Method for generating gingival complementation network and method and device for complementing gingiva

Abstract

The invention provides a method for generating a gingival complementing network and a method and device for complementing gingiva. The method comprises the following steps: acquiring oral cavity scanning data; preprocessing the oral cavity scanning data, and obtaining a real symbol distance field value and a real curvature value corresponding to the target query point; according to voxel data of the oral cavity scanning data, calculating a predicted symbol distance field value and a predicted curvature value corresponding to the target query point by using an initial gingival completion network; and performing parameter adjustment processing on the initial gingival completion network according to the predicted symbol distance field value, the predicted curvature value, the real symbol distance field value and the real curvature value of the target query point to obtain a trained gingival completion network. According to the gingiva complementing network, the gingiva which is attractive and realistic can be constructed for the dental crown.

Description

Technical Field

[0001] The present application relates to the field of digital medical technology, and in particular to a method for generating a gingival completion network, a method for completing gingiva, and a device. Background Art

[0002] Gum generation plays a key role in the digital orthodontic treatment plan viewing process. In orthodontic treatment, doctors can use tooth data to predict gum changes and treatment effects after tooth arrangement; if the plan viewer of digital orthodontic software can display a complete and real oral scan data, it will help doctors diagnose and significantly improve treatment effects.

[0003] At present, the gum part of the oral scan data obtained by a digital scanner is usually incomplete and appears as a non-closed surface. In related technologies, spline curve interpolation and patch subdivision techniques are generally used to construct virtual gums, and this virtual gum is used as the real gum, but the generated gum cannot reproduce the unique gum details and lacks realism. Therefore, a method for generating complete gums is urgently needed. Summary of the invention

[0004] In view of this, one of the technical problems solved by the embodiments of the present application is to provide a method for generating a gingival completion network, a method and device for completing gingivae, so as to overcome the problems in the prior art.

[0005] A first aspect of an embodiment of the present application discloses a method for generating a gingival completion network, the method comprising: acquiring oral scan data; wherein the oral scan data is grid data obtained after scanning the inside of the oral cavity of a target object; preprocessing the oral scan data, and acquiring a true signed distance field value and a true curvature value corresponding to a target query point; wherein the target query point is obtained by query point sampling of the gingival completion data, and the gingival completion data is used to characterize the state of the oral scan data after the gingival completion processing; according to the voxel data of the oral scan data, using the initial gingival completion network to calculate and obtain the predicted signed distance field value and predicted curvature value corresponding to the target query point; adjusting the parameters of the initial gingival completion network according to the predicted signed distance field value, predicted curvature value, true signed distance field value and true curvature value of the target query point to obtain a trained gingival completion network.

[0006] A second aspect of an embodiment of the present application discloses a method for completing gums, the method comprising: obtaining oral scan data to be predicted; preprocessing the oral scan data to be predicted to obtain voxel data of predicted query points and crown data to be predicted; inputting the voxel data into a pre-trained gum completion network to obtain signed distance field values ​​corresponding to the predicted query points; obtaining prediction result data based on the signed distance field values ​​corresponding to the predicted query points; wherein the prediction result data is used to characterize the state of the gums after completion.

[0007] According to a third aspect of an embodiment of the present application, a device for generating a gingival completion network is disclosed. The device includes: a data acquisition module for acquiring oral scan data; wherein the oral scan data is grid data obtained after scanning the inside of the oral cavity of a target object; a real value acquisition module for preprocessing the oral scan data and acquiring a real signed distance field value and a real curvature value corresponding to a target query point; wherein the target query point is obtained by sampling the gingival completion data, and the gingival completion data is used to characterize the state of the oral scan data after the gingival completion processing; a predicted value acquisition module for obtaining a predicted signed distance field value and a predicted curvature value corresponding to the target query point by using an initial gingival completion network calculation according to voxel data of the oral scan data; and a model training module for adjusting parameters of the initial gingival completion network according to the predicted signed distance field value, predicted curvature value, real signed distance field value and real curvature value of the target query point to obtain a trained gingival completion network.

