Method for identifying abnormal teeth in dental model, electronic equipment and storage medium

Abstract

The embodiment of the invention relates to the field of orthodontic digital design, and discloses a method for identifying abnormal teeth in a dental model, electronic equipment and a storage medium. The method for recognizing the abnormal teeth in the dental model comprises the steps that a training sample library is obtained, the sample library comprises a plurality of digital dental models, preset features of all the dental models in the sample library are trained through a machine learning algorithm, a recognition model is obtained, and the preset features at least comprise feature parameters representing the size of a dental crown; and according to the trained identification model, identifying the tooth to be identified, and determining whether the tooth is an abnormal tooth. According to the method for identifying the abnormal teeth in the dental model, provided by the embodiment of the invention, model training and identification are carried out by adopting the volume-related characteristics, a good effect is obtained, and the identification accuracy of the abnormal teeth in the dental model is improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of dental orthodontics, and in particular to a method for identifying abnormal teeth in a dental model, an electronic device, and a storage medium. Background Art

[0002] In the process of orthodontic digital design, the design basis is the patient's dental model. In the process of establishing the dental model, it is necessary to mark the tooth classification and whether there are abnormal teeth, such as whether it is caries or whether it is a tooth with attachments. Among them, caries teeth will be treated as immovable teeth in orthodontic design, which is important information in orthodontic digital design. At present, it is only judged by the naked eye, which is inefficient. There are also some automated judgment methods, mainly based on two-dimensional images, including CBCT images, X-ray panoramic films, and optical intraoral photos. The recognition basis is related to the direction of the teeth collected in the two-dimensional image. The data features collected in different directions are quite different, so the accuracy is not high when used for automatic recognition.

[0003] The dental models used in orthodontic digital design are generally tooth models without accessories. If the patient is already in the orthodontic process and needs to adjust or change the plan, the accessories have been pasted on the tooth surface, and because it is not yet certain whether the plan needs to be changed, the accessories on the patient's teeth cannot be removed. Therefore, the dental model obtained by oral scanning is a dental model with accessories. However, the accessories pasted on the tooth surface affect the tooth contour and affect the orthodontic design. The current method of identifying accessories is mainly through manual visual judgment, which is inefficient, so an automatic, efficient and accurate identification method is needed.

[0004] There are currently no accurate and effective methods for the above two recognition scenarios. Manual recognition is not only inefficient, but may also have large errors due to fatigue. In addition, there is a large demand in orthodontic scenarios, so it is necessary to further develop an automatic, efficient and accurate recognition method. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a method for identifying abnormal teeth in a dental model, an electronic device and a storage medium, which can improve the accuracy of identifying abnormal teeth in a dental model.

[0006] To solve the above technical problems, an embodiment of the present application provides a method for identifying abnormal teeth in a dental model, comprising: obtaining a training sample library, the sample library including multiple digital dental models, using a machine learning algorithm to train preset features of each dental model in the sample library to obtain a recognition model, wherein the preset features at least include: characteristic parameters characterizing the volume of the crown, each tooth in each dental model in the sample library is marked with a label indicating whether it is an abnormal tooth; identifying the tooth to be identified according to the trained recognition model to confirm whether it is an abnormal tooth.

[0007] An embodiment of the present application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for identifying abnormal teeth in a dental model as described above.

[0008] An embodiment of the present application further provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-mentioned method for identifying abnormal teeth in a dental model.

[0009] In the method for identifying abnormal teeth in the dental model of the present application, volume-related features are used for model training, and the recognition model obtained by training is used for recognition. Since the inventors of the present application found that abnormal teeth vary greatly in volume, the volume features in the three-dimensional tooth model are used for recognition, which has achieved good results. In addition, the recognition method can be more automated, which speeds up the recognition efficiency.

[0010] In addition, the characteristic parameters characterizing the crown volume include: crown spatial scale characteristics and / or crown mesh vertex spatial distribution characteristics. Further limiting the specific parameter types of volume-related parameters, through spatial scale or vertex distribution calculation, not only is the calculation simple, but also makes the identification accurate and effective.

[0011] In addition, the characteristic parameters characterizing the volume of the crown are obtained by using a bounding box algorithm and a software extraction method. The use of a bounding box algorithm and a software extraction method can simplify the calculation of volume-related parameters and speed up the execution of the overall algorithm.

