Machine learning-based three-dimensional dental and maxillofacial deformity assisted identification method and system

Abstract

The application discloses a kind of three-dimensional dental jaw craniofacial deformity auxiliary identification method and system based on machine learning.The face three-dimensional photo set obtained and the face three-dimensional photo to be predicted are respectively marked and the coordinate system is unified, after the respective feature information corresponding to face three-dimensional photo set is preprocessed, it is trained by machine learning model, and prediction model is obtained, the landmark point coordinates of the face three-dimensional photo to be predicted are predicted using the prediction model, and the face feature is obtained, to assist dentist to carry out dental jaw craniofacial deformity discrimination.The present application can avoid / reduce the amount of exposure to electric radiation, the process is friendly to patients, and is more acceptable.Based on the learning of larger sample size, the prediction is more accurate.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of orthodontics, and particularly to a three-dimensional dental and maxillofacial deformity auxiliary identification method based on machine learning and a three-dimensional dental and maxillofacial deformity auxiliary identification system based on machine learning. BACKGROUND

[0002] The vertical and sagittal diagnosis of craniofacial and maxillofacial bones in orthodontic treatment is very important, which affects the facial appearance, anchorage design in orthodontic treatment, establishment of tooth arrangement scheme, overall treatment scheme and final treatment effect. At present, the relevant diagnostic results are obtained based on cephalometric analysis or CBCT three-dimensional measurement in clinical practice, however, the quality requirement of cephalometric radiograph is high, the technical sensitivity of landmark calibration is high, and CBCT has the problem of additional ionizing radiation. Experienced orthodontists can give more accurate related diagnosis according to the soft tissue facial type, which indicates that the corresponding diagnostic results can be obtained based on the soft tissue facial type. At present, the related three-dimensional dental and maxillofacial deformity auxiliary identification method based on three-dimensional facial photographs is still blank. SUMMARY

[0003] The present application aims to provide a three-dimensional dental and maxillofacial deformity auxiliary identification method based on machine learning to provide a method for auxiliary diagnosis of maxillofacial deformity in orthodontics without or with reduced ionizing radiation.

[0004] The technical solution adopted by the present application is as follows:

[0005] A three-dimensional dental and maxillofacial deformity auxiliary identification method based on machine learning, comprising:

[0006] obtaining a set of three-dimensional facial photographs;

[0007] for each three-dimensional facial photograph, performing the following steps: calibrating the landmark points in the three-dimensional facial photograph; obtaining the three-dimensional coordinates of each landmark point, and unifying the coordinate systems of the obtained three-dimensional coordinates of each landmark point;

[0008] preprocessing the feature information corresponding to each three-dimensional facial photograph to obtain corresponding training samples, wherein the feature information includes the three-dimensional coordinates of each landmark point after unifying the coordinate systems of the three-dimensional facial photograph;

[0009] training a machine learning model using each training sample to obtain a prediction model;

[0010] using the prediction model to predict the facial features corresponding to the three-dimensional coordinates of each landmark point of the patient's three-dimensional facial photograph.

[0011] Further, the calibration of the landmark points in the three-dimensional facial photograph comprises:

[0012] The calibration tool is used to automatically mark the mark points in the three-dimensional face photo, and all or part of the mark points are manually adjusted.

[0013] Further, the three-dimensional coordinates of the acquired mark points are unified in a coordinate system, comprising:

[0014] The origin of the new coordinate system is set as the subnasale point;

[0015] A reference plane A is established: the left and right antitragus points and the right ala nasi point are taken to establish the reference plane A;

[0016] A horizontal plane is established: taking the left and right antitragus points as the axis, the reference plane A is rotated by 7.5°, and a plane parallel to the plane passing through the subnasale point is made, that is, the horizontal plane of the new coordinate system;

[0017] A sagittal plane is established: a plane passing through the midpoint of the left and right antitragus points and the subnasale point is established, which is perpendicular to the horizontal plane;

[0018] A coronal plane is established: a plane perpendicular to the horizontal plane and the sagittal plane is established;

[0019] Coordinate system transformation: the coordinates in the original coordinate system are converted to the new coordinate system.