[0008] A fourth aspect of an embodiment of the present application discloses a device for completing gums, the device comprising: an acquisition module for acquiring oral scan data to be predicted; a processing module for preprocessing the oral scan data to be predicted to obtain voxel data of predicted query points and crown data to be predicted; a prediction module for inputting the voxel data into a pre-trained gum completion network to obtain signed distance field values ​​corresponding to the predicted query points; a completion module for obtaining prediction result data based on the signed distance field values ​​corresponding to the predicted query points; wherein the prediction result data is used to characterize the state of the gums after completion.

[0009] The fifth aspect of an embodiment of the present application discloses 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 described in the first aspect or the second aspect.

[0010] A sixth aspect of an embodiment of the present application discloses a non-transitory computer-readable storage medium storing computer instructions, where the computer instructions are used to enable a computer to execute the method described in the first aspect or the second aspect.

[0011] A seventh aspect of an embodiment of the present application discloses a computer program product, including a computer program, which implements the method described in the first aspect or the second aspect when executed by a processor.

[0012] In an embodiment of the present application, oral scan data is first acquired, which is grid data obtained after scanning the inside of the oral cavity of the target object; then, the oral scan data is preprocessed, and the real signed distance field value and the real curvature value corresponding to the target query point are acquired; wherein the target query point is obtained by sampling the query point of the gingival completion data, and the gingival completion data is used to characterize the state of the oral scan data after the gingival completion processing; then, according to the voxel data of the oral scan data, the initial gingival completion network is used to calculate and obtain the predicted signed distance field value and the predicted curvature value corresponding to the target query point; then, according to the predicted signed distance field value, predicted curvature value, real signed distance field value and real curvature value of the target query point, the initial gingival completion network is adjusted and processed to obtain the trained gingival completion network, and the real signed distance field value and the real curvature value can be used as the label data of the gingival completion network respectively, to obtain the trained gingival completion network; the gingival completion network can construct gingiva for the crown that is both aesthetically pleasing and realistic. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in 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.

[0014] Figure 1 It is a flow chart of a method for generating a gingival completion network disclosed in Example 1 of the present application;

[0015] Figure 2 It is a schematic flow chart of a method for gum filling disclosed in Example 2 of the present application;

[0016] Figure 3 It is a schematic diagram of the complete oral scan data;

[0017] Figure 4 It is a schematic block diagram of the structure of a device for generating a gingival completion network disclosed in Embodiment 3 of the present application;

[0018] Figure 5 This is a schematic block diagram of the structure of a device for gum filling disclosed in Example 3 of the present application. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0020] It should be noted that the terms "first", "second", "third" and "fourth" in the specification and claims of the present application are used to distinguish different objects rather than to describe a specific order. The terms "including" and "having" in the embodiments of the present application and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment including a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0021] Embodiment 1

[0022] like Figure 1 As shown, Figure 1 This is a schematic flow chart of a method for generating a gingival completion network disclosed in Example 1 of the present application. The method for generating a gingival completion network includes:

[0023] Step S101, obtaining oral scan data; wherein the oral scan data is grid data obtained after scanning the inside of the oral cavity of the target object.

[0024] In this embodiment, the grid data may be obtained by scanning the inside of the oral cavity of different target objects using an oral scanning device, for example, the grid data may be obtained by scanning the inside of the oral cavity of 100 real patients.

[0025] Here, oral scanning equipment can generate 3D point cloud data of oral structures, and then convert these point cloud data into mesh data. For example, CBCT (cone beam computed tomography), intraoral scanners, etc., can capture 3D point cloud data of oral structures such as teeth, gums, and jaws.

[0026] In stomatology, the three-dimensional coordinate system is mainly used to represent and process the spatial position information of the oral structure. Through the three-dimensional coordinate system, the arrangement of teeth, occlusion relationship, jaw morphology, etc. can be intuitively seen, so as to make more accurate diagnosis and treatment planning.

[0027] Step S102, pre-processing the oral scan data, and obtaining the true signed distance field value and the true curvature value corresponding to the target query point; wherein the target query point is obtained by sampling the query point of the gingival completion data, and the gingival completion data is the data after the oral scan data is processed for gingival completion.

[0028] The oral scan data in step S101 may be data before gum completion.

[0029] In one example, after acquiring the oral scan data, the oral scan data may be preprocessed. The following description explains the data obtained by preprocessing the oral scan data, wherein the preprocessing may include at least one of the following: facet-level crown segmentation, gingival completion annotation, data normalization, voxelization, query point sampling, vertex sampling, etc.

