Tooth segmentation method and related equipment

Abstract

The embodiment of the present application provides a tooth segmentation method and related equipment, the method comprising: segmenting oral data containing data of multiple teeth to obtain multiple sub-block data, wherein each sub-block data includes at least one voxel of at least one tooth; inputting the multiple sub-block data into a preset feature prediction model, the feature prediction model predicts multiple feature parameters of each voxel in each sub-block data based on a preset virtual deformer corresponding to each tooth; fusing the multiple sub-block data according to the position information of each sub-block data in the oral data to obtain fused data; performing tooth instance segmentation on the teeth in the fused data based on multiple feature parameters to obtain tooth instance segmentation results corresponding to each preset virtual deformer in the fused data. The above method can realize automatic, fast and accurate segmentation of tooth instances in oral data.

Description

Technical Field

[0001] The present application relates to the field of digital dentistry, and in particular, to a tooth segmentation method and related equipment. Background Art

[0002] Identifying and segmenting the patient's teeth, bones and other structures based on oral data can help doctors make medical diagnoses and plan treatment plans for patients. For example, oral data can be high-resolution three-dimensional oral CT data and cone beam CT data (CBCT data) of patients obtained through computed tomography (CT) technology. However, due to the large volume of oral data, it is difficult for related technologies to achieve accurate instance segmentation of teeth while reducing the computational burden of oral data, which affects the accuracy and reliability of medical diagnosis. For example, in the solution using a three-dimensional convolutional neural network, since the amount of data input to the neural network is very large, the computational cost will be very large, resulting in a high amount of video memory space required for calculation, which is difficult to deploy on ordinary computers and hardware devices. If you want to reduce the amount of calculation, you need to reduce the resolution of the input oral data, which will result in reduced accuracy of semantic segmentation and instance segmentation. Summary of the invention

[0003] The present application discloses a tooth segmentation method and related equipment, which can be based on a method of block prediction and re-fusion of oral data, and by approximating each tooth deformation to a preset virtual deformation body, it is possible to improve the accuracy of tooth instance segmentation while reducing the computational burden.

[0004] A first aspect of an embodiment of the present application provides a tooth segmentation method, comprising: segmenting oral data to obtain multiple sub-block data, wherein the oral data includes data of multiple teeth, and each sub-block data includes at least one voxel of at least one tooth; inputting the multiple sub-block data into a preset feature prediction model, the feature prediction model predicts multiple feature parameters of each voxel in each sub-block data based on a preset virtual deformer corresponding to each tooth; fusing the multiple sub-block data according to position information of each sub-block data in the oral data to obtain fused data; performing tooth instance segmentation on the teeth in the fused data based on the multiple feature parameters to obtain a tooth instance segmentation result corresponding to each preset virtual deformer in the fused data.

[0005] According to an embodiment of the present application, segmenting the oral data to obtain multiple sub-block data includes: determining the data of the multiple teeth in the oral data based on the semantic segmentation result of the oral data; segmenting the oral data to obtain the multiple sub-block data based on the data of the multiple teeth.

[0006] According to an embodiment of the present application, the preset virtual shape body includes a virtual cylinder, the preset virtual shape body corresponds to the tooth, and the voxels in the virtual cylinder correspond to the voxels in the corresponding tooth.

[0007] According to an embodiment of the present application, the multiple characteristic parameters include: a center degree value of each voxel, the center degree value is used to indicate the distance between each voxel and the center of the preset virtual deformer; an expression parameter of the preset virtual deformer corresponding to each voxel, the expression parameter includes: a deformation offset vector of each voxel, an axial direction of the preset virtual deformer, an axial length of the preset virtual deformer and a radial width of the preset virtual deformer, wherein the deformation offset vector is used to indicate the offset of each voxel from the corresponding voxel in the preset virtual deformer.

[0008] According to an embodiment of the present application, the fusing of the multiple sub-block data to obtain fused data includes: if there are at least two sub-block data with overlapping block data, determining at least two different results corresponding to each feature parameter of each voxel in the overlapping block; obtaining updated feature parameters of each feature parameter of each voxel in the overlapping block by averaging the at least two different results, and obtaining the feature parameters of the corresponding voxels in the fused data based on the updated feature parameters.

[0009] According to an embodiment of the present application, the tooth instance segmentation of the multiple teeth in the fused data based on the multiple feature parameters includes: traversing the voxels in the fused data, selecting two voxels in the fused data that have not been subjected to tooth instance segmentation; determining the probability that the teeth corresponding to the two selected voxels are the same tooth based on the multiple feature parameters; determining a set of voxels corresponding to the same tooth based on the probability, and obtaining a tooth instance segmentation result corresponding to each tooth based on the set of voxels corresponding to the same tooth.

[0010] According to an embodiment of the present application, the two voxels include a first voxel and a second voxel; the method of determining the probability that the teeth corresponding to the two selected voxels are the same tooth based on the multiple characteristic parameters includes: determining the axial offset parameter and radial offset parameter of the second voxel compared to the first voxel based on the multiple characteristic parameters, wherein the axial offset parameter is used to indicate the offset distance of the second voxel in the axial direction of the preset virtual deformation body compared to the first voxel, and the radial offset parameter is used to indicate the offset distance of the second voxel in the radial direction of the preset virtual deformation body compared to the first voxel; based on the axial offset parameter and the radial offset parameter, determining the probability that the teeth corresponding to the two voxels are the same tooth.

[0011] According to an embodiment of the present application, the loss function of the feature prediction model includes a first loss function and a second loss function, wherein: the first loss function includes a center degree loss function, and the center degree loss function is used to determine the difference between the center degree value of each voxel in each sub-block data and a preset target center degree value; the second loss function includes an instance segmentation loss function, and the instance segmentation loss function is used to determine the difference between the tooth instance segmentation result corresponding to each preset virtual deformer and the target tooth segmentation result.

[0012] According to an embodiment of the present application, the method is executed and processed by a cloud platform, wherein: the cloud platform receives the oral data uploaded by an electronic device, or the cloud platform retrieves the oral data stored in the cloud.