[0020] Further, the feature information corresponding to each three-dimensional face photo is preprocessed, comprising: orthogonalizing the feature information corresponding to each three-dimensional face photo.

[0021] Further, the three-dimensional coordinates of the mark points of the patient's three-dimensional face photo are the three-dimensional coordinates of the mark points automatically marked by the calibration tool, or the three-dimensional coordinates after manual adjustment.

[0022] The application also provides a three-dimensional dental and maxillofacial deformity auxiliary identification system based on machine learning, comprising an image acquisition module, a sample processing module, a model construction module and a prediction module, wherein:

[0023] The image acquisition module is used to acquire the three-dimensional face photo to be predicted and a set of three-dimensional face photos for training the model;

[0024] The sample processing module is used to mark the mark points in each three-dimensional face photo, acquire the three-dimensional coordinates of the mark points, unify the three-dimensional coordinates of the mark points in the coordinate system, preprocess the feature information corresponding to each three-dimensional face photo in the set of three-dimensional face photos to obtain a training sample set, and the feature information comprises the three-dimensional coordinates of the mark points after unifying the coordinate system;

[0025] The model construction module trains the machine learning model using the training sample set to obtain a prediction model;

[0026] The prediction module evaluates the facial features corresponding to the three-dimensional coordinates of each landmark point of the to-be-predicted facial three-dimensional photo by using the prediction model.

[0027] Further, the sample processing module automatically marks the landmark points in the facial three-dimensional photo by using a calibration tool, and adjusts all or part of the landmark points in response to an adjustment instruction.

[0028] Further, the sample processing module is configured with a computer program for coordinate system unification of the three-dimensional coordinates of each landmark point, and the computer program is run to perform the following method:

[0029] The origin of the new coordinate system is set as the subnasale point;

[0030] A reference plane A is established: the reference plane A is established by taking the left and right antitragus points and the right ala nasi point;

[0031] A horizontal plane is established: the horizontal plane of the new coordinate system is established by rotating the reference plane A by 7.5° about the left and right antitragus points and passing through the subnasale point;

[0032] A sagittal plane is established: a plane passing through the midpoint of the left and right antitragus points and the subnasale point is established, which is perpendicular to the horizontal plane;

[0033] A coronal plane is established: a plane perpendicular to the horizontal plane and the sagittal plane is established;

[0034] Coordinate system transformation: the coordinates in the original coordinate system are converted to the new coordinate system.

[0035] Further, the sample processing module is configured with a computer program for preprocessing of the feature information, and the computer program is run to perform the following method: orthogonalization processing is performed on the feature information.

[0036] Further, the sample processing module transmits the landmark point coordinates of the to-be-predicted facial three-dimensional photo to the prediction module, which are three-dimensional coordinates of the landmark points automatically marked by the calibration tool, or three-dimensional coordinates adjusted in response to an adjustment instruction.

[0037] In summary, due to the adoption of the above technical solutions, the present application has the following advantages:

[0038] 1. The three-dimensional dental and maxillofacial deformity auxiliary identification scheme based on machine learning can train an identification model by using a facial three-dimensional photo, and similarly, can predict facial features by using a facial three-dimensional photo. Compared with traditional imaging methods, the facial three-dimensional photo is easier to obtain, can avoid / reduce the amount of exposure to electric radiation, and has better experience and higher acceptance for patients.

[0039] 2、The application can correct improper data of machine calibration when adjusting the position of the landmark point, and can also realize the expansion of sample data, in addition, the use of expanded data training can improve the recognition accuracy of the measured data of machine calibration.

[0040] 3、The application unifies the coordinate system of the calibration point coordinates, which is convenient for comparing the differences of different faces, and also facilitates statistical analysis of related research. BRIEF DESCRIPTION OF DRAWINGS

[0041] The application will be illustrated by examples and with reference to the accompanying drawings, in which:

[0042] Figure 1 It is a three-dimensional dental jaw and craniofacial deformity auxiliary identification method flow chart based on machine learning.