[0030] Among them, crown segmentation is a common technical process in the fields of stomatology, dental treatment and three-dimensional reconstruction. Its main purpose is to accurately separate the crown from the overall structure of the tooth.

[0031] In one example, a crown segmentation algorithm may be used to process the oral scan data to obtain target crown data.

[0032] The gum completion annotation is used to complete the missing gums in the oral scan data to obtain completed gum data. The completed gum data can be used as a training target for the gum completion network to detect the training effect of the gum completion network.

[0033] Among them, data normalization is performed to complete the gingival data, for example, at least one of the number, size and direction of the crowns is unified so that the gingival completion network can learn unified features (i.e., the shape and size of the crowns are kept consistent before and after gingival completion).

[0034] Voxelization is the process of converting crown data in three-dimensional space (i.e., target crown data) into discrete voxel (VolumetricPixel, volume pixel) data. Each voxel represents a small cube in three-dimensional space and may contain information about the tissue characteristics in the area (such as density, color, etc.).

[0035] In this embodiment, the voxel data of the target crown data and the target query point are used as the input of the gingival completion network to regress and infer the predicted signed distance field value from a given query point (i.e., the target query point) to the surface (i.e., the surface of the completed gingival data). and the predicted curvature value

[0036] It should be noted that the oral scan data surface refers to the data representation of the three-dimensional structure of the oral cavity obtained through digital oral scanning technology. This technology usually uses optical scanning, laser scanning or cone beam computed tomography (CBCT) to capture accurate three-dimensional images of the oral cavity.

[0037] In one example, a resolution N can be preset, and the target crown data can be voxelized to obtain the voxel data of the target crown data. At the same time, the query point sampling is performed on the completed gingival data, and the true signed distance field value d of the target query point in the three-dimensional coordinates is calculated, and the true signed distance field value d is used as the label data of the gingival completion network.

[0038] It should be noted that query point sampling is performed on the completed gingival data to obtain the true signed distance field value and the true curvature value corresponding to the target query point, wherein the target query point can be any query point.

[0039] In this embodiment, the true curvature value c of the target query point in the three-dimensional coordinate system and the true signed distance field value d of the target query point in the three-dimensional coordinate system are respectively used as label data of the gum completion network. The label data is used to complete the gums in the oral scanning data to approach the completed gum data.

[0040] Step S103, according to the voxel data of the oral scan data, the predicted signed distance field value and the predicted curvature value corresponding to the target query point are calculated using the initial gingival completion network.

[0041] In this embodiment, the voxel data of the oral scan data is input into the initial gum completion network to obtain the predicted signed distance field value and the predicted curvature value corresponding to the target query point.

[0042] Here, the predicted signed distance field value corresponding to the target query point is the predicted value of the distance surface of the target query point.

[0043] Step S104, adjusting parameters of the initial gingival completion network according to the predicted signed distance field value, predicted curvature value, true signed distance field value and true curvature value of the target query point to obtain a trained gingival completion network.

[0044] In this embodiment, the loss value between the predicted signed distance field value of the target query point and the true signed distance field value of the target query point is determined, and the loss value between the predicted curvature value of the target query point and the true curvature value of the target query point is determined; then, the parameters of the gingival completion network are adjusted according to the two loss values, and multiple iterations are performed until the above two loss values ​​respectively meet their corresponding iteration cutoff conditions, thereby obtaining a trained gingival completion network.

[0045] In an embodiment of the present application, oral scan data is first acquired, which is grid data obtained after scanning the inside of the oral cavity of a target object; then, the oral scan data is preprocessed, and a true signed distance field value and a true curvature value corresponding to a target query point are acquired; wherein the target query point is obtained by sampling the query point of the gingival completion data, and the gingival completion data is used to characterize the state of the oral scan data after the gingival completion processing; then, based on the voxel data of the oral scan data, an initial gingival completion network is used to calculate and obtain the predicted signed distance field value and predicted curvature value corresponding to the target query point; then, the initial gingival completion network is adjusted according to the predicted signed distance field value, predicted curvature value, true signed distance field value and true curvature value of the target query point to obtain a trained gingival completion network, which can construct gingiva for the crown that is both aesthetically pleasing and realistic. During the training process of the gingival completion network, neighborhood features (i.e., features obtained from data within a preset range from the gum line) and multi-scale features are spliced ​​to further highlight the characteristics of the gum line. In actual digital orthodontic applications, the gingival completion network is used to construct gingiva for crowns that are both beautiful and realistic during the automated tooth arrangement stage.