[0013] A second aspect of an embodiment of the present application provides a tooth segmentation device, which includes: a segmentation module, used to segment oral data to obtain multiple sub-block data, wherein the oral data includes data of multiple teeth, and each sub-block data includes at least one voxel of at least one tooth; a feature prediction module, used to input the multiple sub-block data into a preset feature prediction model, and the feature prediction model predicts multiple feature parameters of each voxel in each sub-block data based on a preset virtual deformer corresponding to each tooth; a block fusion module, used to fuse the multiple sub-block data according to the position information of each sub-block data in the oral data to obtain fused data.

[0014] A third aspect of an embodiment of the present application provides an electronic device, comprising: one or more processors; a memory; and one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the tooth segmentation method.

[0015] A fourth aspect of the embodiments of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the tooth segmentation method is implemented.

[0016] The tooth segmentation method provided in the embodiment of the present application can divide oral data into sub-block data, thereby reducing the data input amount of the algorithm, reducing the video memory consumption required for calculation, and reducing the calculation difficulty and implementation cost of the algorithm; by approximating the shape of each tooth to a preset virtual deformer, based on the feature recognition result of each voxel in the sub-block data, the voxel belonging to the tooth is determined; based on the fusion data of all sub-block data, the tooth instance segmentation result of the complete oral data is determined, which can achieve fast and accurate segmentation of tooth instances in the oral data. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0018] Figure 1 A schematic diagram of an application environment of a tooth segmentation method provided in an embodiment of the present application.

[0019] Figure 2 A schematic flow chart of a tooth segmentation method provided in an embodiment of the present application.

[0020] Figure 3 This is an example diagram of a slice image of oral data provided in an embodiment of the present application.

[0021] Figure 4 This is an example diagram of the semantic segmentation results of a slice image provided in an embodiment of the present application.

[0022] Figure 5 This is an example diagram of the semantic segmentation results of oral data provided in one embodiment of the present application.

[0023] Figure 6 A detailed flowchart of step S204 provided in an embodiment of the present application.

[0024] Figure 7 This is an example diagram of the tooth instance segmentation result provided in one embodiment of the present application.

[0025] Figure 8 A detailed flowchart of step S204 provided for another embodiment of the present application.

[0026] Fig. 9 This is a principle block diagram of a tooth segmentation device provided in one embodiment of the present application.

[0027] Fig.10 A schematic diagram of the structure of an electronic device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0028] In order to make the objectives, technical solutions and advantages of the present application clearer, the present application is described in detail below with reference to the accompanying drawings and specific embodiments.

[0029] It should be noted that in this application, "at least one" means one or more, and "more than one" means two or more than two. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and drawings of this application are used to distinguish similar objects, rather than to describe a specified order or sequence.

[0030] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a concrete way. The following embodiments and features in the embodiments may be combined with each other without conflict.

[0031] Oral data includes computed tomography (CT) data and cone beam CT data (CBCT data). In some embodiments, oral data includes three-dimensional oral data and denture data obtained by scanning with an oral scanner or a desktop scanner. Identifying and segmenting the patient's teeth, bones and other structures based on oral data can help doctors make medical diagnoses and plan treatment plans for patients. In the three-dimensional oral data, the center position of each voxel (volume element) represents a point in three-dimensional space.

[0032] In related technologies, deep neural network technology is usually used to input the original three-dimensional oral data into a preset model (such as a three-dimensional convolutional neural network) to obtain the semantic segmentation of human teeth, bones and other tissues in the three-dimensional oral space and the instance segmentation of teeth. However, due to the large volume of oral data, the computational cost of the model will be very high, and the algorithm requires a large amount of video memory, which is difficult to deploy on ordinary computers and hardware devices. If the amount of calculation is to be reduced, the resolution of the input oral data needs to be reduced, which will result in reduced accuracy of semantic segmentation and instance segmentation.

[0033] Therefore, it is difficult for related technologies to achieve accurate instance segmentation of teeth while reducing the computational burden of oral data, which affects the accuracy and reliability of medical diagnosis.

[0034] In order to solve the above problems, the tooth segmentation method provided in the embodiment of the present application can divide the oral data into sub-block data, thereby reducing the data input amount of the algorithm, reducing the calculation difficulty and implementation cost of the algorithm; by approximating the shape of each tooth to a preset virtual deformer, based on the feature recognition result of each voxel in the sub-block data, the voxel belonging to the tooth is determined; based on the fusion data of all sub-block data, the tooth instance segmentation result of the complete oral data is determined. The above method can be used to achieve fast and accurate segmentation of tooth instances in oral data.

[0035] See also Figure 1 , is a schematic diagram of an application environment of a tooth segmentation method provided in an embodiment of the present application.

[0036] The tooth segmentation method provided in the embodiment of the present application can be applied to an electronic device to perform tooth segmentation on oral cavity data. Figure 1 As shown, the electronic device is taken as the server 30 for example, and the scanning device 20 and the server 30 communicate through a network. The network can be a wired network communication or a wireless network communication. The wired network can be any one of a local area network, a metropolitan area network and a wide area network, a wireless fidelity (Wireless Fidelity, Wi-Fi), a self-organizing network wireless communication (ZigBee Wireless Networks, ZigBee) technology, an ultra-wideband (Ultra Wideband, UWB) technology, a wireless universal serial bus (Universal Serial Bus, USB) and the like.

[0037] In one embodiment of the present application, the scanning device 20 may be a CT scanning device for obtaining high-resolution three-dimensional oral CT data of the patient. In some embodiments, the scanning device 20 may be a CBCT scanning device. The CBCT scanning device may use cone beam X-ray projection technology during the scanning process to perform circular digital projection around the patient's head with a lower X-ray generator radiation dose. The data obtained from multiple projections are recombined to ultimately generate three-dimensional oral CBCT data.

[0038] In one embodiment of the present application, the server 30 can be a single server, a server cluster consisting of multiple servers, or a cloud computing service center (cloud platform). The tooth segmentation method in the embodiment of the present application is run and processed by the cloud platform, wherein the cloud platform receives oral data uploaded by a computer, or the cloud platform retrieves oral data stored in the cloud, and the oral data is segmented into teeth instances on the cloud platform to obtain the tooth instance segmentation result.

