WO2023202143A1 - 基于深度学习的牙齿修复体自动设计方法及系统 - Google Patents

基于深度学习的牙齿修复体自动设计方法及系统 Download PDF

Info

Publication number
WO2023202143A1
WO2023202143A1 PCT/CN2022/142661 CN2022142661W WO2023202143A1 WO 2023202143 A1 WO2023202143 A1 WO 2023202143A1 CN 2022142661 W CN2022142661 W CN 2022142661W WO 2023202143 A1 WO2023202143 A1 WO 2023202143A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
model
restoration
deep learning
processing model
Prior art date
Application number
PCT/CN2022/142661
Other languages
English (en)
French (fr)
Inventor
沈震
杨静
董西松
熊刚
张骜
方启航
王庆
万力
Original Assignee
东莞中科云计算研究院
十维(广东)科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 东莞中科云计算研究院, 十维(广东)科技有限公司 filed Critical 东莞中科云计算研究院
Publication of WO2023202143A1 publication Critical patent/WO2023202143A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61C—DENTISTRY; APPARATUS OR METHODS FOR ORAL OR DENTAL HYGIENE
    • A61C5/00—Filling or capping teeth
    • A61C5/20—Repairing attrition damage, e.g. facets
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61C—DENTISTRY; APPARATUS OR METHODS FOR ORAL OR DENTAL HYGIENE
    • A61C9/00—Impression cups, i.e. impression trays; Impression methods
    • A61C9/004—Means or methods for taking digitized impressions
    • A61C9/0046—Data acquisition means or methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/16—Customisation or personalisation