[0046] In one embodiment of the present application, the initial gingival completion network includes a multi-scale sampling module, a feature aggregation module, a first decoding module and a second decoding module; and according to the voxel data, the initial gingival completion network is used to calculate and obtain the predicted signed distance field value and the predicted curvature value corresponding to the target query point, including:

[0047] The voxel data of the oral scan data is processed by a multi-scale sampling module to obtain the splicing features of the target query point; the splicing features of the target query point are input into a feature aggregation module to obtain the curvature features of each vertex; wherein each vertex is obtained by vertex sampling of the completed gingival data; the splicing features of the target query point are input into a first decoding module to obtain a predicted signed distance field value corresponding to the target query point; and the curvature features of each vertex are input into a second decoding module to obtain a predicted curvature value.

[0048] Here, the multi-scale sampling module is used to extract features from the voxel data to obtain multi-scale features of the target query point; then, the multi-scale features of the target query point are spliced ​​with its domain features to obtain spliced ​​features. Optionally, data within different ranges from the gum line can be sampled to obtain query points within different ranges from the gum line.

[0049] In one example, vertex sampling is performed on the completed gingival data to extract a vertex set and obtain the average curvature value of each vertex (ie, the true curvature value) as label data of the curvature enhancement module (ie, label data of the gingival completion network).

[0050] In one example, according to each vertex p vt The three-dimensional coordinates of the target query point and the splicing features F(p) are calculated by the feature aggregation module to obtain p vt The curvature characteristic F c (p vt ) to better capture features at the gum line.

[0051] In this embodiment, p is calculated by the feature aggregation module. vt The curvature characteristic F c (p vt ) is only effective in the training of the gum completion network to guide more accurate reconstruction of gum line features.

[0052] In one example, in the process of training the gum completion network, a more accurate reconstruction of the gum line feature is guided. The feature aggregation module is implemented using the inverse distance weighting method to calculate the weight distribution of the target query point under the coordinates of a given vertex, and complete the aggregation of shape features to obtain the curvature feature F c (p vt ).

[0053] Here, a feature aggregation module with aggregation function in the prior art can also be used to calculate the curvature features of the vertices according to the coordinates of the vertices and the splicing features of the target query points, so as to better capture the features at the gum line.

[0054] In one example, the concatenation feature of the target query point is input into a first decoding module to obtain a predicted signed distance field value corresponding to the target query point; and the curvature feature of each vertex is input into a second decoding module to obtain a predicted curvature value, including:

[0055] In the process of training the gingival completion network, the concatenation feature F(P) of the target query point and the curvature feature F(P) of each vertex are used to calculate the c (p vt ) are sent to two decoding modules D consisting of four fully connected layers with no shared parameters. s and D c , to get the predicted symbol distance field value and the predicted curvature value

[0056] In one embodiment of the present application, the voxel data of the oral scan data is processed by a multi-scale sampling module to obtain the splicing features of the target query point, including:

[0057] The voxel data of the oral scan data is input into the multi-scale sampling module to obtain the shape features;

[0058] Perform trilinear interpolation on the shape features to obtain multi-scale features of the target query point;

[0059] The multi-scale features of the target query point and the neighborhood features of the target query point are spliced ​​to obtain the splicing features of the target query point; wherein the neighborhood features of the target query point are features obtained by processing data within a preset range from the gum line in the distance completion gum data.

[0060] In this embodiment, the voxel data X is input into the multi-scale downsampling module E(X), which extracts shape feature sets of different scales step by step {F i}. After downsampling, compared with the input feature F i , output feature F i+1 The resolution is reduced to half of the original one, and the receptive field is doubled, so that the shape features gradually transition from local detail features to global structure features.

[0061] In this embodiment, feature extraction is performed on the voxel data of the target crown data to obtain shape features, which are discrete shape features in a three-dimensional coordinate system. Since the target query point p is continuous in the three-dimensional coordinate system, this embodiment uses trilinear interpolation to determine the multi-scale features of the target query point f(p)={F1(p), F2(p),…, F n (p)}.