[0012] In addition, the spatial distribution characteristics of the crown mesh vertices are obtained by: establishing a three-dimensional bounding box of a single crown to be identified, wherein the center of the bounding box is the crown center, and each face of the bounding box is perpendicular to one of the three tooth axes of the crown; dividing the bounding box into several equal parts along the three tooth axes to obtain multiple sub-bounding boxes; calculating the ratio of the number of mesh vertices of the single crown in each sub-bounding box to the total number of vertices of the crown; splicing the ratio of the number of mesh vertices in all sub-bounding boxes to the total number of vertices, and using the splicing result as the spatial distribution characteristics of the crown mesh vertices. The specific method of calculating the spatial distribution characteristics of the crown mesh vertices through the bounding box algorithm is accurate and effective, and is easy to promote and apply.

[0013] In addition, the said dividing the bounding box into several equal parts along the three tooth axis directions includes: the number of equal parts along the three tooth axis directions is the same. Limiting the use of the same number of equal parts in different directions can simplify the calculation and speed up the overall speed of algorithm execution.

[0014] In addition, the said dividing the bounding box into a number of equal parts along the three tooth axis directions includes: the number of equal parts in at least one tooth axis direction is greater than or equal to 2. It is limited that the bounding box in at least one direction is divided into at least two parts, so that the number of sub-bounding boxes is sufficient when calculating the vertex distribution, so that the distribution of the vertices is distinguishable.

[0015] In addition, the dividing the bounding box into a number of equal parts along the three tooth axis directions includes: if the length of the bounding box in a tooth axis direction is less than a first threshold, the number of equal parts on the tooth axis is 1. Some embodiments limit the direction with too small size to not be divided, so as to increase the difference in vertex distribution in each sub-bounding box when the size in each direction is different.

[0016] In addition, the first threshold is 1 / 3 of the size of the non-abnormal tooth on the corresponding tooth axis. The ratio of the size to the normal tooth is used to determine whether division is not required, so as to adapt to teeth of different sizes at different positions, and facilitate the accurate implementation of the identification method in this application.

[0017] In addition, the method for acquiring the crown spatial scale feature includes: establishing a three-dimensional bounding box of a single crown to be identified, wherein the center of the bounding box is the crown center, and each surface of the bounding box is perpendicular to one of the three tooth axes of the crown; acquiring the length, width and height of the bounding box as the crown spatial scale feature. The crown spatial scale is determined by establishing a three-dimensional bounding box, so that the determination process is simple and fast.

[0018] In addition, the preset features also include: mesh normal features, and / or curvature distribution features. It is limited that more feature types can be used as preset features, so as to combine multiple features to further make the recognition result more accurate.

[0019] In addition, the curvature distribution feature is obtained by: calculating the curvature of each grid on the single tooth crown to be identified; statistically analyzing the histogram distribution of each curvature; and using the histogram distribution as the curvature distribution feature.

[0020] In addition, when the preset features include multiple features, each feature is combined in a splicing manner; using a machine learning algorithm to train the preset features in the sample library includes: using a machine learning algorithm to train the spliced ​​preset features of each dental model in the sample library. It is further defined that the multiple features are combined in a splicing manner so that the details of each feature are more completely retained, and the recognition result is more accurate and reliable.

[0021] In addition, the machine learning algorithm includes one of the following: support vector machine, XGBoost, random forest, multi-layer perceptron. It is further limited that the model can be trained by different machine learning algorithms, so that different types of algorithm models can be selected for different needs to adapt to the needs of different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] One or more embodiments are exemplarily described by pictures in the corresponding drawings, and these exemplified descriptions do not constitute limitations on the embodiments. Elements with the same reference numerals in the drawings represent similar elements, and unless otherwise stated, the figures in the drawings do not constitute proportional limitations.

