Facial form intelligent prediction method and system based on double-discriminant generative adversarial network model

Through the intelligent prediction method of facial morphology based on the double discrimination generative adversarial network model, the problem of relying on subjective judgment in full-mouth denture design is solved, and fast and accurate facial morphology prediction and digital display are achieved, which improves the efficiency and accuracy of denture design.

CN119026474APending Publication Date: 2024-11-26PEKING UNIV SCHOOL OF STOMATOLOGY

Patent Information

Application Number
CN202411124077.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-15
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

The prior art relies on subjective judgment and personal experience in the repair of full dentures, resulting in inefficient design, long time-consuming and inability to achieve digital display, and the inaccurate prediction of facial morphological changes after wearing dentures.

Method used

Using the intelligent prediction method of facial morphology based on the double discrimination generative adversarial network model, the double discrimination generative adversarial network model is constructed, combined with the generator and the global-local discriminator, and the expansion convolution and adversarial network loss function are used to train the denture after the descent is realized. Fast and accurate prediction and digital display of facial morphology.

Benefits of technology

This method can learn more intelligently about the deformation of the front and back of the dentures, capture facial details, ensure the naturalness and rationality of the predicted results, improve the accuracy and efficiency of the denture design, and allow doctors to communicate directly with patients and make local adjustments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119026474A_ABST
    Figure CN119026474A_ABST
Patent Text Reader

Abstract

The invention discloses a face form intelligent prediction method and system based on a double-discriminant generative adversarial network model, and belongs to the technical field of oral medicine. Facial form prediction is carried out through a double-discrimination generative adversarial network model, the double-discrimination generative adversarial network model comprises a generator based on expansion convolution and a global-local discrimination model, and the global-local discrimination model comprises a global discriminator and a local discriminator. Expansion convolution is used in the generator, so that more contextual information can be captured, and meanwhile, the calculation complexity is not increased; a double-discriminator adversarial learning network model is utilized, and details of the lower one third of the face can be focused while the consistency of the overall form of the generated image and a real image is evaluated and the naturalness and reasonability of the overall structure of the face are ensured, so that whether tiny convex-concave changes caused by the complete denture conform to the real situation or not is accurately judged.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of oral medicine, and more specifically to a facial morphology intelligent prediction method and system based on a dual-discriminant generative adversarial network model. Background Art

[0002] Full denture restoration is a complex and tedious treatment process, usually for patients with complete tooth loss. In this operation, the doctor must design a complete treatment plan before the operation to restore the patient's chewing function and facial beauty. As an important aesthetic data required for denture design, the restoration of facial fullness requires a method that can be accurately applied to clinical practice to assist clinicians and technicians in formulating the best denture design and restoration plan.

[0003] At present, doctors mainly rely on subjective judgment and personal experience to predict the facial appearance and facial soft tissue changes of patients after surgery, usually through complex wax-up accumulation to speculate the facial morphology of patients after wearing full dentures. However, this traditional design method has some disadvantages, such as consuming a lot of clinical time, low efficiency, heavy reliance on doctors' personal experience, patients cannot directly participate in the design process, and cannot be digitally displayed.

[0004] Therefore, how to propose a facial morphology intelligent prediction method and system to quickly and accurately achieve clinical facial shape prediction is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0005] In view of this, the present invention provides a facial morphology intelligent prediction method and system based on a dual-discriminant generative adversarial network model, which can quickly realize facial morphology prediction and digital display to assist in formulating the best denture design and restoration plan.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] On the one hand, the present invention discloses a facial morphology intelligent prediction method based on a dual-discriminant generative adversarial network model, comprising the following steps:

[0008] Acquire patient data to establish a data set, wherein the patient data includes a patient's edentulous jaw image, a denture image, and a corresponding facial state image of the patient with teeth;

[0009] Constructing a dual-discriminant generative adversarial network model, wherein the dual-discriminant generative adversarial network model includes a generator and a global-local discriminant model, wherein the global-local discriminant model includes a global discriminator and a local discriminator;

[0010] Inputting the data set into the dual-discriminant generative adversarial network model, training the dual-discriminant generative adversarial network model in combination with the objective function of the discriminator and the adversarial network loss function to obtain an optimized model;

[0011] Obtain the edentulous jaw pictures and denture pictures of the patient to be tested and input them into the optimization model, and output the facial morphology prediction results.

