloud feature encoder to obtain the first local feature data corresponding to the three-dimensional dental scanning data.

[0074] In addition, the number of feature dimensions represented by the first local feature data and the number of feature dimensions represented by the first global feature data are not limited and can be reasonably selected according to actual application requirements.

[0075] Furthermore, in order to reduce the amount of data calculation while ensuring good training effects, it is preferred that the first local feature data characterize the features of each tooth before tooth arrangement through 512 dimensions; and the first global feature data characterize the features of all teeth before tooth arrangement through 512 dimensions.

[0076] Step B: based on the first width data corresponding to the three-dimensional tooth scanning data, the initial tooth arrangement model is used to extract and obtain the width embedding feature data corresponding to the three-dimensional tooth scanning data.

[0077] Among them, the initial tooth arrangement model may include a functional module for obtaining width embedded feature data. The specific algorithm adopted by the functional module is not limited and can be reasonably selected according to actual application requirements. In addition, the number of specific features represented by the width embedded feature data is not limited, but it needs to be the same as the number of overall features represented by the first global feature data. For example, when the overall feature represented by the first global feature data is greater than or equal to 1, the execution order of step A and step B is not limited, and they can be executed sequentially or simultaneously.

[0078] Furthermore, step B may include: according to the first width data corresponding to the three-dimensional tooth scanning data, through a linear layer network with a preset number of layers, using Leaky ReLU as an activation function, to obtain width embedding feature data corresponding to the three-dimensional tooth scanning data.

[0079] The specific number of preset layers is not limited and can be reasonably selected according to actual application requirements. For example, in order to obtain a better feature extraction effect and balance the amount of data processing, a 2-layer linear layer network can be preferably used.

[0080] In addition, the activation function is a function added to the artificial neural network to help the network learn complex patterns in the data. Similar to the neuron-based model in the human brain, the activation function ultimately determines what to emit to the next neuron. Introducing nonlinear functions as activation functions can make deep neural networks more expressive.

[0081] Since the first width data represents a directed width, which may have positive and negative values, the Leaky ReLU function is different from the more general ReLU function. ReLU sets all negative values ​​to zero, while Leaky ReLU assigns a non-zero slope to all negative values. Therefore, using Leaky ReLU as the activation function in step B not only has all the advantages of ReLU, but also helps avoid the problem of neuron death and enables the network to learn more features.

[0082] Step C: using the initial tooth arrangement model to fuse the width embedding feature data corresponding to the three-dimensional tooth scanning data and the first global feature data to obtain the second global feature data corresponding to the three-dimensional tooth scanning data.

[0083] Among them, the initial tooth arrangement model may include a functional module for fusing the width embedded feature data and the first global feature data. The specific algorithm adopted by the functional module is not limited and can be reasonably selected according to actual application requirements.

[0084] For example, when the feature dimensions represented by the width embedding feature data and the first global feature data are the same, the two may be added element by element to obtain the second global feature data.

[0085] Step D: perform feature propagation processing on the first local feature data corresponding to the three-dimensional tooth scanning data using the initial tooth arrangement model to obtain second local feature data corresponding to the three-dimensional tooth scanning data.

[0086] The reason for performing feature propagation processing on the first local feature data corresponding to the three-dimensional tooth scanning data is to better capture the feature relationship between adjacent teeth. The initial tooth arrangement model may include a functional module for performing feature propagation processing on the first local feature data. The specific algorithm adopted by the functional module is not limited and can be reasonably selected according to actual application requirements.

[0087] Furthermore, step D may include: using a multi-layer Transformer encoder module as a feature propagation model, performing feature propagation processing on first local feature data corresponding to the three-dimensional tooth scanning data, and obtaining second local feature data corresponding to the three-dimensional tooth scanning data.

[0088] Among them, the Transformer model achieves efficient sequence-to-sequence conversion through a multi-layer stacked encoder and decoder structure. In the encoder, the relationship between data in the input sequence is captured through a multi-head self-attention mechanism. The attention mechanism is more efficient, converges quickly, and has higher accuracy.

[0089] Step E: using the initial tooth arrangement model, the second global feature data and the second local feature data corresponding to the three-dimensional tooth scanning data are spliced ​​and feature mapped to obtain predicted position feature data corresponding to the three-dimensional tooth scanning data.

