WO2021253917A1 - 产生表示牙齿正畸治疗的目标牙齿布局的数字数据集的方法 - Google Patents

产生表示牙齿正畸治疗的目标牙齿布局的数字数据集的方法 Download PDF

Info

Publication number
WO2021253917A1
WO2021253917A1 PCT/CN2021/083968 CN2021083968W WO2021253917A1 WO 2021253917 A1 WO2021253917 A1 WO 2021253917A1 CN 2021083968 W CN2021083968 W CN 2021083968W WO 2021253917 A1 WO2021253917 A1 WO 2021253917A1
Authority
WO
WIPO (PCT)
Prior art keywords
tooth
data set
layout
digital data
orthodontic treatment
Prior art date
Application number
PCT/CN2021/083968
Other languages
English (en)
French (fr)
Inventor
沈恺迪
Original Assignee
杭州朝厚信息科技有限公司
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 杭州朝厚信息科技有限公司 filed Critical 杭州朝厚信息科技有限公司
Priority to US18/011,511 priority Critical patent/US20230334771A1/en
Publication of WO2021253917A1 publication Critical patent/WO2021253917A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61CDENTISTRY; APPARATUS OR METHODS FOR ORAL OR DENTAL HYGIENE
    • A61C13/00Dental prostheses; Making same
    • A61C13/34Making or working of models, e.g. preliminary castings, trial dentures; Dowel pins [4]
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61CDENTISTRY; APPARATUS OR METHODS FOR ORAL OR DENTAL HYGIENE
    • A61C7/00Orthodontics, i.e. obtaining or maintaining the desired position of teeth, e.g. by straightening, evening, regulating, separating, or by correcting malocclusions
    • A61C7/002Orthodontic computer assisted systems
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00Three dimensional [3D] modelling, e.g. data description of 3D objects
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T19/00Manipulating 3D models or images for computer graphics
    • G06T19/20Editing of 3D images, e.g. changing shapes or colours, aligning objects or positioning parts
    • AHUMAN NECESSITIES
    • A61MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61CDENTISTRY; APPARATUS OR METHODS FOR ORAL OR DENTAL HYGIENE
    • A61C7/00Orthodontics, i.e. obtaining or maintaining the desired position of teeth, e.g. by straightening, evening, regulating, separating, or by correcting malocclusions
    • A61C7/002Orthodontic computer assisted systems
    • A61C2007/004Automatic construction of a set of axes for a tooth or a plurality of teeth
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2210/00Indexing scheme for image generation or computer graphics
    • G06T2210/41Medical
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2219/00Indexing scheme for manipulating 3D models or images for computer graphics
    • G06T2219/20Indexing scheme for editing of 3D models
    • G06T2219/2021Shape modification