[0043] Figure 2 It is a three-dimensional dental jaw and craniofacial deformity auxiliary identification system structure diagram based on machine learning.

[0044] Figure 3 It is a schematic diagram of landmark calibration.

[0045] Figure 4 It is a schematic diagram of the three-dimensional face image after unifying the coordinate system. DETAILED DESCRIPTION

[0046] All features disclosed in this specification, or all steps of any methods or processes disclosed in this specification, can be combined in any manner, except where features or steps are mutually exclusive.

[0047] Any feature disclosed in this specification, unless stated otherwise, can be replaced by an alternative feature or equivalent feature having the same or similar effect. That is, unless stated otherwise, each feature is one example only of a number of equivalent or similar features.

[0048] A three-dimensional dental jaw and craniofacial deformity auxiliary identification method based on machine learning, as shown in Figure 1 , comprising:

[0049] 1) The step of obtaining a training data set

[0050] The training data set contains a large number of training samples for training the machine model. The method for obtaining the training data set includes the steps of obtaining a set of three-dimensional face photos and obtaining a set of training samples based on the obtained set of three-dimensional face photos.

[0051] A large number of subjects (volunteers) face three-dimensional photos can be obtained based on three-dimensional optical scanning or three-dimensional stereophotogrammetry technology, which can be one or more, and finally one is selected for use. In this way, a large number of three-dimensional face photos are obtained.

[0052] For each face three-dimensional photo, the following operations are performed:

[0053] In the photo, mark the landmark points (or feature points) in a machine or manual or machine combined with manual manner, the landmark points are points on the face with characteristic identification, such as the alae nasi point, the rhinion point, the zygomatic point, the temple, etc.

[0054] Respectively, obtain the three-dimensional coordinates of each landmark point. Then, unify the three-dimensional coordinates of each landmark point, that is, unify to a coordinate system. Before the coordinate unification, the three-dimensional coordinates may be based on different coordinate systems. After the coordinate unification, the three-dimensional coordinates of each landmark point are re-marked / converted based on the unified three-dimensional coordinate system, for example, the subnasale point (or other landmark point) is unified as the origin of the coordinate system. For the coordinates of each landmark point before the unification of the coordinate system, the coordinates are converted to the new coordinate system according to the relative position of the subnasale point, so as to complete the unification of the three-dimensional coordinates of each landmark point.

[0055] After the three-dimensional coordinates of each landmark point of each face three-dimensional photo are unified, the position information of each landmark point, that is, the X, Y, Z coordinates of the landmark point, is obtained. Then, the other feature information data of the subject corresponding to the face three-dimensional photo is combined as the feature information of the subject. The feature information data can also include the three-dimensional coordinates before the adjustment of the landmark point position, so as to naturally expand the sample.

[0056] The feature information of each subject is preprocessed, that is, the corresponding training sample is obtained. All the training samples constitute the training data set.

[0057] 2) The step of training the machine learning model using the training data set

[0058] For the preprocessed training sample, 80% is used as the training set and 20% is used as the test set to train the machine learning model, and a pre-stored model is obtained. The machine learning model is a multilayer perceptron (MLP, Multilayer Perceptron) learning model.

[0059] The multilayer perceptron is a feedforward network containing simple neurons, which maps the input data set to a set of outputs. The multilayer perceptron contains multiple layers of directed graph fully connected nodes, and each node is a neuron with a nonlinear activation function.

[0060] The basic component of MLP is neuron. In MLP, each pair of neurons of adjacent two layers is connected by an edge with weight coefficient. As shown in the figure, MLP is composed of at least three layers of neurons, including an input layer, one or more hidden layers, and an output layer. The number of input neurons depends on the dimension of input features, and the number of output neurons is determined by the number of categories. MLP adopts a supervised learning method called back propagation to train the network. The perceptron calculates the output by processing the linear combination of weighted real-valued inputs through the following nonlinear activation function.

