Method and device for generating a three-dimensional dental model

Abstract

The specification provides a method and device for generating a three-dimensional dental model. The method comprises: preprocessing point cloud data of a three-dimensional dental model after orthodontic treatment in an existing data set, and extracting geometric features from the preprocessed point cloud data; obtaining a vector quantization variational autoencoder model to be trained, and training the model by: encoding the extracted geometric features into a hidden space based on the encoder of the vector quantization variational autoencoder model, and further reconstructing the encoded information in the hidden space into the point cloud data of the three-dimensional dental model after orthodontic treatment through the decoder; and training a diffusion model in the hidden space of the vector quantization variational autoencoder model, wherein the diffusion model is used for diffusion processing of the encoded information input into the hidden space to generate new encoded information, so that the decoder generates new point cloud data of the three-dimensional dental model after orthodontic treatment according to the new encoded information.

Description

TECHNICAL FIELD

[0001] Embodiments of the present specification relate to the technical field of computer technology, and particularly relate to a three-dimensional dental model generation method and device. BACKGROUND

[0002] Currently, malocclusion has become one of the world's three major oral diseases, and malocclusion can cause many oral-related diseases. Orthodontic treatment is the main means to solve malocclusion, and orthodontic treatment refers to arranging irregular or deformed teeth into good arrangement.

[0003] Traditional orthodontic treatment usually requires experienced dentists or dental technicians to manually design treatment plans. With the continuous development of machine learning technology, in recent years, automatic tooth alignment technology based on neural networks (such as TANet, PSTN, TADPM, Tooth Motion Diffusion, etc.) has also appeared to predict ideal tooth arrangement, thereby assisting dentists in developing orthodontic treatment plans.

[0004] In related technologies, automatic tooth alignment technology requires paired orthodontic three-dimensional dental models (i.e., pre-orthodontic three-dimensional dental models and post-orthodontic three-dimensional dental models) as training data. However, due to the difficulty and time-consuming nature of data collection and organization of pre- and post-orthodontic three-dimensional dental models, existing automatic tooth alignment technology faces the problem of insufficient training data, especially for complex cases (such as reverse jaw, reverse overbite, etc.) three-dimensional dental model data is very small; thereby leading to the prediction results of the existing automatic tooth alignment are not ideal.

[0005] In summary, there is a need to provide a simple and low-cost three-dimensional dental model generation scheme. SUMMARY

[0006] The embodiments of the present specification provide a three-dimensional dental model generation method and device.

[0007] According to a first aspect of the embodiments of the present specification, a three-dimensional dental model generation method is provided, the method comprising:

[0008] preprocessing point cloud data of a post-orthodontic three-dimensional dental model in an existing data set, and extracting geometric features of the post-orthodontic three-dimensional dental model from the preprocessed point cloud data of the post-orthodontic three-dimensional dental model;

[0009] obtaining a vector quantization variational autoencoder model to be trained, and training the vector quantization variational autoencoder model in the following manner:

[0010] The encoder based on the vector quantization variational autoencoder model encodes the extracted geometric features of the post-orthodontic three-dimensional dental model into a latent space, and further reconstructs the point cloud data of the post-orthodontic three-dimensional dental model from the encoded information in the latent space through the decoder of the vector quantization variational autoencoder model.

[0011] Wherein, in the latent space of the vector quantization variational autoencoder model, a diffusion model is also trained, which is used to diffuse the encoded information input into the latent space to generate new encoded information, so that the decoder generates new point cloud data of the post-orthodontic three-dimensional dental model according to the new encoded information.

[0012] Optionally, the point cloud data of the post-orthodontic three-dimensional dental model in the existing data set is preprocessed, and the geometric features of the post-orthodontic three-dimensional dental model are extracted from the preprocessed point cloud data of the post-orthodontic three-dimensional dental model, comprising:

[0013] The point cloud data of the post-orthodontic three-dimensional dental model in the existing data set is centralized and normalized to obtain a tooth sequence corresponding to the post-orthodontic three-dimensional dental model, which is formed by 32 teeth; wherein the tooth sequence is a 2x2x8 spatial structure.

[0014] Each tooth is divided into a voxel structure of rxrxr, and the PointNet network is used to extract the features of each tooth voxel, and all the extracted features of the teeth are aggregated to obtain the geometric features of the post-orthodontic three-dimensional dental model with a shape of [2r, 2r, 8r]; wherein r is the resolution of the tooth.

[0015] Optionally, the vector quantization variational autoencoder model includes a VQ-VAE model using a 3D U-Net network architecture;

[0016] The encoder based on the vector quantization variational autoencoder model encodes the extracted geometric features of the post-orthodontic three-dimensional dental model into a latent space, and further reconstructs the point cloud data of the post-orthodontic three-dimensional dental model from the encoded information in the latent space through the decoder of the vector quantization variational autoencoder model, comprising:

[0017] The encoder of the VQ-VAE model is used to perform multi-layer convolution operation on the extracted geometric features of the post-orthodontic three-dimensional dental model to encode the geometric features into a latent space to obtain a latent vector corresponding to the post-orthodontic three-dimensional dental model;

[0018] Further, the decoder of the VQ-VAE is used to perform multi-layer deconvolution operation on the latent vector in the latent space to obtain the point cloud data of the post-orthodontic three-dimensional dental model reconstructed based on the latent vector;

[0019] The point cloud data of the reconstructed post-orthodontic three-dimensional dental model is used to calculate a reconstruction loss function of the decoder with the point cloud data of the three-dimensional dental model in the existing data set.

[0020] Optionally, the reconstruction loss function of the decoder is calculated by using the point cloud data of the reconstructed post-orthodontic three-dimensional dental model and the point cloud data of the three-dimensional dental model in the existing data set, comprising:

[0021] The chamfer distance between each tooth in the point cloud data of the reconstructed post-orthodontic three-dimensional dental model and the point cloud data of the post-orthodontic three-dimensional dental model in the existing data set is calculated respectively.

[0022] The average value of the chamfer distance of each pair of teeth is calculated to obtain the reconstruction loss function.

[0023] Optionally, the diffusion model is used to diffuse the encoded information input into the latent space to generate new encoded information, so that the trained decoder generates new point cloud data of the post-orthodontic three-dimensional dental model according to the new encoded information, comprising:

[0024] The diffusion model adds noise to the latent vector of the feature of the real post-orthodontic three-dimensional dental model encoded into the latent space by the encoder, and uses the denoiser of the diffusion model for denoising to obtain a new latent vector; further, the trained decoder generates new point cloud data of the post-orthodontic three-dimensional dental model according to the new latent vector.

[0025] Optionally, it further comprises:

[0026] Obtain the pre-orthodontic three-dimensional dental model of the specified dental arrangement style in the existing data set;

[0027] Extract the style features of the dental arrangement in the pre-orthodontic three-dimensional dental model of the specified dental arrangement style, and extract the shape features of the point cloud data of the new post-orthodontic three-dimensional dental model generated by the trained decoder;

[0028] The style features and shape features are used as double conditions of the diffusion model to guide the diffusion model to generate the rigid transformation parameters of each tooth in the required post-orthodontic three-dimensional dental model;

[0029] The generated rigid transformation parameters are applied to each tooth in the post-orthodontic three-dimensional dental model to obtain the point cloud data of the pre-orthodontic three-dimensional dental model corresponding to the post-orthodontic three-dimensional dental model, and the point cloud data of the pre-orthodontic three-dimensional dental model has the specified dental arrangement style.