Definitions

  • This application relates to the technical field of dental computer-aided design, and in particular to an automatic design method and system for dental restorations based on deep learning.
  • the traditional dental restoration design method mainly relies on the experience and technology of dental experts. Not only is the restoration accuracy difficult to control, it takes a long time, but it also brings a lot of inconvenience to the patient.
  • the more popular computer-aided design method uses the crown in the standard tooth database as the initial model of the defective tooth, and uses a series of deformation algorithms to perform appropriate deformation operations to reconstruct the target crown model. Although this is to a certain extent It solves the problem of designing restorations entirely based on experience and technology, but this design method lacks personalization, is less robust, and still requires a lot of manual operations.
  • the purpose of this application is to solve the above technical problems and provide a deep learning-based automatic design method and system for dental restorations that can perform personalized design of dental restorations fully automatically and without manual labor.
  • this application discloses an automatic design method of dental restorations based on deep learning, which includes: segmenting the corresponding three-dimensional segmentation data from the patient's dentition data, and obtaining the target restoration data to form a method including: Data set C of the three-dimensional segmentation data and the target restoration data;
  • the data set C is used to train a data processing model based on deep learning, so that the data processing model can generate a restoration model that matches the patient's three-dimensional segmentation data.
  • the method for obtaining the three-dimensional segmentation data includes:
  • the basic data set is used to train a data-assisted processing model based on deep learning, and the corresponding three-dimensional segmentation data is segmented from the input dentition data through the data-assisted processing model.
  • the basic data collection method includes:
  • Obtain a data set A which includes several pairs of corresponding maxillary and mandibular dentition data and target restoration data respectively from each patient;
  • the maxillary and mandibular dentition data in the data set A are annotated to distinguish the data corresponding to each tooth, thereby forming a data set B used as the basic data.
  • the output restoration model is also evaluated according to a preset evaluation index, an evaluation result is generated, and the data processing model is optimized according to the evaluation result.
  • the number and type of restoration designs are input into the data processing model as conditions to generate a corresponding restoration model.
  • explicit constraints are set on the loss function of the data processing model.
  • the data auxiliary processing model is constructed based on a convolutional neural network, and the data processing model is constructed based on a convolutional neural network or a generative adversarial network.
  • the data auxiliary processing model and the data processing model are trained sequentially through transfer learning according to the sample data size from large to small.
  • the data processing model is built based on a generative adversarial network.
  • the data processing model includes a generative model and a discriminant model.
  • One or more generators are provided in the generative model, and each generator is used to learn a A specific tooth repair mode
  • the discriminant model includes a discriminator or a classifier
  • the classifier is used to classify the restoration model, so that the generation model generates a corresponding restoration model.
  • the output restoration model is also evaluated according to a preset evaluation index, an evaluation result is generated, and the data processing model is optimized according to the evaluation result;
  • the method for optimizing the data processing model according to the evaluation results includes:
  • the results with qualified evaluation results are added to the data set C
  • the restoration model is saved as the target restoration
  • the results with unqualified evaluation results are added to the data set C.
  • the results are stored in the data set D, and the data processing model is optimized according to the data set C and the data set D.
  • This application also discloses an automatic design system for dental restorations based on deep learning, which includes:
  • processors one or more processors
  • the present application also discloses a computer-readable storage medium, which includes a computer program that can be executed by a processor to complete the deep learning-based automatic design method of dental restorations as described above.
  • the automatic design method of dental restorations in this application provides a data processing model.
  • designing a dental restoration for a patient only the patient's dentition data can be obtained to output the corresponding restoration based on the data processing model. It can be seen from this that according to the above design method, dental restorations can be designed fully automatically without any manual interaction during the design process.
  • Figure 1 is a flow chart of an automatic design method for dental restorations according to one embodiment of the present application.
  • FIG. 2 is a specific flow chart of step S1 in Figure 1.
  • Figure 3 is a flow chart of an automatic design method for dental restorations according to another embodiment of the present application.
  • Figure 4 is a schematic flow chart of optimizing the data processing model in the design method of dental restorations based on generative adversarial networks in the embodiment of the present application.
  • Figure 5 is a schematic diagram of the sample format included in data set A in the embodiment of the present application.
  • Figure 6 is a schematic diagram of the sample format included in data set B in the embodiment of the present application.
  • Figure 7 is a schematic diagram of the sample format included in data set C in the embodiment of the present application.
  • Figure 8 is a schematic diagram of the sample format included in the data set D in the embodiment of the present application.
  • Figure 9 is a schematic diagram of multiple types of dental restorations in an embodiment of the present application.
  • Figure 10 is a schematic diagram of the distance between the upper and lower teeth in the embodiment of the present application.
  • Figure 11 is a schematic diagram of a network model trained based on the idea of transfer learning in an embodiment of the present application.
  • Figure 12 is a schematic diagram of a state in which the data processing model generates a restoration model in one embodiment of the present application.
  • Figure 13 is a schematic diagram of a state in which the data processing model generates a restoration model in another embodiment of the present application.
  • Figure 14 is a schematic diagram of a simple network model of the generator and discriminator in the embodiment of the present application.
  • Figure 15 is a schematic diagram of a complex network model in which the generative model includes multiple generators and discriminators in this embodiment of the present application.
  • Figure 16 is a structural diagram of the automatic design system for dental restorations based on deep learning in the embodiment of the present application.
  • This embodiment discloses an automatic design method for dental restorations based on deep learning, which is used for the design of dental restorations during dental treatment.
  • Dental restorations include inlays, inner crowns, full crowns, bridge crowns, etc., As shown in Figure 1, the design method includes the following steps:
  • Dataset C includes three-dimensional segmentation data (corresponding to the target teeth related to dental restoration) and its corresponding target restoration data.
  • the target restoration data in this embodiment is a dental restoration designed and tested by a dentist or technician.
  • the target restoration data can be inlay data, endocrown data, full crown data, and bridge crown data. any one or more.
  • the three-dimensional segmentation data related to tooth restoration includes prepared teeth, adjacent teeth, symmetrical teeth, and opposing teeth.
  • the so-called prepared teeth are the teeth that are prepared for restoration.
  • the adjacent teeth are the teeth adjacent to the prepared teeth.
  • the symmetrical teeth are the teeth that are in a symmetrical position with the prepared teeth in the oral cavity.
  • the opposing teeth are the teeth that are opposite to the prepared teeth in the upper and lower jaws. .
  • S2 Use data set C to train the data processing model, so that the data processing model can generate a restoration model that matches the three-dimensional segmentation data.
  • S4 Input the three-dimensional segmentation data into the data processing model to obtain a restoration model that matches the patient's dentition.
  • a restoration model that matches the patient's dentition.
  • step S1 also includes a method for collecting basic data, which is specifically as follows:
  • Data set A includes several pairs of corresponding maxillary and mandibular dentition data and target restoration data respectively from each patient.
  • the data set A can be represented by three-dimensional point clouds, three-dimensional voxels, three-dimensional grid models, multi-view image sets, 2.5-dimensional depth maps, octree structures, etc.
  • the data processing model in this embodiment can not only directly output the restoration model (as shown in Figure 12), but also generate a three-dimensional model of the repaired teeth (as shown in Figure 13) based on the input three-dimensional segmentation data, and then The difference between the tooth three-dimensional model and the dentition data is taken to obtain the restoration model.
  • the above design method also includes:
  • S5 Evaluate the appearance and function of the output restoration model according to the preset evaluation indicators, and generate an evaluation result (passed or failed).
  • S6 After generating a restoration model that matches the patient's dentition, determine whether the evaluation result of the restoration model is qualified. If not, return the three-dimensional segmentation data corresponding to the restoration model to data processing again. model to obtain the restoration model again, and this cycle continues until a restoration model with qualified evaluation results is obtained.
  • the evaluation indicators in this implementation include occlusal interference strength, contact points, tooth shape curvature, adjacent tooth gap (as shown in Figure 10), minimum thickness and other matters.
  • the restoration made by the output restoration model can better meet the usage requirements, and the accuracy of the restoration model output by the data processing model can be effectively improved.
  • the restoration models include many types (inlays, endocrowns, full crowns, bridge crowns), and there are many public large-scale 3D model data sets, such as the ShapeNet data set. , Completion3D dataset, KITTI dataset. Therefore, as shown in Figure 11, the network model is trained sequentially from large to small according to the amount of data that can be obtained through transfer learning.
  • the network model in this embodiment includes a data-assisted processing model and a data processing model.
  • Transfer learning in this embodiment is to apply knowledge or patterns learned in a certain field or task to different but related fields or problems. It can be divided into instance-based transfer learning, feature-based transfer learning and feature-based transfer learning. Transfer learning of shared parameters. The specific process of transfer learning belongs to common knowledge in this field and will not be described again here.
  • the number and type of restoration designs can be input into the data processing model as conditions to generate a corresponding restoration model. For example, if a crown, an inlay and a bridge crown are input into the data processing model, then the data processing model can output three restoration models based on the input three-dimensional segmentation data related to the patient's dentition environment. Crowns, inlays and bridge crowns, as shown in Figure 9, thus improving design efficiency and quality.
  • the restoration model output by the data processing model that is suitable for the patient's dentition is converted into a grid model, and the grid model is used to manufacture the dental restoration.
  • the mesh model in this embodiment is a model that uses a series of polygons (usually triangles) of similar size and shape to approximately represent a three-dimensional object, and is a three-dimensional model.
  • explicit constraints can also be set on the loss function of the data processing model.
  • the explicit constraints include the statistical characteristics of the interdental space, the generated restoration and the real restoration.
  • the gap feature can be counted by histograms, and related formulas such as chi-square distance and Bhattacharyya distance can be used to measure the distance between two histograms.
  • the occlusal interference strength of upper molars and lower molars is 0.02 to 0.05 mm
  • the occlusal gap provided by full crown tooth preparation is generally 0.5 to 1 mm.
  • the data-assisted processing model in the above embodiment is constructed based on a convolutional neural network
  • the data processing model is constructed based on a convolutional neural network or a generative adversarial network.
  • Network models based on adversarial networks include simple network models and complex network models, as shown in Figure 14.
  • the simple network model includes a single generator and a discriminator.
  • the data processing model is built based on a generative adversarial network, as shown in Figure 15.
  • the data processing model includes a generative model and a discriminant model.
  • One or more generators are set in the generative model, and each generator is used to learn a specific tooth.
  • Restoration modes (such as cusps, ridges, pits and fissures, etc.)
  • the discriminant model includes a discriminator or a classifier, and the classifier is used to classify the restoration model, so that the generated model generates a corresponding restoration model.