[0023] Figure 1 is a flow chart of a method for identifying abnormal teeth in a dental model provided according to an embodiment of the present application;

[0024] Figure 2 It is a flow chart of a method for acquiring spatial distribution features of tooth crown mesh vertices in a method for identifying abnormal teeth in a dental model provided according to an embodiment of the present application;

[0025] Figure 3a It is a schematic diagram of a tooth crown mesh model in an abnormal tooth identification method in a dental model provided by an embodiment of the present application;

[0026] Figure 3b It is a schematic diagram of a bounding box established according to a tooth crown mesh model in a method for identifying abnormal teeth in a dental model provided by an embodiment of the present application;

[0027] Figure 4 is a flow chart of a method for acquiring the spatial scale feature of a tooth crown in a method for identifying abnormal teeth in a dental model provided according to another embodiment of the present application;

[0028] Figure 5 is a flow chart of a method for obtaining mesh normal features in a method for identifying abnormal teeth in a dental model provided according to another embodiment of the present application;

[0029] Figure 6 is a schematic diagram of the distribution characteristics of the curvature histogram obtained in the dental model provided according to another embodiment of the present application;

[0030] Figure 7 It is a structural schematic diagram of an electronic device provided according to another embodiment of the present application. DETAILED DESCRIPTION

[0031] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below in conjunction with the accompanying drawings. However, it will be appreciated by those skilled in the art that in the present application, many technical details are proposed in order to enable the reader to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical scheme claimed in the present application can also be implemented. The division of the following embodiments is for the convenience of description, and the specific implementation of the present application should not be construed as any limitation, and the various embodiments can be combined and referenced with each other under the premise of no contradiction.

[0032] One embodiment of the present application relates to a method for identifying abnormal teeth in a dental model, which is applied to an electronic device, wherein the electronic device may be a computer terminal or a server. In this embodiment and the following embodiments, the electronic device is described by taking a computer terminal as an example. The following is a specific description of the implementation details of the method for identifying abnormal teeth in a dental model of this embodiment. The following content is only provided for the convenience of understanding and is not necessary for the implementation of this solution.

[0033] In this embodiment, the abnormal tooth is a carious tooth. The specific process can be as follows: Figure 1 As shown, including:

[0034] Step 101: Obtain a training sample library.

[0035] Specifically, the sample library obtained in this step includes multiple digital dental models, and each tooth in each dental model in the sample library is marked with a label indicating whether it is caries. Further, the sample library in this step can come from historical case data, and the process of processing the historical case data can include using a large amount of case data previously accumulated as samples, screening out teeth marked as caries to form a caries tooth set, and screening out non-caries teeth to form a non-caries tooth set. The two types of labels can be represented by labels 1 and 0 respectively. The specific representation method of the label can be set according to actual application requirements without affecting the protection scope.

[0036] In some embodiments, in order to ensure that the two sets of data are balanced, the amount of data in the two data sets can be limited to be roughly equal, that is, the total number of caries and the total number of non-caries are roughly equal, and the number of teeth of each type can also be limited to be equal. For example, if the total number of all teeth in the historical case data is N, then the total number of caries and the total number of non-caries are both N / 2. Calculated as 32 teeth for an adult, the number of samples in the data set corresponding to each tooth in the caries data or non-caries data is approximately N / 2 / 32=N / 64. Furthermore, when the number of caries and non-caries teeth screened out is inconsistent, the data enhancement method can be used to increase the sample data in the data set with less data volume. The data enhancement here includes but is not limited to random scaling, grid smoothing, grid compression and other methods.

[0037] Step 102: Use a machine learning algorithm to train the preset features of each dental model in the sample library to obtain a recognition model.

[0038] Specifically, the preset feature in this embodiment can be a feature parameter that characterizes the volume of the crown. More specifically, in some embodiments, the feature parameter that characterizes the volume of the crown is selected from the spatial distribution feature of the crown mesh vertices. The spatial distribution feature of the crown mesh vertices can effectively characterize the spatial feature of each crown and accurately distinguish whether it is a carious tooth morphology. Further, in this embodiment, the spatial distribution feature of the crown mesh vertices can be calculated by a bounding box algorithm, such as Figure 2 As shown, the specific acquisition methods include:

[0039] Step 1021, establish a three-dimensional bounding box of a single crown to be identified, wherein the center of the bounding box is the center of the crown, and each face of the bounding box is perpendicular to one of the three axes of the crown. It can be understood that although the three axes of the crown are used as the length, width and height directions of the three-dimensional bounding box in this embodiment, in actual application, other directions can also be selected to set the bounding box, which does not affect the inventive concept of the present application, and the direction selection does not affect the protection scope of the present application. The schematic diagram of the crown mesh model 10 is shown in FIG. Figure 3a shown.