[0039] In some other embodiments of the present application, the electronic device may also be a mobile phone, a tablet computer, a smart wearable device or other electronic device. For the relevant introduction of the electronic device, please refer to the following for Fig.10 Description of the process.

[0040] In one embodiment of the present application, the server 30 is used to receive the oral data sent by the scanning device 20, segment the oral data, and obtain a plurality of sub-block data, wherein the oral data includes data of a plurality of teeth, and each sub-block data includes at least one voxel of at least one tooth. The server 30 is used to input the plurality of sub-block data into a preset feature prediction model, and the feature prediction model predicts a plurality of feature parameters of each voxel in each sub-block data based on a preset virtual deformer corresponding to each tooth. The training of the feature prediction model can be performed in the server 30. The server 30 is used to fuse the plurality of sub-block data according to the position information of each sub-block data in the oral data to obtain fused data. The server 30 is used to perform tooth instance segmentation on the teeth in the fused data based on the plurality of feature parameters, and obtain a tooth instance segmentation result corresponding to each preset virtual deformer in the fused data.

[0041] The following will implement the tooth segmentation method of the embodiment of the present application with a computer program product running on a server (such as Figure 1 The server 30 shown in FIG. Figure 2 FIG. 1 is a flow chart of a tooth segmentation method provided in an embodiment of the present application. In an embodiment of the present application, the method includes the following steps:

[0042] Step S201, segmenting the oral cavity data to obtain a plurality of sub-block data.

[0043] In some embodiments of the present application, the oral data may include high-resolution three-dimensional data of the patient's oral cavity, wherein the oral data may include data of the patient's teeth, periodontal tissues, alveolar bones, jaws, and other related oral structures. In one example, the oral data generally includes three dimensions: a horizontal dimension (X-axis), a vertical dimension (Y-axis), and a depth dimension (Z-axis). The horizontal dimension may represent the left-right direction of the patient, such as the direction extending from the patient's left ear to the right ear; the vertical dimension may represent the up-and-down direction of the patient, such as the direction extending from the top of the patient's maxilla (or skull top) to the bottom of the mandible; the depth dimension may represent the front-to-back direction of the patient, such as the direction extending from the front of the patient's face (such as the direction of the nose tip) through the oral cavity to the bottom of the skull.

[0044] In some embodiments of the present application, the center position of each voxel (volume element) in the oral data corresponds to a point in three-dimensional space, and each voxel may correspond to a certain category. For example, a voxel may belong to the tooth category; another voxel may belong to the maxillary bone category.

[0045] In some embodiments of the present application, oral data may be obtained by receiving input from a target user (eg, a dentist), acquiring from an oral scanner, etc., and oral data may also be presented using a display interface.

[0046] In some embodiments of the present application, the oral data contains data of multiple teeth, and performing tooth segmentation on the oral data can be understood as determining the set of voxels corresponding to each tooth in the oral data, thereby achieving segmentation of different teeth in the oral data. For example, each tooth can have a corresponding unique identifier, such as tooth m, tooth n, etc., and determining the set of all voxels belonging to tooth m can achieve instance segmentation of tooth m. To this end, all voxels belonging to the tooth category in the oral data can be determined, and these voxels can be analyzed to determine the set of voxels corresponding to each tooth.

[0047] In some embodiments of the present application, a semantic segmentation result of the oral data may be determined based on a semantic segmentation algorithm, the semantic segmentation result including the category of each voxel, so that all voxels in the oral data belonging to the tooth category may be determined based on the semantic segmentation result.

[0048] In one example, determining the semantic segmentation result of oral data based on a semantic segmentation algorithm may include: (1) determining three dimensions corresponding to the oral data, selecting a preset dimension from the three dimensions, and slicing the oral data based on the preset dimension to obtain multiple slice images. Figure 3 As shown, the preset dimension may be a vertical dimension, and the oral data is sliced ​​based on the vertical dimension to obtain Figure 3 (2) Use a preset semantic segmentation model (such as a U-Net model, a fully convolutional neural network model, a DeepLab v3+ model, etc.) to perform semantic segmentation on each slice image to obtain a semantic segmentation result for each slice image. The semantic segmentation result may include classification results of teeth, bones, etc. For example Figure 4 As shown, different colors can be used to display different categories of semantic segmentation results in the graphical user interface. (3) According to the position information of each slice in the oral data, multiple slices are spliced ​​to obtain the three-dimensional semantic segmentation results of the three-dimensional image data. For example Figure 5As shown in FIG. 1 , by splicing multiple slices in the order of their positions in the oral data, the semantic segmentation result of the complete oral data can be obtained. The red part indicates the area where the upper teeth are located, and the green part indicates the area where the lower teeth are located.

[0049] Through the above embodiment, by cutting the three-dimensional data into multiple two-dimensional slice images, the complex three-dimensional recognition problem can be simplified into multiple relatively simple two-dimensional image recognition tasks. The above dimensionality reduction strategy not only enables the semantic segmentation model to be more effectively applied to each slice, improving the recognition accuracy, but also restores the complete semantic information of the original three-dimensional data through the slice splicing step. Therefore, the above method can not only simplify the processing flow, but also enhance the efficiency and robustness of the algorithm when processing large-scale, high-complexity three-dimensional data.

[0050] In some embodiments of the present application, the oral data may be divided into a plurality of smaller block data, and the teeth corresponding to the voxels in each smaller block data may be analyzed. Since there may be a smaller block data in which there are no voxels of the tooth category, for such block data, it is obvious that the tooth instance segmentation process may not analyze the voxels therein. Therefore, a smaller block data (such as sub-block data) containing voxels of the tooth category may be selected for tooth instance segmentation.

[0051] In some embodiments of the present application, segmenting the oral data to obtain multiple sub-block data includes: determining the data of the multiple teeth in the oral data based on the semantic segmentation result of the oral data; segmenting the oral data to obtain the multiple sub-block data based on the data of the multiple teeth, and the data set composed of the multiple sub-block data includes all the data of the multiple teeth.

[0052] In one example, based on the semantic segmentation result of the oral data, a set of voxels of the tooth category can be determined as the data of the multiple teeth; the oral data can be segmented to obtain a plurality of smaller block data; based on the multiple tooth data, a smaller block data containing at least one voxel of at least one tooth can be selected from the multiple smaller block data, and the selected smaller block data can be used as sub-block data.