[0012] Preferably, the data set is input into the dual-discriminant generative adversarial network model, and the dual-discriminant generative adversarial network model is trained in combination with a loss function to obtain an optimized model, including:

[0013] Splicing the edentulous jaw image and the denture image to obtain an image to be input and a partial image to be input;

[0014] The image to be input and the partial image to be input are respectively input into the generator, and a tooth wearing state diagram and a tooth wearing state partial diagram are correspondingly obtained;

[0015] The tooth wearing state diagram is used as the input of the global discriminator, and the tooth wearing state local diagram is used as the input of the local discriminator. The dual discriminant generative adversarial network model is trained in combination with the discriminator's objective function and the adversarial network loss function to obtain an optimized model.

[0016] Preferably, the generator is implemented based on a network structure of dilated convolution.

[0017] Preferably, the objective function of the discriminator includes a global objective function and a local objective function;

[0018] The global objective function L DG (G,D G )as follows:

[0019]

[0020] Where D G (·) represents the global discriminator; G(·) represents the generator; represents the expected value; x1 represents the edentulous jaw image; z1 represents the corresponding facial state image of the toothed state; c represents the denture base morphology;

[0021] The local objective function as follows:

[0022]

[0023] Where D L (·) represents the local discriminator, x'1 represents the local facial image of the edentulous jaw, and z'1 represents the corresponding local facial state image after wearing teeth.

[0024] Preferably, the adversarial network loss function L total The formula is as follows:

[0025] L total =L adv +λ L1 L L1 +λ mse L mse +λ per L per

[0026] Where, L per is the perceptual feature loss; L L1 is L1 loss; L mse is the mean square error loss; L adv is the adversarial loss of GAN; L1 ,λ mse and λ per They are the weight factors of L1 loss, mean square error loss, and perceptual feature loss respectively.

[0027] On the other hand, the present invention also discloses a facial morphology intelligent prediction system based on a dual-discriminant generative adversarial network model, comprising:

[0028] A data set establishment module, used to acquire patient data to establish a data set, wherein the patient data includes a patient's edentulous jaw picture, a denture picture, and a corresponding facial state picture of the patient with teeth;

[0029] A network model construction module, used to construct a dual-discriminant generative adversarial network model, wherein the dual-discriminant generative adversarial network model includes a generator and a global-local discriminant model, wherein the global-local discriminant model includes a global discriminator and a local discriminator;

[0030] A network model training module, used for inputting the data set into the dual-discriminant generative adversarial network model, training the dual-discriminant generative adversarial network model in combination with the objective function of the discriminator and the adversarial network loss function, to obtain an optimized model;

[0031] The prediction module is used to obtain the edentulous jaw image and denture image of the patient to be tested and input them into the optimization model to output the facial morphology prediction result.

[0032] Preferably, the network model training module includes:

[0033] An image processing unit is used to splice the edentulous jaw image and the denture image to obtain an image to be input and a partial image to be input;

[0034] A preliminary prediction unit, inputting the image to be input and the partial image to be input into the generator respectively, and correspondingly obtaining a tooth wearing state diagram and a tooth wearing state partial diagram;

[0035] The adversarial training unit is used to use the tooth wearing state diagram as the input of the global discriminator, and the tooth wearing state local diagram as the input of the local discriminator, and train the dual-discriminant generative adversarial network model in combination with the discriminator's objective function and the adversarial network loss function to obtain an optimized model.

[0036] It can be seen from the above technical solutions that the present invention discloses a facial morphology intelligent prediction method based on a dual-discriminant generative adversarial network model, which has the following beneficial effects compared with the prior art:

[0037] 1. The dual discriminator adversarial learning network model can effectively learn the facial deformation before and after wearing teeth. Compared with the existing facial deformation methods based on back propagation neural network, volume mesh deformation, etc., this technology can more intelligently predict the changes in facial morphology after patients wear dentures.

