Orthodontic image classification model determination system based on deep learning

Abstract

This invention discloses a deep learning-based orthodontic image classification model determination system, relating to the field of image data technology. The system includes a construction module for building an orthodontic data annotation library, comprising multiple tooth overlay models and multiple maxillary / mandibular overlay models, and for annotating orthodontic sample images according to the library. A training module, connected to the construction module, determines the feature information of the annotated orthodontic sample images, selects a deep learning model based on the feature information, and trains the selected deep learning model using the annotated orthodontic sample images to obtain a trained deep learning model. This invention enables data integration and annotation of numerous dental problems through the orthodontic data annotation library, achieving extensive data coverage and ensuring the accuracy and consistency of image annotation. It also enables rapid annotation of oral cavity-related images, improving the efficiency of orthodontic analysis.

Description

Technical Field

[0001] This invention relates to the field of image data technology, and more specifically to a system for determining orthodontic image classification models based on deep learning. Background Technology

[0002] With the development of society and the economy and the improvement of people's living standards, the public is paying increasing attention to oral health and aesthetics. More and more people realize that orthodontic treatment can not only improve problems such as malocclusion, but also have a positive impact on chewing function, pronunciation, and facial aesthetics. Therefore, the demand for orthodontic treatment is constantly rising. In the preparation process for orthodontic treatment, it is necessary to analyze the condition of the patient's teeth and upper and lower jaws. The basis of the analysis is oral X-rays and images of the inside of the mouth. As an important branch of artificial intelligence, deep learning has achieved remarkable results in image recognition, classification, and other fields. Currently, deep learning is also involved in orthodontic assisted diagnosis. On this basis, accurate image annotation is the foundation of model training. However, the annotation of orthodontic images requires professional orthodontists or experts in related fields. The annotation process is time-consuming and laborious, and there may be inconsistencies in the annotation. Summary of the Invention

[0003] The purpose of this invention is to provide a deep learning-based orthodontic image classification model determination system to address the shortcomings of the prior art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a deep learning-based orthodontic image classification model determination system, comprising:

[0005] The construction module is used to build an orthodontic data annotation library, which includes multiple tooth variation models and multiple maxillary and mandibular variation models. Orthodontic sample images are annotated according to the orthodontic data annotation library.

[0006] The training module, connected to the construction module, is used to determine the labeled feature information, select a deep learning model based on the feature information, and train the selected deep learning model using labeled orthodontic sample images to obtain a trained deep learning model.

[0007] The evaluation module, connected to the training module, is used to evaluate the trained deep learning model using a validation dataset. Once the accuracy of the trained deep learning model meets the preset conditions, it is used as the target model.

[0008] In a preferred embodiment, the building module includes:

[0009] Construction units are used to construct standard orthodontic models, which include standard models of the upper and lower jaws;

[0010] The associated unit is used to independently establish the association relationship between teeth in the maxillary and mandibular dentitions for individual tooth models in the standard orthodontic model.

[0011] The configuration unit is used to configure a space sphere for each standard tooth model corresponding to a single tooth in a standard orthodontic model. Multiple tooth transformation models are obtained by stacking a single tooth according to the space sphere.

[0012] The positional variation unit is used to perform positional variation of the upper and lower jaws in a standard orthodontic model to obtain multiple upper and lower jaw variation models.

[0013] The binding unit is used to bind multiple tooth overlay models and multiple maxillary overlay models to standard orthodontic models respectively, and to perform image annotation on multiple tooth overlay models and multiple maxillary overlay models respectively to obtain an orthodontic data annotation library.

[0014] The annotation unit is used to annotate orthodontic sample images according to the orthodontic data annotation library.

[0015] In a preferred embodiment, the building unit includes:

[0016] The first model building unit is used to collect the appearance information of the standard upper and lower jaws and to build a three-dimensional model of the upper and lower jaws based on the appearance information.

[0017] The second model building unit is used to collect the appearance information of standard teeth and build a three-dimensional model of each tooth based on the appearance information of standard teeth.

[0018] The combined construction unit is used to combine the three-dimensional models of all individual teeth with the corresponding three-dimensional models of the upper and lower jaws to construct an orthodontic model;

[0019] The contour binding unit is used to cover multiple adjustment points on the outer contour surface of the orthodontic model, and bind the adjustment points to the outer contour surface of the orthodontic model to obtain a standard orthodontic model.

[0020] In a preferred embodiment, the associated unit includes:

[0021] Independent units are used to separate the 3D models of all individual teeth in a standard orthodontic model from the 3D models of the upper and lower jaws.