[0090] The predicted position feature data is used to characterize the position features of all teeth after tooth arrangement predicted by the network. The initial tooth arrangement model may include a functional module for splicing and feature mapping the second global feature data and the second local feature data. The specific algorithm adopted by the functional module is not limited and can be reasonably selected according to actual application requirements.

[0091] For example, three linear layers can be used to construct a feature mapping module in the initial tooth arrangement model, so that the second global feature data and the second local feature data corresponding to the three-dimensional tooth scanning data are spliced ​​and processed, and the obtained data is then feature mapped through the feature mapping module to finally obtain the predicted position feature data corresponding to the three-dimensional tooth scanning data.

[0092] Step F: using the initial tooth arrangement model, the predicted position feature data and centroid coordinate data corresponding to the three-dimensional tooth scanning data are spliced ​​and predicted, and the transformation parameter data corresponding to the three-dimensional tooth scanning data are calculated.

[0093] The transformation parameter data is used to characterize the translation parameter values ​​and rotation parameter values ​​of all teeth after tooth arrangement compared to before tooth arrangement. The initial tooth arrangement model may include a functional module for splicing the predicted position feature data and the centroid coordinate data. The specific algorithm adopted by the functional module is not limited and can be reasonably selected according to actual application requirements.

[0094] Furthermore, step F may include: using the initial tooth arrangement model to splice the predicted position feature data and centroid coordinate data corresponding to the three-dimensional tooth scanning data, and using a multi-layer perceptron as a decoder to calculate and obtain the transformation parameter data corresponding to the three-dimensional tooth scanning data.

[0095] The transformation parameter data can be used to characterize the parameters of translation and / or rotation of each tooth, and the specific characterization method is not limited and can be reasonably selected according to actual application requirements. For example, the rotation parameter can be represented by a normalized quaternion, and the translation parameter can be a three-dimensional parameter.

[0096] In addition, the multilayer perceptron (MLP) is a feedforward artificial neural network model that introduces one or more hidden layers on the basis of a single-layer neural network, so that the neural network has multiple network layers, and transforms the output of the hidden layer through an activation function. It is usually used to solve classification and regression problems. For each tooth, since the rotation transformation also depends on the rotation center, it is necessary to splice the predicted position feature data and the centroid coordinate data corresponding to the three-dimensional tooth scanning data, and send the spliced ​​data to the multilayer perceptron to obtain the transformation parameter data corresponding to the three-dimensional tooth scanning data.

[0097] Step G, using a preset loss function to calculate the loss calculation value based on the transformation parameter data and the tooth arrangement sample data corresponding to the three-dimensional tooth scanning data.

[0098] Among them, step G is mainly to evaluate the effect of tooth arrangement using the initial tooth arrangement model by calculating the loss calculation value. The specific number and type of preset loss functions are not limited and can be reasonably selected according to actual application requirements.

[0099] Furthermore, in order to make the network more robust, the tooth arrangement effect of the initial tooth arrangement model can be supervised by comparing the mean square error loss of the point cloud corresponding to the tooth arrangement result obtained using the initial tooth arrangement model and the tooth arrangement result as the training set. Specifically, step G may include sub-steps g1-g2:

[0100] Sub-step g1, obtaining tooth arrangement calculation data corresponding to the three-dimensional tooth scanning data according to the transformation parameter data corresponding to the three-dimensional tooth scanning data.

[0101] The tooth arrangement calculation data is used to characterize the state of all teeth in the oral cavity after tooth arrangement, that is, the tooth arrangement result obtained after tooth arrangement using the initial tooth arrangement model.

[0102] Sub-step g2, according to the tooth arrangement calculation data and the tooth arrangement sample data corresponding to the three-dimensional tooth scanning data, a second loss value is calculated using a mean square error loss function.

[0103] Among them, the tooth arrangement calculation data and / or tooth arrangement sample data can directly use point cloud to represent the state of all teeth in the mouth after arrangement; or the tooth arrangement calculation data and / or tooth arrangement sample data can be processed, and the obtained data can be used to represent the state of all teeth in the mouth after arrangement in the form of point cloud.

[0104] In addition, the specific function type of the mean square error loss function is not limited and can be reasonably selected according to actual application requirements.

[0105] For example, the formula The second loss value is calculated.