[0030] Optionally, the extraction of the style features of the dental arrangement in the pre-orthodontic three-dimensional dental model of the specified dental arrangement style comprises:

[0031] The point cloud data of the orthodontic pre-three-dimensional dental model of the specified dental arrangement style is centralized and normalized to obtain a dental sequence arranged by 32 teeth; wherein, the dental sequence is a spatial structure of 2x2x8;

[0032] Each tooth is divided into a voxel structure of rxrxr, and the PointNet network is used to extract the features of the voxels of each tooth, and the features extracted from all teeth are aggregated to obtain geometric features with a shape of [2r, 2r, 8r]; wherein, r is the resolution of the tooth;

[0033] The average value pooling and the variance pooling are used inside the voxel, and after linear transformation, the style features with a shape of [2r, 2r, 8r] are obtained.

[0034] Optionally, the trained decoder generates new shape features of the point cloud data of the orthodontic post-three-dimensional dental model, including:

[0035] The point cloud data of the orthodontic post-three-dimensional dental model generated by the trained decoder is centralized and normalized to obtain a dental sequence arranged by 32 teeth; wherein, the dental sequence is a spatial structure of 2x2x8;

[0036] Each tooth is divided into a voxel structure of rxrxr, and the PointNet network is used to extract the features of the voxels of each tooth, and the features extracted from all teeth are aggregated to obtain geometric features with a shape of [2r, 2r, 8r]; wherein, r is the resolution of the tooth;

[0037] The maximum pooling is used inside each tooth to aggregate the features of all voxels to obtain the shape features.

[0038] Optionally, the diffusion model includes a diffusion model based on a 3D U-Net network architecture;

[0039] The style features and the shape features are used as the double conditions of the diffusion model to guide the diffusion model to generate the rigid transformation parameters of each tooth of the orthodontic post-three-dimensional dental model that meets the requirements, including:

[0040] The style features and the shape features are input into each layer of the 3D U-Net network through a multi-layer perception, to guide the diffusion model to generate the rigid transformation parameters of the teeth, and to output the rigid transformation parameters of each tooth of the orthodontic post-three-dimensional dental model.

[0041] Optionally, the generated rigid transformation parameters are applied to each tooth in the post-orthodontic three-dimensional dental model to obtain a pre-orthodontic three-dimensional dental model corresponding to the post-orthodontic three-dimensional dental model, and the pre-orthodontic three-dimensional dental model has the specified dental arrangement style, including:

[0042] Each tooth in the post-orthodontic three-dimensional dental model is translated and rotated according to the translation parameters and rotation parameters in the generated rigid transformation parameters to obtain point cloud data of a pre-orthodontic three-dimensional dental model corresponding to the post-orthodontic three-dimensional dental model, and the point cloud data of the pre-orthodontic three-dimensional dental model has the specified dental arrangement style.

[0043] According to a second aspect of the embodiments of the present specification, a three-dimensional dental model generation device is provided, the device comprising:

[0044] A feature extraction unit pre-processes point cloud data of a post-orthodontic three-dimensional dental model in an existing data set, and extracts geometric features of the post-orthodontic three-dimensional dental model from the pre-processed point cloud data of the post-orthodontic three-dimensional dental model;

[0045] A model training unit obtains a vector quantization variational autoencoder model to be trained, and trains the vector quantization variational autoencoder model by: encoding the extracted geometric features of the post-orthodontic three-dimensional dental model to a latent space based on an encoder of the vector quantization variational autoencoder model, and further reconstructing the encoded information in the latent space to the point cloud data of the post-orthodontic three-dimensional dental model by a decoder of the vector quantization variational autoencoder model;

[0046] A data generation unit further trains a diffusion model in the latent space of the vector quantization variational autoencoder model, the diffusion model being used for diffusion processing of the encoded information input to the latent space to generate new encoded information, so that the trained decoder generates new point cloud data of the post-orthodontic three-dimensional dental model according to the new encoded information.

[0047] The embodiments of the present specification provide a three-dimensional dental model generation scheme. The geometric features of the post-orthodontic three-dimensional dental model point cloud data are extracted through the inherent structure of the post-orthodontic three-dimensional dental model point cloud data, then the geometric features are encoded to the latent space through the encoder of the vector quantization variational autoencoder model, and the encoded information in the latent space is reconstructed back to the post-orthodontic three-dimensional dental model point cloud data through the decoder to train the model, so as to learn the relationship between the encoded information of the geometric features in the latent space and the post-orthodontic three-dimensional dental model point cloud data; in this way, the diffusion model can be applied in the latent space to generate new encoded information for the existing encoded information, and then the trained decoder is used to reconstruct the point cloud data of the new post-orthodontic three-dimensional dental model according to the new encoded information. The above diffusion processing can increase the number of post-orthodontic three-dimensional dental models in the data set.

[0048] In another aspect, the point cloud data of the new post-orthodontic three-dimensional dental model generated by the diffusion model and the decoder is used to apply a rigid transformation to each tooth in the new post-orthodontic three-dimensional dental model to change the tooth arrangement, thereby obtaining the point cloud data of the pre-orthodontic three-dimensional dental model. In this way, the size and shape of the teeth in the pre- and post-orthodontic three-dimensional dental models can be ensured to be consistent. In this way, the corresponding pre-orthodontic three-dimensional dental model can be added to the newly added post-orthodontic three-dimensional dental model, thereby increasing the volume of the existing data set.

[0049] In another aspect, the style features of the pre-orthodontic three-dimensional dental model with a specified tooth arrangement style in the existing data set are used to control the generated pre-orthodontic three-dimensional dental model to also have the specified tooth arrangement style. In this way, pre-orthodontic three-dimensional dental models with multiple different tooth arrangement styles can be generated, thereby increasing the diversity of the tooth arrangement styles of the pre-orthodontic three-dimensional dental models in the data set. BRIEF DESCRIPTION OF DRAWINGS

[0050] Figure 1 A system architecture diagram of generating a three-dimensional dental model is provided for an embodiment of the present specification;

[0051] Figure 2 A flowchart of a method for generating a three-dimensional dental model is provided for an embodiment of the present specification;

[0052] Figure 3 A schematic diagram of a post-orthodontic three-dimensional dental model and a tooth sequence is provided for an embodiment of the present specification;

[0053] Figure 4 A schematic diagram of a style extractor and a shape extractor is provided for an embodiment of the present specification;

[0054] Figure 5 A schematic diagram of three-dimensional dental models with different tooth arrangement styles is provided for an embodiment of the present specification;

[0055] Figure 6 A comparison schematic diagram of a three-dimensional dental model is provided for an embodiment of the present specification;

[0056] Figure 7 A hardware structure diagram of a three-dimensional dental model generation device is provided for an embodiment of the present specification;

[0057] Figure 8 A module of a three-dimensional dental model generation device is provided for an embodiment of the present specification. DETAILED DESCRIPTION

[0058] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The description of the exemplary embodiments is intended to apply to various alternative embodiments as well. The following description is not limited to the exemplary embodiments, but rather, is applicable to any apparatus and / or method which falls within the scope of the present description. Although the following description can use terms like "first", "second", "third", etc., such terms should not be construed as limiting the description to a particular embodiment. Rather, such terms can be used merely for distinguishing between similar processes, structures, or entities, etc. For example, a process described as a "second" process can also be a "first" process without departing from the scope of the present description. As used herein, the term "if' can be interpreted to mean "when" or "upon" or "in response to determining" taking into account the context in which the term is used.