[0022] The arrangement and association unit is used to sequentially associate the three-dimensional models of individual teeth in the maxillary and mandibular dentition according to the standard tooth arrangement order.

[0023] In a preferred embodiment, the configuration unit includes:

[0024] The copying unit is used to keep the standard orthodontic model unchanged and copy a standard orthodontic model as an adjustment model. The adjustment model and the standard orthodontic model are in an overlapping state.

[0025] The spatial configuration unit is used to configure a spatial sphere for each standard tooth model in the adjustment model. In the spatial sphere, the standard tooth model of a single tooth is angularly transformed according to the root point and the preset orientation to obtain multiple transformed standard tooth models.

[0026] The transformation unit is used to configure the corresponding model space for multiple transformed standard tooth models. The model spaces of the multiple transformed standard tooth models are interconnected. The multiple transformed standard tooth models are superimposed through the model space according to the root point to obtain multiple tooth superimposed models.

[0027] In a preferred embodiment, the space configuration unit includes:

[0028] The delineation unit is used to delineate a spatial sphere of a preset diameter for each standard tooth model, with the lowest point of the tooth root of each standard tooth model as the root point and the root point as the center point.

[0029] The transformation unit is used to define a preset orientation in a space sphere for the standard tooth model corresponding to each tooth, and to perform angular transformation of the standard tooth model within the preset orientation according to a preset difference angle using the root point as the rotation point, so as to obtain multiple transformed standard tooth models.

[0030] In a preferred embodiment, the annotation unit includes:

[0031] The sample construction unit is used to construct a three-dimensional model from orthodontic sample images to obtain the sample model.

[0032] The attachment adjustment unit is used to attach the contour of the sample model to the adjustment model by adjusting the adjustment points on the adjustment model, adjust the contour of the adjustment model, and make the adjustment model coincide with the sample model.

[0033] The matching unit is used to obtain the transformed maxillary and mandibular models and the transformed standard tooth models in the adjusted models with the highest degree of overlap in the corresponding sample models.

[0034] The image annotation unit is used to use the image annotations of the transformed maxillary and mandibular models and the transformed standard tooth model as the image annotations of the sample model.

[0035] In a preferred embodiment, the training module includes:

[0036] The information acquisition unit is used to determine the feature information of the corresponding orthodontic sample image after annotation, wherein the feature information is the data size of the orthodontic sample image;

[0037] The selection unit is used to select a deep learning model based on feature information.

[0038] The training unit is used to divide the labeled orthodontic sample images into a training dataset and a validation dataset. The selected deep learning model is trained using the training dataset to obtain a trained deep learning model.

[0039] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0040] This invention enables data integration and annotation of numerous dental problems through an orthodontic data annotation library, achieving extensive data coverage and ensuring the accuracy and consistency of image annotation. It also allows for rapid annotation of oral images of individuals, improving the efficiency of orthodontic analysis. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0042] Figure 1 This is a system block diagram of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Example 1, please refer to Figure 1 As shown in this embodiment, the orthodontic image classification model determination system based on deep learning includes:

[0045] The construction module is used to build an orthodontic data annotation library, which includes multiple tooth variation models and multiple maxillary and mandibular variation models. Orthodontic sample images are annotated according to the orthodontic data annotation library.

[0046] The training module, connected to the construction module, is used to determine the labeled feature information, select a deep learning model based on the feature information, and train the selected deep learning model using labeled orthodontic sample images to obtain a trained deep learning model.

[0047] The evaluation module, connected to the training module, is used to evaluate the trained deep learning model using a validation dataset. Once the accuracy of the trained deep learning model meets the preset conditions, it is used as the target model.

[0048] In one embodiment, the building module includes:

[0049] Construction units are used to construct standard orthodontic models, which include standard models of the upper and lower jaws;

[0050] The associated unit is used to independently establish the association relationship between teeth in the maxillary and mandibular dentitions for individual tooth models in the standard orthodontic model.

[0051] The configuration unit is used to configure a space sphere for each standard tooth model corresponding to a single tooth in a standard orthodontic model. Multiple tooth transformation models are obtained by stacking a single tooth according to the space sphere.

[0052] The positional variation unit is used to perform positional variation of the upper and lower jaws in a standard orthodontic model to obtain multiple upper and lower jaw variation models.

[0053] The binding unit is used to bind multiple tooth overlay models and multiple maxillary overlay models to standard orthodontic models respectively, and to perform image annotation on multiple tooth overlay models and multiple maxillary overlay models respectively to obtain an orthodontic data annotation library.