[0106] in, is the second loss value; c * is the coordinate value of a single point in the point cloud corresponding to the tooth arrangement sample data; c' is the coordinate value of a single point in the point cloud corresponding to the tooth arrangement calculation data; in the point cloud corresponding to the tooth arrangement calculation data and the point cloud corresponding to the tooth arrangement sample data, c* and c' correspond to the coordinate value of the same point; For c' and c * The mean square error between i The coordinate value set of all points in the point cloud corresponding to the tooth arrangement calculation data; |C′ i | is the total number of points in the point cloud corresponding to the tooth arrangement calculation data; |L| is the number of all teeth.

[0107] Furthermore, in order to form positive and negative example pairs using tooth prior knowledge, step G may include the following sub-steps g3 to g5:

[0108] Sub-step g3, performing staggered arrangement processing on the tooth arrangement sample data corresponding to the three-dimensional tooth scanning data to obtain sample staggered arrangement data corresponding to the three-dimensional tooth scanning data.

[0109] The staggered arrangement of the tooth arrangement sample data corresponding to the three-dimensional tooth scanning data refers to rearranging the order of all teeth represented by the sample staggered arrangement data.

[0110] Sub-step g4, according to the sample misalignment data and the tooth arrangement sample data corresponding to the three-dimensional tooth scanning data, the initial tooth arrangement model is used to calculate and obtain the third local feature data and the fourth local feature data corresponding to the three-dimensional tooth scanning data.

[0111] Among them, the third local feature data and the fourth local feature data are both used to characterize the features of several dimensions of each tooth after tooth arrangement. In fact, they are also data after local feature extraction for two different tooth arrangement methods. The third local feature data corresponding to the three-dimensional tooth scanning data is obtained based on the sample staggered arrangement data corresponding to the three-dimensional tooth scanning data; the fourth local feature data corresponding to the three-dimensional tooth scanning data is obtained based on the tooth arrangement sample data corresponding to the three-dimensional tooth scanning data.

[0112] It should be noted that the function module for extracting the third local feature data and the fourth local feature data is the same as the function module for extracting the first local feature data. For example, if the point cloud feature encoder is used to extract local features from the point cloud data corresponding to the three-dimensional tooth scanning data in the aforementioned sub-step A, then the point cloud feature encoder is also used to extract local features from the sample misalignment data and the tooth arrangement sample data corresponding to the three-dimensional tooth scanning data in sub-step g4.

[0113] Sub-step g5, obtaining a first loss value by calculating using a cosine similarity function based on the predicted position feature data, the third local feature data and the fourth local feature data corresponding to the three-dimensional tooth scanning data.

[0114] Among them, the specific calculation formula of the first loss value is not limited and can be reasonably selected according to actual application requirements.

[0115] For example, we can use the formula Calculate and obtain the first loss value.

[0116] in, is the first loss value; |L| is the number of all teeth; L is the set of all tooth labels; is the data corresponding to the i-th tooth in the predicted position feature data; is the data corresponding to the i-th tooth in the third local feature data in order; is the data corresponding to the i-th tooth in order in the fourth local feature data.

[0117] It should be noted that the second loss value calculation method described in the aforementioned sub-steps g1 to g2 and the first loss value calculation method described in g3 to g5 may only execute the second loss value calculation method, or may execute both.

[0118] When only the calculation method of the second loss value is executed, the calculated second loss value is equal to the loss calculation value in step G.

[0119] When two calculation methods are executed, the loss calculation value in step G is obtained by jointly calculating the first loss value and the second loss value. There is no limit to the method of jointly calculating the loss calculation value, and a reasonable selection can be made according to actual application requirements.

[0120] Furthermore, in order to better supervise the tooth arrangement effect of the initial tooth arrangement model, it is preferred that the aforementioned sub-steps g1 to g5 are all performed, that is, the corresponding loss calculation values ​​in step G and step H are preferably determined jointly according to the two loss value calculation methods.

[0121] Step H, back-propagation processing is performed on the initial tooth arrangement model according to the loss calculation value.

[0122] Among them, step H mainly updates the relevant parameters of the initial tooth arrangement model through back propagation, so as to use the initial tooth arrangement model with updated parameters for the next round of training.