[0038] 2. A facial deformation prediction method based on dilated convolution generative adversarial network is proposed, which uses dilated convolution to capture more contextual information without increasing computational complexity. In scenarios where accurate simulation of facial convexity and concavity changes is required, dilated convolution plays an important role in extracting detailed facial images.

[0039] 3. The dual-discriminator generative adversarial network can evaluate the consistency of the overall morphology of the generated image with the real image, ensure the naturalness and rationality of the overall facial structure, and focus on the details of the lower third of the face to accurately determine whether the subtle convex and concave changes caused by full dentures are consistent with the actual situation. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] 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.

[0041] Figure 1 A schematic diagram of the overall structure of the dual-discriminant generative adversarial network model provided by the present invention;

[0042] Figure 2(a) is a schematic diagram of the generator network structure, Figure 2(b) is a schematic diagram of the global discriminator network structure, and Figure 2(c) is a schematic diagram of the local discriminator network structure;

[0043] Figure 3 This is a system architecture diagram provided by the present invention. DETAILED DESCRIPTION

[0044] 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.

[0045] On the one hand, an embodiment of the present invention discloses a facial morphology intelligent prediction method based on a dual-discriminant generative adversarial network model, comprising the following steps:

[0046] S1. Acquire patient data to establish a data set, where the patient data includes the patient's edentulous jaw image, denture image, and corresponding facial state image of the patient with teeth.

[0047] Collect data from 1,000 patients who have completed full denture treatment before, including pictures of each patient's edentulous jaw, dentures, and facial state of the patient wearing teeth after treatment. The patient's mandible needs to be in the resting jaw position. 700 cases of data are used as training sets, 200 cases of data are used as validation sets, and 100 cases of data are used as test sets.

[0048] S2. Construct a dual-discriminant generative adversarial network model (reference Figure 1 ), the dual-discriminant generative adversarial network model includes a generator and a global-local discriminant model, and the global-local discriminant model includes a global discriminator and a local discriminator.

[0049] The network model in this embodiment uses a convolution operation with a fractional step size (1 / 4) to reduce the resolution by a factor of two, which can generate a clearer facial morphology prediction. The network consists of multiple layers, in which a set of filters are convolved with the input image to generate an occlusal image. After each convolution layer (Conv), except for the last layer, a batch normalization layer (BN) and a rectified linear unit (ReLU) activation function are added.

[0050] Specifically, the generator G is implemented based on the network structure of dilated convolution, and its network structure is shown in Figure 2(a).

[0051] To ensure that the network can handle complex occlusal reconstructions, dilated convolutional layers are added to the generator network to enhance the learning ability of the network, mainly using dispersed convolutional kernels and allowing each output pixel to be calculated over a larger input area.

[0052] The dilated convolution layer connects the previous feature map to the subsequent output in the channel respectively. In addition, the dilated convolution layer obtains fine-grained structural information and expands the receptive field to achieve effective feature representation. In order to reconstruct a clearer facial morphological structure, different dilation rates are set to obtain a receptive field with an appropriate scale. In addition, this connection allows each feature map to undergo multiple convolution operations with different dilation rates to capture different types of feature information. The feature map of each layer is integrated into the output of the last dilated convolution layer, so that the generated model can extract more effective and richer biometric features. More specifically, if a 2D layer is h of the i-th channel i × i image, and the next layer is h of the i+1th channel i+1 × i+1 image, then the dilated convolution operator for each pixel can be expressed as

[0053]

[0054] Among them, k h+1 =(k h-1 ) / 2 and k w+1 =(k w-1 ) / 2 are the width and height of the convolution kernel, respectively; x and y are the pixel components of the layer’s input and output; F(·) represents the component nonlinear transfer function; b is the bias vector; M represents the i+1×i matrix of the convolution kernel; and λ is the dilation coefficient.