[0054] The annotation unit is used to annotate orthodontic sample images according to the orthodontic data annotation library.

[0055] In one embodiment, the building unit includes:

[0056] The first model building unit is used to collect the appearance information of the standard upper and lower jaws and to build a three-dimensional model of the upper and lower jaws based on the appearance information.

[0057] The second model building unit is used to collect the appearance information of standard teeth and build a three-dimensional model of each tooth based on the appearance information of standard teeth.

[0058] The combined construction unit is used to combine the three-dimensional models of all individual teeth with the corresponding three-dimensional models of the upper and lower jaws to construct an orthodontic model;

[0059] The contour binding unit is used to cover multiple adjustment points on the outer contour surface of the orthodontic model, and bind the adjustment points to the outer contour surface of the orthodontic model to obtain a standard orthodontic model.

[0060] In one embodiment, the associated unit includes:

[0061] Independent units are used to separate the 3D models of all individual teeth in a standard orthodontic model from the 3D models of the upper and lower jaws.

[0062] The arrangement and association unit is used to sequentially associate the three-dimensional model of a single tooth in the maxillary and mandibular dentition according to the standard tooth arrangement order;

[0063] In one embodiment, the configuration unit includes:

[0064] The copying unit is used to keep the standard orthodontic model unchanged and copy a standard orthodontic model as an adjustment model. The adjustment model and the standard orthodontic model are in an overlapping state.

[0065] The spatial configuration unit is used to configure a spatial sphere for each standard tooth model in the adjustment model. In the spatial sphere, the standard tooth model of a single tooth is angularly transformed according to the root point and the preset orientation to obtain multiple transformed standard tooth models.

[0066] The transformation unit is used to configure the corresponding model space of multiple transformed standard tooth models respectively. The model spaces of multiple transformed standard tooth models are interconnected. The multiple transformed standard tooth models are superimposed through the model space according to the root point to obtain multiple tooth superimposed models.

[0067] In one embodiment, the space configuration unit includes:

[0068] The delineation unit is used to delineate a spatial sphere of a preset diameter for each standard tooth model, with the lowest point of the tooth root of each standard tooth model as the root point and the root point as the center point.

[0069] The transformation unit is used to define a preset orientation in a space sphere for the standard tooth model corresponding to each tooth, and to perform angular transformation of the standard tooth model within the preset orientation according to a preset difference angle using the root point as the rotation point, so as to obtain multiple transformed standard tooth models.

[0070] It's important to note that orthodontic treatment involves not only tooth alignment and movement but also the morphology and positional relationship of the maxilla and mandible. For example, some orthodontic problems may be caused by developmental discrepancies between the maxilla and mandible, such as underbite (mandibular protrusion) and overbite (maxillary protrusion). In these cases, observing and analyzing images of the maxilla and mandible is crucial for accurate diagnosis and treatment planning. Therefore, orthodontic images include X-rays and CT scans of the maxilla and mandible to comprehensively assess the orthodontic problem. Detailed images of the teeth are also an important component, used to observe their morphology, position, inclination, and occlusal relationship, such as periapical radiographs and panoramic radiographs. This provides comprehensive analytical data for orthodontic treatment.

[0071] To better annotate the image samples used in training the model, a standard orthodontic model needs to be constructed. Specifically: First, standard maxillary and mandibular appearance information is collected, including the shape, contour, and size of the maxilla and mandible. Based on this appearance information, a 3D model of the maxilla and mandible is constructed using a 3D reconstruction algorithm. Second, standard tooth appearance information is collected, including the shape, contour, and size of the standard teeth. Based on this appearance information, a 3D model of each tooth is constructed using a 3D reconstruction algorithm. All the individual tooth 3D models are then combined with the corresponding maxillary and mandibular 3D models to form the orthodontic model. This combination is not a fused model, but rather a combination of models, facilitating subsequent positional transformations of the teeth and jaws. This model is an active, assembled model. Next, multiple adjustment points are covered on the outer contour surface of the orthodontic model. These adjustment points are then bound to the outer contour surface of the orthodontic model to obtain the standard orthodontic model. These adjustment points are attachment points. When annotating actual orthodontic sample images later, these adjustment points allow the standard orthodontic model to fit the contours of the maxilla and teeth in the orthodontic sample images. The adjustment points enable contour adjustments to the standard orthodontic model, demonstrating good adaptability.

