Full-process intelligent orthodontic tooth movement plan generation method and system

Abstract

The invention discloses a full-process intelligent orthodontic tooth movement plan generation method and system, and relates to the technical field of orthodontics, and the method comprises the steps: collecting a three-dimensional medical image of the oral cavity of a user, and generating a corresponding two-dimensional medical image according to the three-dimensional medical image; performing three-dimensional reconstruction based on the acquired image data to obtain an oral cavity three-dimensional model; obtaining target prediction data according to the oral cavity three-dimensional model of the user; and inputting the target prediction data into a trained orthodontic tooth movement planning generation model for analysis to obtain a tooth movement planning result. According to the method, a manual planning mode of an orthodontic doctor can be learned from a large-scale data set containing various clinical orthodontic cases, and an orthodontic path planning result more conforming to clinical practice can be quickly generated, so that the planning efficiency and accuracy in traditional orthodontic tooth movement planning are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of oral orthodontics, and more specifically to a full-process intelligent orthodontic tooth movement planning generation method and system. Background Art

[0002] At present, in the field of orthodontics, tooth movement planning is crucial to the success of orthodontic treatment. For a long time, traditional orthodontic tooth movement planning has mainly relied on the professional knowledge and experience of orthodontists, and there are many limitations. In the traditional way, doctors need to manually measure and analyze the patient's plaster model and X-rays (such as panoramic films, cephalographic films, etc.). When processing plaster models, a variety of measuring tools are used to determine parameters such as tooth size, shape, arrangement, and length and width of the dental arch. When interpreting X-rays, it is necessary to observe the length, angle, height and density of the tooth root with the naked eye. This manual operation is extremely time-consuming, and the measurement results are easily affected by the doctor's subjective factors and operational proficiency. Different doctors may have very different analysis results for the same patient, resulting in a lack of consistency in treatment plans. At the same time, as separate data sources, plaster models and X-rays are difficult to effectively integrate the two by traditional methods. When formulating tooth movement planning, only observing tooth arrangement based on plaster models may ignore the actual situation of the tooth root in the jawbone, and fail to fully consider various factors, affecting the scientificity and accuracy of tooth movement planning, and thus having an adverse effect on the effect of orthodontic treatment. Moreover, with the increasing demand for orthodontic treatment and the higher expectations for treatment effects, the low efficiency and high error rate of traditional methods can no longer meet the requirements of modern orthodontic treatment. In addition, in the traditional planning process, the individual differences of different patients are not fully considered, and it is difficult to formulate the best personalized plan for each patient's age, gender, medical history and other factors. There is also a lack of effective mechanisms for dynamic adjustment of changes that may occur during the treatment process.

[0003] Therefore, how to improve the planning efficiency and accuracy in traditional orthodontic tooth movement planning is an urgent problem that technical personnel in this field need to solve. Summary of the invention

[0004] In view of this, the present invention provides a full-process intelligent orthodontic tooth movement planning generation method and system to solve the problems existing in the background technology.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A full-process intelligent orthodontic tooth movement planning generation method, comprising:

[0007] Collecting a three-dimensional medical image of the user's oral cavity, and generating a corresponding two-dimensional medical image based on the three-dimensional medical image;

[0008] Perform three-dimensional reconstruction based on the acquired image data to obtain a three-dimensional model of the oral cavity;

[0009] Obtain target prediction data based on the user's oral 3D model;

[0010] The target prediction data is input into a trained orthodontic tooth movement planning generation model for analysis to obtain a tooth movement planning result.

[0011] Optionally, the oral cavity three-dimensional model acquisition process specifically includes:

[0012] Inputting the two-dimensional medical image into a three-dimensional reconstruction network to obtain a three-dimensional reconstructed segmented image corresponding to the two-dimensional medical image, wherein the three-dimensional reconstructed segmented image is a reconstructed and segmented three-dimensional medical image;

[0013] Based on the three-dimensional reconstructed segmented image, a three-dimensional model of the oral cavity is obtained.