Definitions

  • the present application generally relates to a method of generating a digital data set representing a target tooth layout for orthodontic treatment.
  • a set of shell-shaped appliances usually includes a dozen or even dozens of successive shell-shaped appliances, which are used to reposition the patient’s teeth from the initial layout to the target layout one by one. N successive intermediate layouts from the first intermediate layout to the last intermediate layout.
  • a commonly used method of manufacturing shell-shaped appliances is to obtain them by hot pressing film forming process on a series of dental molds successively from the first intermediate layout to the target layout. These dental molds can be made using a series of successive three-dimensional digital model control equipment representing the first intermediate layout to the target layout.
  • a commonly used method to obtain these successive three-dimensional digital models is to first scan to obtain a three-dimensional digital model representing the patient's original tooth layout (that is, the patient's tooth layout before orthodontic treatment), and then manually manipulate the three-dimensional digital model to obtain the target A three-dimensional digital model of the tooth layout (that is, the tooth layout that the orthodontic treatment hopes to achieve), and then interpolate based on these two three-dimensional digital models to obtain a series of successive three-dimensional digital models in the middle.
  • a computer-implemented method for generating a target tooth layout for orthodontic treatment is provided.
  • One aspect of the present application provides a method for generating a digital data set representing a target tooth layout for orthodontic treatment, including: acquiring a first three-dimensional digital model that represents a jaw in the initial tooth layout; and comparing the first three-dimensional digital data set Each tooth in the model is sampled to obtain a corresponding sampling point set; the first deep artificial neural network trained is used to generate a corresponding geometric code based on the sampling point set of each tooth; the geometric codes of all teeth are combined Obtain the overall geometric code of the dental jaw; and use the trained second deep artificial neural network to generate a digital data set representing the target tooth layout of the orthodontic treatment based on the overall geometric code of the dental jaw.
  • the first deep artificial neural network is a deep artificial neural network capable of processing point clouds.
  • the first deep artificial neural network may be one of the following: PointNet, PointNet++, PointCNN, and DGCNN networks.
  • the second deep artificial neural network may be a regression network based on a multilayer perceptron.
  • the second deep artificial neural network may include an SE module for adjusting the weight of each channel of the current feature according to global information.
  • the dental jaw may include an upper jaw and a lower jaw.
  • the geometric code of each tooth is an M-dimensional vector, where the M'-dimensional vector is selected by the first deep artificial neural network during the training process, where M and M'are both Natural numbers, M>M'.
  • the M-dimensional vector further includes a pose representing the corresponding tooth.
  • the M'-dimensional vector represents the geometric shape of the corresponding tooth.
  • the feature extracted for each sampling point of the tooth includes the positional relationship between the sampling point and the adjacent tooth.
  • the feature of the positional relationship with the adjacent tooth may be the shortest distance from the adjacent tooth.
  • the method for generating a digital data set representing a target tooth layout for orthodontic treatment may further include: performing at least one iteration based on the digital data set representing the target tooth layout to obtain a new target tooth. Layout of the digital data set.
  • Fig. 1 is a schematic flowchart of a computer-executed method for generating a digital data set representing a target tooth layout for orthodontic treatment in an embodiment of the application;
  • Fig. 2 schematically shows the structure of the second deep learning artificial neural network in an embodiment of the present application.
  • One aspect of the present application provides a computer-implemented method for generating a digital data set representing a target tooth layout for orthodontic treatment.
  • Orthodontic treatment is the process of repositioning teeth from the original layout to the target layout.
  • the target layout is the desired tooth layout for orthodontic treatment;
  • the original layout may be the patient's tooth layout before orthodontic treatment, or the patient's current tooth layout based on the method of the present application to generate the target layout.
  • FIG. 1 is a schematic flowchart of a method 100 for generating a target tooth layout for orthodontic treatment executed by a computer in an embodiment of the application.
  • the method for generating a target tooth layout for orthodontic treatment of the present application may only generate a target tooth layout for a single jaw (for example, the upper jaw or the lower jaw); in another embodiment, the method for generating the teeth of the present application
  • the method of the target tooth layout of the orthodontic treatment can also process the upper jaw and the lower jaw as a whole, and at the same time generate the target tooth layout of the upper jaw and the lower jaw. In the following, the latter is taken as an example to describe in detail the method of generating the target tooth layout of the orthodontic treatment of the present application.
  • a first three-dimensional digital model and a second three-dimensional digital model are obtained, which respectively represent the upper and lower teeth of the patient in the original layout.
  • the patient's jaw can be scanned directly to obtain a three-dimensional digital model representing the teeth in the original layout.
  • a solid model of the patient's jaw such as a plaster model, can be scanned to obtain a three-dimensional digital model representing the teeth in the original layout.
  • a three-dimensional digital model representing the teeth in the original layout can be obtained by scanning the bite mold of the patient's jaw.
  • the three-dimensional digital model representing the teeth in the original layout after obtaining the three-dimensional digital model representing the teeth in the original layout, it can be segmented so that the teeth in the three-dimensional digital model are independent of each other, so that each of the three-dimensional digital models can be moved independently. Teeth.
  • the first three-dimensional digital model and the second three-dimensional digital model can be expressed in combination with a global coordinate system and a local coordinate system, that is, each tooth has its own local coordinate system, and its local coordinate system
  • the pose of the coordinate system in the global coordinate system represents the pose of the tooth.
  • each tooth in the first three-dimensional digital model and the second three-dimensional digital model is sampled to obtain a corresponding set of sampling points.
  • the three-dimensional digital model of each tooth can be uniformly sampled.
  • the farthest point sampling can be used as the sampling strategy. Under the enlightenment of this application, it can be understood that the sampling method is not limited to those listed above, and any other applicable sampling method can be adopted.
  • 1024 points can be sampled from its vertices.
  • the number of sampling points is not limited to 1024, as long as the number of sampling points is large enough to retain the geometric characteristics of the teeth.
  • the trained first deep learning artificial neural network is used to perform self-encoding based on the sampling point set of each tooth to obtain a corresponding geometric code.
  • the first deep learning artificial neural network may be any deep learning artificial neural network capable of processing point clouds, for example, PointNet, PointNet++, PointCNN, and DGCNN networks.
  • PointNet PointNet
  • PointNet++ PointNet++
  • PointCNN PointCNN
  • DGCNN DGCNN networks