[0055] The global discriminator network structure and the local discriminator network structure are shown in Figure 2(b) and Figure 2(c) respectively. The global discriminator uses a convolution kernel of size 5×5 and a convolution operation of stride 2. The local discriminator follows the same structural pattern, consisting of five convolutional layers and one fully connected layer. Finally, the outputs of the two discriminators are fused together through a splicing layer.

[0056] S3. Input the data set into the dual-discriminant generative adversarial network model, and train the dual-discriminant generative adversarial network model by combining the discriminator's objective function and the adversarial network loss function to obtain an optimized model.

[0057] S31. The edentulous jaw image and the denture image are spliced ​​to obtain an image to be input and a partial image to be input, wherein the partial image is the lower third of the face.

[0058] The data obtained in S1 were preprocessed using Geomagic software, and the patient data were adjusted to the same coordinate system. The post-treatment model was matched to the pre-treatment model based on the positions of the forehead and nose where the changes were not obvious, and the images to be input and partial images to be input were obtained accordingly.

[0059] S32. Input the image to be input and the partial image to be input into the generator respectively, and obtain the tooth wearing state diagram and the tooth wearing state partial diagram accordingly.

[0060] During the training process, the edentulous jaw images and denture images are spliced ​​and input into the generator, downsampled through convolution operations, and further used with multiple layers of dilated convolutions to enhance the network's ability to extract contextual information. During the decoding process, dilated convolutions are also used to enhance the network's ability to perceive contextual information during the image generation process, and finally a tooth wearing state map and a tooth wearing state partial map are generated.

[0061] S33. The tooth wearing state diagram is used as the input of the global discriminator, and the tooth wearing state local diagram is used as the input of the local discriminator. The dual discriminant generative adversarial network model is trained by combining the objective function of the discriminator and the adversarial network loss function to obtain an optimized model.

[0062] In order to obtain more realistic facial change details, a dual discriminator network is used as the GAN discriminator to distinguish high-resolution real and generated facial images. During the training process, the global discriminator D G and the local discriminator D L They share the same generator G, but their outputs are tested in different ways. Global Discriminator D G The tooth wearing state diagram is used as input, and the local discriminator D L Only the local image of the lower third of the face, i.e. the state of wearing teeth, is used as input.

[0063] Global Discriminator D G The objective function of adversarial learning (i.e., global objective function) is expressed as:

[0064] Global objective function as follows:

[0065]

[0066] Where D G (·) represents the global discriminator; G(·) represents the generator; represents the expected value; x1 represents the edentulous jaw picture; z1 represents the corresponding facial state picture with teeth; c represents the denture base morphology.

[0067] The generator G strives to minimize the global objective function, while the global discriminator D G Trying to To maximize the global objective function.

[0068] Local objective function as follows:

[0069]

[0070] Where DL (·) represents the local discriminator, x'1 represents the local facial image of the edentulous jaw, and z'1 represents the corresponding local facial state image after wearing teeth.

[0071] Effective adversarial training is a necessary condition to ensure the quality of facial deformation prediction. The purpose of successful prediction by the global discriminator is that the generated picture after tooth wearing (i.e., the tooth wearing state picture) should not only be realistic, but also consistent with the facial morphological features before tooth wearing. The local discriminator is used to identify whether the changes in the local tooth wearing state picture are real. Finally, the outputs of the two discriminators are fused together through a splicing layer to predict a continuous value corresponding to the probability that the facial reconstructed image is real. The ultimate goal is to learn a generative network conditioned on the facial morphology of completely missing teeth and the facial biometric morphology. Therefore, the generator G is encouraged to generate a facial reconstruction picture after tooth wearing that is realistic enough to deceive the two discriminators D. G and D L . So that the adversarial learning problem becomes a multi-objective learning problem, which is defined as follows:

[0072]

[0073] in,

[0074] Furthermore, in order to train the network to reconstruct the real crown, we add the perceptual feature loss (L per ), L1 loss (L L1 ) and mean square error (L mse ) to improve the adversarial loss of GAN (L adv ), based on these four losses, we construct the adversarial network loss function L total The formula is as follows:

[0075] L total =L adv +λ L1 L L1 +λ mse L mse +λ per L per ;

[0076] Where, L per is the perceptual feature loss; L L1 is L1 loss; L mse is the mean square error loss; L adv is the adversarial loss of GAN; L1 ,λ mse and λ per They are the weight factors of L1 loss, mean square error loss, and perceptual feature loss respectively.