[0040] Step 1022, divide the bounding box into several equal parts along the three tooth axis directions, and obtain multiple sub-bounding boxes. Specifically, the calculation process of steps 1021-1022 includes: calculating the crown tooth axis and their 3D bounding boxes, and at the same time limiting each tooth axis to be perpendicular to these bounding boxes. In some embodiments, the bounding box is divided into several equal parts along the three tooth axis directions, specifically, the number of equal parts along the three tooth axis directions is the same. Limiting the use of the same number of equal parts in different directions can simplify calculations and speed up the overall execution of the algorithm. In other embodiments, the bounding box is divided into several equal parts along the three tooth axis directions, specifically, the number of equal parts in at least one tooth axis direction is greater than or equal to 2. Limit the bounding box in at least one direction to be divided into at least two parts, so that the number of sub-bounding boxes is sufficient when calculating the vertex distribution, so that the distribution of vertices is distinguishable. For example, such as Figure 3b As shown in the figure, a tooth crown mesh model 10 is used as an example to establish a bounding box and each sub-bounding box. In this embodiment, the direction from the lingual side to the buccal side is the z-axis direction, the direction from the root to the crown is the y-axis direction, and the mesiodistal direction is the x-axis direction. The bounding box is divided into k equal parts along the three-axis directions according to the distance, for example, k=3, so that the entire bounding box is divided into 27 sub-bounding boxes. It can be understood that the AABB (Axis-aligned bounding box) algorithm is used in this embodiment, which is more conducive to obtaining the distribution of each vertex. In actual applications, other bounding box algorithms can also be used, such as the OBB (Oriented bounding box) algorithm, the bounding sphere algorithm, etc., which are not listed here one by one.

[0041] Step 1023, calculate the ratio of the number of mesh vertices of a single crown in each sub-bounding box to the total number of vertices of the crown. Specifically, based on the three-dimensional model data of a single crown, confirm the positions of all mesh vertices, which belong to different sub-bounding boxes. Count the total number of mesh vertices X of the crown, and the number of mesh fixed points Xs1, Xs2, Xs3, etc. in each sub-bounding box, and continue to calculate the ratio of the number of mesh fixed points in each sub-bounding box to the total number of vertices of the crown, which are respectively: Xs1 / X, Xs2 / X, Xs3 / X, etc. Taking 27 sub-bounding boxes as an example, calculate the ratio of the number of mesh vertices in each sub-bounding box to the total number of vertices, and expand the 27 sub-bounding boxes into a 27-dimensional vector in a fixed order as a new feature.

[0042] Furthermore, if the length of the bounding box in a tooth axis direction is less than the first threshold, the number of equal parts on the tooth axis is 1. The first threshold can be 1 / 3 of the size of the non-abnormal tooth on the corresponding tooth axis. Determine whether division is not required by the size ratio with the normal tooth, so as to adapt to teeth of different sizes in different positions, and facilitate the accurate implementation of the recognition method in this embodiment. Some embodiments limit the direction of too small size to not be divided, so as to increase the difference in the distribution of vertices in each sub-bounding box when the size in each direction is different. In some embodiments, when confirming the number of crown mesh vertices in each sub-bounding box, the part of the sub-bounding box in the direction of the too small size is directly assigned a value of 0. For example, if the crown height is lower than the first threshold, the number of mesh vertices of the remaining two layers of sub-bounding boxes along the height is 0; if the crown height is greater than the first threshold but less than 2 times the first threshold, the number of mesh vertices of the remaining layer of sub-bounding boxes along the height is 0.

[0043] Step 1024 , concatenate the ratio of the number of mesh vertices in all sub-bounding boxes to the total number of vertices, and use the concatenation result as the spatial distribution feature of the mesh vertices of the tooth crown.

[0044] Specifically, splicing can be simpler and retain the original features of the data. For example, the splicing result can be s=[Xs1 / X, Xs2 / X, Xs3 / X...Xsn / X], where n is the number of sub-bounding boxes, such as 27 mentioned above.