  • the trained PointNet network is used to self-encode the sampling point set of each tooth to obtain a code c about the three-dimensional geometric form of the corresponding tooth, where c is an n-dimensional vector.
  • the value of n can be determined according to the experimental results and the computing power of the computing system. For example, the value of n can be in the range of 100-300, for example, n can be set to 100.
  • each tooth its code c can be combined with its position p and posture q information to obtain the geometric code (c, p, q) of the tooth.
  • the position information p may be a three-dimensional vector, representing a displacement of the tooth relative to the world coordinate system
  • the posture information q may be a rotation amount (for example, a quaternion, Euler angle, rotation matrix or rotation vector ), represents a rotation of the tooth relative to the world coordinate system.
  • p and q may be the position and posture of the local coordinate system of the corresponding tooth in the global coordinate system.
  • the overall geometric code of the tooth jaw is the overall geometric code of the single tooth jaw.
  • the overall geometric coding of the dental jaw is the overall geometric coding of the upper and lower jaws.
  • the overall geometric code of the dental jaw can be a two-dimensional matrix of N*(n+7), where N represents the total number of teeth (for the plan that takes the upper and lower jaws as a whole, N is usually 28), n is the geometrically coded dimension of each tooth.
  • N represents the total number of teeth (for the plan that takes the upper and lower jaws as a whole, N is usually 28), n is the geometrically coded dimension of each tooth.
  • the dimension of p may be 3, and the dimension of q may be 4.
  • the trained second deep learning artificial neural network is used to generate a digital data set representing the target tooth layout based on the overall geometric coding of the dental jaw.
  • FIG. 2 schematically shows the second deep learning artificial neural network 200 in an embodiment of the present application.
  • the second deep learning artificial neural network 200 can be regarded as a regression network based on a multilayer perceptron, that is, a regression problem is solved based on a multilayer perceptron.
  • the second deep learning artificial neural network 200 includes an input module 201, a parameter-sharing fully connected layer module 203, a jump connection module 205, a parameter-sharing fully connected layer module 207 and 209, and an SE network module 211 (Squeeze-and-Excitation Network) , SE output module 213, vector flattening module 215, multilayer perceptron module 217, vector reconstruction module 219, and output module 221.
  • SE network module 211 Seeze-and-Excitation Network
  • the input module 201 receives the output of the first deep learning artificial neural network, for example, a two-dimensional matrix representing the overall geometric coding of the dental jaw.
  • the SE network module 211 is used to adjust the weight of each channel of the current feature according to the global information, so as to highlight useful features and suppress less useful features.
  • the implementation of the SE network module 211 can refer to "Squeeze-and-Excitation Networks" published by JieHu, LiShen, and GangSun.
  • the output of the jump connection module 205 and the output of the SE network module 211 are combined and then input to the vector flattening module 215 for vector flattening operation to convert the data into a one-dimensional vector, so that the multi-layer perceptron connected to the vector flattening module 215
  • the module 217 can receive and process these data.
  • the multilayer perceptron module 217 may include several fully connected layers.
  • the vector reconstruction module 219 reconstructs the one-dimensional vector output by the multilayer perceptron module 217 into a matrix form, which is output by the output module 221.
  • the second deep learning artificial neural network 200 may output a digital data set representing the pose of each tooth under the target tooth layout, for example, the coordinates and angle of each tooth under the target tooth layout.
  • the second deep learning artificial neural network 200 may also output the spatial transformation matrix of each tooth, which is used to move each tooth under the current tooth layout to the position and posture under the target tooth layout.
  • the first deep learning artificial neural network and the second deep learning artificial neural network may be trained as a whole. For each tooth, what feature is the n-dimensional feature in its geometric encoding? This is determined by the first deep learning artificial neural network during training, so that the n-dimensional feature can capture the geometric features of the tooth as accurately as possible The impact on tooth arrangement, which guides the network to achieve more accurate tooth arrangement.
  • the first deep learning artificial neural network and the second deep learning artificial neural network may also be separately trained.
  • the first deep learning artificial neural network may be a PointNet network, which may be trained using a sampling point set (as input) and a reconstruction point set (as output) to sample the point set and reconstruct The chamfer distance between the point sets is used as the loss function.
  • the coordinates of the sampling point on the tooth may be the coordinate value of the global coordinate system.
  • the extracted features may include pose information, normal information, and each adjacent tooth.
  • the shortest distance between The inventor of the present application has discovered through a lot of experiments that the feature of the shortest distance to each adjacent tooth is of great help in improving the accuracy of the generated target tooth layout. Under the enlightenment of this application, it can be understood that this feature can be replaced with other features of the positional relationship with the adjacent tooth, for example, the furthest distance from the adjacent tooth or the average distance from the adjacent tooth.
  • the geometric coding of each tooth is obtained based on the features extracted from the sampling point set of the tooth.
  • the coordinates of the sampling point on the tooth may be the coordinate value of the local coordinate system.
  • the upper and lower jaws can be performed as a whole. In this case, for each sampling point, it is extracted from each adjacent In the feature of the shortest distance between teeth, not only the adjacent teeth in the jaw, but also the adjacent teeth in the opposite jaw must be considered.
  • the method of the present application is used to generate a digital data set representing the target tooth layout, which can be performed based on a single jaw, for example, a digital data set representing the target tooth layout is generated only for the upper jaw or the lower jaw.
  • At least one iteration is performed based on the obtained digital data set representing the target tooth layout, and an iterated digital data set representing the target tooth layout is obtained.
  • the obtained three-dimensional digital model of the dental jaw representing the target tooth layout may be used as a basis, and the operation may be repeated more than once to iteratively obtain a new digital data set representing the target tooth layout.
  • the inventor of the present application has found through a lot of experiments that, under normal circumstances, the result obtained by one iteration is good enough.
  • this operation can be regarded as an augmentation of the existing data used for training.
  • the second deep learning artificial neural network may also be a deep learning convolutional neural network (Convolutional Neural Network, CNN for short).
  • CNN convolutional Neural Network
  • the various diagrams may show exemplary architectures or other configurations of the disclosed methods and systems, which are helpful in understanding the features and functions that can be included in the disclosed methods and systems.
  • the claimed content is not limited to the exemplary architecture or configuration shown, and the desired features can be implemented with various alternative architectures and configurations.
  • the order of the blocks given here should not be limited to the various embodiments that are implemented in the same order to perform the functions, unless clearly indicated in the context .