[0072] In the standard orthodontic model, the 3D models of all individual teeth are decoupled from the 3D models of the upper and lower jaws (the 3D models of individual teeth and the 3D models of the upper and lower jaws are separate models; the relationship between them is a model combination used for subsequent activity transformations, improving the adaptability of subsequent image annotation). Within the maxillary and mandibular dentitions, the 3D models of individual teeth are sequentially associated according to the standard tooth arrangement order. Maxillary teeth: commonly referred to as the maxillary dentition. From the center outwards, they are: maxillary central incisor, maxillary lateral incisor, maxillary canine, maxillary first premolar, maxillary second premolar, maxillary first molar, maxillary second molar, and maxillary third molar. Mandibular teeth: referred to as the mandibular dentition. From the center outwards, they are: mandibular central incisor, mandibular lateral incisor, mandibular canine, mandibular first premolar, mandibular second premolar, mandibular first molar, mandibular second molar, and mandibular third molar. Because the teeth are arranged in an ordered manner, a sequential association of the teeth is necessary.

[0073] The standard orthodontic model remains unchanged during use. Here, a standard orthodontic model is copied from the standard model to serve as an adjustment model. The adjustment model and the standard orthodontic model are aligned. Subsequent iterations are set for the adjustment model, which also has adjustment points. The adjustment model is essentially the standard orthodontic model, serving as the transformable entity. The standard orthodontic model remains constant. The lowest point of the root of each standard tooth model in the adjustment model is taken as the root point. A space sphere of a preset diameter is defined with the root point as the center point for each standard tooth model. This space sphere can be used for changes in tooth angles to cover existing tooth inclinations, occlusal relationships, etc., allowing for the use of a large database. A preset orientation is defined within the space sphere for each standard tooth model. This preset orientation represents the range of angles involved in the current tooth inclinations. Using the root point as the rotation point, the standard tooth model is angularly transformed within the preset orientation according to preset difference angles. These preset difference angles are achieved by meshing the surface of the space sphere. The process involves uniformly dividing the space into grids corresponding to the surface of a sphere in a preset orientation, using these grids as transformation grids. The center point of each grid is used as the angle of transformation, and the preset difference angle is based on the difference angle between the center points of each grid, resulting in multiple transformed standard tooth models. These transformed standard tooth models are then configured with corresponding model spaces, allowing them to be stored in different model spaces. These model spaces are cloud servers configured for each transformed standard tooth model, and multiple cloud servers are connected. Using the root point as a reference point, the multiple transformed standard tooth models are overlapped (a process of overlay). The model spaces of the multiple transformed standard tooth models are interconnected, and the models are overlaid through the model spaces according to the root point, resulting in multiple overlay tooth models. This allows for data integration and annotation of various dental problems through an orthodontic data annotation library, achieving large-scale data coverage and ensuring the accuracy and consistency of image annotation. It also enables rapid annotation of oral images, improving the efficiency of orthodontic analysis.

[0074] Multiple mandibular overlapping models are obtained by altering the position of the upper and lower jaws in the standard orthodontic model. Specifically, the upper and lower jaws are transformed in the adjustment model, such as underbite (mandibular protrusion) and overbite (maxillary protrusion), resulting in multiple transformed mandibular and upper jaw models. Corresponding model spaces are configured for the multiple transformed mandibular and upper jaw models. The model spaces of the multiple transformed mandibular and upper jaw models are interconnected, and the points connecting the upper and lower jaws are used as reference points for coincidence, thus obtaining multiple mandibular overlapping models.

[0075] Multiple tooth folding models and multiple maxillary folding models were bound to standard orthodontic models respectively. Image annotations were then performed on the multiple tooth folding models and multiple maxillary folding models to obtain an orthodontic data annotation library. Image annotation refers to the process of marking and annotating orthodontic images. Professional orthodontists or experts in related fields classify images into different orthodontic types or annotate specific feature information based on the tooth condition and orthodontic features presented in the images, such as different orthodontic conditions such as crowded teeth, sparse teeth, deep overbite, and open bite. This provides accurate labels and classification basis for the subsequent training of deep learning models, enabling the subsequent models to learn the relationship between image features and corresponding annotations.

[0076] In one embodiment, the annotation unit includes:

[0077] The sample construction unit is used to construct a three-dimensional model from orthodontic sample images to obtain the sample model.

[0078] The attachment adjustment unit is used to attach the contour of the sample model to the adjustment model by adjusting the adjustment points on the adjustment model, adjust the contour of the adjustment model, and make the adjustment model coincide with the sample model.