[0123] From the above description of the present embodiment, it can be seen that the present embodiment performs farthest point sampling processing on the teeth represented by the three-dimensional tooth scanning data, obtains the first width data according to the obtained point cloud data measurement and trains the automatic tooth arrangement model, so as to balance the difficulty and complexity of training the automatic tooth arrangement model while ensuring that the automatic tooth arrangement model has a good tooth arrangement effect; by utilizing a multi-layer Transformer encoder module as a feature propagation model, the first local feature data corresponding to the three-dimensional tooth scanning data is subjected to feature propagation processing, so as to better capture the feature relationship between adjacent teeth, make the attention mechanism more efficient, and converge quickly with higher accuracy.

[0124] Embodiment 3

[0125] Embodiment 3 of the present application provides a device for constructing an automated tooth arrangement model. Figure 8 This is a schematic diagram of the structure of a device for constructing an automated tooth arrangement model disclosed in Example 3 of the present application, the device comprising:

[0126] A data acquisition module is used to obtain a plurality of three-dimensional tooth scanning data and tooth arrangement sample data corresponding to each three-dimensional tooth scanning data; wherein the three-dimensional tooth scanning data is used to characterize the state of all teeth in the oral cavity before tooth arrangement; and the tooth arrangement sample data is used to characterize the state of all teeth in the oral cavity after tooth arrangement;

[0127] A width acquisition module, used to obtain first width data corresponding to each three-dimensional tooth scanning data; wherein the first width data is used to represent the directional width between the teeth of the upper jaw and the directional width between the teeth of the lower jaw;

[0128] The model training module is used to obtain an automated tooth arrangement model by training with a preset deep learning algorithm based on all three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data; wherein the input of the automated tooth arrangement model is the three-dimensional tooth scanning data and the first width data, and the output of the automated tooth arrangement model is the tooth arrangement calculation data, which is used to characterize the state of all teeth in the oral cavity after tooth arrangement.

[0129] Optionally, the width acquisition module includes a sampling processing unit and a measuring unit, wherein:

[0130] The sampling processing unit is used to perform farthest point sampling processing on the teeth represented by each three-dimensional tooth scanning data to obtain point cloud data corresponding to each three-dimensional tooth scanning data; wherein the point cloud data is used to represent the state of all teeth in the oral cavity before tooth arrangement in the form of point cloud;

[0131] The measuring unit is used to measure and obtain first width data corresponding to each three-dimensional tooth scanning data according to the point cloud data corresponding to each three-dimensional tooth scanning data.

[0132] Optionally, the model training module is also used to: construct an initial tooth arrangement model;

[0133] According to all the three-dimensional tooth scanning data and the corresponding first width data and the tooth arrangement sample data, the initial tooth arrangement model is trained for a preset number of rounds to obtain an automated tooth arrangement model;

[0134] According to all the three-dimensional tooth scanning data and the corresponding first width data and tooth arrangement sample data, a round of training of the initial tooth arrangement model includes:

[0135] According to the point cloud data corresponding to the three-dimensional tooth scanning data, the initial tooth arrangement model is used to calculate the centroid coordinate data, the first local feature data and the first global feature data corresponding to the three-dimensional tooth scanning data; wherein the centroid coordinate data is used to characterize the centroid coordinate values ​​of all teeth before tooth arrangement; the first local feature data is used to characterize the features of several dimensions of each tooth before tooth arrangement; the first global feature data is used to characterize the features of several dimensions of all teeth before tooth arrangement;

[0136] According to the first width data corresponding to the three-dimensional tooth scanning data, the initial tooth arrangement model is used to extract and obtain the width embedding feature data corresponding to the three-dimensional tooth scanning data;

[0137] Using the initial tooth arrangement model, the width embedding feature data corresponding to the three-dimensional tooth scanning data and the first global feature data are fused to obtain the second global feature data corresponding to the three-dimensional tooth scanning data;

[0138] Using the initial tooth arrangement model, feature propagation processing is performed on first local feature data corresponding to the three-dimensional tooth scanning data to obtain second local feature data corresponding to the three-dimensional tooth scanning data;

[0139] The second global feature data and the second local feature data corresponding to the three-dimensional tooth scanning data are spliced ​​and feature mapped using the initial tooth arrangement model to obtain predicted position feature data corresponding to the three-dimensional tooth scanning data; wherein the predicted position feature data is used to characterize the position features of all teeth after tooth arrangement predicted by the network;

[0140] The predicted position feature data and centroid coordinate data corresponding to the three-dimensional tooth scanning data are spliced ​​and predicted using the initial tooth arrangement model, and the transformation parameter data corresponding to the three-dimensional tooth scanning data are calculated; wherein the transformation parameter data is used to characterize the translation parameter values ​​and rotation parameter values ​​of all teeth after tooth arrangement compared with before tooth arrangement;

[0141] According to the transformation parameter data and tooth arrangement sample data corresponding to the three-dimensional tooth scanning data, a loss calculation value is obtained by using a preset loss function;

[0142] The initial tooth arrangement model is back-propagated according to the loss calculation value.