[0045] The spatial distribution characteristics of the crown mesh vertices are obtained through the above steps 1021-1024, and the specific method of calculating the spatial distribution characteristics of the crown mesh vertices through the bounding box algorithm is limited, which is accurate, effective and easy to promote and apply.

[0046] After acquiring the features, a machine learning algorithm is used for training. In some embodiments, the machine learning algorithm may use a support vector machine algorithm. A support vector machine (SVM) is a generalized linear classifier that performs binary classification of data in a supervised learning manner, and its decision boundary is the maximum-margin hyperplane for solving the learning samples. Specifically, SVM can use a hinge loss function to calculate empirical risk and add a regularization term to the solution system to optimize structural risk.

[0047] It should be noted that in addition to the support vector machine algorithm used as the machine learning algorithm, algorithms such as XGBoost, LightGBM, and CatBoost can also be used. Among them, XGBoost is an optimized distributed gradient boosting library designed to achieve high efficiency, flexibility, and portability. It implements machine learning algorithms under the Gradient Boosting framework. LightGBM (Light Gradient Boosting Machine) is a distributed gradient boosting framework based on the decision tree algorithm. CatBoost is a GBDT framework based on symmetric decision trees (oblivious trees) as the base learner with fewer parameters, support for categorical variables, and high accuracy. CatBoost, XGBoost, and LightGBM are all improved implementations under the GBDT algorithm framework, which effectively improve the computational efficiency and algorithm accuracy of GBDT. It is further limited that different machine learning algorithms can be used for model training, so that different types of algorithm models can be selected for different needs to adapt to the needs of different application scenarios.

[0048] In addition to the above algorithms, other algorithms such as random forest and multi-layer perceptron can also be used, which are not listed here one by one.

[0049] Step 103: Identify the tooth to be identified according to the trained recognition model to determine whether it is an abnormal tooth.

[0050] Specifically, the tooth to be identified may come from a dental model in the case information, and may specifically be a single crown obtained after cutting the dental model.

[0051] It can be seen that the method for identifying abnormal teeth in the dental model in this embodiment uses volume-related features for model training, and uses the recognition model obtained by training for recognition. Since the inventors of this application found that the volume of abnormal teeth varies greatly, the volume features in the three-dimensional tooth model are used for recognition, which has achieved good results, and the recognition method can be more automated, which speeds up the recognition efficiency. In addition, in this embodiment, the spatial distribution features of the vertices of the crown mesh are used for feature training. Since the inventors of this application found that most caries are incomplete crowns, the spatial distribution of the triangular mesh or mesh vertices of caries will be very different from that of a complete crown. The spatial distribution features of the vertices of the crown mesh can effectively reflect the difference between caries and not caries. At the same time, the bounding box subdivision method is further used to describe this spatial distribution feature, so that the difference can be accurately expressed by numbers, making the calculation more direct and clear.

[0052] In addition, the bounding box algorithm is used in this embodiment to obtain characteristic parameters that characterize the crown volume. In practical applications, other methods can be used to obtain the parameters, such as software extraction, such as geomagic, mimics, etc. A segmentation method can also be used, specifically, the entire crown model is first divided into a volume mesh, such as using tetrahedral units for meshing, and then the volume of each tetrahedron is counted and superimposed to obtain the volume of the final crown model. It can be understood that there are many ways in the art to obtain various characteristic parameters that characterize the crown volume, which will not be listed here one by one.

[0053] Another embodiment of the present application relates to a method for identifying abnormal teeth in a dental model. The present embodiment is substantially the same as the first embodiment, with the main difference being that the spatial distribution features of the crown mesh vertices are used for feature training in the previous embodiment, while the spatial scale features of the crown are used for feature training in the present embodiment. Different features are used to characterize the crown volume so that the present application can be applied to different needs and adapt to more application scenarios.

[0054] Specifically, the spatial scale characteristics of the crown are obtained as follows Figure 4 As shown, specifically including:

[0055] Step 401, establish a three-dimensional bounding box of a single crown to be identified, wherein the center of the bounding box is the center of the crown, and each face of the bounding box is perpendicular to one of the three tooth axes of the crown. Specifically, the process of establishing a three-dimensional bounding box for a single crown in step 401 is substantially the same as step 1021, and will not be described in detail in this embodiment to avoid repetition.

[0056] Step 402, obtaining the length, width and height of the bounding box as the spatial scale features of the tooth crown.