[0059] The terminology used in the present description is for the purpose of describing particular embodiments only and is not intended to be limiting of the present description. As used in the description of the embodiments and the appended claims, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It also will be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0060] It will be understood that, although the terms first, second, third, etc. can be used herein to describe various information, the information should not be limited by these terms. These terms are used only to distinguish one piece of information from another. For example, a first information could be termed a second information, and, similarly, a second information could be termed a first information without departing from the scope of the present description. As used herein, the word "if' can be interpreted to mean "when" or "upon" or "in response to determining" taking into account the context in which the term is used.

[0061] In the related art, automatic tooth alignment technology needs a pair of orthodontic three-dimensional dental models (i.e., a three-dimensional dental model before orthodontics and a three-dimensional dental model after orthodontics) as training data, but since the data collection and arrangement of the three-dimensional dental models before and after orthodontics is a difficult and time-consuming work, the existing automatic tooth alignment technology faces the problem of insufficient training data, especially the three-dimensional dental model data of complex cases (such as reverse jaw, reverse overbite, etc. dental arrangement style) is very small; thus, the prediction result of the existing automatic tooth alignment network is not ideal.

[0062] The solution to this problem is to use AI technology to automatically generate point cloud data for paired orthodontic 3D dental models. Several general data generation techniques already exist. For example, PointFlow technology uses a normalizing flow method to hierarchically generate 3D point cloud data, first generating the shape distribution, then generating the distribution of sampled points for a given shape. Another example is DPM (Denoising Diffusion Probabilistic Models), which applies a diffusion model to 3D data to iteratively generate 3D point cloud data. Yet another example is LION technology, which first uses a VAE to encode the original data into a latent space, then applies a diffusion model to the latent space to generate new encodings based on the original data's encoding results, thus achieving the goal of generating new data.

[0063] However, when these general data generation techniques are applied to generate point cloud data for paired orthodontic 3D dental models, the resulting 3D dental model generation task differs from traditional 3D data generation task, leading to unsatisfactory tooth shapes in the generated 3D dental models that fail to reflect the precise shape of the teeth.

[0064] Specifically, when applied to generating paired orthodontic 3D dental model point cloud data, the task objective can be:

[0065] 1. All individual teeth in the generated three-dimensional dental model must be separated so that the features of each tooth can be easily extracted and transformations applied. Otherwise, each tooth would need to be further segmented.

[0066] 2. The number of teeth in the generated 3D dental model is not fixed. At the same time, it is necessary to maintain the coordination of the size and shape of different teeth in the entire 3D dental model. Therefore, it is not possible to generate each tooth separately and then stitch them together.

[0067] 3. The generated three-dimensional dental models need to be paired, representing the three-dimensional dental models before and after orthodontics respectively, and the shape and size of the same tooth need to be consistent before and after orthodontics.

[0068] Common point cloud generation methods, such as PointFlow and LION, generate data that is an unsegmented whole and requires additional segmentation processing when applied to downstream tasks. Furthermore, common point cloud generation methods do not utilize prior knowledge specific to the dental jaw, resulting in an unsatisfactory shape of the generated 3D dental jaw model that fails to reflect the precise shape of the teeth.

[0069] In view of this, this invention proposes a simple and low-cost scheme for generating three-dimensional dental and jaw models. Geometric features are extracted from the inherent structure of the point cloud data of the post-orthodontic three-dimensional dental and jaw models. Then, a vector quantization variational autoencoder model is used to encode these geometric features into the latent space, and a decoder reconstructs the encoded information from the latent space back into the post-orthodontic three-dimensional dental and jaw model point cloud data. This process trains the model to learn the relationship between the encoded information of geometric features in the latent space and the post-orthodontic three-dimensional dental and jaw model point cloud data. A diffusion model can then be applied to the latent space to generate new encoded information from the existing encoded information. Finally, the trained decoder is used to reconstruct new post-orthodontic three-dimensional dental and jaw model point cloud data based on the new encoded information. This diffusion process can increase the number of post-orthodontic three-dimensional dental and jaw models in the dataset.

[0070] On the other hand, for the new post-orthodontic 3D dental model point cloud data generated by the diffusion model and decoder, a rigid transformation is applied to each tooth in the new post-orthodontic 3D dental model to change the tooth alignment, resulting in new pre-orthodontic 3D dental model point cloud data. This ensures that the size and shape of the teeth in the pre- and post-orthodontic 3D dental models are consistent. This allows for the addition of corresponding pre-orthodontic 3D dental models to the newly added post-orthodontic 3D dental models, thereby increasing the size of the existing dataset.

[0071] On the other hand, by utilizing the style features of pre-orthodontic 3D dental models with a specified tooth arrangement style in the existing dataset, the generated pre-orthodontic 3D dental models can also be controlled to have the specified tooth arrangement style. In this way, a variety of pre-orthodontic 3D dental models with different tooth arrangement styles can be generated, thereby increasing the diversity of tooth arrangement styles in the pre-orthodontic 3D dental models in the dataset.

[0072] Please refer to the following: Figure 1 The system architecture diagram shown is for a 3D dental model generation system. This system can include a two-stage architecture. The first stage generates point cloud data of the 3D dental model after orthodontic treatment, which is arranged in a relatively regular pattern. Figure 1 The first stage is shown in the figure. Then, according to the specified tooth alignment style, the pre-orthodontic 3D dental model point cloud data corresponding to the post-orthodontic 3D dental model point cloud data reconstructed in the first stage is generated. Figure 1 (The second phase shown in the diagram).

[0073] like Figure 1 As shown, a vector quantization variational autoencoder model (such as...) can be used in the first stage. Figure 1 The VQ-VAE shown in the figure generates more different post-orthodontic 3D dental model point cloud data based on the input post-orthodontic 3D dental model point cloud data. The generation process of the vector quantization variational autoencoder model will be described in subsequent embodiments.

[0074] In the second stage, based on the post-orthodontic 3D dental model point cloud data generated in the first stage, and combining it with the tooth arrangement style of other specified pre-orthodontic 3D dental models, a corresponding pre-orthodontic 3D dental model point cloud data with the specified tooth arrangement style is generated. Since the second stage actually generates rigid body transformation parameters for each tooth, the tooth morphology and size of the pre- and post-orthodontic 3D dental models can remain consistent.

[0075] Please refer to the following. Figure 2 The diagram shows a method for generating a three-dimensional dental model, which may include the following steps:

[0076] Step 110: Preprocess the point cloud data of the post-orthodontic 3D dental model in the existing dataset, and extract the geometric features of the post-orthodontic 3D dental model from the preprocessed point cloud data.

[0077] Since the orthodontic post-orthodontic 3D dental models collected in the dataset may come from different data sources, and these different data sources have different data formats or data types, in order to facilitate subsequent processing, the orthodontic post-orthodontic 3D dental models in the dataset can be preprocessed to transform them into a unified form of 3D dental models.

[0078] In one exemplary embodiment, the preprocessing may include centralization and normalization processes;

[0079] Centering refers to the operation of adjusting the center point of the 3D dental model to the origin (0, 0, 0). This is achieved by subtracting the existing vector between the center point and the origin from the coordinates of each vertex, thus removing data offset. Centering eliminates data offset, centered the data distribution around zero; it also facilitates subsequent operations such as rotation and translation of the model, avoiding additional calculations and adjustments.

[0080] Normalization refers to scaling the vertex coordinates of a 3D dental model to a specific range ([-1,1]). By dividing the coordinates of each vertex by the modulus from the vertex coordinates to the origin, the goal is to unify all 3D dental models to the same order of magnitude, allowing neural networks to process data from different ranges more efficiently. Normalization maps data to a fixed range, preventing excessively large or small feature values ​​from negatively impacting model training. It is particularly important for algorithms that require distance or similarity calculations (such as K-means clustering and neural networks), ensuring that each feature contributes equally to the model.