[0014] Optionally, the three-dimensional reconstruction network includes an encoding network, a conversion network and a decoding network;

[0015] extracting two-dimensional features of the two-dimensional medical image using the encoding network;

[0016] Using the conversion network to convert the two-dimensional features of the two-dimensional medical image into three-dimensional features;

[0017] Using the decoding network to convert the three-dimensional features into a three-dimensional reconstructed segmentation image corresponding to the two-dimensional medical image;

[0018] The 3D reconstruction network is trained based on the 3D reconstructed segmentation image and the 3D segmentation data marked on the 3D medical image. Specifically, the 3D segmentation data marked on the 3D medical image is used as a 3D segmentation label, and the 3D reconstruction network is trained according to the difference between the 3D reconstructed segmentation image and the 3D segmentation label.

[0019] Optionally, obtaining target prediction data according to the user's oral three-dimensional model includes:

[0020] Extracting crown position information and dentition shape information from the user's oral three-dimensional model;

[0021] Digitally marking the dentition in the three-dimensional oral model according to the crown position information to obtain dentition marking information;

[0022] Performing standardization on the oral three-dimensional model;

[0023] Setting anatomical feature points of the user, and obtaining position information of the anatomical feature points according to the normalization processing result;

[0024] The target prediction data is obtained according to the user's dentition shape information, dentition marking information and position information of anatomical feature points.

[0025] Optionally, the orthodontic tooth movement planning generation model includes an embedded coding layer for feature encoding the target prediction data, and a convolutional neural network for feature extraction of the feature-encoded prediction data.

[0026] Optionally, the embedded coding layer includes:

[0027] A bidirectional long short-term memory network, wherein the bidirectional long short-term memory network is used to encode input target prediction data;

[0028] A residual network, wherein the residual network is used to extract features of the encoded prediction data by stacking residual blocks layer by layer;

[0029] A multi-layer self-attention network is provided, wherein the multi-layer self-attention network is used to perform representation learning and layer-by-layer feature extraction on the correlation between different positions in the prediction data.

[0030] Optionally, the multi-layer self-attention network is used to perform representation learning and layer-by-layer feature extraction on the correlation between different positions in the prediction data, specifically including:

[0031] The feature vector output by the residual network is used as the input sequence to pass to the self-attention layer; by calculating the attention weight, the correlation between each position and other positions in the input sequence is established to obtain the attention weight;

[0032] The obtained attention weights are weighted summed with the representation of the input sequence to obtain the output feature vector.

[0033] Optionally, the network structure of the convolutional neural network includes:

[0034] An input layer, the input layer being used to receive the encoded prediction data;

[0035] Convolution layer, which slides the convolution kernel on the predicted data to extract local relevant features. The sliding process of each convolution kernel is realized by performing a series of product and sum operations on the predicted data. For the output of each convolution kernel, nonlinearity is introduced through the activation function.

[0036] A pooling layer, wherein the pooling layer is used to reduce the dimension of the feature map through a pooling operation and retain the target features;

[0037] Fully connected layer, the output of the pooling layer is connected to one or more fully connected layers, which map the output of the convolutional layer to the output layer to generate the final tooth movement planning result.

[0038] Optionally, the residual network specifically includes at least one residual block, and each residual block includes a main path and a skip path;

[0039] The main path is used to extract feature information of input prediction data, and the jump path is used to provide a bypass path that bypasses the main path;

[0040] The residual network extracts features from the encoded prediction data by stacking residual blocks layer by layer, which specifically includes:

[0041] Input prediction data, perform convolution and pooling operations on the input prediction data, extract features through the main path, and generate intermediate feature maps;

[0042] The skip path is used to add the input prediction data to the intermediate feature map output by the main path to form a residual;

[0043] The residual is transformed nonlinearly through the activation function and enters the next residual block until it loops to the last residual block;

[0044] The data output by the last residual block is globally pooled to obtain the final feature vector.