[0053] In one example, the oral data can be segmented based on a preset segmentation size, and the segmentation size can be set according to actual needs, such as 100 voxels × 100 voxels × 100 voxels, and the present application does not impose specific restrictions on this. By segmenting the oral data based on the preset segmentation size, the size of each sub-block data obtained can be made consistent. Among them, there may or may not be overlapping blocks between adjacent sub-block data, which can be set according to actual needs, and the present application does not impose restrictions on this. If there are overlapping blocks, the size of the overlapping blocks can be set according to actual needs, and the present application does not impose restrictions on this.

[0054] In some embodiments of the present application, a corresponding identifier may be set for each sub-block data, such as sub-block data A, sub-block data B, etc. The position information of each sub-block in the oral data may be determined. In one example, the position information of sub-block data A may be expressed as (12, 30, 24), indicating that sub-block data A is a smaller block located at the 12th in the horizontal dimension, the 30th in the vertical dimension, and the 24th in the depth dimension in the oral data. In another example, the position information of the sub-block in the oral data may also be determined based on the coordinates of the voxel of the center point of the sub-block data.

[0055] In one example, each voxel has a three-dimensional coordinate in the oral data, and the coordinate set of the voxels belonging to each sub-block data can be determined. If there are overlapping blocks, at least two sub-blocks corresponding to the overlapping blocks and the coordinate set of the voxels in the overlapping blocks can also be determined. In another example, a corresponding identifier can also be set for each voxel in the sub-block data, for example, voxel i in sub-block data A represents the i-th voxel in sub-block data A, i represents a positive integer, for example, the value range of i is [1,1000].

[0056] Through the above-mentioned embodiments, oral data can be segmented into multiple small blocks, and sub-blocks containing voxels of tooth categories can be accurately selected for instance segmentation based on the semantic segmentation results, which can effectively avoid invalid processing of data in areas without teeth, effectively reduce the difficulty of the tooth instance segmentation algorithm, and improve the efficiency and accuracy of tooth instance segmentation.

[0057] Step S202: input the plurality of sub-block data into a preset feature prediction model, wherein the feature prediction model predicts a plurality of feature parameters of each voxel in each sub-block data based on a preset virtual deformable body corresponding to each tooth.

[0058] In some embodiments of the present application, a pre-trained feature prediction model may be used to perform feature recognition on voxels in sub-block data, wherein each tooth may be approximately fitted to a preset virtual shape body, and instance segmentation of each tooth may be achieved by determining voxels belonging to the same preset virtual shape body. In one example, the preset virtual shape body may be an approximate shape of a tooth after deformation, for example, the preset virtual shape body may be a virtual cylinder, and there is a corresponding relationship (e.g., a one-to-one corresponding relationship) between the preset virtual shape body and the tooth, and there is a corresponding relationship (e.g., a one-to-one corresponding relationship) between the voxels in the virtual cylinder and the voxels in the corresponding tooth.

[0059] In some embodiments of the present application, each sub-block data can be input into a preset feature prediction model, and the feature prediction model is used to output multiple feature parameters of each voxel in each sub-block data. The multiple feature parameters include, but are not limited to, one or more of the following parameters: the center degree value of each voxel, the center degree value is used to indicate the distance between each voxel and the center of the preset virtual deformer; the expression parameters of the preset virtual deformer corresponding to each voxel, the expression parameters include: the deformation offset vector of each voxel, the axial direction of the preset virtual deformer, the axial length of the preset virtual deformer, and the radial width of the preset virtual deformer, wherein the deformation offset vector is used to indicate the offset of each voxel inside the tooth from the corresponding voxel in the preset virtual deformer.

[0060] In one example, the center value c_i of voxel i in the sub-block data A refers to the distance between voxel i and the center of tooth m corresponding to the sub-block data A (e.g., the center of a virtual cylinder), and the value range is [0,1]. If voxel i is not inside tooth m, the center value c_i=0; if voxel i is located at the center of tooth m, the center value c_i=1.

[0061] In one example, if voxel i is inside tooth m, the deformation offset F_i of voxel i is used to indicate the offset of voxel i after tooth m is deformed, and the deformation offset F_i is a three-dimensional vector. For example, the deformation offset F_i can be a vector from voxel i in tooth m to voxel I in virtual cylinder M, wherein tooth m and virtual cylinder M have a corresponding relationship, and voxel i and voxel I have a corresponding relationship.

[0062] In one example, if the voxel point i is inside the tooth m, the axial direction A_i of the preset virtual deformer can represent the axial direction of the virtual cylinder M corresponding to the tooth m after deformation; the axial direction A_i is a three-dimensional vector; the axial direction A_i can be used to indicate the long axis or main growth direction of the tooth. In another example, the direction perpendicular to the axial direction can be called a radial direction, and the expression parameter can also include the radial direction.

[0063] In one example, if the voxel point i is inside the tooth m, the axial length h_i of the preset virtual deformer may represent the axial length of the virtual cylinder M corresponding to the tooth m after deformation; the axial length h_i may be used to indicate the length of the tooth.

[0064] In one example, if the voxel point i is inside the tooth m, the radial width w_i of the preset virtual deformer may represent the radial width of the tooth after deformation, which is approximately a cylinder; the radial width w_i may be used to indicate the diameter or width of the tooth.

[0065] Through the above embodiment, the pre-trained feature prediction model can be used to perform feature recognition on the voxels in the sub-block data, and a preset virtual shape body (such as a virtual cylinder) can be used to approximate each tooth, and the instance segmentation of the tooth can be achieved by determining the voxels belonging to the same virtual shape body. The model output includes characteristic parameters such as the voxel center degree value, deformation offset vector, axial direction, axial length and radial width. These parameters can accurately describe the shape, position and deformation of the tooth, so that the subsequent process can facilitate the high-precision recognition and segmentation of the tooth instance.

[0066] Step S203 , fusing the plurality of sub-block data according to the position information of each sub-block data in the oral data to obtain fused data.