[0077] Specifically, L1 loss is a commonly used regression loss function, which is the sum of the absolute values ​​of the difference between the target value and the predicted value, indicating the average error amplitude of the predicted value without considering the direction of the error. The formula is as follows:

[0078]

[0079] The mean square error (MSE) is one of the most widely used fidelity metrics in medical image analysis research. It evaluates the difference between the generated crown image G(x) and its corresponding target crown z at the pixel level and objectively quantifies the strength of the error signal. Therefore, the MSE loss is used to stabilize the training:

[0080]

[0081] Among them, ||·|| F represents the Frobenius norm.

[0082] Compared with MSE loss, the perceptual feature loss L per Evaluating the two crown images at the feature level rather than the pixel level helps preserve the generated occlusal surface details and makes it clearer in terms of structure. per Added to the network’s training to further improve the ability to reconstruct the biomorphic features of the missing tooth surface. The perceptual feature loss can be expressed as:

[0083]

[0084] S4. Obtain the edentulous jaw and denture images of the patient to be tested and input them into the optimization model, output the facial morphology prediction results, and visualize the facial effect of the patient after wearing the denture. The doctor can communicate directly with the patient in the clinic, modify the local position, and send it to the technician for denture design and production after confirmation.

[0085] On the other hand, reference Figure 3 The present invention also discloses a facial morphology intelligent prediction system based on a dual-discriminant generative adversarial network model, comprising:

[0086] A data set building module is used to acquire patient data to build a data set, wherein the patient data includes a patient's edentulous jaw image, a denture image, and a corresponding facial state image of the patient with teeth;

[0087] A network model building module is used to build a dual-discrimination generative adversarial network model, which includes a generator and a global-local discriminant model, and the global-local discriminant model includes a global discriminator and a local discriminator;

[0088] The network model training module is used to input the data set into the dual-discriminant generative adversarial network model, and train the dual-discriminant generative adversarial network model by combining the objective function of the discriminator and the adversarial network loss function to obtain an optimized model;

[0089] The prediction module is used to obtain the edentulous jaw and denture images of the patient to be tested and input them into the optimization model to output the facial morphology prediction results.

[0090] Preferably, the network model training module includes:

[0091] An image processing unit is used to splice the edentulous jaw image and the denture image to obtain the image to be input and the partial image to be input;

[0092] The preliminary prediction unit inputs the image to be input and the partial image to be input into the generator, and obtains a tooth wearing state diagram and a tooth wearing state partial diagram accordingly;

[0093] The adversarial training unit is used to use the tooth wearing state diagram as the input of the global discriminator and the tooth wearing state local diagram as the input of the local discriminator, and to train the dual-discriminant generative adversarial network model by combining the objective function of the discriminator and the adversarial network loss function to obtain an optimized model.

[0094] 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.

[0095] 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 facial morphology intelligent prediction method based on a dual-discriminant generative adversarial network model, characterized in that: The following steps are involved: Acquire patient data to establish a data set, wherein the patient data includes a patient's edentulous jaw image, a denture image, and a corresponding facial state image of the patient with teeth; Constructing a dual-discriminant generative adversarial network model, wherein the dual-discriminant generative adversarial network model includes a generator and a global-local discriminant model, wherein the global-local discriminant model includes a global discriminator and a local discriminator; Inputting the data set into the dual-discriminant generative adversarial network model, training the dual-discriminant generative adversarial network model in combination with the objective function of the discriminator and the adversarial network loss function to obtain an optimized model; Obtain the edentulous jaw pictures and denture pictures of the patient to be tested and input them into the optimization model, and output the facial morphology prediction results.