[0045] A full-process intelligent orthodontic tooth movement planning generation system, including:

[0046] A data acquisition module, which acquires a three-dimensional medical image of the user's oral cavity and generates a corresponding two-dimensional medical image based on the three-dimensional medical image;

[0047] A three-dimensional reconstruction module performs three-dimensional reconstruction based on the acquired image data to obtain a three-dimensional model of the oral cavity;

[0048] A data processing module obtains target prediction data based on the user's oral 3D model;

[0049] The model analysis module inputs the target prediction data into the trained orthodontic tooth movement planning generation model for analysis to obtain the tooth movement planning results.

[0050] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses a full-process intelligent orthodontic tooth movement planning generation method and system, which collects three-dimensional medical images of the user's oral cavity, and generates corresponding two-dimensional medical images based on the three-dimensional medical images; performs three-dimensional reconstruction based on the acquired image data to obtain a three-dimensional model of the oral cavity; obtains target prediction data based on the user's three-dimensional model of the oral cavity; inputs the target prediction data into the trained orthodontic tooth movement planning generation model for analysis to obtain tooth movement planning results. The present invention can learn the manual planning patterns of orthodontists from a large-scale data set containing various clinical orthodontic cases, and can quickly generate orthodontic path planning results that are more in line with clinical practice, thereby improving the planning efficiency and accuracy in traditional orthodontic tooth movement planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0052] Figure 1 A schematic diagram of the method provided by the present invention;

[0053] Figure 2 This is a schematic diagram of the system structure provided by the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] To facilitate understanding, the following is a brief introduction to the relevant terms and concepts that may be involved in the embodiments of the present application.

[0056] Three-dimensional reconstruction refers to obtaining a comprehensive three-dimensional image of human organs through two-dimensional images. It can provide doctors with more intuitive and accurate information on the location, spatial structure, morphology, volume, etc. of lesions, and help in formulating surgical plans, individualized postoperative reconstruction, and postoperative effect evaluation. It also helps to improve the accuracy and safety of surgery.

[0057] The embodiment of the present invention discloses a method for generating a full-process intelligent orthodontic tooth movement plan. Figure 1 As shown, including:

[0058] Collecting a three-dimensional medical image of the user's oral cavity, and generating a corresponding two-dimensional medical image based on the three-dimensional medical image;

[0059] The above-mentioned two-dimensional medical image may be a two-dimensional projection medical image, and the three-dimensional medical image may be an image showing information such as the spatial structure, morphology, and volume of human organs.

[0060] It should be understood that the two-dimensional medical image may be a magnetic resonance image (MRI), a digital radiographic image (DR), a CT image or other medical image.

[0061] Perform three-dimensional reconstruction based on the acquired image data to obtain a three-dimensional model of the oral cavity;

[0062] Obtain target prediction data based on the user's oral 3D model;

[0063] The target prediction data is input into the trained orthodontic tooth movement planning generation model for analysis to obtain the tooth movement planning results.

[0064] In a specific embodiment, the process of acquiring the oral cavity three-dimensional model specifically includes:

[0065] Inputting the two-dimensional medical image into a three-dimensional reconstruction network to obtain a three-dimensional reconstructed segmented image corresponding to the two-dimensional medical image, wherein the three-dimensional reconstructed segmented image is a reconstructed and segmented three-dimensional medical image;

[0066] Based on the 3D reconstruction segmentation image, a 3D model of the oral cavity is obtained.

[0067] In a specific embodiment, the 3D reconstruction network includes an encoding network, a conversion network, and a decoding network;

[0068] Extracting 2D features from 2D medical images using encoding networks;

[0069] Using a conversion network to convert the two-dimensional features of a two-dimensional medical image into three-dimensional features;

[0070] The decoding network is used to convert the three-dimensional features into a three-dimensional reconstructed segmentation image corresponding to the two-dimensional medical image;

[0071] The 3D reconstruction network is trained based on the 3D reconstructed segmentation image and the 3D segmentation data marked on the 3D medical image. Specifically, the 3D segmentation data marked on the 3D medical image is used as the 3D segmentation label, and the 3D reconstruction network is trained according to the difference between the 3D reconstructed segmentation image and the 3D segmentation label.