[0143] Optionally, the model training module is also used to: according to the first width data corresponding to the three-dimensional dental scanning data, obtain the width embedding feature data corresponding to the three-dimensional dental scanning data through a linear layer network with a preset number of layers and using Leaky ReLU as an activation function.

[0144] Optionally, the model training module is also used to: use a multi-layer Transformer encoder module as a feature propagation model to perform feature propagation processing on first local feature data corresponding to the three-dimensional dental scanning data to obtain second local feature data corresponding to the three-dimensional dental scanning data.

[0145] Optionally, the model training module is also used to: use the initial tooth arrangement model to splice the predicted position feature data and center of mass coordinate data corresponding to the three-dimensional tooth scanning data, and use a multi-layer perceptron as a decoder to calculate and obtain the transformation parameter data corresponding to the three-dimensional tooth scanning data.

[0146] Optionally, the model training module is further used to: perform misalignment processing on the tooth arrangement sample data corresponding to the three-dimensional tooth scanning data to obtain sample misalignment data corresponding to the three-dimensional tooth scanning data;

[0147] According to the sample misalignment data and the sample tooth arrangement data corresponding to the three-dimensional tooth scanning data, the initial tooth arrangement model is used to calculate and obtain the third local feature data and the fourth local feature data corresponding to the three-dimensional tooth scanning data; wherein the third local feature data and the fourth local feature data are both used to characterize the features of several dimensions of each tooth after tooth arrangement;

[0148] The first loss value is obtained by calculating the cosine similarity function according to the predicted position feature data, the third local feature data and the fourth local feature data corresponding to the three-dimensional tooth scanning data.

[0149] Optionally, the model training module is further used to: according to the predicted position feature data, the third local feature data and the fourth local feature data corresponding to the three-dimensional tooth scanning data, use the formula

[0150] Calculate and obtain a first loss value;

[0151] in, is the first loss value; |L| is the number of all teeth; L is the set of all tooth labels; is the data corresponding to the i-th tooth in the predicted position feature data; is the data corresponding to the i-th tooth in the third local feature data in order; is the data corresponding to the i-th tooth in order in the fourth local feature data.

[0152] Optionally, the measuring unit is further used to fit the median plane and occlusal plane of the dentition corresponding to the three-dimensional tooth scanning data according to the point cloud data corresponding to the three-dimensional tooth scanning data;

[0153] The median plane and the occlusal plane of the dentition corresponding to the three-dimensional tooth scanning data are used as references to measure and obtain the first width data corresponding to each three-dimensional tooth scanning data.

[0154] Optionally, the data acquisition module is further used to: acquire a plurality of original three-dimensional scanning data using an oral scanning device;

[0155] Perform tooth crown segmentation on each original 3D scan data to obtain digital mesh data corresponding to each tooth;

[0156] The digital mesh data corresponding to all teeth are completed to obtain multiple three-dimensional tooth scanning data.

[0157] Through the automatic tooth arrangement model construction device of this embodiment, the corresponding automatic tooth arrangement model construction methods in the aforementioned multiple method embodiments can be implemented, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0158] So far, specific embodiments of the present application have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recorded in the claims can be performed in a different order and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing can be advantageous.