  • each generator in the generative model focuses on learning a specific tooth repair mode, so that the final restoration model obtained by the generative model comes from multiple generators that have learned different modes to alleviate the problem of mode collapse.
  • the discriminant model can not only include a discriminator but also a classifier. In this way, the discriminant model not only undertakes the task of distinguishing true and false restorations, but can also classify real restorations to improve the convergence speed of the network.
  • the classifier classifies teeth into incisors, canines, premolars, and molars according to their morphology to better guide the generation of models to generate dental restorations.
  • methods for optimizing the data processing model based on evaluation results include:
  • the process of optimizing the data processing model includes the following steps:
  • Step S60 Generate model G and output restoration model G(z) through data set C;
  • Step S61 Train the discrimination model through the restoration model G(z) and its corresponding target restoration x, plus the restoration model in the data set D, so that it can distinguish the restoration model G(z) and the target restoration. x, and identify the restoration model in data set D as a generated restoration;
  • Step S62 Based on the discriminant model, use the data set C to retrain the generation model G;
  • Step S63 Repeat steps S60 to S62.
  • the final discriminator cannot distinguish between the restoration model G(z) generated by data set C and the target restoration x, but it can better distinguish the unqualified restoration model as a generated restoration.
  • the loss function using a generative adversarial network to construct a data processing model is as follows,
  • x represents the target restoration
  • z represents random noise
  • y represents the conditional data guiding the generative adversarial network.
  • the above embodiment discloses an automatic design method for dental restorations, which forms a data-assisted processing model and a data processing model by collecting dental restoration data of dental patients, and uses an evaluation method to evaluate the data.
  • the restoration model output by the processing model is evaluated, and the data processing model is optimized based on the evaluation results, so as to obtain an automatic design model of dental restorations with relatively high accuracy.
  • the matching restoration model is then converted into a mesh model. Through this mesh model, additive manufacturing or other manufacturing methods can be used to evaluate the quality of the restoration according to the requirements of the patient or doctor (such as manufacturing materials).
  • the mesh model of the restoration is created. The doctor bonds it to the patient's prepared tooth.
  • dental restorations can be designed fully automatically without any manual interaction during the design process. This not only solves the problem of relying on the experience and technology of doctors to design dental restorations, but also effectively improves design efficiency and saves time. and labor. Furthermore, the restoration model is generated based on the patient's dentition data, making the final dental restoration more personalized. In addition, with the increase in the number of restoration designs, data-assisted processing models and data processing models have been continuously optimized, and design capabilities have been continuously improved, thus better meeting the needs of dentists and patients.
  • the invention also discloses an automatic design system for dental restorations based on deep learning, as shown in Figure 16, which includes a central processor 100, a dental database 101, a data acquisition module 102, an automatic design module 103 and a user interaction interface 104. Doctors or technicians can modify and visualize the design results through the user interaction interface.
  • the dental database 101 is used to store data related to the patient's dentition.
  • the user interaction interface 104 is used to facilitate users to input data and obtain result data.
  • the data acquisition module 102 is used to segment the corresponding three-dimensional segmentation data from the patient's dentition data and obtain the corresponding target restoration data to form a data set C including the three-dimensional segmentation data and the target restoration data.
  • the data set C is stored in the dental database.
  • the automatic design module 103 is used to use the data set C to train a data processing model based on deep learning, so that the data processing model can generate a restoration model adapted to the patient's three-dimensional segmentation data.
  • the above-mentioned system also includes a data-assisted processing module 105, which is used to train a data-assisted processing model based on deep learning using a basic data set.
  • the data-assisted processing model can be derived from the input dentition data. Segment the corresponding three-dimensional segmentation data.
  • the data-assisted processing model is constructed based on a convolutional neural network
  • the data processing model is constructed based on a convolutional neural network or a generative adversarial network.
  • the above system also includes a basic data generation module 106, which is used to annotate the maxillary and mandibular dentition data in the data set A to distinguish the data corresponding to each tooth, thereby forming Data set B used as basic data, said data set A includes several pairs of corresponding maxillary and mandibular dentition data and target restoration data respectively from each patient.
  • a basic data generation module 106 which is used to annotate the maxillary and mandibular dentition data in the data set A to distinguish the data corresponding to each tooth, thereby forming Data set B used as basic data
  • said data set A includes several pairs of corresponding maxillary and mandibular dentition data and target restoration data respectively from each patient.
  • the above system also includes an evaluation module 107, which is used to evaluate the output restoration model according to preset evaluation indicators, generate evaluation results, and process the data according to the evaluation results.
  • the model is optimized.
  • the above system also includes a data conversion module 108 for converting the restoration model output by the data processing model that is adapted to the patient's dentition into a grid model.
  • the grid model Used in the manufacture of dental restorations.
  • the above-mentioned system also includes a transfer learning module 109, which is used to sequentially train the data auxiliary processing model and the data processing model through transfer learning according to the amount of sample data from large to small.
  • This application also discloses another automatic design system for dental restorations based on deep learning, which includes one or more processors, memories, and one or more programs, wherein one or more programs are stored in the memory, and Configured to be executed by the one or more processors, the program includes instructions for performing the deep learning-based automated design method of dental restorations as described above.
  • the processor can use a general central processing unit (Central Processing Unit, CPU), a microprocessor, an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits for executing related programs to implement
  • the modules in the deep learning-based automatic design system for dental restorations in the embodiments of the present application are required to perform functions, or execute the deep learning-based automatic design method for dental restorations in the method embodiments of the present application.
  • the program includes instructions for executing an automatic design method for dental restorations based on deep learning as described below:
  • Segment the corresponding three-dimensional segmentation data from the patient's dentition data and obtain the target restoration data to form a data set C including the three-dimensional segmentation data and the target restoration data;
  • the data set C is used to train a data processing model based on deep learning, so that the data processing model can generate a restoration model that matches the patient's three-dimensional segmentation data.
  • the instruction of the program to execute the automatic design method of dental restorations based on deep learning also includes: evaluating the output restoration model according to a preset evaluation index, and Generate evaluation results, and optimize the data processing model based on the evaluation results.
  • the data-assisted processing model is constructed based on a convolutional neural network
  • the data processing model is constructed based on a convolutional neural network or a generative adversarial network.
  • the data processing model is built based on a generative adversarial network.
  • the data processing model includes a generative model and a discriminant model.
  • One or more generators are provided in the generative model, each of which The generator is used to learn a specific tooth repair mode, and the discriminant model includes a discriminator or a classifier, and the classifier is used to classify the restoration model, so that the generation model generates a corresponding restoration model.
  • the instruction of the program to execute the automatic design method of dental restorations based on deep learning also includes: evaluating the output restoration model according to a preset evaluation index, and Generate evaluation results, and optimize the data processing model according to the evaluation results;
  • the method for optimizing the data processing model according to the evaluation results includes:
  • the results with qualified evaluation results are added to the data set C
  • the restoration model is saved as the target restoration
  • the results with unqualified evaluation results are added to the data set C.
  • the results are stored in the data set D, and the data processing model is optimized according to the data set C and the data set D.
  • the present application also discloses a computer-readable storage medium, which includes a computer program that can be executed by a processor to complete the deep learning-based automatic design method of dental restorations as described above.
  • the computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center integrated with one or more available media.
  • the available media may be read-only memory (ROM), random access memory (RAM), or magnetic media, such as floppy disks, hard disks, tapes, disks, or optical media, such as, Digital versatile disc (digital versatile disc, DVD), or semiconductor media, such as solid state drive (solid state disk, SSD), etc.
  • the computer program can be executed by a processor to complete the following deep learning-based automatic design method for dental restorations including:
  • Segment the corresponding three-dimensional segmentation data from the patient's dentition data and obtain the target restoration data to form a data set C including the three-dimensional segmentation data and the target restoration data;
  • the data set C is used to train a data processing model based on deep learning, so that the data processing model can generate a restoration model that matches the patient's three-dimensional segmentation data.
  • the computer program can be executed by a processor to complete the automatic design method of dental restorations based on deep learning, which further includes: evaluating the output restoration model according to a preset evaluation index. Conduct evaluation, generate evaluation results, and optimize the data processing model based on the evaluation results.
  • the data-assisted processing model is constructed based on a convolutional neural network
  • the data processing model is constructed based on a convolutional neural network or a generative adversarial network.
  • the data processing model is built based on a generative adversarial network.
  • the data processing model includes a generative model and a discriminant model.
  • One or more generators are provided in the generative model, each of which The generator is used to learn a specific tooth repair mode, and the discriminant model includes a discriminator or a classifier, and the classifier is used to classify the restoration model, so that the generation model generates a corresponding restoration model.
  • the computer program can be executed by a processor to complete the automatic design method of dental restorations based on deep learning, which further includes: evaluating the output restoration model according to a preset evaluation index. Conduct evaluation, generate evaluation results, and optimize the data processing model according to the evaluation results;
  • the method for optimizing the data processing model according to the evaluation results includes:
  • the results with qualified evaluation results are added to the data set C
  • the restoration model is saved as the target restoration
  • the results with unqualified evaluation results are added to the data set C.
  • the results are stored in the data set D, and the data processing model is optimized according to the data set C and the data set D.
  • the embodiment of the present application also discloses a computer program product or computer program.
  • the computer program product or computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium.
  • the processor of the electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device executes the above-mentioned automatic design method of dental restorations based on deep learning.
  • What is disclosed above is only the preferred embodiment of the present application. Of course, it cannot be used to limit the scope of rights of the present application. Therefore, equivalent changes made based on the patent scope of the present application still fall within the scope of the present application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Veterinary Medicine (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Public Health (AREA)
  • Molecular Biology (AREA)
  • Animal Behavior & Ethology (AREA)
  • Epidemiology (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Dentistry (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Geometry (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Dental Tools And Instruments Or Auxiliary Dental Instruments (AREA)
  • Dental Prosthetics (AREA)