Landscapes

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

Abstract

一种产生表示正畸治疗的目标牙齿布局的数字数据集的方法,包括:获取第一三维数字模型,表示处于初始牙齿布局的牙颌;对所述第一三维数字模型中的每一牙齿进行采样获得一个对应的采样点集;利用经训练的第一深度人工神经网络,基于每一牙齿的采样点集,产生一个对应的几何编码;将所有牙齿的几何编码组合得到所述牙颌的整体几何编码(107);以及利用经训练的第二深度人工神经网络,基于所述牙颌的整体几何编码,产生表示所述牙颌正畸治疗的目标牙齿布局的数字数据集。

Description

产生表示牙齿正畸治疗的目标牙齿布局的数字数据集的方法 技术领域
本申请总体上涉及产生表示牙齿正畸治疗的目标牙齿布局的数字数据集的方法。
背景技术
当今,由于美观、便捷以及利于清洁等优点,基于高分子材料的壳状矫治器越来越受欢迎。一套壳状矫治器通常包括十几个甚至几十个逐次的壳状矫治器,用于将患者牙齿从初始布局逐次地重新定位到目标布局,其中,在初始布局到目标布局之间包括从第一中间布局到最后中间布局的N个逐次的中间布局。
一种常用的制作壳状矫治器的方法是在一系列逐次的从第一中间布局到目标布局的牙模上,以热压膜成型工艺压膜获得。可以利用表示从第一中间布局到目标布局的一系列逐次的三维数字模型控制设备制作这些牙模。一个常用的获得这些逐次的三维数字模型的方法是,先扫描获得表示患者原始牙齿布局(即进行正畸治疗前患者的牙齿布局)的三维数字模型,接着,人工操作该三维数字模型获得表示目标牙齿布局(即正畸治疗希望达到的牙齿布局)的三维数字模型,然后,基于这两个三维数字模型进行插值得到中间的一系列逐次的三维数字模型。
人工操作表示患者原始牙齿布局的三维数字模型获得表示目标牙齿布局的三维数字模型费时费力,并且其结果高度依赖操作人员的专业水平和认知,较难保证结果的一致性,鉴于此,有必要提供一种计算机执行的产生牙齿正畸治疗的目标牙齿布局的方法。
发明内容
本申请的一方面提供了一种产生表示正畸治疗的目标牙齿布局的数字数据集的方法,包括:获取第一三维数字模型,表示处于初始牙齿布局的牙颌;对所述第一三维数字模型中的每一牙齿进行采样获得一个对应的采样点集;利用经训练的第一深度人工神经网络,基于每一牙齿的采样点集,产生一个对应的几何编码;将所有牙齿的几何编码组合得到所述牙颌的整体几何编码;以及利用经训练的第二深度人工神经网络,基于所述牙颌的整体几何编码,产生表示所述牙颌正畸治疗的目标牙齿布局的数字数据集。
在一些实施方式中,所述第一深度人工神经网络是能够处理点云的深度人工神经网络。
在一些实施方式中,所述第一深度人工神经网络可以是以下之一:PointNet、PointNet++、PointCNN以及DGCNN网络。
在一些实施方式中,所述第二深度人工神经网络可以是基于多层感知器的回归网络。
在一些实施方式中,所述第二深度人工神经网络可以包括SE模块,用于根据全局信息对当前特征的各通道的权值进行调整。
在一些实施方式中,所述牙颌可以包括上颌与下颌。
在一些实施方式中,所述每一牙齿的几何编码为M维向量,其中的M’维向量是由所述第一深度人工神经网络在训练过程中选定,其中,M和M’均为自然数,M>M’。
在一些实施方式中,所述M维向量还包括表示对应牙齿的位姿。
在一些实施方式中,所述M’维向量表示对应牙齿的几何形态。
在一些实施方式中,所述第一深度人工神经网络对每一牙齿进行几何编码时,对该牙齿每一采样点提取的特征包括该采样点与邻牙的位置关系。
在一些实施方式中,所述与邻牙的位置关系特征可以是与邻牙的最短距离。
在一些实施方式中,所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法还可以包括:基于所述表示目标牙齿布局的数字数据集进行至少一次迭代,获得新的表示目标牙齿布局的数字数据集。
附图说明
以下将结合附图及其详细描述对本申请的上述及其他特征作进一步说明。应当理解的是,这些附图仅示出了根据本申请的若干示例性的实施方式,因此不应被视为是对本申请保护范围的限制。除非特别指出,附图不必是成比例的,并且其中类似的标号表示类似的部件。
图1为本申请一个实施例中的计算机执行的产生表示牙齿正畸治疗的目标牙齿布局的数字数据集的方法的示意性流程图;以及
图2示意性地展示了本申请一个实施例中第二深度学习人工神经网络的结构。
具体实施方式
以下的详细描述中引用了构成本说明书一部分的附图。说明书和附图所提及的示意性实施方式仅仅出于是说明性之目的,并非意图限制本申请的保护范围。在本申请的启示下,本领域技术人员能够理解,可以采用许多其他的实施方式,并且可以对所描述实施方式做出各种改变,而不背离本申请的主旨和保护范围。应当理解的是,在此说明并图示的本申请的各个方面可以按照很多不同的配置来布置、替换、组合、分离和设计,这些不同配置都在本申请的保护范围之内。