[0159] The present application is described with reference to the flowcharts and / or block diagrams of the methods according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0160] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0161] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods and devices. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0162] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0163] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for constructing an automated tooth arrangement model, characterized in that: The method comprises: Obtaining a plurality of three-dimensional tooth scanning data and tooth arrangement sample data corresponding to each of the three-dimensional tooth scanning data; wherein the three-dimensional tooth scanning data is used to characterize the state of all teeth in the oral cavity before tooth arrangement; and the tooth arrangement sample data is used to characterize the state of all teeth in the oral cavity after tooth arrangement; Obtaining first width data corresponding to each of the three-dimensional tooth scanning data; wherein the first width data is used to characterize the directional width between the teeth of the upper jaw and the directional width between the teeth of the lower jaw; According to all the three-dimensional tooth scanning data and the corresponding first width data and the tooth arrangement sample data, an automated tooth arrangement model is obtained by training with a preset deep learning algorithm; wherein, the input of the automated tooth arrangement model is the three-dimensional tooth scanning data and the first width data, and the output of the automated tooth arrangement model is the tooth arrangement calculation data, and the tooth arrangement calculation data is used to characterize the state of all teeth in the oral cavity after tooth arrangement.

2. The method according to claim 1, characterized in that The obtaining of the first width data corresponding to each of the three-dimensional tooth scanning data comprises: Performing farthest point sampling processing on each tooth represented by the three-dimensional tooth scanning data to obtain point cloud data corresponding to each three-dimensional tooth scanning data; wherein the point cloud data is used to represent the state of all teeth in the oral cavity before tooth arrangement in the form of point cloud; According to the point cloud data corresponding to each of the three-dimensional tooth scanning data, first width data corresponding to each of the three-dimensional tooth scanning data is measured and obtained.

3. The method according to claim 2, characterized in that The method of obtaining an automated tooth arrangement model by training with a preset deep learning algorithm based on all the three-dimensional tooth scanning data and the corresponding first width data and the tooth arrangement sample data comprises: Construct the initial tooth arrangement model; According to all the three-dimensional tooth scanning data and the corresponding first width data and the tooth arrangement sample data, the initial tooth arrangement model is trained for a preset number of rounds to obtain the automated tooth arrangement model; The step of training the initial tooth arrangement model according to all the three-dimensional tooth scanning data and the corresponding first width data and the tooth arrangement sample data comprises: According to the point cloud data corresponding to the three-dimensional tooth scanning data, the initial tooth arrangement model is used to calculate the centroid coordinate data, the first local feature data and the first global feature data corresponding to the three-dimensional tooth scanning data; wherein the centroid coordinate data is used to characterize the centroid coordinate values ​​of all teeth before tooth arrangement; the first local feature data is used to characterize the features of several dimensions of each tooth before tooth arrangement; the first global feature data is used to characterize the features of several dimensions of all teeth before tooth arrangement; According to the first width data corresponding to the three-dimensional tooth scanning data, the width embedding feature data corresponding to the three-dimensional tooth scanning data is obtained by extracting the initial tooth arrangement model; Using the initial tooth arrangement model, the width embedding feature data corresponding to the three-dimensional tooth scanning data and the first global feature data are fused to obtain second global feature data corresponding to the three-dimensional tooth scanning data; Using the initial tooth arrangement model, performing feature propagation processing on the first local feature data corresponding to the three-dimensional tooth scanning data to obtain second local feature data corresponding to the three-dimensional tooth scanning data; The second global feature data and the second local feature data corresponding to the three-dimensional tooth scanning data are spliced ​​and feature mapped using the initial tooth arrangement model to obtain predicted position feature data corresponding to the three-dimensional tooth scanning data; wherein the predicted position feature data is used to characterize the position features of all teeth after tooth arrangement predicted by the network; The predicted position feature data and the centroid coordinate data corresponding to the three-dimensional tooth scanning data are spliced ​​and predicted by using the initial tooth arrangement model, and the transformation parameter data corresponding to the three-dimensional tooth scanning data are calculated; wherein the transformation parameter data is used to characterize the translation parameter values ​​and rotation parameter values ​​of all teeth after tooth arrangement compared with before tooth arrangement; According to the transformation parameter data corresponding to the three-dimensional tooth scanning data and the tooth arrangement sample data, a loss calculation value is obtained by using a preset loss function; The initial tooth arrangement model is back-propagated according to the loss calculation value.

4. The method according to claim 3, characterized in that The extracting and obtaining the width embedding feature data corresponding to the three-dimensional tooth scanning data by using the initial tooth arrangement model according to the first width data corresponding to the three-dimensional tooth scanning data comprises: According to the first width data corresponding to the three-dimensional tooth scan data, a linear layer network with a preset number of layers is used, and Leaky ReLU is used as an activation function to obtain width embedding feature data corresponding to the three-dimensional tooth scan data.