Abstract

本申请公开了一种基于深度学习的牙齿修复体自动设计方法及系统，该方法包括：从患者的牙列数据中分割出相应的三维分割数据，并获取对应的目标修复体数据，以形成包括所述三维分割数据和所述目标修复体数据的数据集C；采用所述数据集C训练基于深度学习的数据处理模型，使得该数据处理模型可根据患者的三维分割数据生成与之相适配的修复体模型；根据上述设计方法，可全自动进行牙齿修复体设计，不需要任何人工交互，有效提高设计效率，节约时间和人工，而且使得最终制作的牙齿修复体个性化更强；另外，随着修复体设计量的增加，数据处理模型不断得到优化，设计能力不断得到提高，从而更好地满足了牙科医生和患者的需求。

Description

基于深度学习的牙齿修复体自动设计方法及系统
本申请要求于2022年4月22日提交中国专利局、申请号为2022104363193、发明名称为“基于深度学习的牙齿修复体自动设计方法及系统”的中国专利申请的优先权，其全部内容通过引用结合在申请中。
技术领域
本申请涉及牙齿计算机辅助设计技术领域，尤其涉及一种基于深度学习的牙齿修复体自动设计方法及系统。
背景技术
牙齿缺损后，不但容易影响面部形态的美观，而且咀嚼能力的衰退会影响人体对营养成分的吸收，甚至导致一些牙科的疾病。对于严重牙体缺失的症状，常采用修复体治疗方法进行修复，如嵌体或人造冠等。
传统的牙齿修复体设计方法主要依靠牙科专家的经验和技术，不但修复精度不易控制、耗时较长，而且给患者带来诸多不便。近来，比较流行的计算机辅助设计的方法，以标准牙数据库中的牙冠作为缺损牙齿的初始模型，采用一系列变形算法进行合适的变形操作，重建出目标牙冠模型，这虽然在一定程度上解决了完全靠经验和技术设计修复体的问题，但是这种设计方法缺乏个性化程度，鲁棒性较差，而且仍然需要大量的手工操作。
发明内容
本申请的目的是为解决上述技术问题而提供一种可全自动、免人工进行牙齿修复体的个性化设计的基于深度学习的牙齿修复体自动设计方法及系统。
为了实现上述目的，本申请公开了一种基于深度学习的牙齿修复体自动设计方法，其包括：从患者的牙列数据中分割出相应的三维分割数据，并获取目标修复体数据，以形成包括所述三维分割数据和所述目标修复体数据的数据集 C；
采用所述数据集C训练基于深度学习的数据处理模型，使得该数据处理模型可根据患者的三维分割数据生成与之相适配的修复体模型。
较佳地，所述三维分割数据的获取方法包括：
采用基础数据集训练基于深度学习的数据辅助处理模型，通过该数据辅助处理模型从所输入的牙列数据中分割出相应的的三维分割数据。
较佳地，所述基础数据的收集方法包括：
获取数据集A，所述数据集A包括若干对分别来自于每一患者的相对应的上下颌牙列数据和目标修复体数据；
对所述数据集A中的上下颌牙列数据进行标注，以区分出与每一牙齿相对应的数据，从而形成用作所述基础数据的数据集B。
较佳地，还根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化。
较佳地，将修复体设计的数量和类型作为条件输入所述数据处理模型，以生成相应的修复体模型。
较佳地，对所述数据处理模型的损失函数设置显式约束条件。
较佳地，所述数据辅助处理模型基于卷积神经网络构建，所述数据处理模型基于卷积神经网络或生成对抗网络构建。
较佳地，根据样本数据量由大到小，通过迁移学习依次训练所述数据辅助处理模型和所述数据处理模型。
较佳地，所述数据处理模型基于生成对抗网络构建，所述数据处理模型包括生成模型和判别模型，所述生成模型中设置有一个或多个生成器，每一所述生成器用于学习一种特定牙齿修复模式，所述判别模型包括判别器或分类器，所述分类器用于对修复体模型进行分类，以使得所述生成模型生成相应的修复体模型。
较佳地，还根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化；
根据所述评价结果对所述数据处理模型进行优化的方法包括：
当生成与患者牙列相适配的修复体模型后，将评价结果为合格的结果增加到所述数据集C中，将所述修复体模型作为目标修复体保存，将评价结果为不合格的结果存储到数据集D中，根据所述数据集C和所述数据集D对所述数据处理模型进行优化。
本申请还公开一种基于深度学习的牙齿修复体自动设计系统，其包括：
一个或多个处理器；
存储器；
以及一个或多个程序，其中一个或多个程序被存储在所述存储器中，并且被配置成由所述一个或多个处理器执行，所述程序包括用于执行如上所述的基于深度学习的牙齿修复体自动设计方法的指令。
本申请还公开一种计算机可读存储介质，其包括计算机程序，所述计算机程序可被处理器执行以完成如上所述的基于深度学习的牙齿修复体自动设计方法。
与现有技术相比，本申请牙齿修复体自动设计方法，提供有数据处理模型，在针对患者进行牙齿修复体设计时，只需获取患者的牙列数据即可根数据处理模型输出相应的修复体模型；由此可知，根据上述设计方法，可全自动进行牙齿修复体设计，在设计过程中不需要任何人工交互，不仅解决了需要靠医生的经验和技术设计牙齿修复体的问题，还可有效提高设计效率，节约时间和人工；再者，修复体模型是根据患者的牙列数据生成，从而使得最终制作的牙齿修复体个性化更强；另外，随着修复体设计量的增加，数据处理模型不断得到优化，设计能力不断得到提高，从而更好地满足了牙科医生和患者的需求。
附图说明
图1为本申请其中一实施例的牙齿修复体自动设计方法流程图。
图2为图1中步骤S1的具体流程图。
图3为本申请另一实施例的牙齿修复体自动设计方法流程图。
图4为本申请实施例中基于生成对抗网络的牙齿修复体的设计方法中对数据处理模型进行优化的流程示意图。
图5为本申请实施例中数据集A包含的样本形式示意图。
图6为本申请实施例中数据集B包含的样本形式示意图。
图7为本申请实施例中数据集C包含的样本形式示意图。
图8为本申请实施例中数据集D包含的样本形式示意图。
图9为本申请实施例中包含多个类型的牙齿修复体示意图。
图10为本申请实施例中上下牙间隙距离示意图。
图11为本申请实施例中基于迁移学习思想训练网络模型的示意图。
图12为本申请其中一实施例中数据处理模型生成修复体模型的状态示意图。
图13为本申请另一实施例中数据处理模型生成修复体模型的状态示意图。
图14为本申请实施例中生成器和判别器的简单网络模型示意图。
图15为本申请实施例中生成模型包括多个生成器和判别器的复杂网络模型示意图。
图16为本申请实施例中基于深度学习的牙齿修复体自动设计系统结构图。
具体实施方式
为详细说明本申请的技术内容、构造特征、所实现目的及效果，以下结合实施方式并配合附图详予说明。
本实施例公开了一种基于深度学习的牙齿修复体自动设计方法，以用于牙齿治疗过程中的牙齿修复体的设计，牙齿修复体包括嵌体、内冠、全冠以及连桥冠等，如图1，该设计方法包括如下步骤：
S1：采用基础数据集训练数据辅助处理模型，使得该数据辅助处理模型可从所输入的牙列数据中分割出相应的目标牙齿相对应的三维分割数据，并形成数据集C(如图7)，数据集C包括三维分割数据(与牙齿修复相关的目标牙齿相对应)和与其对应的目标修复体数据。本实施例中的目标修复体数据，为口 腔医生或技师设计并经过测试的牙齿修复体，该目标修复体数据可以是嵌体数据、内冠数据、全冠数据、以及连桥冠数据中的任一种或多种。另外，在本实施例中，与牙齿修复相关的三维分割数据包括预备牙、邻牙、对称牙以及对颌牙。所谓预备牙，即为预备修复的牙齿，邻牙是与预备牙相邻的牙齿，对称牙是在口腔中处于与预备牙左右对称位置的牙齿，对颌牙是与预备牙上下颌相对的牙齿。