[0067] In some embodiments of the present application, multiple sub-block data can be spliced ​​according to the position information of each sub-block data in the oral data to achieve fusion of multiple sub-block data. If there are at least two sub-block data with overlapping block data, at least two different results corresponding to each feature parameter of each voxel in the overlapping block are determined; by averaging the at least two different results, updated feature parameters of each feature parameter of each voxel in the overlapping block are obtained, and based on the updated feature parameters, the feature parameters of the corresponding voxels in the fused data are obtained.

[0068] In one example, if there is an overlapping block C between sub-block data A and sub-block data B, taking the characteristic parameter as the center degree value, the at least two different results corresponding to the voxel i in the overlapping block C include: the center degree value c_i_A of the voxel i in the sub-block data A, and the center degree value c_i_B of the voxel i in the sub-block data B. The at least two different results are averaged to obtain an updated center degree value: (c_i_A+c_i_B) / 2. Therefore, the center degree value of the voxel i in the fused data is (c_i_A+c_i_B) / 2.

[0069] Through the above-mentioned embodiments, multiple sub-block data in the oral data can be accurately spliced ​​according to their position information, and the characteristic parameters of the voxels in the overlapping blocks can be specially processed. The characteristic parameters of the voxel parameters in the overlapping blocks can be updated by averaging and other methods, thereby achieving seamless fusion of sub-block data, ensuring the integrity and accuracy of the fused data, and providing more accurate data support for subsequent tooth instance segmentation.

[0070] Step S204, performing tooth instance segmentation on the teeth in the fused data based on the multiple feature parameters, and obtaining a tooth instance segmentation result corresponding to each preset virtual deformable body in the fused data.

[0071] In some embodiments of the present application, each voxel in each sub-block data may be analyzed in the order of the position information of each sub-block data in the fused data. Based on multiple characteristic parameters of each voxel, voxels belonging to the same preset virtual image variant may be determined, thereby implementing tooth instance segmentation of the oral cavity data. In one example, the refinement process of step S204 includes the following steps: Figure 6 The process shown. Figure 6 As shown, step S204 includes the following process:

[0072] Step S2041, traversing the voxels in the fused data, and selecting two voxels in the fused data that have not been segmented into tooth instances.

[0073] In some embodiments of the present application, when traversing the voxels in the fused data, the same sub-block data or at least two adjacent sub-block data can be selected according to the position information of the sub-block data. From the selected same sub-block data or at least two adjacent sub-block data, any two voxels that have not been segmented by tooth instances are selected. The voxels that have not been segmented by tooth instances can represent voxels that have not been determined to belong to a certain tooth voxel set.

[0074] Through the above-mentioned embodiment, the position information of the sub-block data can be used to traverse the voxels in the fused data that are as adjacent as possible and have not been segmented for tooth instances, and select the voxel pairs that are most likely to belong to the same tooth or adjacent teeth for instance segmentation, thereby improving the accuracy, consistency and efficiency of instance segmentation and enhancing the adaptability and robustness of the algorithm when processing complex tooth morphologies.

[0075] In other embodiments of the present application, when selecting two voxels in the fusion data that have not been segmented for tooth instances, a voxel that has not been segmented for tooth instances and is relatively close to the center of the tooth can be selected as the first voxel. For example, based on the center degree value corresponding to the voxel, a voxel that has not been segmented for tooth instances and whose center degree value is greater than a preset center degree threshold is determined as the first voxel. Based on the position of the sub-block data corresponding to the first voxel, a voxel that has not been segmented for tooth instances and is located in the same sub-block as the first voxel can be selected, or a voxel that has not been segmented for tooth instances and is located in an adjacent sub-block to the first voxel can be selected, and the selected voxel will be used as the second voxel of the two voxels that have not been segmented for tooth instances.

[0076] Through the above embodiment, the first voxel close to the center of the tooth can be selected as the starting point, and the adjacent unsegmented voxel can be selected as the second voxel based on its position information, which can further improve the accuracy and efficiency of tooth instance segmentation, ensure that the segmentation process can be carried out around the core area of ​​the tooth, and take into account the coherence of the local area, thereby enhancing the reliability and practicality of the instance segmentation results.

[0077] Step S2042: Based on the multiple feature parameters, determine the probability that the teeth corresponding to the two selected voxels are the same tooth.

[0078] In some embodiments of the present application, the two selected voxels include a first voxel and a second voxel, and the probability that the teeth corresponding to the two selected voxels are the same tooth can be determined by determining the offset distance between the first voxel and the second voxel. The first voxel can represent a voxel that has not been segmented by a tooth instance and whose center value is greater than a preset center threshold.

[0079] In some embodiments of the present application, the axial offset parameter and radial offset parameter of the second voxel compared to the first voxel can be determined based on the multiple characteristic parameters, wherein the axial offset parameter is used to indicate the offset distance of the second voxel in the axial direction of the preset virtual deformation body compared to the first voxel, and the radial offset parameter is used to indicate the offset distance of the second voxel in the radial direction of the preset virtual deformation body compared to the first voxel; based on the axial offset parameter and the radial offset parameter, the probability that the teeth corresponding to the two voxels are the same tooth is determined.

[0080] In some embodiments of the present application, the probability that the teeth corresponding to the two selected voxels are the same tooth is determined based on the multiple feature parameters, and the formula used may include:

[0081] ,

[0082] ,

[0083] ,

[0084] Here, i represents the first voxel of the two selected voxels, and j represents the second voxel of the two selected voxels. represents the coordinates of the voxel (for example, voxel I) corresponding to the first voxel i in the corresponding preset virtual deformation body, and the deformation offset between voxel i and voxel I is F_i. represents the coordinates of the voxel (for example, voxel J) corresponding to the second voxel j in the corresponding preset virtual deformation body, and the deformation offset between voxel j and voxel J is F_j. represents the axial length of the preset virtual deformation body corresponding to the first voxel i, represents the radial length of the preset virtual deformation body corresponding to the first voxel i, represents the axial direction of the preset virtual deformation body corresponding to the first voxel i, Represents element-wise multiplication of vectors. represents the probability that the teeth corresponding to the two voxels are the same tooth, represents the axial offset parameter of the second voxel compared to the first voxel, represents a radial offset parameter of the second voxel compared to the first voxel.