[0062] Here, the neighborhood features of the target query point are features obtained by processing the data within a preset range from the gum line in the distance-completed gum data; that is, the data within a preset range from the gum line in the distance-completed gum data is first sampled to obtain the target query point within the preset range from the gum line; then, the features of the data within the preset range from the gum line are encoded and extracted to obtain the neighborhood features of the target query point.

[0063] In one example, query point sampling is performed on the surface of the completed gum data and its neighborhood data (i.e., data within a preset range from the gum line), and the true signed distance field values ​​of the query points under the three-dimensional coordinates are calculated as label data for the gum completion network. In order to make the network pay more attention to the gum line.

[0064] Correspondingly, for a target query point, not only the multi-scale features of the target query point itself are collected, but also the neighborhood features of the target query point are included, and the neighborhood features are spliced ​​with the multi-scale features of point p to form the spliced ​​features F(P) at the target query point.

[0065] In this embodiment, points near the gum line are selected so that the gum completion network pays more attention to the gum line area, enhances the geometric details of the area near the gum line, and highlights the gum line features. Optionally, the gum line features can be highlighted for sampling points (ie, query points) within different ranges from the gum line.

[0066] It should be noted that directly using points on the gum line may lead to poor results in the gum completion network. Because the gum line itself is relatively small or not prominent enough in three-dimensional space, query points sampled directly from the gum line may not be sufficient to represent the geometric structure of the surrounding area. If you only rely on points on the gum line, it is not enough to support the smooth transition of the gum line area to the adjacent non-gum line area, which will cause discontinuities or jumps in the gum completion network near the gum line. In addition, selecting points near the gum line helps capture the neighborhood details of the gum line, thereby generating a more realistic and accurate gum completion network.

[0067] In one embodiment of the present application, the initial gingival completion network is adjusted according to the predicted signed distance field value, predicted curvature value, true signed distance field value and true curvature value of the target query point to obtain a trained gingival completion network, including:

[0068] The predicted signed distance field value of the target query point, the predicted curvature value of the target query point, the true signed distance field value of the target query point, and the true curvature value of the target query point are respectively input into a preset loss function to obtain a loss value corresponding to the distance field value and a loss value corresponding to the curvature value; wherein the preset loss function is a loss function represented by an implicit function;

[0069] The initial gum completion network is adjusted according to the loss value corresponding to the distance field value and the loss value corresponding to the curvature value to obtain the trained gum completion network.

[0070] In this embodiment, the parameters of the gum filling network are adjusted by the loss value corresponding to the distance field value and the loss value corresponding to the curvature value.

[0071] Here, the loss function represented by the implicit function is used to calculate the loss value between the predicted signed distance field value and the true signed distance field value, as well as the predicted curvature value and the true curvature value.

[0072] In one embodiment of the present application, the formula of the preset loss function is:

[0073] L=L SDF +L BCE +L CUR

[0074] Among them, L is the loss function expressed by implicit function; is the predicted symbol distance field value, d is the true symbol distance field value; L BCE For values ​​in the range y∈{0,1} Binary cross entropy is used for supervision, where 1 means the target query point is within the surface of the completed gingival data, and 0 means the target query point is outside the surface of the completed gingival data; is the predicted curvature value, c is the true curvature value, and x is the loss value.

[0075] Among them, L SDF Loss calculation predicts the signed distance field value The L2 distance from the true signed distance field value d. y∈{0,1}, the target query point is inside the surface of the completed gingival data, and 0 means that the target query point is outside the surface of the completed gingival data. The judgment of whether the target query point is inside or outside the surface is determined by the decision surface of the specified threshold.

[0076] Among them, for the predicted curvature value And the true curvature value c, use Smooth L1Loss to calculate L CUR .