S2：采用数据集C训练数据处理模型，使得该数据处理模型可根据三维分割数据生成与之相适配的修复体模型。
S3：将某一患者的牙列数据输入数据辅助处理模型，以得到三维分割数据；
S4：将该三维分割数据输入数据处理模型，以得到与该患者牙列相适配的修复体模型，通过该修复体模型即可制作出与该患者牙列相适配的牙齿修复体。
如图2，在上述步骤S1中，还包括对基础数据的收集方法，其具体如下：
S10：获取数据集A(如图5)，数据集A包括若干对分别来自于每一患者的相对应的上下颌牙列数据和目标修复体数据。该数据集A可由三维点云、三维体素、三维网格模型、多视角图像集、2.5维深度图、八叉树结构等表示。
S11：对数据集A中的上下颌牙列数据进行标注，以区分出与每一牙齿相对应的数据，从而形成用作基础数据的数据集B(如图5)。
S12：采用数据集B训练数据辅助处理模型，使得该数据辅助处理模型可分割出相应的三维数据，以得到上述数据集C。
根据上述设计方法，当需要对某一患者进行牙齿修复体设计时，只需获取该患者的牙列数据，然后通过数据辅助处理模型和数据处理模型自动输出与该患者牙列相适配的修复体模型。另外需要说明的是，本实施例中的数据处理模型既可直接输出修复体模型(如图12)，还可根据输入的三维分割数据生成修复后的牙齿三维模型(如图13)，然后对牙齿三维模型和牙列数据取差，以得到修复体模型。
为进一步提高牙齿修复模型的精准度，如图3，上述设计方法还包括:
S5：根据预设的评价指标对输出的修复体模型的外观和功能等进行评价， 并生成评价结果(合格或不合格)。
S6：当生成与患者牙列相适配的修复体模型后，判断该修复体模型的评价结果是否合格，如果不合格，则将与该修复体模型相对应的三维分割数据再次返回到数据处理模型，以再次获得修复体模型，以此循环，直到获得评价结果合格的修复体模型。本实施中的评价指标包括咬合干涉强度、接触点、牙齿形状弧度、邻牙间隙(如图10)、最小厚度等事项。
S7：根据评价结果对数据处理模型进行优化。
通过上述评价方法，使得通过输出的修复体模型做出的修复体能更好地满足使用要求，并能有效提高数据处理模型输出的修复体模型的精准度。
进一步地，由于牙齿数据集样本少，标注困难，修复体模型包含种类多(嵌体、内冠、全冠、连桥冠)，而且，公开的大规模三维模型数据集多，比如ShapeNet数据集、Completion3D数据集、KITTI数据集。因此，如图11，通过迁移学习根据可获取的数据量由大到小依次训练网络模型，本实施例中的网络模型包括数据辅助处理模型和数据处理模型。本实施例中的迁移学习是将某个领域或任务上学习到的知识或模式应用到不同但相关的领域或问题中，可将其分为基于实例的迁移学习、基于特征的迁移学习和基于共享参数的迁移学习，关于迁移学习的具体过程属于本领域的公知常识，在此不再赘述。
进一步地，可根据当前患者的具体情况，将修复体设计的数量和类型作为条件输入数据处理模型，以生成相应的修复体模型。例如，在数据处理模型中输入牙冠一个、嵌体一个以及桥冠一个，那么数据处理模型即可根据输入的与该患者牙列环境相关的三维分割数据输出三个修复体模型，分别为牙冠、嵌体和桥冠，如图9，从而提高设计效率和质量。
为便于修复体的制作，进一步地，将数据处理模型输出的与患者牙列相适配的修复体模型转换成网格模型，网格模型用于制造牙齿修复体。本实施例中的网格模型为采用一系列大小和形状接近的多边形(通常是三角形)近似表示三维物体的模型，为一种三维模型。
再者，为进一步提高数据处理模型输出的精准度，还可对数据处理模型的 损失函数设置显式约束条件，例如该显式约束条件包括对颌牙间隙的统计特征、生成修复体与真实修复体的距离以及邻牙间隙的统计特征等。其中，间隙特征可用直方图进行统计，使用卡方距离、巴氏距离等相关公式进行两个直方图之间的距离度量。根据口腔技师建议，上磨牙和下磨牙的咬合干涉强度为0.02～0.05mm，全冠牙体预备提供的合面间隙一般为0.5～1mm。
进一步地，上述实施例中的数据辅助处理模型基于卷积神经网络构建，数据处理模型基于卷积神经网络或生成对抗网络构建。
基于对抗网络的网络模型包括简单网络模型和复杂网络模型，如图14，简单网络模型包括单个生成器和判别器。本实施例中，数据处理模型基于生成对抗网络构建，如图15，数据处理模型包括生成模型和判别模型，生成模型中设置有一个或多个生成器，每一生成器用于学习一种特定牙齿修复模式(如牙尖、嵴、窝沟等)，判别模型包括判别器或分类器，分类器用于对修复体模型进行分类，以使得生成模型生成相应的修复体模型。本实施例中，生成模型中的每个生成器专注于学习特定的牙齿修复模式，使得生成模型最终得到的修复体模型来自多个学习到了不同模式的生成器，以缓解模式崩溃的问题。另外，判别模型不仅可包含判别器，还包含分类器，这样判别模型不仅承担判别真假修复体的任务，也可以对真实的修复体进行分类，以提升网络的收敛速度。比如，分类器按照形态，将牙齿分为切牙、尖牙、前磨牙、磨牙，以更好地指导生成模型生成牙齿修复体。
在基于生成对抗网络模型构建数据处理模型时，根据评价结果对数据处理模型进行优化的方法包括：
将评价结果为合格的修复体模型所对应的三维分割数据增加到数据集C中，并将该修复体模型作为目标修复体体数据保存，并将评价结果为不合格的修复体模型所对应的三维分割数据保存到数据集D(如图8)中，根据数据集C和数据D重新优化生成数据处理模型的参数。
具体地，在基于生成对抗网络模型构建数据处理模型中，如图4，对数据处理模型进行优化的过程包括如下步骤：
步骤S60：生成模型G通过数据集C输出修复体模型G(z)；
步骤S61：通过修复体模型G(z)及其对应的目标修复体x，再加上数据集D中的修复体模型训练判别模型，使其能够分辨修复体模型G(z)和目标修复体x，并将数据集D中的修复体模型判别为生成修复体；
步骤S62：基于该判别模型，使用数据集C重新训练生成模型G；
步骤S63：不断重复步骤S60至步骤S62。
理想状态下，最终判别器无法区分数据集C产生的修复体模型G(z)和目标修复体x，但能更好地将评价不合格的修复体模型判别为生成修复体。
更具体地，采用生成对抗网络构建数据处理模型的损失函数如下所示，
Figure PCTCN2022142661-appb-000001
其中，x表示目标修复体，z表示随机噪声，y表示指导生成对抗网络的条件数据。
综上，如图1至图15，上述实施例公开了一种牙齿修复体自动设计方法，通过采集牙病患者的牙齿修复数据训练形成数据辅助处理模型和数据处理模型，并采用评价方法对数据处理模型输出的修复体模型进行评价，且根据评价结果对数据处理模型进行优化，从而得到精准度比较高的牙齿修复体自动设计模型。当需要针对某一牙病患者进行牙齿修复体的设计时，只需采集该患者的牙列数据，然后经过数据辅助处理模型和数据处理模型的自动处理，从数据处理模型输出与该患者牙列相适配的修复体模型，接着，将该修复体模型转换成网格模型，通过该网格模型即可采用增材制造或其他制造方式根据病人或医生的要求(如制造材料)将评价合格的修复体网格模型制造出来。医生将其粘合到患者的预备牙上。
根据上述设计方法，可全自动进行牙齿修复体设计，在设计过程中不需要任何人工交互，不仅解决了需要靠医生的经验和技术设计牙齿修复体的问题，还可有效提高设计效率，节约时间和人工。再者，修复体模型是根据患者的牙列数据生成，从而使得最终制作的牙齿修复体个性化更强。另外，随着修复体 设计量的增加，数据辅助处理模型和数据处理模型不断得到优化，设计能力不断得到提高，从而更好地满足了牙科医生和患者的需求。