[0057] Specifically, the inventor of the present application considers that the tooth body of some caries is seriously missing, so that the volume of caries is greatly reduced compared with that of intact teeth, so the length, width and height of the three-dimensional space bounding box of caries can be used to characterize the volume of caries. The specific features can be expressed as Vol = [a, b, h], where a, b, h represent the length, width and height of the bounding box respectively.

[0058] It can be seen that through steps 401-402, the spatial scale characteristics of the tooth crown can be accurately obtained, and the spatial scale of the tooth crown is determined by establishing a three-dimensional bounding box, making the determination process simple and fast.

[0059] Furthermore, after obtaining the spatial scale characteristics of the tooth crown, the process of model training and recognition is roughly the same as that of the previous implementation method, and will not be described in detail here.

[0060] It should also be explained that in addition to using characteristic parameters that characterize the crown volume as preset features when training the machine learning algorithm, other features can also be combined, such as mesh normal features and curvature distribution features. The above two features can also be combined at the same time to combine multiple features into new features.

[0061] Specifically, in some embodiments, the method for obtaining the mesh normal feature may include: obtaining the normals of all triangular facets on the crown model to be identified; obtaining the long axis direction of the crown model to be identified; calculating the first angle θ between each normal and the long axis direction, such as Figure 5 As shown, taking the triangular facet 1001 as an example, its normal is n, and the angle between n and the direction y from the root to the crown in the tooth axis is the first angle θ, and the first angle corresponding to all triangular facets is used as the normal feature of the crown model. In addition, in practical applications, the normal feature can use the distribution characteristics of the normal value in multiple intervals, specifically including the characteristics of obtaining the histogram distribution of all first angles in p intervals. The mesh normal feature can express the three-dimensional morphology of the crown mesh, and when the histogram distribution characteristics of the normal feature are further used, the morphological difference of whether it is caries can be further revealed.

[0062] Specifically, the method of obtaining the curvature distribution feature may include: calculating the curvature of each grid on the crown of a single tooth to be identified; statistically analyzing the histogram distribution of each curvature; and using the histogram distribution as the curvature distribution feature. Specifically, the curvature of each grid is calculated, and the histogram distribution of the curvature is statistically analyzed. The curvature histogram can be divided into m intervals, for example, m=100, and the ratio of the count of each interval to the total count is calculated. Then, the curvature can also generate an m-dimensional feature. For example Figure 6 What is displayed is a curvature histogram distribution. It should be noted that the inventor of the present application has taken into account that compared with the non-caries crown, the surface of the caries is bumpy and not as smooth as the intact crown, so the curvature variation of the caries surface is also relatively large. Therefore, the curvature histogram distribution of the caries grid should also be statistically different from the curvature histogram distribution of the intact crown. Furthermore, the above curvature can use Gaussian curvature or average curvature. In addition, since the surface curvature of non-caries also varies to a certain extent, for caries and non-caries, using the curvature histogram distribution as a training feature can better show the surface differences between caries and non-caries.

[0063] It should also be noted that in some embodiments, when the preset features include multiple features, each feature is combined in a splicing manner, such as the crown mesh vertex spatial distribution feature s, the crown spatial scale feature Vol and the mesh normal histogram distribution feature v_n can be considered simultaneously, and these three features are spliced ​​to obtain a new feature Y = [v_n, s, Vol], where the mesh normal histogram distribution feature v_n = (bn1, bn2, ..., bnp), the crown mesh vertex spatial distribution feature s = (r1, r2, ..., rk^3), the crown spatial scale feature Vol = (a, b, h), where k is the number of divisions in each direction in the bounding box, and after splicing, a p+k^3+3-dimensional feature vector can be formed. Some other embodiments can also use the above four features, and the feature Z formed by splicing the four features is used to train the machine learning algorithm model. The new feature Z = [v_n, v_c, s, Vol], where the histogram distribution feature with added curvature v_c = (bc1, bc2, ... bcm). This concatenation method will form a vector of p+m+k^3+3 dimensions.

[0064] Correspondingly, the process of using a machine learning algorithm to train the preset features in the sample library may include: using a machine learning algorithm to train the new features Y or Z after splicing of the dental models in the sample library.