本申请的一方面提供了一种计算机执行的产生表示牙齿正畸治疗的目标牙齿布局的数字数据集的方法。
牙科正畸治疗是把牙齿从原始布局重新定位到目标布局的过程。可以理解, 目标布局是正畸治疗期望达到的牙齿布局;原始布局可以是进行正畸治疗之前患者的牙齿布局,也可以是利用本申请的方法产生目标布局时所基于的患者当前牙齿布局。
请参图1,为本申请一个实施例中的计算机执行的产生牙齿正畸治疗的目标牙齿布局的方法100的示意性流程图。
在一个实施例中,本申请的产生牙齿正畸治疗的目标牙齿布局的方法可以仅针对单个牙颌(例如,上颌或下颌)产生目标牙齿布局;在又一实施例中,本申请的产生牙齿正畸治疗的目标牙齿布局的方法也可以把上颌与下颌作为一个整体进行处理,同时产生上颌与下颌的目标牙齿布局。下面以后者为例对本申请的产生牙齿正畸治疗的目标牙齿布局的方法进行详细说明。
在101中,获取第一三维数字模型和第二三维数字模型,分别表示处于原始布局的患者上颌牙齿与下颌牙齿。
在一个实施例中,可以通过直接扫描患者的牙颌,以获取表示处于原始布局的牙齿的三维数字模型。在又一实施例中,可以通过扫描患者牙颌的实体模型,例如石膏模型,以获取表示处于原始布局的牙齿的三维数字模型。在又一实施例中,可以通过扫描患者牙颌的咬模,以获取表示处于原始布局的牙齿的三维数字模型。
在一个实施例中,在获得表示处于原始布局的牙齿的三维数字模型后,可以将其进行分割,使得该三维数字模型中各牙齿之间相互独立,从而可以单独移动该三维数字模型中的每颗牙齿。
在一个实施例中,为了便于计算,可以结合全局坐标系和局部坐标系来表达所述第一三维数字模型和第二三维数字模型,即每一颗牙齿拥有自己的局部坐标系,以其局部坐标系在全局坐标系中的位姿表示该牙齿的位姿。
在103中,对第一三维数字模型和第二三维数字模型中的每一牙齿进行采样获得一个对应的采样点集。
在一个实施例中,可以对每一牙齿的三维数字模型进行均匀采样,在一个实施例中,可以采用最远点采样作为采样策略。在本申请的启示下,可以理解,采样方法并不限于以上所列,可以采用任何其他适用的采样方法。
在一个实施例中,对于每一颗牙齿的三维数字模型,可以在其顶点中采样获得1024个点。在本申请的启示下,可以理解,采样点的数量并不仅限于1024个,只要采样点的数量足够多,能够保留牙齿的几何特征即可。
采样完成后,对于所述第一三维数字模型和第二三维数字模型中的每一颗牙齿,获得一个对应的采样点集。
在105中,利用经训练的第一深度学习人工神经网络,基于每一牙齿的采样点集进行自编码,获得一个对应的几何编码。
在一个实施例中,第一深度学习人工神经网络可以是任何能够处理点云的深度学习人工神经网络,例如,PointNet、PointNet++、PointCNN以及DGCNN网络等。在下面的实施例中,将以PointNet网络为例进行说明。
在一个实施例中,以经训练的PointNet网络对每一颗牙齿的采样点集进行自编码,得到一个关于对应牙齿的三维几何形态的编码c,其中,c是一个n维向量。在一个实施例中,n的值可以根据实验结果和计算系统的计算力来确定,例如,可以在100~300的范围内为n取值,例如,可以将n设为100。
在一个实施例中,对于每一颗牙齿,可以将其编码c与其位置p及姿态q信息进行组合,得到该牙齿的几何编码(c,p,q)。在一个实施例中,位置信息p可以是一个三维向量,表示牙齿相对于世界坐标系的一个位移,姿态信息q可以是一个旋转量(例如,四元数、欧拉角、旋转矩阵或旋转向量),表示牙齿相对于世界坐标系的一个旋转。在一个实施例中,p和q可以是对应牙齿的局部坐标系在全局坐标系中的位置和姿态。
在107中,将所有牙齿的几何编码进行组合得到牙颌的整体几何编码。
对于针对单个牙颌产生目标牙齿布局的方案,牙颌的整体几何编码是该单个 牙颌的整体几何编码。对于将上、下颌作为一个整体产生目标牙齿布局的方案,牙颌的整体几何编码是上、下颌的整体几何编码。
在一个实施例中,牙颌的整体几何编码可以是一个N*(n+7)的二维矩阵,其中,N表示牙齿的总数量(对于将上、下颌作为一个整体的方案,N通常为28),n为每一颗牙齿的几何编码的维度,在一个实施例中,对于每一颗牙齿,p的维度可以是3,q的维度可以是4。
在109中,利用经训练的第二深度学习人工神经网络,基于牙颌的整体几何编码,产生表示目标牙齿布局的数字数据集。
请参图2,示意性地展示了本申请一个实施例中第二深度学习人工神经网络200。
第二深度学习人工神经网络200可以被认为是基于多层感知器的回归网络,即基于多层感知器解决回归问题。
第二深度学习人工神经网络200包括输入模块201、参数共享的全连接层模块203、跳跃连接模块205、参数共享的全连接层模块207和209、SE网络模块211(Squeeze-and-Excitation Network)、SE输出模块213、向量整平模块215、多层感知器模块217、向量重构模块219以及输出模块221。