发明还公开一种基于深度学习的牙齿修复体自动设计系统，如图16，其包括中央处理器100、牙齿数据库101、数据获取模块102、自动设计模块103以及用户交互界面104。医生或技师可以通过用户交互界面修改设计结果，并可视化设计结果。
牙齿数据库101，用于存储病人牙列相关数据。用户交互界面104，用于方便用户输入数据，并获取结果数据。
数据获取模块102，用于从患者的牙列数据中分割出相应的三维分割数据，并获取相应的目标修复体数据，以形成包括所述三维分割数据和所述目标修复体数据的数据集C，所述数据集C存储于所述牙齿数据库中。
所述自动设计模块103，用于采用所述数据集C训练基于深度学习的数据处理模型，使得该数据处理模型可根据患者的三维分割数据生成与之相适配的修复体模型。
进一步地，上述系统还包括数据辅助处理模块105，所述数据辅助处理模块105用于采用基础数据集训练基于深度学习的数据辅助处理模型，该数据辅助处理模型可从所输入的牙列数据中分割出相应的的三维分割数据。
在本实施例中，所述数据辅助处理模型基于卷积神经网络构建，所述数据处理模型基于卷积神经网络或生成对抗网络构建。
进一步地，上述系统还包括基础数据生成模块106，所述基础数据生成模块106用于对数据集A中的上下颌牙列数据进行标注，以区分出与每一牙齿相对应的数据，从而形成用作基础数据的数据集B，所述数据集A包括若干对分别来自于每一患者的相对应的上下颌牙列数据和目标修复体数据。
进一步地，上述系统还包括评价模块107，所述评价模块107用于根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化。
进一步地，上述系统还包括数据转换模块108，用于数据转换模块108用于 将所述数据处理模型输出的与患者牙列相适配的修复体模型转换成网格模型，所述网格模型用于制造牙齿修复体。
进一步地，上述系统还包括迁移学习模块109，所述迁移学习模块109，用于根据样本数据量由大到小，通过迁移学习依次训练所述数据辅助处理模型和所述数据处理模型。
另外需要说明的是，本实施例的基于深度学习的牙齿修复体自动设计系统的工作原理和工作过程详见上述基于深度学习的牙齿修复体自动设计方法，在此不再赘述。
本申请还公开另一种基于深度学习的牙齿修复体自动设计系统，其包括一个或多个处理器、存储器以及一个或多个程序，其中一个或多个程序被存储在所述存储器中，并且被配置成由所述一个或多个处理器执行，所述程序包括用于执行如上所述的基于深度学习的牙齿修复体自动设计方法的指令。处理器可以采用通用的中央处理器(Central Processing Unit，CPU)，微处理器，应用专用集成电路(Application Specific Integrated Circuit，ASIC)，或者一个或多个集成电路，用于执行相关程序，以实现本申请实施例的基于深度学习的牙齿修复体自动设计系统中的模块所需执行的功能，或者执行本申请方法实施例的基于深度学习的牙齿修复体自动设计方法。
可选的，在一具体实施例中，所述程序包括用于执行如下所述的基于深度学习的牙齿修复体自动设计方法的指令包括：
从患者的牙列数据中分割出相应的三维分割数据，并获取目标修复体数据，以形成包括所述三维分割数据和所述目标修复体数据的数据集C；
采用所述数据集C训练基于深度学习的数据处理模型，使得该数据处理模型可根据患者的三维分割数据生成与之相适配的修复体模型。
可选的，在一具体实施例中，所述程序执行所述基于深度学习的牙齿修复体自动设计方法的指令还包括：根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化。
可选的，在一具体实施例中，所述数据辅助处理模型基于卷积神经网络构 建，所述数据处理模型基于卷积神经网络或生成对抗网络构建。
可选的，在一具体实施例中，所述数据处理模型基于生成对抗网络构建，所述数据处理模型包括生成模型和判别模型，所述生成模型中设置有一个或多个生成器，每一所述生成器用于学习一种特定牙齿修复模式，所述判别模型包括判别器或分类器，所述分类器用于对修复体模型进行分类，以使得所述生成模型生成相应的修复体模型。
可选的，在一具体实施例中，所述程序执行所述基于深度学习的牙齿修复体自动设计方法的指令还包括：根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化；
根据所述评价结果对所述数据处理模型进行优化的方法包括：
当生成与患者牙列相适配的修复体模型后，将评价结果为合格的结果增加到所述数据集C中，将所述修复体模型作为目标修复体保存，将评价结果为不合格的结果存储到数据集D中，根据所述数据集C和所述数据集D对所述数据处理模型进行优化。
本申请还公开一种计算机可读存储介质，其包括计算机程序，所述计算机程序可被处理器执行以完成如上所述的基于深度学习的牙齿修复体自动设计方法。该计算机可读存储介质可以是计算机能够存取的任何可用介质或者是包含一个或多个可用介质集成的服务器、数据中心等数据存储设备。该可用介质可以是只读存储器(read-onlymemory，ROM)，或随机存取存储器(random access memory，RAM)，或磁性介质，例如，软盘、硬盘、磁带、磁碟、或光介质，例如，数字通用光盘(digital versatile disc，DVD)、或者半导体介质，例如，固态硬盘(solid state disk，SSD)等。
可选的，在一具体实施例中，所述计算机程序可被处理器执行以完成如下所述的基于深度学习的牙齿修复体自动设计方法包括：
从患者的牙列数据中分割出相应的三维分割数据，并获取目标修复体数据，以形成包括所述三维分割数据和所述目标修复体数据的数据集C；
采用所述数据集C训练基于深度学习的数据处理模型，使得该数据处理模 型可根据患者的三维分割数据生成与之相适配的修复体模型。
可选的，在一具体实施例中，所述计算机程序可被处理器执行完成所述基于深度学习的牙齿修复体自动设计方法还包括：根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化。
可选的，在一具体实施例中，所述数据辅助处理模型基于卷积神经网络构建，所述数据处理模型基于卷积神经网络或生成对抗网络构建。
可选的，在一具体实施例中，所述数据处理模型基于生成对抗网络构建，所述数据处理模型包括生成模型和判别模型，所述生成模型中设置有一个或多个生成器，每一所述生成器用于学习一种特定牙齿修复模式，所述判别模型包括判别器或分类器，所述分类器用于对修复体模型进行分类，以使得所述生成模型生成相应的修复体模型。
可选的，在一具体实施例中，所述计算机程序可被处理器执行完成所述基于深度学习的牙齿修复体自动设计方法还包括：根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化；
根据所述评价结果对所述数据处理模型进行优化的方法包括：
当生成与患者牙列相适配的修复体模型后，将评价结果为合格的结果增加到所述数据集C中，将所述修复体模型作为目标修复体保存，将评价结果为不合格的结果存储到数据集D中，根据所述数据集C和所述数据集D对所述数据处理模型进行优化。
本申请实施例还公开了一种计算机程序产品或计算机程序，该计算机程序产品或计算机程序包括计算机指令，该计算机指令存储在计算机可读存储介质中。电子设备的处理器从计算机可读存储介质读取该计算机指令，处理器执行该计算机指令，使得该电子设备执行上述基于深度学习的牙齿修复体自动设计方法。以上所揭露的仅为本申请的优选实施例而已，当然不能以此来限定本申请之权利范围，因此依本申请申请专利范围所作的等同变化，仍属本申请所涵 盖的范围。