[0072] The 3D reconstruction network is trained based on the 3D reconstruction segmentation image and the 3D segmentation data labeled on the 3D medical image samples.

[0073] Specifically, the 3D reconstruction network may be trained using a training method such as supervised learning, unsupervised learning or semi-supervised learning. The present invention does not specifically limit the training method of the 3D reconstruction network.

[0074] According to the technical solution provided by the embodiment of the present invention, a two-dimensional medical image corresponding to a three-dimensional medical image is obtained; the two-dimensional medical image is input into a three-dimensional reconstruction network to obtain a three-dimensional reconstructed segmentation image corresponding to the two-dimensional medical image; based on the three-dimensional reconstructed segmentation image and the three-dimensional segmentation data marked on the three-dimensional medical image, the three-dimensional reconstruction network is trained, so that the trained three-dimensional reconstruction network can directly reconstruct the two-dimensional medical image into a segmented three-dimensional medical image, enabling doctors to diagnose from a 3D perspective while providing segmentation results, thereby providing richer and more intuitive three-dimensional information.

[0075] The 3D reconstruction network includes an encoding network, a conversion network and a decoding network. Specifically, the encoding network is used to extract the 2D features of the 2D medical image, such as the position and size of teeth; the conversion network is used to convert the 2D features of the 2D medical image into 3D features; and the decoding network is used to convert the 3D features into a 3D reconstructed segmented image corresponding to the 2D medical image.

[0076] Specifically, the encoding network can be composed of 5 2D convolution residual blocks. Each 2D convolution residual block can be composed of: 2D convolution layer (4×4 convolution kernel, step size is 2) → 2D batch normalization layer → ReLU (RectifiedLinearUnit, linear rectifier function) layer → 2D convolution layer (3×3 convolution kernel, step size is 1) → 2D batch normalization layer → ReLU layer, where the output of the first ReLU layer is cross-layer connected with the output of the second ReLU layer to obtain the final output. For example, taking the reconstruction of a 3D oral segmentation image as an example, a DR image of size 1×512×512 (where 1 represents the number of channel layers) is input into the first 2D convolution residual block, and an output of size 512×256×256 can be obtained. After that, after four 2D convolution residual blocks, the output sizes can be 1024×128×128, 2048×64×64, 4096×32×32, and 8192×16×16 respectively.

[0077] The conversion network is used to expand the data from 2D to 3D. The convolution operation with a 3×3 convolution kernel and ReLU activation function can be used to learn the transformation of 2D features. Then, the 8192×16×16 is reshaped to 4096×2×16×16. Then, a 1×1×1 convolution kernel and a 3D deconvolution layer with a stride of 1 are used to learn the transformation of 3D features while keeping the feature size unchanged.

[0078] The decoding network is used to convert the 3D features into a 3D segmentation image. The 3D deconvolution layer can be used to continuously reduce the number of channels and increase the size, and then pass through a 3D convolution layer with a 1×1×1 convolution kernel and a stride of 1, and finally obtain a 25×256×256×256 3D segmentation result.

[0079] In a specific embodiment, obtaining target prediction data according to the three-dimensional model of the user's oral cavity includes:

[0080] Extract crown position information and dentition shape information from the user's oral 3D model;

[0081] According to the crown position information, the dentition in the three-dimensional oral model is digitally marked to obtain the dentition marking information;

[0082] Standardize the oral 3D model;

[0083] Setting the user's anatomical feature points, and obtaining the position information of the anatomical feature points according to the standardized processing results;

[0084] Target prediction data is obtained based on the user's dentition shape information, dentition marking information, and position information of anatomical feature points.

[0085] In a specific embodiment, the orthodontic tooth movement planning generation model includes an embedded coding layer for feature encoding of target prediction data, and a convolutional neural network for feature extraction of feature encoded prediction data.