[0085] In one example, represents the displacement vector between voxel I and voxel J, represents the element-by-element product of the displacement vector and the axial direction of the virtual deformer where the voxel i is located, and can be used to indicate the relative offset vector of the deformed voxel J compared to the voxel I in the axial direction.

[0086] In one example, It represents the remaining component after removing the axial component from the displacement vector between voxel I and voxel J. The remaining component can be used to represent the relative offset vector of the deformed voxel J compared with voxel I in the radial direction.

[0087] In one example, represents the square of the normalized Euclidean distance between the deformed voxel J and the voxel I. As the standard deviation in the axial direction, use As the radial standard deviation, the distance between different virtual shape bodies can be measured.

[0088] In one example, Indicates that based on parameters The exponential function can map the calculation result to the interval (0,1), thereby obtaining the probability that the teeth corresponding to the two voxels are the same tooth model. The specific physical meaning of the exponential function includes: after fitting the tooth deformation into a virtual deformation body (such as a virtual cylinder), the eigenvalue at voxel i parameterizes a virtual cylinder, Indicates the distance between voxel j and the virtual cylinder. Obviously, the larger the offset distance between voxel i and voxel j, the greater the distance between voxel j and the virtual cylinder, the smaller the probability that the teeth corresponding to the two voxels are the same tooth, and the closer the distance between voxel j and the virtual cylinder is. The closer the value of is to 1, the farther the voxel j is from the virtual cylinder. The closer the value of is to 0. This is consistent with the actual situation of tooth instance segmentation.

[0089] Through the above embodiment, by calculating the axial and radial offsets between two voxels and combining the axial and radial dimensions of the virtual deformer for standardization, it is possible to accurately evaluate the probability of whether the teeth corresponding to the two voxels are the same tooth. This can not only improve the accuracy of tooth instance segmentation, but also make the tooth instance segmentation result more consistent with the actual situation, which is helpful for diagnosis and treatment in stomatology.

[0090] Step S2043: determining a set of voxels corresponding to the same tooth according to the probability, and obtaining a tooth instance segmentation result corresponding to each tooth based on the set of voxels corresponding to the same tooth.

[0091] In some embodiments of the present application, it can be determined that two voxels corresponding to a probability greater than or equal to a preset probability threshold belong to the same tooth, thereby determining a set of voxels corresponding to the same tooth. The probability threshold can be set according to actual needs. When the threshold is high, the segmented tooth model will be smaller (occupying fewer voxels), and when the threshold is low, the segmented tooth model will be larger (occupying more voxels). For example, the probability threshold is 1, 0.999, 0.5, etc., and the present application does not impose specific restrictions on this. By dividing all voxels in the fused data into a set of voxels for each tooth, the position of the corresponding tooth model can be determined according to each set, thereby realizing the segmentation of tooth instances in the oral data.

[0092] In some embodiments of the present application, the probability threshold is set to 0.5, and the size of the segmented tooth model is consistent with the actual tooth instance segmentation result.

[0093] In some embodiments of the present application, the tooth instance segmentation result can be displayed to the user in a graphical user interface, for example, using different colors to indicate voxels belonging to different sets. Figure 7As shown, it is an example diagram of the segmentation result of a tooth instance provided by an embodiment of the present application, and a set of voxels with the same color indicates the segmentation result of a single tooth instance. In another example, the segmentation result can also be displayed by text (such as tooth number) or other means. In another example, it can be selected whether to display the segmentation result, for example, it can be selected not to display the segmentation result.

[0094] Through the above embodiment, voxels with high probability association can be effectively classified as a set of the same tooth based on the set probability threshold, thereby achieving automatic and accurate tooth instance segmentation. The segmentation results of different teeth are also intuitively displayed through color distinction, providing strong support for subsequent oral medicine analysis and treatment.

[0095] The tooth segmentation method provided in the embodiment of the present application can divide oral data into sub-block data, thereby reducing the data input amount of the algorithm, reducing the calculation difficulty and implementation cost of the algorithm; by approximating the shape of each tooth to a preset virtual deformer, based on the feature recognition result of each voxel in the sub-block data, the voxel belonging to the tooth is determined; based on the fusion data of all sub-block data, the tooth instance segmentation result of the complete oral data is determined, which can realize accurate, fast and fully automatic segmentation of tooth instances in the oral data.

[0096] In some embodiments of the present application, the feature prediction model may include a Unet model, such as a 3D-Unet model. The feature prediction model may be trained based on a supervised training method using a preset loss function. According to the application reasoning process of the feature prediction model in the above embodiment, it can be known that the loss function of the feature prediction model includes two objectives: (a) making the difference between the center degree value of the voxel output by the model (e.g., c_i) and the actual center degree value of the voxel compared to the center point of the tooth (e.g., the target center degree value) as small as possible, so that the model can accurately predict the center degree value of the voxel. Based on the actual tooth instance segmentation result obtained, the center point of the tooth can be determined according to the average position of all voxels in the tooth; the actual distance of the voxel compared to the center point of the tooth (e.g., Euclidean distance) is determined, and the target center degree value is determined according to the actual distance (e.g., the actual distance is normalized based on the maximum radius of the tooth). In this way, the target center degree value of c_i can be predefined. The closer the voxel i is to the center of the tooth where it is located, the closer the target center degree value is to 1. Conversely, the farther the voxel i is from the center of the tooth where it is located, the closer the target center degree value is to 0. (b) Make the difference between the virtual shape body corresponding to the expression parameters of the virtual shape body output by the model and the actual shape of the tooth as small as possible, so that the model can accurately predict the virtual shape body corresponding to the tooth, and improve the accuracy of the expression parameters of the virtual shape body and the prediction probability The accuracy of tooth instance segmentation based on virtual deformable bodies can be improved.