[0081] When using centralized and normalized processing, step 210 above may include:

[0082] Step 211: Center and normalize the point cloud data of the orthodontic post-three-dimensional dental model in the existing dataset to obtain a tooth sequence of 32 teeth corresponding to the orthodontic post-three-dimensional dental model; wherein the tooth sequence is a 2×2×8 spatial structure.

[0083] In this step, the tooth sequence is as follows: Figure 3 As shown in (b), it can be a sequence of 32 teeth; where, for the position of a missing tooth, the missing tooth can be filled in its sequence with 0 (or other preset characters or numbers). This arrangement of teeth sequence allows for good interaction of information between adjacent pairs of teeth, symmetrical pairs of teeth in the same jaw (maxilla or mandible), and symmetrical pairs of teeth in both the upper and lower jaws.

[0084] Step 213: Divide each tooth into an r×r×r voxel structure, use the PointNet network to extract features from the voxels of each tooth, and aggregate the extracted features of all teeth to obtain the geometric features of the orthodontic 3D dental model with shape [2r, 2r, 8r]; where r is the resolution of the tooth.

[0085] It should be noted that the three-dimensional dental models mentioned in this manual, whether before or after orthodontic treatment, generally refer to a large number of three-dimensional spatial coordinate points. These three-dimensional spatial coordinate points can constitute the shape of the three-dimensional dental model. Generally, these three-dimensional spatial coordinate points can be referred to as point cloud data, that is, the point cloud data of the three-dimensional dental model.

[0086] The voxel structure of a tooth refers to the discretized representation of a tooth in three-dimensional space. The voxel technique is used to divide the point cloud of each tooth into many small cubic units in the form of a mesh to describe its details and composition. Each voxel represents a small region in space.

[0087] PointNet is a deep learning network architecture specifically designed for processing point cloud data. It can effectively handle irregular and unordered point cloud data without requiring the conversion of point cloud data into a mesh or voxel representation as in traditional methods. By extracting and aggregating features from each tooth voxel using the PointNet network, a 3D dental model with a shape of [2r, 2r, 8r] can be obtained.

[0088] Step 120: Obtain the vector quantization variational autoencoder model to be trained, and train the vector quantization variational autoencoder model in the following manner:

[0089] The encoder based on the vector quantization variational autoencoder model encodes the extracted geometric features of the post-orthodontic 3D dental model into the latent space, and further reconstructs the point cloud data of the post-orthodontic 3D dental model from the encoded information in the latent space through the decoder of the vector quantization variational autoencoder model.

[0090] In this step, after obtaining the geometric features of the three-dimensional dental model with shape [2r, 2r, 8r] in step 210, it can be used to train the vector quantization variational autoencoder model.

[0091] The vector quantization variational autoencoder model can include an encoder and a decoder. The encoder is used to encode the geometric features of the input three-dimensional dental model into the latent space, while the decoder is used to decode the encoded information obtained by the encoder in the latent space to reconstruct the point cloud data of the corresponding orthodontic three-dimensional dental model.

[0092] In one exemplary embodiment, the vector quantization variational autoencoder model may include a VQ-VAE model employing a 3D U-Net network architecture.

[0093] Among them, VQ-VAE (Vector Quantized Variational Autoencoder) is a variant of variational autoencoder (VAE). VQ-VAE discretizes the latent space by introducing vector quantization, thereby generating a higher quality 3D dental model.

[0094] In addition, since the dental data involved in this embodiment is three-dimensional data, in order for VQ-VAE to generate point cloud data of three-dimensional dental models, a VQ-VAE model based on the 3D U-Net network architecture can be used.

[0095] The 3D U-Net network mentioned here is a deep learning model that is widely used in the processing of 3D image data (such as CT, MRI and other medical images) and 3D voxel models.

[0096] By using the 3D U-Net network as the base network of the VQ-VAE model, high-quality three-dimensional dental models can be generated using the VQ-VAE model.

[0097] When using a VQ-VAE model with a 3D U-Net network architecture, in step 220 above, the encoder based on the vector quantization variational autoencoder model encodes the extracted geometric features of the post-orthodontic 3D dental model into the latent space, and further reconstructs the point cloud data of the post-orthodontic 3D dental model from the encoded information in the latent space through the decoder of the vector quantization variational autoencoder model, which may further include:

[0098] Step 221: Use the encoder of the VQ-VAE model to perform multi-layer convolution operation on the extracted geometric features of the three-dimensional dental model after orthodontic treatment, so as to encode the geometric features into the latent space and obtain the latent vector corresponding to the three-dimensional dental model after orthodontic treatment.

[0099] After obtaining the geometric features of a 3D dental model with shape [2r, 2r, 8r], a VQ-VAE encoder based on a 3D U-Net network is used to encode the geometric features of the entire 3D dental model into a small-dimensional latent space through multi-layer convolution operations, thus obtaining latent vectors.

[0100] Step 223: Further use the decoder of the VQ-VAE to perform multi-layer deconvolution operation on the latent vectors in the latent space to obtain point cloud data of the orthodontic three-dimensional dental model reconstructed based on the latent vectors.

[0101] The VQ-VAE decoder uses multiple deconvolution operations to decode the latent vectors into point cloud data of the original orthodontic 3D dental model. Additionally, it can output a mask. The reconstructed 3D dental model's point cloud data can contain a fixed 32 teeth, and this mask can be used to indicate which of these 32 teeth should appear in the final output.

[0102] Step 225: Calculate the reconstruction loss function of the decoder using the point cloud data of the reconstructed three-dimensional dental model after orthodontic treatment and the point cloud data of the three-dimensional dental model in the existing dataset.

[0103] Since the default or initial model parameters are usually not optimal, it is necessary to use a dataset to perform several rounds of iterative training to achieve model convergence and thus determine the optimal model parameters.

[0104] Taking the VQ-VAE model, a vector quantization variational autoencoder model with a 3D U-Net network architecture, as an example, the point cloud data of the post-orthodontic 3D dental model in the existing dataset can be regarded as the real result; while the point cloud data of the post-orthodontic 3D dental model reconstructed after the decoder can be regarded as the prediction result. Usually, there is an error between the prediction result and the real result, and the reconstruction loss function is a function used to measure this error between the prediction result and the real result. The reconstruction loss function can be used to adjust the encoder model parameters according to the measured error.

[0105] For example, calculating the reconstruction loss function of the decoder may include:

[0106] Calculate the chamfer distance between the point cloud data of the reconstructed post-orthodontic 3D dental model and the point cloud data of the post-orthodontic 3D dental model in the existing dataset for each tooth; then, calculate the average chamfer distance for each pair of teeth to obtain the reconstruction loss function.

[0107] In this embodiment, the reconstruction loss is calculated during model training by calculating the chamfer distance between corresponding individual teeth between the actual input result and the predicted output result. The calculation formula is as follows:

[0108]

[0109] Where S1 represents the predicted result (i.e., the point cloud data of the reconstructed three-dimensional dental model after orthodontic treatment), and S2 represents the true result (i.e., the point cloud data of the three-dimensional dental model after orthodontic treatment in the dataset). This represents the sum of the minimum distances from any point x in the prediction result S1 to S2; Let S1 be the sum of the minimum distances from any point y in the true result S2 to S1.

[0110] The larger the value of the chamfer distance Chamfer(S1,S2), the greater the difference between the actual and predicted results; conversely, the smaller the value of the chamfer distance, the smaller the difference between the actual and predicted results, which means a better reconstruction effect of the three-dimensional dental model point cloud data after orthodontic treatment. Finally, the reconstruction loss function is obtained by averaging the chamfer distance of each pair of teeth.