2. According to claim 1, a facial morphology intelligent prediction method based on a dual-discriminant generative adversarial network model is characterized in that: Inputting the data set into the dual-discriminant generative adversarial network model, training the dual-discriminant generative adversarial network model in combination with a loss function to obtain an optimized model, including: Splicing the edentulous jaw image and the denture image to obtain an image to be input and a partial image to be input; The image to be input and the partial image to be input are respectively input into the generator, and a tooth wearing state diagram and a tooth wearing state partial diagram are correspondingly obtained; The tooth wearing state diagram is used as the input of the global discriminator, and the tooth wearing state local diagram is used as the input of the local discriminator. The dual discriminant generative adversarial network model is trained in combination with the discriminator's objective function and the adversarial network loss function to obtain an optimized model.

3. According to claim 2, a facial morphology intelligent prediction method based on a dual-discriminant generative adversarial network model is characterized in that: The generator is implemented based on a network structure of dilated convolution.

4. According to claim 2, a facial morphology intelligent prediction method based on a dual-discriminant generative adversarial network model is characterized in that: The objective function of the discriminator includes a global objective function and a local objective function; The global objective function as follows: Where D G (·) represents the global discriminator; G(·) represents the generator; represents the expected value; x1 represents the edentulous jaw picture; z1 represents the corresponding facial state picture of the toothed state; c represents the denture base morphology; The local objective function as follows: Where D L (·) represents the local discriminator, x'1 represents the local facial image of the edentulous jaw, and z'1 represents the corresponding local facial state image after wearing teeth.

5. The facial morphology intelligent prediction method based on a dual-discriminant generative adversarial network model according to claim 2 is characterized in that: The adversarial network loss function L total The formula is as follows: L total =L adv +λ L1 L L1 +λ mse L mse +λ per L per Where, L per is the perceptual feature loss; L L1 is L1 loss; L mse is the mean square error loss; L adv is the adversarial loss of GAN; L1 ,λ mse and λ per They are the weight factors of L1 loss, mean square error loss, and perceptual feature loss respectively.

6. A facial morphology intelligent prediction system based on a dual-discriminant generative adversarial network model, characterized in that: include: A data set establishment module, used to acquire patient data to establish a data set, wherein the patient data includes a patient's edentulous jaw picture, a denture picture, and a corresponding facial state picture of the patient with teeth; A network model construction module, used to construct a dual-discriminant generative adversarial network model, wherein the dual-discriminant generative adversarial network model includes a generator and a global-local discriminant model, wherein the global-local discriminant model includes a global discriminator and a local discriminator; A network model training module, used for inputting the data set into the dual-discriminant generative adversarial network model, training the dual-discriminant generative adversarial network model in combination with the objective function of the discriminator and the adversarial network loss function, to obtain an optimized model; The prediction module is used to obtain the edentulous jaw image and denture image of the patient to be tested and input them into the optimization model to output the facial morphology prediction result.

7. The facial morphology intelligent prediction system based on the dual-discriminant generative adversarial network model according to claim 6, characterized in that: The network model training module includes: An image processing unit is used to splice the edentulous jaw image and the denture image to obtain an image to be input and a partial image to be input; A preliminary prediction unit, inputting the image to be input and the partial image to be input into the generator respectively, and correspondingly obtaining a tooth wearing state diagram and a tooth wearing state partial diagram; The adversarial training unit is used to use the tooth wearing state diagram as the input of the global discriminator, and the tooth wearing state local diagram as the input of the local discriminator, and train the dual-discriminant generative adversarial network model in combination with the objective function of the discriminator and the adversarial network loss function to obtain an optimized model.

Citation Information

Patent Citations

  • A face image synthesis method and device based on adversarial learning

    CN109815928A

  • Deformable medical image registration method and system

    CN113763442A

Cited By

  • Point cloud-based face capacity change prediction method, system and equipment after edentulous jaw implantation

    CN119559160A