[0097] In some embodiments of the present application, according to the above two objectives, the loss function of the feature prediction model includes a first loss function and a second loss function, wherein: the first loss function includes a center degree loss function, and the center degree loss function is used to determine the difference between the center degree value of each voxel in each sub-block data and the preset target center degree value, such as the absolute value. The second loss function includes an instance segmentation loss function, and the instance segmentation loss function is used to determine the difference between the tooth instance segmentation result corresponding to each preset virtual morph body and the target tooth segmentation result (such as the actual tooth instance segmentation result). In one example, the probability calculated based on the expression parameters (F_i, A_i, h_i, w_i) of the virtual morph body , determine the difference between the instance segmentation result based on the preset virtual shape and the real (target) tooth segmentation result. You can use BCE Loss function or Dice Loss function to make The value of is consistent with the actual tooth segmentation situation, that is, if voxel i and voxel j belong to the same tooth, then approaches 1, if voxel i and voxel j do not belong to the same tooth, Approaching 0.

[0098] Through the above embodiments, the supervised training method and the preset loss function can be combined to effectively improve the accuracy of the model in predicting the voxel center value and the expression parameters of the virtual tooth shape. By optimizing the center loss function and the instance segmentation loss function at the same time, the model can not only locate the center point of the tooth more accurately, but also fit the tooth model into a virtual shape that is closer to the real tooth shape, thereby significantly improving the accuracy and efficiency of tooth instance segmentation based on the virtual shape.

[0099] refer to Figure 8 As shown, it is a detailed flow chart of step S204 provided by another embodiment of the present application. The voxel i that is not segmented and has the largest center value in the fused data can be determined, and it can be determined whether the center value of voxel i is greater than a preset center value threshold (for example, 0.5). If the center value of voxel i is greater than the preset center value threshold, for each unsegmented voxel j, the probability that voxel i and voxel j belong to the same tooth is calculated. . The probability Voxels j with probability greater than a preset probability threshold (eg, 0.5) are aggregated into an instance segmentation result of a tooth, and the above process is repeated until there are no unsegmented voxels.

[0100] Through the above embodiment, the voxel i which is not segmented and has the highest center value in the fused data can be selected iteratively, and based on the comparison between its center value and the preset center value threshold, and the association probability between it and other unsegmented voxels j, , effectively aggregating voxels with a high probability of belonging to the same tooth together to form the instance segmentation result of the tooth. This process is repeated until all voxels are segmented, thus achieving efficient and accurate tooth instance segmentation.

[0101] Fig. 9 It is a structural diagram of a tooth segmentation device provided in one embodiment of the present application.

[0102] In some embodiments, the tooth segmentation device 90 may include a plurality of functional modules composed of computer program segments. The computer programs of the various program segments in the tooth segmentation device 90 may be stored in a memory of an electronic device and executed by at least one processor to perform (see Figure 2 Description) Functionality for tooth instance segmentation.

[0103] In this embodiment, the tooth segmentation device 90 can be divided into multiple functional modules according to the functions it performs. The functional modules may include: a block segmentation module 901, a feature prediction module 902, a block fusion module 903, and an instance segmentation module 904. The module referred to in this application refers to a series of computer program segments that can be executed by at least one processor and can complete fixed functions, which are stored in a memory. In this embodiment, the functional implementation of each module in the tooth segmentation device 90 can refer to the above definition of the tooth segmentation method, and will not be repeated here.

[0104] The segmentation module 901 is used to segment the oral cavity data to obtain a plurality of sub-block data, wherein the oral cavity data includes data of a plurality of teeth, and each sub-block data includes at least one voxel of at least one tooth.

[0105] The feature prediction module 902 is used to input the multiple sub-block data into a preset feature prediction model, and the feature prediction model predicts multiple feature parameters of each voxel in each sub-block data based on a preset virtual deformation body corresponding to each tooth.

[0106] The block fusion module 903 is used to fuse the multiple sub-block data according to the position information of each sub-block data in the oral data to obtain fused data.

[0107] The instance segmentation module 904 is used to perform tooth instance segmentation on the teeth in the fused data based on the multiple feature parameters, and obtain a tooth instance segmentation result corresponding to each preset virtual shape body in the fused data.

[0108] Another embodiment of the present application also provides an electronic device. Figure 1 The application environment is only an example. In other exemplary embodiments, the computer program product implementing the tooth segmentation method of the embodiment of the present application can also be run on any electronic device with sufficient computing power (such as Fig.10 In the electronic device shown in the figure), the various steps of the tooth segmentation method are executed to provide the function of tooth segmentation.

[0109] See also Fig.10 , is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device 10 can be a mobile phone, a tablet computer, a smart wearable device, an augmented reality (AR) / virtual reality (VR) device, a laptop, a netbook, an energy storage device, a power distribution device, a vehicle-mounted device, a self-mobile device, and other electronic devices. The embodiment of the present application does not impose any restrictions on the specific type of the electronic device.

[0110] like Fig.10 As shown, the electronic device 10 may include a communication module 101, a memory 102, a processor 103, an input / output (I / O) interface 104 and a bus 105. The processor 103 is coupled to the communication module 101, the memory 102 and the I / O interface 104 through the bus 105.

[0111] The communication module 101 may include a wired communication module and / or a wireless communication module. The wired communication module may provide one or more wired communication solutions such as universal serial bus (USB), controller area network bus (CAN, Controller Area Network). The wireless communication module may provide one or more wireless communication solutions such as wireless fidelity (wireless fidelity, Wi-Fi), bluetooth (bluetooth, BT), mobile communication network, frequency modulation (frequency modulation, FM), near field communication technology (near field communication, NFC), infrared technology (infrared, IR), etc.

[0112] The memory 102 may include one or more random access memories (RAM) and one or more non-volatile memories (NVM). The random access memory can be directly read and written by the processor 103, and can be used to store executable programs (such as machine instructions) of the operating system or other running programs, and can also be used to store user and application data. The random access memory may include static random-access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), etc.

[0113] The non-volatile memory may also store executable programs and user and application data, etc., and may be loaded into the random access memory in advance for direct reading and writing by the processor 103. The non-volatile memory may include a disk storage device and a flash memory.

[0114] The memory 102 is used to store one or more computer programs. The one or more computer programs are configured to be executed by the processor 103. The one or more computer programs include a plurality of instructions, and when the plurality of instructions are executed by the processor 103, the tooth segmentation method executed on the electronic device 10 can be implemented.

[0115] In other embodiments, the electronic device 10 further includes an external memory interface for connecting to an external memory to expand the storage capacity of the electronic device 10 .