[0111] By calculating the reconstruction loss function, the model parameters are iteratively optimized, and the model is then trained again based on the adjusted parameters. Through multiple iterations, the model parameters are gradually optimized, minimizing the error between the predicted and actual results, until the model converges or meets the convergence criteria.

[0112] Step 230: A diffusion model is also trained in the latent space of the vector quantization variational autoencoder model. The diffusion model is used to diffuse the encoded information input into the latent space to generate new encoded information, so that the trained decoder can generate new point cloud data of the orthodontic three-dimensional dental model based on the new encoded information.

[0113] In this embodiment, the vector quantization variational autoencoder model also includes a diffusion model in the latent space. To adapt to 3D data such as 3D dental model point cloud data, the diffusion model can also be a diffusion model based on a 3D U-Net network architecture. The mechanism of the diffusion model is to gradually add noise to the input data, and then use a denoiser to recover the original data from the noise. The denoiser is trained by repeatedly performing this process.

[0114] In an exemplary embodiment, after the vector quantization variational autoencoder model is trained, geometric features of the post-orthodontic 3D dental model or other real post-orthodontic 3D dental models in the existing dataset can be input into the trained vector quantization variational autoencoder model, and the encoder encodes the geometric features into the latent space. At this time, the diffusion model in the latent space can add noise to the latent vectors encoded into the latent space, and use the denoiser of the diffusion model to denoise them to obtain new latent vectors. Then, the trained decoder is further used to generate new point cloud data of the post-orthodontic 3D dental model based on the new latent vectors.

[0115] The above noise-adding process can add noise to the latent vectors in the latent space. This noise can include random noise sampled from an isotropic Gaussian distribution.

[0116] Through the above embodiments, geometric features are extracted using the inherent structure of the point cloud data of the post-orthodontic 3D dental model. Then, a vector quantization variational autoencoder model is used to encode the geometric features of the 3D dental model into the latent space, and a decoder reconstructs the encoded information from the latent space back into the point cloud data of the post-orthodontic 3D dental model. This model training learns the relationship between the encoded information of the geometric features of the 3D dental model in the latent space and the point cloud data of the post-orthodontic 3D dental model. In this way, a diffusion model can be applied to the latent space to generate new encoded information from the existing encoded information. Then, the trained decoder is used to reconstruct new point cloud data of the post-orthodontic 3D dental model based on the new encoded information. This diffusion process can increase the number of post-orthodontic 3D dental models in the dataset.

[0117] In practical applications, as mentioned above, automatic tooth alignment technology requires paired orthodontic 3D dental model point cloud data (i.e., pre-orthodontic 3D dental model point cloud data and post-orthodontic 3D dental model point cloud data) as training data. Although the above embodiments can generate more post-orthodontic 3D dental model point cloud data, it is still necessary to generate paired pre-orthodontic 3D dental model point cloud data for it. Furthermore, in order to improve the diversity of the dataset, these pre-orthodontic 3D dental model point cloud data also need to have diverse tooth arrangement styles (e.g., tooth arrangement styles of various complex cases).

[0118] In view of the above, this specification states that... Figure 2 Based on the method embodiments shown, the following steps may also be included (corresponding to): Figure 1 The second phase (in the process):

[0119] Step 240: Obtain a pre-orthodontic 3D dental model with a specified tooth alignment style from an existing dataset;

[0120] Step 250: Extract the style features of the tooth arrangement in the pre-orthodontic 3D dental model with the specified tooth arrangement style, and extract the shape features of the point cloud data of the new post-orthodontic 3D dental model generated by the trained decoder.

[0121] For example, the style features and shape features can be respectively achieved by means of, for example, Figure 4 The style feature extractor and shape feature extractor shown in the figure are used for extraction.

[0122] In an exemplary embodiment, extracting the style features of the tooth arrangement in the pre-orthodontic three-dimensional dental model of the specified tooth arrangement style includes:

[0123] The point cloud data of the pre-orthodontic three-dimensional dental model with the specified tooth arrangement style is centered and normalized to obtain a tooth sequence consisting of 32 teeth; wherein the tooth sequence is a 2×2×8 spatial structure.

[0124] Each tooth is divided into an r×r×r voxel structure. The PointNet network is used to extract features from each tooth voxel. The extracted features from all teeth are aggregated to obtain geometric features of shape [2r, 2r, 8r], where r is the resolution of the tooth.

[0125] Within the voxel, mean pooling and variance pooling are used, and after linear transformation, style features with shape [2r, 2r, 8r] are obtained.

[0126] In this embodiment, the centering and normalization processes, voxel structure, and PointNet network feature extraction are the same as in step 210 above, and will not be repeated here.

[0127] The difference from step 210 above is that, after extracting the dental features, mean pooling and variance pooling (e.g., ...) are used within the voxels. Figure 4 The MP and SP shown in the style feature extractor are transformed linearly to obtain style features with the same shape [2r, 2r, 8r]. This process can be referenced by the following formula:

[0128]

[0129] in, This represents the style feature extracted from the i-th voxel.

[0130] In one exemplary embodiment, the shape features of the point cloud data of the new orthodontic post-three-dimensional dental model generated by the trained decoder are extracted, including:

[0131] The point cloud data of the new orthodontic three-dimensional dental model generated by the trained decoder is centered and normalized to obtain a tooth sequence consisting of 32 teeth; wherein the tooth sequence is a 2×2×8 spatial structure.

[0132] Each tooth is divided into an r×r×r voxel structure. The PointNet network is used to extract features from each tooth voxel. The extracted features from all teeth are aggregated to obtain geometric features of shape [2r, 2r, 8r], where r is the resolution of the tooth.

[0133] Inside each tooth, maximum pooling is used to aggregate the features of all voxels to obtain shape features.

[0134] In this embodiment, the centering and normalization processes, voxel structure, and PointNet network extraction of dental features are the same as in step 210 above, and will not be repeated here.

[0135] The difference from step 210 above is that, after extracting the geometric features, max pooling (e.g., ...) is used inside each tooth. Figure 4 The shape feature extractor shown in MaxP aggregates the features of all voxels to obtain shape features.

[0136] Step 160: Use the style features and shape features as dual conditions for the diffusion model to guide the diffusion model to generate rigid transformation parameters for each tooth in the orthodontic post-three-dimensional dental model that meet the requirements.

[0137] As mentioned above, diffusion models can include diffusion models based on the 3D U-Net network architecture for processing three-dimensional spatial data such as three-dimensional dental models.

[0138] To incorporate style and shape features as conditional inputs to the denoiser of the diffusion model, a multilayer perceptron (MLP) can be used to fuse these two features and pass them as inputs to the 3D U-Net network.

[0139] The output of each layer of the 3D U-Net network carries progressively extracted spatial information. The conditional vectors generated by the MLP are combined with the 3D U-Net features of each layer, allowing style and shape features to be incorporated into the generation process at each layer. To ensure that the features of each layer are effectively influenced by style and shape features, the output of the MLP (as a conditional vector) can be weighted and merged or concatenated with the output of that layer. This approach ensures that each layer can perform denoising guided by features and generate more accurate rigid body transformation parameters.

[0140] The objective of the diffusion model can be set as the rigid transformation parameters of the teeth. Through a multi-step denoising process, the diffusion model gradually recovers the rigid transformation parameters of each tooth from random noise. Therefore, each round of denoising in the diffusion model is influenced by the output of the previous round, and conditional features are used to optimize the generation of rigid transformations.