[0086] In a specific embodiment, the embedded coding layer includes:

[0087] Bidirectional long short-term memory network, the bidirectional long short-term memory network is used to encode the input target prediction data;

[0088] Residual network, which is used to extract features from the encoded prediction data by stacking residual blocks layer by layer;

[0089] Multi-layer self-attention network,The multi-layer self-attention network is used to perform representation learning and layer-by-layer feature extraction on the correlation between different positions in the prediction data.

[0090] The process of using a bidirectional long short-term memory network to encode prediction data specifically includes:

[0091] The predicted data is input into the embedding layer of the bidirectional long short-term memory network, through which the vector representation of each part is mapped to a vector of fixed length;

[0092] The two sub-networks LSTM of the bidirectional long short-term memory network read the input prediction data from left to right and from right to left respectively, and obtain the previous and next context information in the encoding process to obtain the prediction data features;

[0093] Among them, in each time step, the sub-network LSTM takes the input vector of the current time step and the hidden state of the previous time step as input, and calculates the hidden state of the current time step. The hidden state includes the cell state and the hidden state. The cell state is used to retain long-term memory, and the hidden state is used to transmit information.

[0094] In a specific usage scenario, the Bidirectional Long Short-Term Memory Network (BiLSTM) is a deep learning model for processing and encoding sequence data. The BiLSTM network consists of two LSTM networks, one processing the input sequence from left to right and the other processing it from right to left. This structure enables the network to capture contextual information in the sequence and use bidirectional information to encode the sequence. When encoding the prediction data, each tooth can be represented as a vector. The entire prediction data sequence can be viewed as a sequence of vectors. The BiLSTM network receives the prediction data sequence as input and gradually processes the input through multiple time steps. First, the input prediction data sequence passes through an embedding layer to map the vector representation of each tooth to a fixed-length vector. Then, the two sub-networks of the BiLSTM network read the input sequence from left to right and from right to left respectively. At each time step, the LSTM unit takes the input vector of the current time step and the hidden state of the previous time step as input to calculate the hidden state of the current time step. The hidden state consists of a cell state and a hidden state. The cell state is used to retain long-term memory and the hidden state is used to pass information. The key to the bidirectional LSTM network is its ability to process input sequences in reverse order. By using both left-to-right and right-to-left LSTM units, the network is able to capture contextual information before and after the encoding process. For example, the encoding of the left input at the current time step will take into account the information from the beginning of the sequence to the current position, while the encoding of the right input will take into account the information from the end of the sequence to the current position. Therefore, the BiLSTM network can more comprehensively capture the characteristics of the predicted data sequence. Finally, the network merges the outputs of the left and right LSTM networks to form a comprehensive encoding representation. This encoding representation can be used as input for subsequent tasks.

[0095] The residual network specifically includes at least one residual block, each residual block includes a main path and a skip path;

[0096] Among them, the main path is used to extract the feature information of the input prediction data, and the jump path is used to provide a bypass path that bypasses the main path;

[0097] The residual network extracts features from the encoded prediction data by stacking residual blocks layer by layer, which specifically includes:

[0098] Input prediction data, perform convolution and pooling operations on the input prediction data, extract features through the main path, and generate intermediate feature maps;

[0099] The skip path is used to add the input prediction data to the intermediate feature map output by the main path to form a residual;

[0100] The residual is transformed nonlinearly through the activation function and enters the next residual block until it loops to the last residual block;

[0101] The data output by the last residual block is globally pooled to obtain the final feature vector.

[0102] In a specific usage scenario, the residual network is a deep learning network structure. Its main idea is to solve the gradient vanishing and network degradation problems in the deep neural network training process by introducing the "residual block". In the residual network, each residual block consists of two main parts: the main path and the shortcut (also called jump connection, bypass connection). The main path extracts the feature information of the input data by stacking multiple layers of convolution, pooling and other operations. The shortcut connection provides a "jump" path that bypasses the main path.