[0116] The processor 103 may include one or more processing units, for example, the processor 103 may include an application processor (AP), a modem processor, a graphics processor (GPU), an image signal processor (ISP), a controller, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural-network processing unit (NPU), etc. Different processing units may be independent devices or integrated into one or more processors.

[0117] The processor 103 provides computing and control capabilities. For example, the processor 103 is used to execute a computer program stored in the memory 102 to implement the above-mentioned tooth segmentation method.

[0118] The I / O interface 104 is used to provide a channel for user input or output. For example, the I / O interface 104 can be used to connect various input and output devices, such as a mouse, keyboard, touch device, display screen, etc., so that the user can enter information or visualize information.

[0119] The bus 105 is at least used to provide a channel for mutual communication among the communication module 101 , the memory 102 , the processor 103 , and the I / O interface 104 in the electronic device 10 .

[0120] It is to be understood that the structure illustrated in the embodiment of the present application does not constitute a specific limitation on the electronic device 10. In other embodiments of the present application, the electronic device 10 may include more or fewer components than shown in the figure, or combine some components, or split some components, or arrange the components differently. The components shown in the figure may be implemented in hardware, software, or a combination of software and hardware.

[0121] An embodiment of the present application further provides a computer-readable storage medium, on which a computer program is stored. The computer program includes program instructions. The method implemented when the program instructions are executed can refer to the methods in the above-mentioned embodiments of the present application.

[0122] The computer-readable storage medium may be an internal memory of the electronic device of the above-mentioned embodiment, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device.

[0123] In some embodiments, the computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the electronic device, etc.

[0124] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0125] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specified application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specified application, but such implementation should not be considered to be beyond the scope of this application.

[0126] In the embodiments provided in the present application, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0127] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0128] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A tooth segmentation method, characterized in that: The method comprises: Segmenting the oral cavity data to obtain a plurality of sub-block data, wherein the oral cavity data includes data of a plurality of teeth, and each sub-block data includes at least one voxel of at least one tooth; Inputting the plurality of sub-block data into a preset feature prediction model, wherein the feature prediction model predicts a plurality of feature parameters of each voxel in each sub-block data based on a preset virtual deformation body corresponding to each tooth; According to the position information of each sub-block data in the oral data, the plurality of sub-block data are fused to obtain fused data; Based on the multiple feature parameters, tooth instance segmentation is performed on the teeth in the fused data to obtain a tooth instance segmentation result corresponding to each preset virtual shape body in the fused data, including: Traversing the voxels in the fused data, selecting two voxels in the fused data that have not been segmented for tooth instances, including: selecting a first voxel close to the center of the tooth as a starting point, and selecting an adjacent unsegmented voxel as a second voxel based on position information of the first voxel; Determining the probability that the teeth corresponding to the two selected voxels are the same tooth based on the multiple characteristic parameters, including: determining an axial offset parameter and a radial offset parameter of the second voxel compared to the first voxel based on the multiple characteristic parameters, wherein the axial offset parameter is used to indicate an offset distance of the second voxel compared to the first voxel in the axial direction of the preset virtual deformation body, and the radial offset parameter is used to indicate an offset distance of the second voxel compared to the first voxel in the radial direction of the preset virtual deformation body; determining the probability that the teeth corresponding to the two voxels are the same tooth based on the axial offset parameter and the radial offset parameter; A set of voxels corresponding to the same tooth is determined according to the probability, and a tooth instance segmentation result corresponding to each tooth is obtained based on the set of voxels corresponding to the same tooth.

2. The tooth segmentation method according to claim 1, characterized in that: The oral cavity data is segmented to obtain a plurality of sub-block data, including: Determine the data of the plurality of teeth in the oral data based on the semantic segmentation result of the oral data; The oral cavity data is segmented, and the plurality of sub-block data are obtained according to the plurality of teeth data.

3. The tooth segmentation method according to claim 1, characterized in that: The preset virtual shape body includes a virtual cylinder, the preset virtual shape body corresponds to the tooth, and the voxels in the virtual cylinder correspond to the voxels in the corresponding tooth.

4. The tooth segmentation method according to claim 3, characterized in that: The multiple characteristic parameters include: a center degree value of each voxel, wherein the center degree value is used to indicate the distance between each voxel and the center of the preset virtual deformer; The expression parameters of the preset virtual deformation body corresponding to each voxel include: the deformation offset vector of each voxel, the axial direction of the preset virtual deformation body, the axial length of the preset virtual deformation body and the radial width of the preset virtual deformation body, wherein the deformation offset vector is used to indicate the offset of each voxel and the corresponding voxel in the preset virtual deformation body.

5. The tooth segmentation method according to any one of claims 1 to 4, characterized in that: The fusing the plurality of sub-block data to obtain fused data includes: If there are at least two sub-block data with overlapping block data, determining at least two different results corresponding to each characteristic parameter of each voxel in the overlapping block; By averaging the at least two different results, an updated characteristic parameter of each characteristic parameter of each voxel in the overlapping block is obtained, and the characteristic parameter of the corresponding voxel in the fused data is obtained based on the updated characteristic parameter.

6. The tooth segmentation method according to claim 1, characterized in that: The loss function of the feature prediction model includes a first loss function and a second loss function, wherein: The first loss function comprises a center degree loss function, and the center degree loss function is used to determine the difference between the center degree value of each voxel in each sub-block data and a preset target center degree value; The second loss function comprises an instance segmentation loss function, and the instance segmentation loss function is used to determine the difference between the tooth instance segmentation result corresponding to each preset virtual deformable body and the target tooth segmentation result.

7. The tooth segmentation method according to claim 1, characterized in that: The method further comprises: A graphical user interface is used to display the tooth instance segmentation result.

8. A cloud platform, characterized in that: The cloud platform executes the tooth segmentation method according to any one of claims 1 to 4, wherein the cloud platform receives the oral data uploaded by an electronic device, or the cloud platform retrieves the oral data stored in the cloud.

9. An electronic device, characterized in that: The electronic device comprises: one or more processors; A memory; and one or more application programs, wherein the one or more application programs are stored in the memory and configured to be executed by the processor to implement the tooth segmentation method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the tooth segmentation method according to any one of claims 1 to 7 is implemented.

Images