[0061]

[0062] where ω i represents the weight vector, x i is the input vector, b is the offset rate, and is the activation function. Common activation functions include ReLU function, sigmoid function, and tanh function. MLP adjusts the weights of the hidden layer to reduce the output error during the learning process. MLP propagates the input pattern signal forward through the network and starts to propagate the error signal backward at the output. The error function of back propagation is composed of the difference between the true value and the expected value. The goal of the learning process is to minimize the error function. The minimum value of the error function is found by taking the derivative of the error function with respect to the weight matrix. The learning process includes: (1) randomly initializing the weights with values between (-1, 1); (2) sending an input pattern to the network; (3) calculating the output of the network; (4) for each node in the output layer, calculating the error of the output node and increasing the error function value for all weights connected to the node.

[0063] In order to control the convergence speed and reduce the step size of the adapted weights, a learning parameter is introduced. In some embodiments, ReLu activation function is used, and 200 layers of hidden layers are used. The training data set is divided into 80% training set and 20% test set, and the test set is used to verify the MSE and other indicators.

[0064] 3) Auxiliary identification of dental and maxillofacial deformities in patients

[0065] Through the above learning process, an identification model for identifying dental and maxillofacial facial features by landmark point coordinates is obtained. The landmark point coordinates corresponding to the three-dimensional photo of the patient's face are input into the identification model, and the identification model can output the facial features with reference value to assist the doctor in diagnosing the patient's facial deformity. The landmark point coordinates corresponding to the three-dimensional photo of the patient's face are obtained in the same way as the method of obtaining landmark point coordinates from the three-dimensional photo of the face when obtaining the training sample, or the step of adjusting the landmark points can be omitted.

[0066] Example Two

[0067] The embodiment discloses a three-dimensional dental and maxillofacial deformity auxiliary identification method based on machine learning. The embodiment has the same general steps as embodiment one and belongs to further optimization of embodiment one. As shown in Figure 1 The method comprises the following steps:

[0068] 1) Take a three-dimensional face photo of each subject based on three-dimensional optical scanning or three-dimensional stereophotogrammetry technology, and save it in obj format. The shooting requirements are as follows: ① The person being photographed cannot wear glasses or any facial ornaments, and the bilateral ears and forehead must be completely exposed; ② The person being photographed is required to look straight ahead, take a natural head position, relax the lips, and keep the teeth in a natural rest position; ③ The person being photographed remains still until the shooting is completed.

[0069] 2) Calibrate the landmark points. This step is performed by combining machines with manual methods. Import the obj format three-dimensional face photo into Cliniface software, manually adjust the landmark points based on the automatic calibration results of the software, for example, edit, drag, and manually add bilateral antitragus points, as shown in Figure 3 A total of 61 landmark points are calibrated, and the corresponding number description is shown in Table 1. It should be noted that the adjusted landmark points and the landmark points before adjustment can be saved in association and used as sample data. When predicting the patient's facial features later, the landmark points before adjustment can be directly referenced without the need to adjust the landmark point positions and then make judgments.

[0070] Table 1: Landmark point number and name

[0071]

[0072]

[0073]

[0074] In fact, the calibrated landmark points can only select part of the above Table 1, but the landmark point as the origin of the unified coordinate system must be retained. For the calibrated landmark points, adaptive selection and adjustment can be made according to the test.

[0075] 3) Obtain the three-dimensional coordinates of the facial landmark points and unify the coordinate system

[0076] Similarly, in the Cliniface software, the coordinates of each landmark point are exported to a csv file. As shown in Figure 4 Take the subnasale as the origin of the coordinate system, and unify the coordinates of each three-dimensional face photo, as follows:

[0077] ① Define the subnasale as the origin of the coordinate system (of course, other landmark points can also be used);

[0078] 2. Establishing reference plane A: taking the left and right Tragion and the right Alar Curvature Point R to establish the reference plane A;

[0079] 3. Establishing horizontal plane: taking the left and right Tragion as the axis, rotating the reference plane A by 7.5°, and passing through the Subnasale to form a plane parallel to the plane, i.e. the horizontal plane of the unified coordinate system;

[0080] 4. Establishing sagittal plane: taking the midpoint of the left and right Tragion and the Subnasale, and forming a plane perpendicular to the horizontal plane;

[0081] 5. Establishing coronal plane: forming a plane perpendicular to the horizontal plane and the sagittal plane;

[0082] 6. Coordinate system transformation: converting the coordinates in the original coordinate system to the new coordinate system.