输入模块201接收所述第一深度学习人工神经网络的输出,例如,表示牙颌整体几何编码的二维矩阵。
SE网络模块211用于根据全局信息对当前特征的各通道的权重进行调整,以突出有用的特征,压制不太有用的特征。SE网络模块211的实现可参考由JieHu、LiShen和GangSun发表的《Squeeze-and-Excitation Networks》。
跳跃连接模块205的输出与SE网络模块211的输出合并后输入向量整平模块215,进行向量整平操作,以将数据转化为一维向量,使得与向量整平模块215连接的多层感知器模块217能够接收并处理这些数据。在一个实施例中,多层感知器模块217可以包括若干层全连接层。
向量重构模块219将多层感知器模块217输出的一维向量重构为矩阵形式,由输出模块221输出。
在一个实施例中,第二深度学习人工神经网络200可以输出表示目标牙齿布局下各牙齿的位姿的数字数据集,例如,目标牙齿布局下各牙齿的坐标和角度。在又一实施例中,第二深度学习人工神经网络200也可以输出各牙齿的空间变换矩阵,用于将当前牙齿布局下各牙齿移动到目标牙齿布局下的位姿。
在一个实施例中,可以将所述第一深度学习人工神经网络和第二深度学习人工神经网络作为一个整体进行训练。对于每一颗牙齿,其几何编码中的n维特征是什么特征,这是由第一深度学习人工神经网络在训练中所决定,这样能够使得这n维特征能够尽量准确地捕捉牙齿的几何特征对于排牙的影响,从而指导网络实现更准确的排牙。
在又一实施例中,所述第一深度学习人工神经网络和第二深度学习人工神经网络也可以分开分别进行训练。
在一个实施例中,所述第一深度学习人工神经网络可以是PointNet网络,可以利用采样点集(作为输入)和重构点集(作为输出)对其进行训练,以采样点集和重构点集之间的chamfer距离作为损失函数。
在一个实施例中,利用所述第一深度学习人工神经网络对每颗牙齿进行几何编码时,牙齿上采样点的坐标可以是全局坐标系的坐标值。
在一个实施例中,利用所述第一深度学习人工神经网络对每颗牙齿进行几何编码时,对于每个采样点,提取的特征可以包括位姿信息、法向信息以及其与每一邻接牙齿之间的最短距离。本申请的发明人经过大量实验发现,与每一邻接牙之间的最短距离该特征对于提升产生的目标牙齿布局的精确度有较大帮助。在本申请的启示下,可以理解,该特征可以其他与邻接牙的位置关系的特征替换,例如,与邻接牙之间的最远距离或与邻接牙之间的平均距离等。每一牙齿的几何编码是基于对该牙齿的采样点集提取到的特征而获得。
在又一实施例中,利用所述第一深度学习人工神经网络对每颗牙齿进行几何编码时,牙齿上采样点的坐标可以是局部坐标系的坐标值。
在以上实施例中,利用本申请的方法产生表示目标牙齿布局的数字数据集,可以将上、下颌作为一个整体进行,在这种情况下,对于每个采样点,在提取其与每一邻接牙齿之间的最短距离这个特征时,不仅要考虑其所在牙颌内的邻接牙齿,还要考虑对颌牙齿中的与之邻接的牙齿。
在又一实施例中,利用本申请的方法产生表示目标牙齿布局的数字数据集,可以基于单个牙颌进行,例如,仅针对上颌或下颌产生表示其目标牙齿布局的数字数据集。
在111中,基于获得的表示目标牙齿布局的数字数据集进行至少一次迭代,获得迭代后的表示目标牙齿布局的数字数据集。
在一个实施例中,可以将获得的表示目标牙齿布局的牙颌的三维数字模型作为基础,重新进行一次以上操作,迭代获得新的表示目标牙齿布局的数字数据集。
本申请的发明人经过大量实验发现,通常情况下,迭代一次获得的结果已经足够好。
在训练的阶段,该操作可被视为是对用于训练的现有数据的增广。
在一个实施例中,所述第二深度学习人工神经网络也可以是深度学习卷积神经网络(ConvolutionalNeuralNetwork,简称CNN)。
尽管在此公开了本申请的多个方面和实施例,但在本申请的启发下,本申请的其他方面和实施例对于本领域技术人员而言也是显而易见的。在此公开的各个方面和实施例仅用于说明目的,而非限制目的。本申请的保护范围和主旨仅通过后附的权利要求书来确定。
同样,各个图表可以示出所公开的方法和系统的示例性架构或其他配置,其有助于理解可包含在所公开的方法和系统中的特征和功能。要求保护的内容并不 限于所示的示例性架构或配置,而所希望的特征可以用各种替代架构和配置来实现。除此之外,对于流程图、功能性描述和方法权利要求,这里所给出的方框顺序不应限于以同样的顺序实施以执行所述功能的各种实施例,除非在上下文中明确指出。
除非另外明确指出,本文中所使用的术语和短语及其变体均应解释为开放式的,而不是限制性的。在一些实例中,诸如“一个或多个”、“至少”、“但不限于”这样的扩展性词汇和短语或者其他类似用语的出现不应理解为在可能没有这种扩展性用语的示例中意图或者需要表示缩窄的情况。