[0065] It can be seen that some of the above embodiments further limit the combination of multiple features by splicing, so that the details of various features are more completely preserved, making the recognition result more accurate and reliable.

[0066] In summary, each embodiment uses different types of features and combinations of different features for model training in order to more broadly express the differences between caries and non-caries, further making the recognition results of the present application more accurate and reliable, while being able to meet different needs and adapt to more application scenarios.

[0067] Another embodiment of the present application also relates to a method for identifying abnormal teeth in a dental model. This embodiment is substantially the same as the above embodiment, with the main difference being that in the above embodiment, the abnormal teeth are caries, while in this embodiment, the abnormal teeth are teeth with attachments. The features or feature combinations selected in the above embodiments can be used in this embodiment.

[0068] Specifically, when obtaining the training sample library, each tooth in each dental model in the sample library is marked with a label indicating whether it is a tooth with attachments. Teeth marked as having attachments are screened out to form a tooth set with attachments, and teeth without attachments are screened out to form a tooth set without attachments. Subsequent feature acquisition is extracted based on the tooth set with attachments and the tooth set without attachments.

[0069] It should be noted that when a tooth has an attachment, its volume increases significantly, so the characteristic parameters that characterize the crown volume can effectively distinguish whether the tooth to be identified has an attachment.

[0070] In addition, the curvature of the teeth with attachments at the junction of the attachments and the tooth surface will change greatly, and the different shapes of the attachments themselves will also cause different curvature distributions. Therefore, the curvature distribution is also an important feature to distinguish whether a tooth has an attachment. Furthermore, for teeth with attachments, since the attachments protrude from the tooth surface and the attachments themselves also have various shapes, the normal distribution of teeth with attachments will also be greatly different from the normal distribution of teeth without attachments.

[0071] It can be seen that the use of the above-mentioned features or combinations can effectively identify whether the teeth have attachments, and the selection of machine learning algorithms can also refer to the various algorithm models used in caries identification.

[0072] It should be further explained that in the above implementation, the identification of abnormal teeth is limited to caries or teeth with attachments. The abnormal teeth can also be limited to caries and teeth with attachments. Correspondingly, the tooth marks in each jaw model in the acquired training sample library can be divided into at least three categories: caries labels, labels of teeth with attachments, and labels of normal teeth (non-caries and without attachments). After training with a machine learning algorithm, the obtained recognition model can perform three-category recognition on the teeth to be identified, and identify whether it is caries and whether it is a tooth with attachments at one time.

[0073] It should be noted that the above examples in this embodiment are all illustrative descriptions for easy understanding and do not constitute a limitation on the technical solutions of the present invention.

[0074] The step division of the above methods is only for the purpose of clear description. When implemented, they can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this patent; adding insignificant modifications to the algorithm or process or introducing insignificant designs without changing the core design of the algorithm and process are all within the scope of protection of this patent.

[0075] Another embodiment of the present application relates to an electronic device, such as Figure 7 As shown, it includes: at least one processor 701; and a memory 702 that is communicatively connected to the at least one processor 701; wherein the memory 702 stores instructions that can be executed by the at least one processor 701, and the instructions are executed by the at least one processor 701 so that the at least one processor 701 can execute the method for identifying abnormal teeth in the dental model in the above-mentioned embodiments.

[0076] Among them, the memory and the processor are connected in a bus manner, and the bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and memories together. The bus can also connect various other circuits such as peripherals, voltage regulators, and power management circuits, which are well known in the art and are therefore not further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be one element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. The data processed by the processor is transmitted on a wireless medium via an antenna, and further, the antenna also receives data and transmits the data to the processor.

[0077] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory can be used to store data used by the processor when performing operations.

[0078] Another embodiment of the present application relates to a computer-readable storage medium storing a computer program, which implements the above method embodiment when executed by a processor.

[0079] That is, those skilled in the art can understand that all or part of the steps in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a program, and the program is stored in a storage medium, including a number of instructions to enable a device (which can be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as: ROM), random access memory (Random Access Memory, referred to as: RAM), disk or optical disk and other media that can store program codes.