[0083] Specifically, the method for unifying coordinates comprises:

[0084] 1. CSV file preprocessing: unifying the exported CSV file into the format (number, X, Y, Z).

[0085] 2. Establishing reference plane A normal vector: performing cross multiplication on the line connecting the left and right Tragion and the line connecting the right Tragion and the right Alar Curvature Point R, and then normalizing to obtain the normal vector of plane A.

[0086] 3. Establishing horizontal plane: obtaining the normal vector of the selected rotated plane A, which is the normal vector of the horizontal plane.

[0087] Given a rotation axis A = [a x , a y , a z ] with a unit length and a rotation angle θ, the matrix representation of the object rotation transformation around the OA axis can be determined as follows:

[0088]

[0089]

[0090]

[0091] P′ = P · M T

[0092] where P is the normal vector of plane A, P′ is the normal vector of the horizontal plane, and M T represents the transpose matrix of M.

[0093] 4. Establishing sagittal plane: performing cross multiplication on the vector of the line connecting the left and right Tragion and the normal vector of the horizontal plane, and then normalizing to obtain the normal vector of the sagittal plane.

[0094] V = V1 x V2 x V3

[0095] V = V1 x V2 x V3 -1 V = V1 x V2 x V3

[0096] X is the coordinate of each point in the original coordinate system, V is the coordinate of the origin in the new coordinate system, V1, V2, V3 represent the coordinate axis normal vector of the new coordinate.

[0097] Figure 4 The visualization of each landmark point on the face after the coordinate system is shown.

[0098] 4) Implementation of machine learning process

[0099] (1) Obtain the position information of each landmark point after the unified coordinate system through steps 1) to 3) above, including the X, Y, Z coordinates of each landmark point, and save the three-dimensional coordinates of each landmark point in CSV format.

[0100] For each three-dimensional face photo, obtain the corresponding feature information data, including the gender, age, BMI, facial sagittal and vertical assessment results of the subject, and the three-dimensional coordinates of each landmark point after the unified coordinate system.

[0101] (2) Preprocess each feature information data, including orthogonalization processing of the feature information data. After preprocessing, the training data set is obtained, and the training data set is used for machine learning training using Scikit-learn.

[0102] (3) Use 80% of the data as the training set and 20% as the test set to establish a multi-layer perception learning model through Scikit-learn, which serves as a classifier for assisting in the diagnosis of facial features based on three-dimensional coordinate information of landmark points.

[0103] 5) Output of results:

[0104] Take a three-dimensional photo of the patient's face, obtain the landmark point coordinates after the unified coordinate system, and import them into the model trained in step 4) above to input the prediction results of the patient's facial features, which can assist the physician in diagnosing whether the patient has facial dental craniofacial deformity.

[0105] It should be noted that this embodiment is an operation instruction using Cliniface as an example. Without violating the overall inventive concept, other software with the same function can be used to calibrate the landmark points and export data. Similarly, the machine learning model used can be replaced by other multi-layer perception models.

[0106] Example Three

[0107] This embodiment discloses a three-dimensional dentofacial deformity auxiliary recognition method based on machine learning. In this method, the facial sagittal and vertical evaluation results of the training samples specifically include: sagittal division into classes I, II, and III; vertical division into low angle, average angle, and high angle.

[0108] Correspondingly, the prediction results for the patient's three-dimensional facial photographs include sagittal and vertical assessment results.

[0109] Example 4

[0110] This embodiment discloses a machine learning-based three-dimensional dentofacial deformity auxiliary recognition system, such as... Figure 2 As shown, it includes an image acquisition module, a sample processing module, a model building module, and a prediction module, wherein:

[0111] The image acquisition module captures 3D facial photographs of patients and subjects (volunteers). Based on 3D optical scanning or 3D stereophotometry, the module captures one 3D facial photograph for each subject and saves it in .obj format. The shooting requirements are as follows: ① The subject must not wear glasses or any accessories that obscure the face; both ears and the forehead must be fully exposed; ② The subject must look straight ahead, maintain a natural head position, relax the lip muscles, and keep the teeth in a natural occlusal position; ③ The subject must remain still until the shooting is complete.