Claims (12)

  1. 一种产生表示正畸治疗的目标牙齿布局的数字数据集的方法,包括:
    获取第一三维数字模型,表示处于初始牙齿布局的牙颌;
    对所述第一三维数字模型中的每一牙齿进行采样获得一个对应的采样点集;
    利用经训练的第一深度人工神经网络,基于每一牙齿的采样点集,产生一个对应的几何编码;
    将所有牙齿的几何编码组合得到所述牙颌的整体几何编码;以及
    利用经训练的第二深度人工神经网络,基于所述牙颌的整体几何编码,产生表示所述牙颌正畸治疗的目标牙齿布局的数字数据集。
  2. 如权利要求1所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,所述第一深度人工神经网络是能够处理点云的深度人工神经网络。
  3. 如权利要求2所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,所述第一深度人工神经网络是以下之一:PointNet、PointNet++、PointCNN以及DGCNN网络。
  4. 如权利要求1所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,所述第二深度人工神经网络是基于多层感知器的回归网络。
  5. 如权利要求4所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,所述第二深度人工神经网络包括SE模块,用于根据全局信息对当前特征的各通道的权值进行调整。
  6. 如权利要求1所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,所述牙颌包括上颌与下颌。
  7. 如权利要求1所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,所述每一牙齿的几何编码为M维向量,其中的M’维向量是由所述第一深度人工神经网络在训练过程中选定,其中,M和M’均为自然数, M>M’。
  8. 如权利要求7所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,所述M维向量还包括表示对应牙齿的位姿。
  9. 如权利要求7所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,所述M’维向量表示对应牙齿的几何形态。
  10. 如权利要求1所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,所述第一深度人工神经网络对每一牙齿进行几何编码时,对该牙齿每一采样点提取的特征包括该采样点与邻牙的位置关系。
  11. 如权利要求10所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,所述与邻牙的位置关系特征是与邻牙的最短距离。
  12. 如权利要求1所述的产生表示正畸治疗的目标牙齿布局的数字数据集的方法,其特征在于,它还包括:基于所述表示目标牙齿布局的数字数据集进行至少一次迭代,获得新的表示目标牙齿布局的数字数据集。
PCT/CN2021/083968 2020-06-19 2021-03-30 产生表示牙齿正畸治疗的目标牙齿布局的数字数据集的方法 WO2021253917A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US18/011,511 US20230334771A1 (en) 2020-06-19 2021-03-30 Method for generating digital data set representing target tooth arrangement for orthodontic treatment

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010565918.6A CN113822305A (zh) 2020-06-19 2020-06-19 产生表示牙齿正畸治疗的目标牙齿布局的数字数据集的方法
CN202010565918.6 2020-06-19