[0079] The matching unit is used to obtain the transformed maxillary and mandibular models and the transformed standard tooth models in the adjusted models with the highest degree of overlap in the corresponding sample models.

[0080] Image annotation unit, used to use the image annotations of the transformed maxillary and mandibular models and the transformed standard tooth model as the image annotations of the sample model;

[0081] It should be noted that after extracting information from the orthodontic sample images, a 3D model is constructed to obtain the sample model. Specifically, this process involves understanding the morphology using X-rays of the upper and lower jaws and teeth, and obtaining comprehensive contour information of the upper and lower jaws and teeth through CT scans and intraoral images. This contour information is then used to construct the sample model. Next, adjustment points are used to attach the sample model's contour, initially aligning it with the adjustment model. These adjustment points are then attached to the sample model's contour. Further adjustments are made to the adjustment model, which then overlaps with the standard orthodontic model. While the standard orthodontic model remains stationary, the background process matches and modifies the adjustment model. The adjustment model's contour is adjusted to overlap with the sample model. Each model in the adjustment model is an independent combination, thus allowing for changes based on the adjustment points. The adjustment model is essentially a parts library of multiple independent and single models combined with different forms. It does not require building a comprehensive database of the upper and lower jaws and teeth for multiple situations. Here, the upper and lower jaws and a single tooth are treated as small parts. This is equivalent to a combination of multiple small parts with different forms, inclinations, and positions. These small parts are integrated by their corresponding positions to the whole. It is not necessary to list the various tooth states separately, which can reduce the complexity of model construction. At the same time, it can store a large amount of data. It can be independent yet interconnected and integrated. It can obtain the adjusted model with the highest overlap with the sample model, including the transformed upper and lower jaw models and the transformed standard tooth models. Then, the image annotations of the transformed upper and lower jaw models and the transformed standard tooth models are used as the image annotations of the sample models.

[0082] In one embodiment, the training module includes:

[0083] The information acquisition unit is used to determine the feature information of the corresponding orthodontic sample image after annotation, wherein the feature information is the data size of the orthodontic sample image;

[0084] The selection unit is used to select a deep learning model based on feature information.

[0085] The training unit is used to divide the labeled orthodontic sample images into a training dataset and a validation dataset. The selected deep learning model is trained using the training dataset to obtain a trained deep learning model.

[0086] It should be noted that the feature information after determining the annotation of the corresponding orthodontic sample images refers to the data scale of the orthodontic sample images, and the deep learning models used are ResNet and VGGNet. ResNet is used when the orthodontic image dataset is large and the image content is complex. It introduces a residual block structure, which effectively solves the gradient vanishing and gradient exploding problems in deep neural networks, allowing the network to be trained to be deeper. For example, on datasets containing a large number of orthodontic X-rays and intraoral photographs from different angles and under different lighting conditions, ResNet can learn richer and more abstract features through its deep structure, thereby improving classification accuracy. VGGNet is a suitable choice for orthodontic image datasets when they are relatively small. VGGNet has a simple and uniform network structure, consisting of multiple convolutional and pooling layers. It converges quickly on small datasets and can learn the basic features of the images. The labeled orthodontic sample images are divided into training and validation datasets. A selected deep learning model is trained on the training dataset to obtain a trained deep learning model. The trained deep learning model is evaluated on the validation dataset. When the accuracy of the trained deep learning model meets a preset condition, it is used as the target model. For example, in an orthodontic image classification task, assuming the validation dataset has 1000 images, of which 600 are in the positive class (images requiring orthodontic treatment) and 400 are in the negative class (images not requiring orthodontic treatment), if the model correctly predicts 500 positive images and 300 negative images, the accuracy is 80%. Setting the accuracy at 80% is considered acceptable, and the model is used as the target model, demonstrating good model determination and training capabilities.