5. The method according to claim 3, characterized in that: The using the initial tooth arrangement model to perform feature propagation processing on the first local feature data corresponding to the three-dimensional tooth scan data to obtain second local feature data corresponding to the three-dimensional tooth scan data comprises: A multi-layer Transformer encoder module is used as a feature propagation model to perform feature propagation processing on the first local feature data corresponding to the three-dimensional tooth scan data to obtain the second local feature data corresponding to the three-dimensional tooth scan data.

6. The method according to claim 3, characterized in that The step of calculating the loss calculation value using a preset loss function according to the transformation parameter data corresponding to the three-dimensional tooth scanning data and the tooth arrangement sample data comprises: Performing staggered processing on the tooth arrangement sample data corresponding to the three-dimensional tooth scanning data to obtain sample staggered data corresponding to the three-dimensional tooth scanning data; According to the sample misalignment data and the tooth arrangement sample data corresponding to the three-dimensional tooth scanning data, the initial tooth arrangement model is used to calculate and obtain the third local feature data and the fourth local feature data corresponding to the three-dimensional tooth scanning data; wherein the third local feature data and the fourth local feature data are both used to characterize the features of several dimensions of each tooth after tooth arrangement; A first loss value is obtained by calculating the cosine similarity function according to the predicted position feature data, the third local feature data and the fourth local feature data corresponding to the three-dimensional tooth scanning data.

7. The method according to claim 6, characterized in that The calculating and obtaining the first loss value by using a cosine similarity function according to the predicted position feature data, the third local feature data and the fourth local feature data corresponding to the three-dimensional tooth scanning data comprises: According to the predicted position feature data, the third local feature data and the fourth local feature data corresponding to the three-dimensional tooth scanning data, the formula Calculate and obtain the first loss value; in, is the first loss value; |L| is the number of all teeth; L is the set of all tooth labels; is the data corresponding to the i-th tooth in the predicted position feature data; is the data corresponding to the i-th tooth in the third local feature data in order; It is the data corresponding to the i-th tooth in order in the fourth local feature data.

8. The method according to claim 2, characterized in that: The step of measuring and obtaining first width data corresponding to each of the three-dimensional tooth scan data according to the point cloud data corresponding to each of the three-dimensional tooth scan data comprises: Fitting the median plane and occlusal plane of the dentition corresponding to the three-dimensional tooth scanning data according to the point cloud data corresponding to the three-dimensional tooth scanning data; The median plane and the occlusal plane of the dentition corresponding to the three-dimensional tooth scanning data are used as references to measure and obtain first width data corresponding to each three-dimensional tooth scanning data.

9. The method according to claim 1, characterized in that: The obtaining of a plurality of three-dimensional tooth scanning data and tooth arrangement sample data corresponding to each of the three-dimensional tooth scanning data comprises: Using oral scanning equipment to acquire multiple original three-dimensional scanning data; Performing tooth crown segmentation processing on each of the original three-dimensional scanning data to obtain digital grid data corresponding to each tooth; The digital mesh data corresponding to all teeth are completed to obtain multiple three-dimensional tooth scanning data.

10. A device for constructing an automated tooth arrangement model, characterized in that: The device comprises: A data acquisition module, used to obtain a plurality of three-dimensional tooth scanning data and tooth arrangement sample data corresponding to each of the three-dimensional tooth scanning data; wherein the three-dimensional tooth scanning data is used to characterize the state of all teeth in the oral cavity before tooth arrangement; and the tooth arrangement sample data is used to characterize the state of all teeth in the oral cavity after tooth arrangement; A width acquisition module, used for obtaining first width data corresponding to each of the three-dimensional tooth scanning data; wherein the first width data is used for representing the directional width between the teeth of the upper jaw and the directional width between the teeth of the lower jaw; A model training module is used to obtain an automated tooth arrangement model by training using a preset deep learning algorithm based on all the three-dimensional tooth scanning data and the corresponding first width data and the tooth arrangement sample data; wherein the input of the automated tooth arrangement model is the three-dimensional tooth scanning data and the first width data, and the output of the automated tooth arrangement model is tooth arrangement calculation data, and the tooth arrangement calculation data is used to characterize the state of all teeth in the oral cavity after tooth arrangement.

Images