Claims (20)

  1. 一种基于深度学习的牙齿修复体自动设计方法，包括：
    从患者的牙列数据中分割出相应的三维分割数据，并获取目标修复体数据，以形成包括所述三维分割数据和所述目标修复体数据的数据集C；
    采用所述数据集C训练基于深度学习的数据处理模型，使得该数据处理模型可根据患者的三维分割数据生成与之相适配的修复体模型。
  2. 根据权利要求1所述的基于深度学习的牙齿修复体自动设计方法，所述三维分割数据的获取方法包括：
    采用基础数据集训练基于深度学习的数据辅助处理模型，通过该数据辅助处理模型从所输入的牙列数据中分割出相应的三维分割数据。
  3. 根据权利要求2所述的基于深度学习的牙齿修复体自动设计方法，所述基础数据的收集方法包括：
    获取数据集A，所述数据集A包括若干对分别来自于每一患者的相对应的上下颌牙列数据和目标修复体数据；
    对所述数据集A中的上下颌牙列数据进行标注，以区分出与每一牙齿相对应的数据，从而形成用作所述基础数据的数据集B。
  4. 根据权利要求1所述的基于深度学习的牙齿修复体自动设计方法，还根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化。
  5. 根据权利要求1所述的基于深度学习的牙齿修复体自动设计方法，将修复体设计的数量和类型作为条件输入所述数据处理模型，以生成相应的修复体模型。
  6. 根据权利要求1所述的基于深度学习的牙齿修复体自动设计方法，对所述数据处理模型的损失函数设置显式约束条件。
  7. 根据权利要求1所述的基于深度学习的牙齿修复体自动设计方法，所述数据辅助处理模型基于卷积神经网络构建，所述数据处理模型基于卷积神经网络或生成对抗网络构建。
  8. 根据权利要求2所述的基于深度学习的牙齿修复体自动设计方法，根据样本数据量由大到小，通过迁移学习依次训练所述数据辅助处理模型和所述数据处理模型。
  9. 根据权利要求7所述的基于深度学习的牙齿修复体自动设计方法，所述数据处理模型基于生成对抗网络构建，所述数据处理模型包括生成模型和判别模型，所述生成模型中设置有一个或多个生成器，每一所述生成器用于学习一种特定牙齿修复模式，所述判别模型包括判别器或分类器，所述分类器用于对修复体模型进行分类，以使得所述生成模型生成相应的修复体模型。
  10. 根据权利要求9所述的基于深度学习的牙齿修复体自动设计方法，还根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化；
    根据所述评价结果对所述数据处理模型进行优化的方法包括：
    当生成与患者牙列相适配的修复体模型后，将评价结果为合格的结果增加到所述数据集C中，将所述修复体模型作为目标修复体保存，将评价结果为不合格的结果存储到数据集D中，根据所述数据集C和所述数据集D对所述数据处理模型进行优化。
  11. 一种基于深度学习的牙齿修复体自动设计系统，包括：
    一个或多个处理器；
    存储器；
    以及一个或多个程序，其中一个或多个程序被存储在所述存储器中，并且被配置成由所述一个或多个处理器执行，所述程序包括用于执行如下所述的基于深度学习的牙齿修复体自动设计方法的指令包括：
    从患者的牙列数据中分割出相应的三维分割数据，并获取目标修复体数据，以形成包括所述三维分割数据和所述目标修复体数据的数据集C；
    采用所述数据集C训练基于深度学习的数据处理模型，使得该数据处理模型可根据患者的三维分割数据生成与之相适配的修复体模型。
  12. 根据权利要求11所述的基于深度学习的牙齿修复体自动设计系统，所述程序执行所述基于深度学习的牙齿修复体自动设计方法的指令还包括：根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化。
  13. 根据权利要求11所述的基于深度学习的牙齿修复体自动设计系统，所述数据辅助处理模型基于卷积神经网络构建，所述数据处理模型基于卷积神经网络或生成对抗网络构建。
  14. 根据权利要求13所述的基于深度学习的牙齿修复体自动设计系统，所述数据处理模型基于生成对抗网络构建，所述数据处理模型包括生成模型和判别模型，所述生成模型中设置有一个或多个生成器，每一所述生成器用于学习一种特定牙齿修复模式，所述判别模型包括判别器或分类器，所述分类器用于对修复体模型进行分类，以使得所述生成模型生成相应的修复体模型。
  15. 根据权利要求14所述的基于深度学习的牙齿修复体自动设计系统，所述 程序执行所述基于深度学习的牙齿修复体自动设计方法的指令还包括：根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化；
    根据所述评价结果对所述数据处理模型进行优化的方法包括：
    当生成与患者牙列相适配的修复体模型后，将评价结果为合格的结果增加到所述数据集C中，将所述修复体模型作为目标修复体保存，将评价结果为不合格的结果存储到数据集D中，根据所述数据集C和所述数据集D对所述数据处理模型进行优化。
  16. 一种计算机可读存储介质，包括计算机程序，所述计算机程序可被处理器执行以完成如下所述的基于深度学习的牙齿修复体自动设计方法包括：
    从患者的牙列数据中分割出相应的三维分割数据，并获取目标修复体数据，以形成包括所述三维分割数据和所述目标修复体数据的数据集C；
    采用所述数据集C训练基于深度学习的数据处理模型，使得该数据处理模型可根据患者的三维分割数据生成与之相适配的修复体模型。
  17. 根据权利要求16所述的计算机可读存储介质，所述计算机程序可被处理器执行完成所述基于深度学习的牙齿修复体自动设计方法还包括：根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化。
  18. 根据权利要求16所述的计算机可读存储介质，所述数据辅助处理模型基于卷积神经网络构建，所述数据处理模型基于卷积神经网络或生成对抗网络构建。
  19. 根据权利要求18所述的计算机可读存储介质，所述数据处理模型基于生成对抗网络构建，所述数据处理模型包括生成模型和判别模型，所述生成模型 中设置有一个或多个生成器，每一所述生成器用于学习一种特定牙齿修复模式，所述判别模型包括判别器或分类器，所述分类器用于对修复体模型进行分类，以使得所述生成模型生成相应的修复体模型。
  20. 根据权利要求19所述的计算机可读存储介质，所述计算机程序可被处理器执行完成所述基于深度学习的牙齿修复体自动设计方法还包括：根据预设的评价指标对输出的所述修复体模型进行评价，并生成评价结果，根据所述评价结果对所述数据处理模型进行优化；
    根据所述评价结果对所述数据处理模型进行优化的方法包括：
    当生成与患者牙列相适配的修复体模型后，将评价结果为合格的结果增加到所述数据集C中，将所述修复体模型作为目标修复体保存，将评价结果为不合格的结果存储到数据集D中，根据所述数据集C和所述数据集D对所述数据处理模型进行优化。
PCT/CN2022/142661 2022-04-22 2022-12-28 基于深度学习的牙齿修复体自动设计方法及系统 WO2023202143A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210436319.3 2022-04-22
CN202210436319.3A CN114880924A (zh) 2022-04-22 2022-04-22 基于深度学习的牙齿修复体自动设计方法及系统