[0112] The sample processing module performs the following processing on each facial 3D image: 1) Labeling landmark points in the facial 3D images; 2) Obtaining the 3D coordinates of each landmark point; 3) Standardizing the 3D coordinates of each landmark point. Furthermore, the sample processing module combines the standardized 3D coordinates of each landmark point with the subject's feature information data to obtain the subject's feature information. This feature information is then preprocessed to obtain the corresponding training samples. The feature information may also include the coordinates of the landmark points before standardization, thereby expanding the sample size and ensuring good compatibility and recognition of machine-calibrated test data in subsequent prediction processes.

[0113] The sample processing module can calibrate the facial landmark points in a machine, manual or machine-assisted manner. In some embodiments, the sample processing module performs calibration of the landmark points using Cliniface software (or other calibration tools): imports the facial three-dimensional photos in obj format into the Cliniface software, adjusts the positions of the landmark points based on the automatic calibration results of the software, and manually adds the bilateral antitragus points in response to adjustment instructions such as editing and dragging, as shown in the figure. A total of 61 landmark points are calibrated, and the corresponding number is as shown in Table 1 above. In addition, for the landmark points automatically calibrated by the Cliniface software and the manually adjusted landmark points, the sample processing module stores the correlation as sample data, which is equivalent to expanding the sample data.

[0114] Similarly, the sample processing module exports the coordinates of each landmark point in a csv file in the Cliniface software. The subnasale is taken as the origin of the coordinate system, and the coordinates of each facial three-dimensional photo are unified. For specific operation process of coordinate unification, see Example 2, which will not be repeated here.

[0115] The feature information data of the subject includes the gender, age, BMI, facial sagittal and vertical evaluation results of the subject, and the three-dimensional coordinates of each landmark point after unification of the coordinate system. The sample processing module performs orthogonalization processing on the feature information data to obtain training samples.

[0116] The model construction module trains the machine learning model using the training samples to obtain a prediction model of facial features. The machine learning model is a multilayer perceptron (MLP, Multilayer Perceptron) learning model. The multilayer perceptron learning model has been described in detail in the foregoing, and will not be repeated here. In some embodiments, the model construction module uses 80% of the training sample data as a training set and 20% as a test set to establish a multilayer perceptron learning model through Scikit-learn, as a classifier for assisting in diagnosing facial features through landmark three-dimensional coordinate information.

[0117] The prediction module uses the prediction model trained by the model construction module to predict the landmark point coordinates of the facial three-dimensional photos of the patients calibrated by the sample processing module, and outputs the facial features with reference to assist the physician in diagnosing the facial deformity of the patient.

[0118] The present application is not limited to the specific embodiments described above. The present application extends to any new feature or any new combination described in this specification, as well as any new method or process steps or any new combination disclosed.