[0087] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A system for determining orthodontic image classification models based on deep learning, characterized in that, include: The construction module is used to build an orthodontic data annotation library, which includes multiple tooth variation models and multiple maxillary and mandibular variation models. Orthodontic sample images are annotated according to the orthodontic data annotation library. The building module includes: Construction units are used to construct standard orthodontic models, which include standard models of the upper and lower jaws; The associated unit is used to independently establish the association relationship between teeth in the maxillary and mandibular dentitions for individual tooth models in the standard orthodontic model. The configuration unit is used to configure a space sphere for each standard tooth model corresponding to a single tooth in a standard orthodontic model. Multiple tooth transformation models are obtained by stacking a single tooth according to the space sphere. The positional variation unit is used to perform positional variation of the upper and lower jaws in a standard orthodontic model to obtain multiple upper and lower jaw variation models. The binding unit is used to bind multiple tooth overlay models and multiple maxillary overlay models to standard orthodontic models respectively, and to perform image annotation on multiple tooth overlay models and multiple maxillary overlay models respectively to obtain an orthodontic data annotation library. The annotation unit is used to annotate orthodontic sample images according to the orthodontic data annotation library; The training module, connected to the construction module, is used to determine the labeled feature information, select a deep learning model based on the feature information, and train the selected deep learning model using labeled orthodontic sample images to obtain a trained deep learning model. The evaluation module, connected to the training module, is used to evaluate the trained deep learning model using a validation dataset. Once the accuracy of the trained deep learning model meets the preset conditions, it is used as the target model.

2. The orthodontic image classification model determination system based on deep learning according to claim 1, characterized in that: The building unit includes: The first model building unit is used to collect the appearance information of the standard upper and lower jaws and to build a three-dimensional model of the upper and lower jaws based on the appearance information. The second model building unit is used to collect the appearance information of standard teeth and build a three-dimensional model of each tooth based on the appearance information of standard teeth. The combined construction unit is used to combine the three-dimensional models of all individual teeth with the corresponding three-dimensional models of the upper and lower jaws to construct an orthodontic model; The contour binding unit is used to cover multiple adjustment points on the outer contour surface of the orthodontic model, and bind the adjustment points to the outer contour surface of the orthodontic model to obtain a standard orthodontic model.

3. The orthodontic image classification model determination system based on deep learning according to claim 1, characterized in that: The associated unit includes: Independent units are used to separate the 3D models of all individual teeth in a standard orthodontic model from the 3D models of the upper and lower jaws. The arrangement and association unit is used to sequentially associate the three-dimensional models of individual teeth in the maxillary and mandibular dentition according to the standard tooth arrangement order.

4. The orthodontic image classification model determination system based on deep learning according to claim 1, characterized in that: The configuration unit includes: The copying unit is used to keep the standard orthodontic model unchanged and copy a standard orthodontic model as an adjustment model. The adjustment model and the standard orthodontic model are in an overlapping state. The spatial configuration unit is used to configure a spatial sphere for each standard tooth model in the adjustment model. In the spatial sphere, the standard tooth model of a single tooth is angularly transformed according to the root point and the preset orientation to obtain multiple transformed standard tooth models. The transformation unit is used to configure the corresponding model space for multiple transformed standard tooth models. The model spaces of the multiple transformed standard tooth models are interconnected. The multiple transformed standard tooth models are superimposed through the model space according to the root point to obtain multiple tooth superimposed models.

5. The orthodontic image classification model determination system based on deep learning according to claim 4, characterized in that: The spatial configuration unit includes: The delineation unit is used to delineate a spatial sphere of a preset diameter for each standard tooth model, with the lowest point of the tooth root of each standard tooth model as the root point and the root point as the center point. The transformation unit is used to define a preset orientation in a space sphere for the standard tooth model corresponding to each tooth, and to perform angular transformation of the standard tooth model within the preset orientation according to a preset difference angle using the root point as the rotation point, so as to obtain multiple transformed standard tooth models.

6. The orthodontic image classification model determination system based on deep learning according to claim 5, characterized in that: The annotation unit includes: The sample construction unit is used to construct a three-dimensional model of orthodontic sample images to obtain the sample model. The attachment adjustment unit is used to attach the contour of the sample model to the adjustment model by adjusting the adjustment points on the adjustment model, adjust the contour of the adjustment model, and make the adjustment model coincide with the sample model. The matching unit is used to obtain the transformed maxillary and mandibular models and the transformed standard tooth models in the adjusted models with the highest degree of overlap in the corresponding sample models. The image annotation unit is used to use the image annotations of the transformed maxillary and mandibular models and the transformed standard tooth model as the image annotations of the sample model.

7. The orthodontic image classification model determination system based on deep learning according to claim 1, characterized in that: The training module includes: The information acquisition unit is used to determine the feature information of the corresponding orthodontic sample image after annotation, wherein the feature information is the data size of the orthodontic sample image; The selection unit is used to select a deep learning model based on feature information. The training unit is used to divide the labeled orthodontic sample images into a training dataset and a validation dataset. The selected deep learning model is trained using the training dataset to obtain a trained deep learning model.

Images