[0103] In a specific embodiment, a multi-layer self-attention network is used to perform representation learning and layer-by-layer feature extraction on the correlation between different positions in the prediction data, specifically including:

[0104] The feature vector output by the residual network is used as the input sequence to pass to the self-attention layer; by calculating the attention weight, the correlation between each position and other positions in the input sequence is established to obtain the attention weight;

[0105] The obtained attention weights are weighted summed with the representation of the input sequence to obtain the output feature vector.

[0106] Specifically, the multi-layer self-attention network is a deep learning model that can be used to embed information in the prediction data sequence. It mainly uses the self-attention mechanism to learn the correlation between different positions in the sequence and use these correlations to generate more effective representations. In the multi-layer self-attention network, the prediction data sequence is first encoded. A common method is to use an embedding layer to convert each amino acid in the sequence into a vector representation. These vectors will be passed to the self-attention layer as the input sequence.

[0107] The self-attention layer consists of multiple attention heads, and self-attention calculations are performed inside each attention head. In the self-attention calculation, by calculating the attention weights, the model can establish the correlation between each position and other positions in the sequence. The calculation of attention weights is usually completed by calculating the similarity between two positions in the input sequence. The similarity here can be calculated by tensor product, dot product, etc. After obtaining the attention weights, the output feature vector can be calculated by weighted summing these weights with the representation of the input sequence, that is, weighted pooling of the sequence. The advantage of this is that the learned attention weights can be used to focus on important features at different positions and effectively integrate these features. In a multi-layer self-attention network, multiple self-attention layers are usually used for layer-by-layer feature extraction and representation learning. The output of each self-attention layer will be used as the input of the next layer, and self-attention calculations will be performed again on this input. In this way, the model can gradually learn higher-level abstract representations to better capture the characteristics of the predicted data sequence.

[0108] In a specific embodiment, the network structure of the convolutional neural network includes:

[0109] Input layer, the input layer is used to receive the encoded prediction data;

[0110] Convolution layer, which slides the convolution kernel on the predicted data to extract local relevant features. The sliding process of each convolution kernel is realized by performing a series of product and sum operations on the predicted data. For the output of each convolution kernel, nonlinearity is introduced through the activation function.

[0111] Pooling layer: The pooling layer is used to reduce the dimension of the feature map through pooling operation and retain the target features;

[0112] Fully connected layer, the output of the pooling layer is connected to one or more fully connected layers, which map the output of the convolutional layer to the output layer to generate the final tooth movement planning result.

[0113] Convolutional Neural Network (CNN) is a deep learning model mainly used for image processing and recognition. However, CNN can also be applied to other types of data, including user's dentition shape information, dentition marking information, and location information of anatomical feature points, collectively referred to as target prediction data. Convolutional neural networks extract encoded information by performing convolution operations on the prediction data. The following is a detailed description of how convolutional neural networks extract the encoded information of prediction data: Input layer: The encoded prediction data is used as the input data of the model. The prediction data can be represented as a one-hot encoded matrix or other encoded feature information. Convolution layer: The convolution operation is the core part of CNN. The model slides on the prediction data sequence through convolution kernels of different sizes to extract local relevant features. The sliding process of each convolution kernel is achieved by performing a series of product and sum operations on the sequence. Activation function: For the output of each convolution kernel, an activation function (such as ReLU function) is usually applied to introduce nonlinearity. This helps the network capture nonlinear features and makes the network more expressive. Pooling layer: The pooling operation can reduce the dimension of the feature map and retain the most important features. By limiting the size of convolutional features, the pooling layer can reduce the amount of computation and improve the robustness of the model. Fully connected layer: The output of the pooling layer is connected to one or more fully connected layers. The fully connected layer maps the output of the convolutional layer to the final output layer to generate the final tooth movement planning result. When training CNN, the network is optimized through the back-propagation algorithm to gradually improve its encoding ability for the predicted data sequence. Specifically, on a dataset of labeled tooth movement planning, the error between the model prediction result and the true label is calculated, and the model parameters are updated according to the error, so that the model gradually converges to a more accurate prediction ability. Therefore, the convolutional neural network uses convolution, activation function and pooling operations, as well as fully connected layers, to extract encoded information from the predicted data, which can be used to predict the tooth movement planning results.