Claims

1. A method for assisting in identifying three-dimensional dental and maxillofacial deformities based on machine learning, characterized in that, The method comprises the following steps: acquiring a set of facial three-dimensional photos, including: acquiring one facial three-dimensional photo for each of a plurality of subjects, thereby obtaining a plurality of facial three-dimensional photos; for each facial three-dimensional photo, performing the following steps: calibrating landmark points in the facial three-dimensional photo, including: automatically marking the landmark points in the facial three-dimensional photo by using a calibration tool, and adjusting all or part of the landmark points in response to an adjustment instruction; acquiring three-dimensional coordinates of each landmark point, and unifying the coordinate systems of the acquired three-dimensional coordinates of each landmark point; performing orthogonalization processing on feature information corresponding to each facial three-dimensional photo to obtain corresponding training samples, wherein the feature information includes the gender, age, BMI, sagittal and vertical evaluation results of the subject, the three-dimensional coordinates of each landmark point after unifying the coordinate systems, and the three-dimensional coordinates before adjusting the landmark point positions, and the adjusted landmark points are correlated with the unadjusted landmark points; the sagittal and vertical evaluation results specifically include: sagittal classification of I, II, and III; vertical classification of low angle, average angle, and high angle; training a machine learning model using each training sample to obtain a prediction model for identifying facial features through landmark point coordinates; using the prediction model to predict the facial features corresponding to the three-dimensional coordinates of each landmark point in the patient's facial three-dimensional photo without adjustment, wherein the facial features include the sagittal and vertical evaluation results. 2.The machine learning based three-dimensional dental-skeletal-craniofacial deformity auxiliary identification method of claim 1, wherein, The unification of the coordinate systems of the acquired three-dimensional coordinates of each landmark point comprises: setting the origin of the new coordinate system as the subnasale point; establishing a reference plane A: taking the left and right antitragus points and the right ala nasi point to establish the reference plane A; establishing a horizontal plane: taking the left and right antitragus points as the axis, rotating the reference plane A by 7.5°, and passing through the subnasale point to form a plane parallel to the horizontal plane, which is the horizontal plane of the new coordinate system; establishing a sagittal plane: establishing a plane passing through the midpoint of the left and right antitragus points and the subnasale point, which is perpendicular to the horizontal plane; establishing a coronal plane: a plane perpendicular to the horizontal plane and the sagittal plane; coordinate system transformation: converting the coordinates in the original coordinate system to the new coordinate system.

3. A machine learning-based three-dimensional dental-skeletal-craniofacial deformity auxiliary identification system, characterized in that, The method comprises the following steps: The image acquisition module is used to acquire the facial three-dimensional photo to be predicted and the set of facial three-dimensional photos for training the model; wherein the set of facial three-dimensional photos is obtained by the image acquisition module taking one facial three-dimensional photo for each of a plurality of subjects; The sample processing module is configured to calibrate the landmark points in each three-dimensional facial photo, automatically mark the landmark points in the three-dimensional facial photo by using a calibration tool, and adjust all or part of the landmark points in response to an adjustment instruction; obtain three-dimensional coordinates of each landmark point, and unify the three-dimensional coordinates of the landmark points in each three-dimensional facial photo; orthogonalize the feature information corresponding to each three-dimensional facial photo in the set of three-dimensional facial photos, to obtain a training sample set, wherein the feature information includes the gender, age, BMI, sagittal and vertical evaluation results of the subject, the three-dimensional coordinates of each landmark point after being unified in the coordinate system, and the three-dimensional coordinates before the position of each landmark point is adjusted, and each landmark point after being adjusted is correlated with each landmark point before being adjusted; and the sagittal and vertical evaluation results specifically include: sagittal classification I, II, and III; and vertical classification low, average, and high. The model construction module trains a machine learning model by using the training sample set, to obtain a prediction model for identifying facial features by using landmark point coordinates. The prediction module evaluates the facial features corresponding to the three-dimensional coordinates of each landmark point in the three-dimensional facial photo to be predicted, which includes the sagittal and vertical evaluation results of the face. 4.The machine learning based three-dimensional malocclusion identification system of claim 3, wherein, The sample processing module is configured with a computer program for unifying the three-dimensional coordinates of each landmark point in a coordinate system, and the computer program is run to perform the following method: The origin of the new coordinate system is set as the subnasale point; A reference plane A is established: the reference plane A is established by using the left and right antihelical tragus points and the right alar point; A horizontal plane is established: the horizontal plane of the new coordinate system is established by rotating the reference plane A by 7.5° about the left and right antihelical tragus points and passing through the subnasale point; A sagittal plane is established: the sagittal plane is established by passing through the midpoint of the left and right antihelical tragus points and the subnasale point, and is perpendicular to the horizontal plane; A coronal plane is established: the coronal plane is perpendicular to the horizontal plane and the sagittal plane; Coordinate system transformation: the coordinates in the original coordinate system are converted to the new coordinate system.

Images