[0141] Since the points in the paired 3D dental models in the dataset are in one-to-one correspondence—that is, the points constituting the same tooth in the paired pre-orthodontic 3D dental models and post-orthodontic 3D dental models, only the 3D coordinate positions of the points are different in the pre-orthodontic 3D dental models and post-orthodontic 3D dental models—the rigid body transformation parameters of each tooth from post-orthodontic to pre-orthodontic transformation can be calculated based on the correspondence of points in the paired 3D dental models.

[0142] The rigid body transformation parameters can be used to describe how teeth change from their pre-orthodontic position to their post-orthodontic position through translation, rotation, and other transformations.

[0143] Step 270: Apply the generated rigid transformation parameters to each tooth in the post-orthodontic 3D dental model to obtain point cloud data of the pre-orthodontic 3D dental model corresponding to the post-orthodontic 3D dental model, and the point cloud data of the pre-orthodontic 3D dental model has the specified tooth arrangement style.

[0144] In an exemplary embodiment, the rigid body transformation parameters include 9-dimensional rigid body transformation parameters for each of the 32 teeth, including 3-dimensional translation parameters in three-dimensional space (i.e., translation parameters of the X-axis, Y-axis, and Z-axis) and 6-dimensional rotation parameters in three-dimensional space (such as rotation parameters along the X-axis, Y-axis, Z-axis, X and Y-axis, X and Z-axis, and Y and Z-axis).

[0145] For any tooth i, x i =(m i ,r i )∈R 9 The 9-dimensional rigid body transformation parameters of the i-th tooth can be represented, where i is an integer in the range [1, 32], and m...i ∈R 3 Let r be the 3D translation parameter of the i-th tooth, and r be the translation parameter of the i-th tooth. i ∈R 6 Let be the 6-dimensional rotational parameters of the i-th tooth.

[0146] When using 6D rotation parameter r i When representing rotation in three-dimensional space, the 6D rotation parameter r can be used. i ∈R 6 Transformed into a 3×3 rotation matrix R i Then, translate the i-th tooth to the origin, and then apply the rotation matrix R. i Apply a rotational transformation; and translate the i-th tooth back to its original position, then according to the translation parameter m i Perform a translation.

[0147] By using rigid body transformation parameters to translate and rotate each tooth of the post-orthodontic 3D dental model in the manner described above, a pre-orthodontic 3D dental model corresponding to the post-orthodontic 3D dental model can be generated, and the pre-orthodontic 3D dental model has a specified tooth arrangement style (i.e., the tooth arrangement style of the pre-orthodontic 3D dental model specified in step 250).

[0148] Through the above embodiments, on the one hand, a new post-orthodontic 3D dental model is generated using a diffusion model and a decoder. Then, a rigid transformation is applied to each tooth in this new post-orthodontic 3D dental model to change the tooth alignment, resulting in a pre-orthodontic 3D dental model. This ensures that the size and shape of the teeth in the pre- and post-orthodontic 3D dental models are consistent. This allows for the addition of a corresponding pre-orthodontic 3D dental model to the newly added post-orthodontic 3D dental model generated by the diffusion processing, thereby increasing the size of the existing dataset.

[0149] On the other hand, by utilizing the tooth arrangement styles of pre-orthodontic 3D dental models in existing datasets, the tooth arrangement style type of the generated pre-orthodontic 3D dental models can be controlled. This allows for the generation of various pre-orthodontic 3D dental models with different tooth arrangement styles, thereby increasing the diversity of tooth arrangement styles in the dataset. Figure 5 The diagram shows a variety of three-dimensional dental models of different tooth alignment styles before orthodontic treatment.

[0150] In practical applications, the data quality of the generated three-dimensional dental model after orthodontic treatment can also be evaluated. A schematic diagram comparing the three-dimensional dental model generated using the embodiments provided in this specification with existing commonly used data generation techniques is shown below, based on the method for generating the three-dimensional dental model provided in this specification.

[0151] Existing common data generation technologies include PointFlow, DPM, LION, PVD, and DiT-3D. Following the approach of LION, the CD and CMD metrics of 1-NNA are used to evaluate the quality of the generated results. Furthermore, to assess the uniqueness of the generated data, a [specific definition could be needed]. As an indicator, S is defined as:

[0152] S={fi:Chamfer(fi,fj)>r,j <i,i,j∈{1…N}

[0153] Among them, f i Let |S| represent the i-th sample in the dataset, and r be a predefined distance threshold (set to 1 cm in subsequent experiments). In this way, |S| can represent the number of unique samples in the dataset, while U... CD It can represent the proportion of unique samples.

[0154] The final experimental results are shown in the table below (bolded data indicates good performance):

[0155] CD (%, down) EMD (%, down) U CD (%,↑) PointFlow 97.62 83.88 62.22 DPM 89.25 74.50 75.69 PVD 84.87 78.12 49.58 LION 90.41 77.93 52.78 DiT-3D 95.75 82.01 34.03 The methods of the present specification 69.50 71.88 96.25

[0156] As can be seen, the data quality of the three-dimensional dental models generated using the embodiments provided in this specification is significantly higher than that of existing general data generation technologies. Exemplary generation results can be found in [reference needed]. Figure 6 As shown.

[0157] like Figure 1 As shown, since the three-dimensional dental model generation system provided in this manual can generate paired orthodontic three-dimensional dental models (i.e., paired pre-orthodontic three-dimensional dental models and post-orthodontic three-dimensional dental models) in two stages, it is impossible to compare the pre-orthodontic three-dimensional dental model with general data generation technology alone, and therefore it is impossible to evaluate the generation effect of the pre-orthodontic three-dimensional dental model.

[0158] To further evaluate the quality of the generated paired orthodontic 3D dental models, the existing point cloud-based automatic tooth alignment tools TANet and PSTN were trained on real and generated datasets respectively (both training sets were the same size, each containing 720 data pairs). The models were also tested on real datasets, and the experimental results are shown in the table below:

[0159]

[0160] As can be seen from the table, after training on the dataset generated by the three-dimensional dental model generation method provided in this specification, the automatic tooth alignment tools TANet and PSTN showed performance that was only slightly inferior to the results trained on the real dataset. This demonstrates that the dataset generated by the embodiments in this specification does indeed have high quality.

[0161] Corresponding to the aforementioned embodiments of the method for generating three-dimensional dental models, this specification also provides embodiments of a device for generating three-dimensional dental models. These devices can be implemented through software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of its host device reading the corresponding computer program from non-volatile memory into memory and running it. From a hardware perspective, such as... Figure 7 The diagram shown is a hardware structure diagram of the device used to generate the three-dimensional dental model in this specification. (Except for...) Figure 7 In addition to the processor, network interface, memory, and non-volatile memory shown, the device in the embodiment may also include other hardware depending on the actual communication function, which will not be described in detail here.

[0162] Please see Figure 8 This is a block diagram of a three-dimensional dental model generation device provided in one embodiment of this specification, the device corresponding to... Figure 1 The illustrated embodiment includes:

[0163] The feature extraction unit 810 preprocesses the point cloud data of the orthodontic three-dimensional dental model in the existing dataset and extracts the geometric features of the orthodontic three-dimensional dental model from the preprocessed point cloud data.

[0164] The model training unit 820 acquires the vector quantization variational autoencoder model to be trained and trains the vector quantization variational autoencoder model in the following way: the encoder based on the vector quantization variational autoencoder model encodes the extracted geometric features of the post-orthodontic three-dimensional dental model into the latent space, and further reconstructs the point cloud data of the post-orthodontic three-dimensional dental model from the encoded information in the latent space through the decoder of the vector quantization variational autoencoder model.