Publications (1)

Publication Number Publication Date
WO2021253917A1 true WO2021253917A1 (zh) 2021-12-23

Family

ID=78912012

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/083968 WO2021253917A1 (zh) 2020-06-19 2021-03-30 产生表示牙齿正畸治疗的目标牙齿布局的数字数据集的方法

Country Status (3)

Country Link
US (1) US20230334771A1 (zh)
CN (1) CN113822305A (zh)
WO (1) WO2021253917A1 (zh)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090098502A1 (en) * 2006-02-28 2009-04-16 Ormco Corporation Software and Methods for Dental Treatment Planning
CN106618760A (zh) * 2016-12-07 2017-05-10 上海牙典医疗器械有限公司 一种设计正畸矫治方案的方法
CN109363786A (zh) * 2018-11-06 2019-02-22 上海牙典软件科技有限公司 一种牙齿正畸矫治数据获取方法及装置
CN109528323A (zh) * 2018-12-12 2019-03-29 上海牙典软件科技有限公司 一种基于人工智能的正畸方法及装置
CN110473283A (zh) * 2018-05-09 2019-11-19 无锡时代天使医疗器械科技有限公司 牙齿三维数字模型的局部坐标系设定方法
CN111274666A (zh) * 2019-12-06 2020-06-12 上海正雅齿科科技股份有限公司 一种数字化牙齿位姿变化量的设计、模拟排牙方法及装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090098502A1 (en) * 2006-02-28 2009-04-16 Ormco Corporation Software and Methods for Dental Treatment Planning
CN106618760A (zh) * 2016-12-07 2017-05-10 上海牙典医疗器械有限公司 一种设计正畸矫治方案的方法
CN110473283A (zh) * 2018-05-09 2019-11-19 无锡时代天使医疗器械科技有限公司 牙齿三维数字模型的局部坐标系设定方法
CN109363786A (zh) * 2018-11-06 2019-02-22 上海牙典软件科技有限公司 一种牙齿正畸矫治数据获取方法及装置
CN109528323A (zh) * 2018-12-12 2019-03-29 上海牙典软件科技有限公司 一种基于人工智能的正畸方法及装置
CN111274666A (zh) * 2019-12-06 2020-06-12 上海正雅齿科科技股份有限公司 一种数字化牙齿位姿变化量的设计、模拟排牙方法及装置

Also Published As

Publication number Publication date
US20230334771A1 (en) 2023-10-19
CN113822305A (zh) 2021-12-21

Similar Documents

Publication Publication Date Title
US11735306B2 (en) Method, system and computer readable storage media for creating three-dimensional dental restorations from two dimensional sketches
CN112200843B (zh) 一种基于超体素的cbct与激光扫描点云数据牙齿配准方法
Tian et al. DCPR-GAN: dental crown prosthesis restoration using two-stage generative adversarial networks
US20180028294A1 (en) Dental cad automation using deep learning
JP4384026B2 (ja) 既知のデジタルオブジェクトを走査3dモデルに登録するための方法および装置
CA3109245A1 (en) Automated orthodontic treatment planning using deep learning
KR20190140990A (ko) 치과 기구의 제조
CN111152218B (zh) 一种异构仿人机械臂的动作映射方法及系统
US20220008175A1 (en) Method for generating dental models based on an objective function
WO2019030794A1 (ja) 情報処理装置、モデルデータ作成プログラム、モデルデータ作成方法
CN112785609B (zh) 一种基于深度学习的cbct牙齿分割方法
CN113011526B (zh) 基于强化学习和无监督学习的机器人技能学习方法及系统
CN111696068A (zh) 利用人工神经网络产生表示目标牙齿布局的数字数据集的方法及计算机系统
WO2021253917A1 (zh) 产生表示牙齿正畸治疗的目标牙齿布局的数字数据集的方法
CN111265317A (zh) 一种牙齿正畸过程预测方法
CN111260702B (zh) 激光三维点云与ct三维点云配准方法
CN115619773A (zh) 一种三维牙齿多模态数据配准方法及系统
CN115908690A (zh) 产生表示目标牙齿布局的数字数据集的方法
JP2024031920A (ja) 3次元スキャンデータから補綴物を自動で生成する方法、及びこれをコンピュータで実行させるためのプログラムが記録されたコンピュータ読取り可能な記録媒体
TW201544077A (zh) 牙體立體影像建立方法
US20230063677A1 (en) Method for generating a digital data set representing a target tooth arrangement
CN107564094B (zh) 一种基于局部坐标的牙齿模型特征点自动识别算法
CN114612532A (zh) 一种三维牙齿配准方法、系统、计算机设备及存储介质
US11937994B2 (en) Method for generating a digital data set representing a target tooth arrangement
US20240029380A1 (en) Integrated Dental Restoration Design Process and System

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: 21824941

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21824941

Country of ref document: EP

Kind code of ref document: A1