[0080] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

Claims

1. A method for identifying abnormal teeth in a dental model, characterized in that: include: Obtain a training sample library, the sample library includes a plurality of digital dental models, and use a machine learning algorithm to train preset features of each dental model in the sample library to obtain a recognition model, wherein the preset features at least include: characteristic parameters characterizing the volume of a crown, and each tooth in each dental model in the sample library is marked with a label indicating whether it is an abnormal tooth; The teeth to be identified are identified according to the trained recognition model to confirm whether they are abnormal teeth.

2. The method for identifying abnormal teeth in a dental model according to claim 1, characterized in that: The characteristic parameters characterizing the volume of the tooth crown include: the spatial scale characteristics of the tooth crown, and / or the spatial distribution characteristics of the vertices of the tooth crown mesh.

3. The method for identifying abnormal teeth in a dental model according to claim 2, characterized in that: The characteristic parameters characterizing the volume of the tooth crown are obtained by using a bounding box algorithm and software extraction.

4. The method for identifying abnormal teeth in a dental model according to claim 3, characterized in that: The method for acquiring the spatial distribution characteristics of the crown mesh vertices includes: Establishing a three-dimensional bounding box of a single tooth crown to be identified, wherein the center of the bounding box is the center of the tooth crown, and each face of the bounding box is perpendicular to one of the three tooth axes of the tooth crown; Divide the bounding box into several equal parts along the three tooth axis directions to obtain multiple sub-bounding boxes; Calculate the ratio of the number of mesh vertices of a single crown in each sub-bounding box to the total number of vertices of the crown; The ratio of the number of mesh vertices in all sub-bounding boxes to the total number of vertices is concatenated, and the concatenated result is used as the spatial distribution feature of the mesh vertices of the crown.

5. The method for identifying abnormal teeth in a dental model according to claim 4, characterized in that: The dividing the enclosing frame into a plurality of equal parts along the three tooth axis directions comprises: the number of equal parts along the three tooth axis directions is the same.

6. The method for identifying abnormal teeth in a dental model according to claim 4, characterized in that: The enclosing frame is divided into several equal parts along the three tooth axis directions, including: The number of equal parts in at least one tooth axis direction is greater than or equal to 2.

7. The method for identifying abnormal teeth in a dental model according to claim 4, characterized in that: The enclosing frame is divided into several equal parts along the three tooth axis directions, including: If the length of the bounding box in the direction of a tooth axis is less than the first threshold, the number of equal divisions on the tooth axis is 1.

8. The method for identifying abnormal teeth in a dental model according to claim 7, characterized in that: The first threshold is 1 / 3 of the size of the non-abnormal tooth on the corresponding tooth axis.

9. The method for identifying abnormal teeth in a dental model according to claim 3, characterized in that: The method for acquiring the spatial scale feature of the tooth crown includes: Establishing a three-dimensional bounding box of a single tooth crown to be identified, wherein the center of the bounding box is the center of the tooth crown, and each face of the bounding box is perpendicular to one of the three tooth axes of the tooth crown; The length, width and height of the bounding box are obtained as the spatial scale features of the tooth crown.

10. The method for identifying abnormal teeth in a dental model according to claim 1, characterized in that: The preset features also include: mesh normal features, and / or curvature distribution features.

11. The method for identifying abnormal teeth in a dental model according to claim 10, characterized in that: The method for obtaining the curvature distribution feature includes: Calculate the curvature of each grid on the crown of a single tooth to be identified; Statistical histogram distribution of each curvature; The histogram distribution is used as the curvature distribution feature.

12. The method for identifying abnormal teeth in a dental model according to claim 1, characterized in that: When the preset features include multiple features, the features are combined in a splicing manner; The method of using a machine learning algorithm to train the preset features in the sample library includes: using a machine learning algorithm to train the preset features of the spliced ​​dental models in the sample library.

13. The method for identifying abnormal teeth in a dental model according to claim 1, characterized in that: The machine learning algorithm includes one of the following: support vector machine, XGBoost, random forest, and multi-layer perceptron.

14. The method for identifying abnormal teeth in a dental model according to any one of claims 1 to 13, characterized in that: The abnormal teeth are carious teeth and / or teeth with attachments.

15. An electronic device, characterized in that: include: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for identifying abnormal teeth in a dental model according to any one of claims 1 to 14.

16. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for identifying abnormal teeth in a dental model according to any one of claims 1 to 14 is implemented.

Images