[0165] The data generation unit 830 also trains a diffusion model in the latent space of the vector quantization variational autoencoder model. The diffusion model is used to diffuse the encoded information input into the latent space to generate new encoded information, so that the trained decoder can generate point cloud data of the new orthodontic three-dimensional dental model based on the new encoded information.

[0166] Optionally, the feature extraction unit 810 includes:

[0167] The preprocessing subunit centers and normalizes the point cloud data of the orthodontic post-three-dimensional dental model in the existing dataset to obtain a tooth sequence of 32 teeth corresponding to the orthodontic post-three-dimensional dental model; wherein the tooth sequence has a 2×2×8 spatial structure.

[0168] The feature extraction subunit divides each tooth into an r×r×r voxel structure, uses the PointNet network to extract features from each tooth voxel, and aggregates the extracted features from all teeth to obtain the geometric features of the orthodontic 3D dental model with shape [2r, 2r, 8r]; where r is the resolution of the tooth.

[0169] Optionally, the vector quantization variational autoencoder model includes a VQ-VAE model employing a 3D U-Net network architecture;

[0170] The model training unit 820 includes:

[0171] The encoding subunit uses the encoder of the VQ-VAE model to perform multi-layer convolution operations on the extracted geometric features of the three-dimensional dental model after orthodontics, so as to encode the geometric features into the latent space and obtain the latent vector corresponding to the three-dimensional dental model after orthodontics.

[0172] The decoding subunit uses the decoder of the VQ-VAE to perform multi-layer deconvolution operations on the latent vectors in the latent space to obtain point cloud data of the orthodontic three-dimensional dental model reconstructed based on the latent vectors.

[0173] The computational subunit calculates the reconstruction loss function of the decoder using the point cloud data of the reconstructed three-dimensional dental model after orthodontic treatment and the point cloud data of the three-dimensional dental model in the existing dataset.

[0174] Optionally, the computational subunit is further configured to calculate the chamfer distance between each tooth in the point cloud data of the reconstructed post-orthodontic 3D dental model and the point cloud data of the post-orthodontic 3D dental model in the existing dataset; calculate the average chamfer distance of each pair of teeth to obtain the reconstruction loss function.

[0175] Optionally, the data generation unit 830 is further configured to use a diffusion model to add noise to the latent vectors of the encoder that encode the features of the real post-orthodontic 3D dental model into the latent space, and to use the denoiser of the diffusion model to denoise them to obtain new latent vectors; and to use the trained decoder to generate new point cloud data of the post-orthodontic 3D dental model based on the new latent vectors.

[0176] Optionally, the device further includes:

[0177] The acquisition unit retrieves a pre-orthodontic 3D dental model with a specified tooth alignment style from an existing dataset.

[0178] The style and shape extraction unit extracts the style features of the tooth arrangement in the pre-orthodontic 3D dental model with the specified tooth arrangement style, and extracts the shape features of the point cloud data of the new post-orthodontic 3D dental model generated by the trained decoder.

[0179] The parameter generation unit uses the style features and shape features as dual conditions for the diffusion model, guiding the diffusion model to generate rigid transformation parameters for each tooth in the orthodontic post-three-dimensional dental model that meet the requirements.

[0180] The second data generation unit applies the generated rigid transformation parameters to each tooth in the post-orthodontic three-dimensional dental model to obtain point cloud data of the pre-orthodontic three-dimensional dental model corresponding to the post-orthodontic three-dimensional dental model, and the point cloud data of the pre-orthodontic three-dimensional dental model has the specified tooth arrangement style.

[0181] Optionally, when extracting the style features of the tooth arrangement in the pre-orthodontic 3D dental model of the specified tooth arrangement style, the style and shape extraction unit is further used for:

[0182] The point cloud data of the pre-orthodontic three-dimensional dental model with the specified tooth arrangement style is centered and normalized to obtain a tooth sequence consisting of 32 teeth; wherein the tooth sequence is a 2×2×8 spatial structure.

[0183] Each tooth is divided into an r×r×r voxel structure. The PointNet network is used to extract features from each tooth voxel. The extracted features from all teeth are aggregated to obtain geometric features of shape [2r, 2r, 8r], where r is the resolution of the tooth.

[0184] Within the voxel, mean pooling and variance pooling are used, and after linear transformation, style features with shape [2r, 2r, 8r] are obtained.

[0185] Optionally, when extracting the shape features of the point cloud data of the new orthodontic post-3D dental model generated by the trained decoder, the style and shape extraction unit is further used for:

[0186] The point cloud data of the new orthodontic three-dimensional dental model generated by the trained decoder is centered and normalized to obtain a tooth sequence consisting of 32 teeth; wherein the tooth sequence is a 2×2×8 spatial structure.

[0187] Each tooth is divided into an r×r×r voxel structure. The PointNet network is used to extract features from each tooth voxel. The extracted features from all teeth are aggregated to obtain geometric features of shape [2r, 2r, 8r], where r is the resolution of the tooth.

[0188] Inside each tooth, maximum pooling is used to aggregate the features of all voxels to obtain shape features.

[0189] Optionally, the diffusion model includes a diffusion model based on a 3D U-Net network architecture;

[0190] The parameter generation unit is further used to input the style features and shape features into each layer of the 3D U-Net network through a multilayer perceptron, guide the discrete model to generate rigid transformation parameters of the teeth, and output rigid transformation parameters for each tooth in the three-dimensional dental model after orthodontics.

[0191] Optionally, the second data generation unit is further configured to translate and rotate each tooth in the post-orthodontic three-dimensional dental model according to the translation and rotation parameters in the generated rigid transformation parameters, so as to obtain point cloud data of the pre-orthodontic three-dimensional dental model corresponding to the post-orthodontic three-dimensional dental model, and the point cloud data of the pre-orthodontic three-dimensional dental model has the specified tooth arrangement style.

[0192] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which can take the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email sending and receiving device, game console, tablet computer, wearable device, or any combination of these devices.

[0193] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0194] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the solution in this specification according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0195] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0196] Other embodiments of this specification will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This specification is intended to cover any variations, uses, or adaptations that follow the general principles of this specification and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this specification are indicated by the following claims.

[0197] It should be understood that this specification is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this specification is limited only by the appended claims.

Claims

1. A method for generating a three-dimensional dental model, characterized in that, The method includes: The point cloud data of the orthodontic post-3D dental model in the existing dataset is preprocessed, and the geometric features of the orthodontic post-3D dental model are extracted from the preprocessed point cloud data. Obtain the vector quantization variational autoencoder model to be trained, and train the vector quantization variational autoencoder model in the following manner: The encoder based on the vector quantization variational autoencoder model encodes the extracted geometric features of the post-orthodontic three-dimensional dental model into the latent space, and further reconstructs the point cloud data of the post-orthodontic three-dimensional dental model from the encoded information in the latent space through the decoder of the vector quantization variational autoencoder model. In the latent space of the vector quantization variational autoencoder model, a diffusion model is also trained. The diffusion model is used to diffuse the encoded information input into the latent space to generate new encoded information, so that the trained decoder can generate point cloud data of the new orthodontic three-dimensional dental model based on the new encoded information. Obtain a pre-orthodontic 3D dental model of a specified tooth alignment style from an existing dataset; Extract the style features of tooth arrangement in the pre-orthodontic 3D dental model with the specified tooth arrangement style, and extract the shape features of the point cloud data of the new post-orthodontic 3D dental model generated by the trained decoder. The style features and shape features are used as dual conditions for the diffusion model to guide the diffusion model to generate rigid transformation parameters for each tooth in the orthodontic post-three-dimensional dental model that meet the requirements. The generated rigid transformation parameters are applied to each tooth in the post-orthodontic 3D dental model to obtain point cloud data of the pre-orthodontic 3D dental model corresponding to the post-orthodontic 3D dental model, and the point cloud data of the pre-orthodontic 3D dental model has the specified tooth arrangement style.