[0114] In a specific embodiment, after obtaining the tooth movement planning result, further performing correction treatment on the user's teeth according to the tooth movement planning result; specifically, it may include:

[0115] Use 3D printing technology to make the corresponding appliance mold;

[0116] Based on the mold, the mold is pressed to produce an appliance that can be worn in the patient's mouth;

[0117] Orthodontic treatment is performed on target patients using braces.

[0118] The processes of making the mold and making the appliance can be combined into one step, that is, directly printing into the shape of the appliance, reducing the steps of lamination production; and further improving the production efficiency of the appliance.

[0119] Specifically, a dentition model of each dentition morphology is 3D printed, and the solid model obtained by 3D printing is used as a mold. Through heat pressing or other technical means, an appliance (such as a bracketless invisible orthodontic appliance) that is consistent with the dentition morphology and can be tightly covered on the dentition model is obtained. After a certain period of time, the force can move the teeth within the space limited by the appliance, thereby achieving an orthodontic or correction effect. By continuously wearing the appliance of the corresponding stage, the patient's tooth morphology can be moved from one correction stage to another as expected, completing orthodontic treatment.

[0120] A full-process intelligent orthodontic tooth movement planning generation system, such as Figure 2 As shown, including:

[0121] A data acquisition module, which acquires a three-dimensional medical image of the user's oral cavity and generates a corresponding two-dimensional medical image based on the three-dimensional medical image;

[0122] A three-dimensional reconstruction module performs three-dimensional reconstruction based on the acquired image data to obtain a three-dimensional model of the oral cavity;

[0123] A data processing module obtains target prediction data based on the user's oral 3D model;

[0124] The model analysis module inputs the target prediction data into the trained orthodontic tooth movement planning generation model for analysis to obtain the tooth movement planning results.

[0125] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0126] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A full-process intelligent orthodontic tooth movement planning generation method, characterized in that: include: Collecting a three-dimensional medical image of the user's oral cavity, and generating a corresponding two-dimensional medical image based on the three-dimensional medical image; Perform three-dimensional reconstruction based on the acquired image data to obtain a three-dimensional model of the oral cavity; Obtain target prediction data based on the user's oral 3D model; The target prediction data is input into a trained orthodontic tooth movement planning generation model for analysis to obtain a tooth movement planning result.

2. A full-process intelligent orthodontic tooth movement planning generation method according to claim 1, characterized in that: The oral three-dimensional model acquisition process specifically includes: Inputting the two-dimensional medical image into a three-dimensional reconstruction network to obtain a three-dimensional reconstructed segmented image corresponding to the two-dimensional medical image, wherein the three-dimensional reconstructed segmented image is a reconstructed and segmented three-dimensional medical image; Based on the three-dimensional reconstructed segmented image, a three-dimensional model of the oral cavity is obtained.

3. A full-process intelligent orthodontic tooth movement planning generation method according to claim 2, characterized in that: The three-dimensional reconstruction network includes an encoding network, a conversion network and a decoding network; extracting two-dimensional features of the two-dimensional medical image using the encoding network; Using the conversion network to convert the two-dimensional features of the two-dimensional medical image into three-dimensional features; Using the decoding network to convert the three-dimensional features into a three-dimensional reconstructed segmentation image corresponding to the two-dimensional medical image; The 3D reconstruction network is trained based on the 3D reconstructed segmentation image and the 3D segmentation data marked on the 3D medical image. Specifically, the 3D segmentation data marked on the 3D medical image is used as a 3D segmentation label, and the 3D reconstruction network is trained according to the difference between the 3D reconstructed segmentation image and the 3D segmentation label.