Publications (1)

Publication Number Publication Date
WO2023202143A1 true WO2023202143A1 (zh) 2023-10-26

Family

ID=82672580

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/142661 WO2023202143A1 (zh) 2022-04-22 2022-12-28 基于深度学习的牙齿修复体自动设计方法及系统

Country Status (2)

Country Link
CN (1) CN114880924A (zh)
WO (1) WO2023202143A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114880924A (zh) * 2022-04-22 2022-08-09 东莞中科云计算研究院 基于深度学习的牙齿修复体自动设计方法及系统
CN115578518A (zh) * 2022-11-01 2023-01-06 高峰医疗器械(无锡)有限公司 冠桥模型的构建方法、装置、设备及存储介质
CN118078471B (zh) * 2024-04-29 2024-07-02 南京笑领科技有限公司 基于人工智能的三维牙冠建模方法、系统及应用

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180028294A1 (en) * 2016-07-27 2018-02-01 James R. Glidewell Dental Ceramics, Inc. Dental cad automation using deep learning
US20190282344A1 (en) * 2018-03-19 2019-09-19 James R. Glidewell Dental Ceramics, Inc. Dental cad automation using deep learning
US20210153986A1 (en) * 2019-11-25 2021-05-27 Dentsply Sirona Inc. Method, system and computer readable storage media for creating three-dimensional dental restorations from two dimensional sketches
CN113888615A (zh) * 2021-10-21 2022-01-04 先临三维科技股份有限公司 修复体图像生成方法、装置、设备及存储介质
WO2022016294A1 (en) * 2020-07-24 2022-01-27 Intellident Dentaire Inc. Automatic generation of dental restorations using machine learning
CN114342002A (zh) * 2019-09-05 2022-04-12 登士柏希罗纳有限公司 用于定制牙科对象的即时自动化设计的方法、系统和设备
CN114880924A (zh) * 2022-04-22 2022-08-09 东莞中科云计算研究院 基于深度学习的牙齿修复体自动设计方法及系统

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180028294A1 (en) * 2016-07-27 2018-02-01 James R. Glidewell Dental Ceramics, Inc. Dental cad automation using deep learning
US20190282344A1 (en) * 2018-03-19 2019-09-19 James R. Glidewell Dental Ceramics, Inc. Dental cad automation using deep learning
CN114342002A (zh) * 2019-09-05 2022-04-12 登士柏希罗纳有限公司 用于定制牙科对象的即时自动化设计的方法、系统和设备
US20210153986A1 (en) * 2019-11-25 2021-05-27 Dentsply Sirona Inc. Method, system and computer readable storage media for creating three-dimensional dental restorations from two dimensional sketches
WO2022016294A1 (en) * 2020-07-24 2022-01-27 Intellident Dentaire Inc. Automatic generation of dental restorations using machine learning
CN113888615A (zh) * 2021-10-21 2022-01-04 先临三维科技股份有限公司 修复体图像生成方法、装置、设备及存储介质
CN114880924A (zh) * 2022-04-22 2022-08-09 东莞中科云计算研究院 基于深度学习的牙齿修复体自动设计方法及系统

Also Published As

Publication number Publication date
CN114880924A (zh) 2022-08-09

Similar Documents

Publication Publication Date Title
WO2023202143A1 (zh) 基于深度学习的牙齿修复体自动设计方法及系统
Tian et al. Automatic classification and segmentation of teeth on 3D dental model using hierarchical deep learning networks
US11957541B2 (en) Machine learning scoring system and methods for tooth position assessment
US11651494B2 (en) Apparatuses and methods for three-dimensional dental segmentation using dental image data
Tian et al. DCPR-GAN: dental crown prosthesis restoration using two-stage generative adversarial networks
EP3638146A1 (en) Automatic detection of tooth type and eruption status
Yuan et al. Personalized design technique for the dental occlusal surface based on conditional generative adversarial networks
CN107239649B (zh) 一种口腔参数化测量的方法
CN106687068A (zh) 基于数据挖掘的全口义齿制作方法和装置
US11450426B2 (en) Tooth virtual editing method and system, computer device, and storage medium
Tian et al. Efficient computer-aided design of dental inlay restoration: a deep adversarial framework
CN108735292A (zh) 基于人工智能的可摘局部义齿方案决策方法和系统
CN111709959B (zh) 一种口腔正畸数字化智能诊断方法
CN113679500B (zh) 一种基于ai算法的龋病和牙菌斑检测其分布方法
CN112037913A (zh) 一种基于卷积神经网络的牙周炎智能检测方法及系统
CN112790879B (zh) 一种牙齿模型的牙轴坐标系构建方法及系统
WO2021218724A1 (zh) 一种用于口腔数字印模仪的数字模型智能设计方法
Tian et al. A dual discriminator adversarial learning approach for dental occlusal surface reconstruction
Choi et al. Possibilities of artificial intelligence use in orthodontic diagnosis and treatment planning: Image recognition and three-dimensional VTO
Tian et al. Efficient tooth gingival margin line reconstruction via adversarial learning
TW202409874A (zh) 牙齒復原自動化技術
CN116958169A (zh) 一种三维牙颌模型牙齿分割方法
CN116306283A (zh) 一种功能微压口腔黏膜形态预测方法及系统
CN112420171B (zh) 一种基于人工智能的上颌窦底骨质分类方法及系统
CN114529553A (zh) 一种自动牙颌数字模型分割算法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22938361

Country of ref document: EP

Kind code of ref document: A1