2. The method according to claim 1, characterized in that, The process of preprocessing the point cloud data of the post-orthodontic 3D dental model in the existing dataset and extracting the geometric features of the post-orthodontic 3D dental model from the preprocessed point cloud data includes: The point cloud data of the post-orthodontic 3D dental model in the existing dataset is centered and normalized to obtain a tooth sequence of 32 teeth corresponding to the post-orthodontic 3D dental model; wherein the tooth sequence is a 2×2×8 spatial structure. Each tooth is divided into an r×r×r voxel structure. The PointNet network is used to extract features from the voxels of each tooth. The extracted features from all teeth are aggregated to obtain the geometric features of the orthodontic 3D dental model with shape [2r, 2r, 8r], where r is the resolution of the tooth.

3. The method according to claim 2, characterized in that, The vector quantization variational autoencoder model includes a VQ-VAE model using a 3D U-Net network architecture; The encoder based on the vector quantization variational autoencoder model encodes the extracted geometric features of the post-orthodontic 3D dental model into a latent space, and further reconstructs the point cloud data of the post-orthodontic 3D dental model from the encoded information in the latent space using the decoder of the vector quantization variational autoencoder model, including: The encoder of the VQ-VAE model is used to perform multi-layer convolution operation on the extracted geometric features of the three-dimensional dental model after orthodontics, so as to encode the geometric features into the latent space and obtain the latent vector corresponding to the three-dimensional dental model after orthodontics. The VQ-VAE decoder is then used to perform multi-layer deconvolution operations on the latent vectors in the latent space to obtain point cloud data of the orthodontic three-dimensional dental model reconstructed based on the latent vectors. The reconstruction loss function of the decoder is calculated using the point cloud data of the reconstructed three-dimensional dental model after orthodontic treatment and the point cloud data of the three-dimensional dental model in the existing dataset.

4. The method according to claim 3, characterized in that, The step of calculating the reconstruction loss function of the decoder using the point cloud data of the reconstructed post-orthodontic 3D dental model and the point cloud data of the 3D dental model in the existing dataset includes: Calculate the chamfer distance between the point cloud data of the reconstructed orthodontic 3D dental model and the point cloud data of the orthodontic 3D dental model in the existing dataset for each tooth. The average chamfer distance for each pair of teeth is calculated to obtain the reconstruction loss function.

5. The method according to claim 1, characterized in that, The diffusion model is used to diffuse the encoded information input into the latent space to generate new encoded information. The trained decoder then uses this new encoded information to generate point cloud data for a new post-orthodontic 3D dental model, including: The diffusion model adds noise to the latent vectors of the real post-orthodontic 3D dental model encoded into the latent space by the encoder, and then uses the denoiser of the diffusion model to denoise them to obtain new latent vectors; further, the trained decoder is used to generate point cloud data of the new post-orthodontic 3D dental model based on the new latent vectors.

6. The method according to claim 1, characterized in that, The extraction of style features of tooth arrangement in the pre-orthodontic 3D dental model of the specified tooth arrangement style includes: The point cloud data of the pre-orthodontic three-dimensional dental model with the specified tooth arrangement style is centered and normalized to obtain a tooth sequence consisting of 32 teeth; wherein the tooth sequence is a 2×2×8 spatial structure. Each tooth is divided into an r×r×r voxel structure. The PointNet network is used to extract features from each tooth voxel. The extracted features from all teeth are aggregated to obtain geometric features of shape [2r, 2r, 8r], where r is the resolution of the tooth. Within the voxel, mean pooling and variance pooling are used, and after linear transformation, style features with shape [2r, 2r, 8r] are obtained.

7. The method according to claim 1, characterized in that, Extract the shape features of the point cloud data of the new orthodontic post-orthodontic 3D dental model generated by the trained decoder, including: The point cloud data of the new orthodontic three-dimensional dental model generated by the trained decoder is centered and normalized to obtain a tooth sequence consisting of 32 teeth; wherein the tooth sequence is a 2×2×8 spatial structure. Each tooth is divided into an r×r×r voxel structure. The PointNet network is used to extract features from each tooth voxel. The extracted features from all teeth are aggregated to obtain geometric features of shape [2r, 2r, 8r], where r is the resolution of the tooth. Inside each tooth, maximum pooling is used to aggregate the features of all voxels to obtain shape features.

8. The method according to claim 1, characterized in that, The diffusion model includes a diffusion model based on a 3D U-Net network architecture; The step of using the style features and shape features as dual conditions for the diffusion model to guide the generation of rigid transformation parameters for each tooth in the post-orthodontic 3D dental model that meet the requirements includes: The style and shape features are input into each layer of the 3D U-Net network via a multilayer perceptron, guiding the discrete model to generate rigid transformation parameters for the teeth, and outputting rigid transformation parameters for each tooth in the post-orthodontic 3D dental model.

9. The method according to claim 8, characterized in that, The generated rigid transformation parameters are applied to each tooth in the post-orthodontic 3D dental model to obtain a pre-orthodontic 3D dental model corresponding to the post-orthodontic 3D dental model, and the pre-orthodontic 3D dental model has the specified tooth arrangement style, including: Each tooth in the post-orthodontic 3D dental model is translated and rotated according to the translation and rotation parameters in the generated rigid transformation parameters to obtain point cloud data of the pre-orthodontic 3D dental model corresponding to the post-orthodontic 3D dental model, and the point cloud data of the pre-orthodontic 3D dental model has the specified tooth arrangement style.

10. A device for generating a three-dimensional dental model, characterized in that, The device includes: The feature extraction unit preprocesses the point cloud data of the orthodontic post-3D dental model in the existing dataset and extracts the geometric features of the orthodontic post-3D dental model from the preprocessed point cloud data. The model training unit acquires the vector quantization variational autoencoder model to be trained and trains the vector quantization variational autoencoder model in the following way: the encoder based on the vector quantization variational autoencoder model encodes the geometric features of the extracted orthodontic three-dimensional dental model into the latent space, and further reconstructs the point cloud data of the orthodontic three-dimensional dental model from the encoded information in the latent space through the decoder of the vector quantization variational autoencoder model. The data generation unit also trains a diffusion model in the latent space of the vector quantization variational autoencoder model. The diffusion model is used to diffuse the encoded information input into the latent space to generate new encoded information, so that the trained decoder can generate point cloud data of the new orthodontic three-dimensional dental model based on the new encoded information. The acquisition unit retrieves a pre-orthodontic 3D dental model of a specified tooth alignment style from an existing dataset. The style and shape extraction unit extracts the style features of the tooth arrangement in the pre-orthodontic 3D dental model with the specified tooth arrangement style, and extracts the shape features of the point cloud data of the new post-orthodontic 3D dental model generated by the trained decoder. The parameter generation unit uses the style features and shape features as dual conditions for the diffusion model, guiding the diffusion model to generate rigid transformation parameters for each tooth in the orthodontic post-three-dimensional dental model that meet the requirements. The second data generation unit applies the generated rigid transformation parameters to each tooth in the post-orthodontic three-dimensional dental model to obtain point cloud data of the pre-orthodontic three-dimensional dental model corresponding to the post-orthodontic three-dimensional dental model, and the point cloud data of the pre-orthodontic three-dimensional dental model has the specified tooth arrangement style.

11. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to perform the method according to any one of claims 1-9.

Images