4. A full-process intelligent orthodontic tooth movement planning generation method according to claim 1, characterized in that: The obtaining of target prediction data according to the user's oral three-dimensional model includes: Extracting crown position information and dentition shape information from the user's oral three-dimensional model; Digitally marking the dentition in the three-dimensional oral model according to the crown position information to obtain dentition marking information; Performing standardization on the oral three-dimensional model; Setting anatomical feature points of the user, and obtaining position information of the anatomical feature points according to the normalization processing result; The target prediction data is obtained according to the user's dentition shape information, dentition marking information and position information of anatomical feature points.

5. A full-process intelligent orthodontic tooth movement planning generation method according to claim 1, characterized in that: The orthodontic tooth movement planning generation model includes an embedded coding layer for feature encoding the target prediction data, and a convolutional neural network for feature extraction of the feature-encoded prediction data.

6. A full-process intelligent orthodontic tooth movement planning generation method according to claim 5, characterized in that: The embedded coding layer comprises: A bidirectional long short-term memory network, wherein the bidirectional long short-term memory network is used to encode input target prediction data; A residual network, wherein the residual network is used to extract features of the encoded prediction data by stacking residual blocks layer by layer; A multi-layer self-attention network is provided, wherein the multi-layer self-attention network is used to perform representation learning and layer-by-layer feature extraction on the correlation between different positions in the prediction data.

7. A full-process intelligent orthodontic tooth movement planning generation method according to claim 6, characterized in that: The multi-layer self-attention network is used to perform representation learning and layer-by-layer feature extraction on the correlation between different positions in the prediction data, specifically including: The feature vector output by the residual network is used as the input sequence to pass to the self-attention layer; by calculating the attention weight, the correlation between each position and other positions in the input sequence is established to obtain the attention weight; The obtained attention weights are weighted summed with the representation of the input sequence to obtain the output feature vector.

8. A full-process intelligent orthodontic tooth movement planning generation method according to claim 5, characterized in that: The network structure of the convolutional neural network includes: An input layer, the input layer being used to receive the encoded prediction data; Convolution layer, which slides the convolution kernel on the predicted data to extract local relevant features. The sliding process of each convolution kernel is realized by performing a series of product and sum operations on the predicted data. For the output of each convolution kernel, nonlinearity is introduced through the activation function. A pooling layer, wherein the pooling layer is used to reduce the dimension of the feature map through a pooling operation and retain the target features; Fully connected layer, the output of the pooling layer is connected to one or more fully connected layers, which map the output of the convolutional layer to the output layer to generate the final tooth movement planning result.

9. A full-process intelligent orthodontic tooth movement planning generation method according to claim 6, characterized in that: The residual network specifically includes at least one residual block, each residual block includes a main path and a skip path; The main path is used to extract feature information of input prediction data, and the jump path is used to provide a bypass path that bypasses the main path; The residual network extracts features from the encoded prediction data by stacking residual blocks layer by layer, which specifically includes: Input prediction data, perform convolution and pooling operations on the input prediction data, extract features through the main path, and generate intermediate feature maps; The skip path is used to add the input prediction data to the intermediate feature map output by the main path to form a residual; The residual is transformed nonlinearly through the activation function and enters the next residual block until it loops to the last residual block; The data output by the last residual block is globally pooled to obtain the final feature vector.

10. A full-process intelligent orthodontic tooth movement planning generation system, characterized in that: A full-process intelligent orthodontic tooth movement planning generation method according to any one of claims 1 to 9 is applied, comprising: A data acquisition module, which acquires a three-dimensional medical image of the user's oral cavity and generates a corresponding two-dimensional medical image based on the three-dimensional medical image; A three-dimensional reconstruction module performs three-dimensional reconstruction based on the acquired image data to obtain a three-dimensional model of the oral cavity; A data processing module obtains target prediction data based on the user's oral 3D model; The model analysis module inputs the target prediction data into the trained orthodontic tooth movement planning generation model for analysis to obtain the tooth movement planning results.

Images