[0077] In this embodiment, oral scan data is first acquired; then, the oral scan data is preprocessed, and the true signed distance field value and the true curvature value corresponding to the target query point are acquired; wherein the target query point is obtained by sampling the query point of the gingival completion data, and the gingival completion data is used to characterize the state of the oral scan data after the gingival completion processing; and the voxel data of the oral scan data is subjected to multi-scale feature extraction using multiple downsampling modules to extract the shape features of the local details and the global structure; then, according to the coordinates of the sampled target query point, the coordinates of the target query point are continuous coordinates in the three-dimensional space, and the coordinates of the target query point are continuous coordinates in the distance field. The multi-scale features of the target query point are interpolated on the scattered shape features; the multi-scale features are spliced ​​with their domain features to obtain the spliced ​​features; in the training stage, the curvature features are aggregated through the curvature enhancement module to obtain the curvature features of the gum line; then, the spliced ​​features and curvature features are feature decoded respectively to obtain the predicted signed distance field value and the predicted curvature value of the target query point; then, the parameters of the gum completion network are adjusted according to the loss value between the predicted signed distance field value and the true signed distance field value of the target query point, as well as the loss value between the predicted curvature value and the true curvature value, to obtain the trained gum completion network.

[0078] Embodiment 2

[0079] like Figure 2 As shown, Figure 2 This is a schematic flow chart of a method for gum filling disclosed in Example 2 of the present application. The method for gum filling includes:

[0080] Step S201, obtaining oral scan data to be predicted;

[0081] In this embodiment, a plurality of oral cavity scanning data to be predicted in a three-dimensional coordinate system may be collected by an oral cavity scanning device.

[0082] Step S202, pre-processing the oral scan data to be predicted to obtain voxel data of the prediction query point and the crown data to be predicted.

[0083] In this embodiment, the oral scan data to be predicted is preprocessed, for example, preprocessing operations such as facet-level crown segmentation, voxelization, and query point sampling are performed to obtain prediction query points and voxel data of the crown data to be predicted.

[0084] Among them, crown segmentation is a common technical process in the fields of stomatology, dental treatment and three-dimensional reconstruction. Its main purpose is to accurately separate the crown from the overall structure of the tooth.

[0085] In one example, a tooth crown segmentation algorithm may be used to process the oral cavity scan data to be predicted to obtain the tooth crown data to be predicted.

[0086] Voxelization is the process of converting the crown data to be predicted in three-dimensional space (i.e., three-dimensional crown data) into discrete voxel (VolumetricPixel, volume pixel) data. Each voxel represents a small cube in three-dimensional space and may contain information about the tissue characteristics in the area (such as density, color, etc.).

[0087] In one example, a resolution N may be preset, and the crown data to be predicted may be voxelized to obtain voxel data.

[0088] Step S203: input the voxel data into a pre-trained gingival completion network to obtain a signed distance field value corresponding to the predicted query point.

[0089] Here, the gingival filling network can be Figure 1 Gingiva completion network generated by the method in the paper.

[0090] Step S204, obtaining prediction result data according to the signed distance field value corresponding to the prediction query point, wherein the prediction result data is used to characterize the state of the gums after completion.

[0091] It should be noted that in the inference stage, the curvature enhancement module is not effective and only D s Output the prediction result data.

[0092] The embodiment of the present application can input the crown data to be predicted into the gum completion network, and can generate complete three-dimensional gum tooth data with realism and aesthetics (i.e., complete the gums in the oral scan data to be predicted). In actual digital orthodontic applications, the method of the present application can be applied to the orthodontic process, including the completion of the oral scan data to be predicted and the generation of gums for the crowns after tooth arrangement, and has high practical value.

[0093] In this embodiment, the gums in the oral scan data to be predicted are completed by predicting the signed distance field value corresponding to the query point, so that the complete oral scan data can be obtained.

[0094] In one embodiment of the present application, the method further includes:

[0095] The prediction result data is combined with the crown data to be predicted to obtain complete oral scan data; wherein the combination is used to ensure that the crown shape and size of the complete oral scan data are consistent with the crown data to be predicted.

[0096] In one example, a three-dimensional Boolean union operation is used to union the prediction result data with the crown data to be predicted.

[0097] In this embodiment, the prediction result data and the crown data to be predicted (for example Figure 3 (a1) or (b1)) shown in the figure is combined to obtain complete oral scan data, that is, oral scan data with the missing gums completed, for example Figure 3 (a2) or (b2) shown; wherein, is used to ensure that the shape and size of the crown of the complete oral scan data are consistent with the crown data to be predicted.

[0098] Embodiment 3

[0099] Embodiment 3 of the present application provides a device for generating a gingival filling network. Figure 4 This is a schematic diagram of a structure of a device for generating a gingival filling network disclosed in Example 3 of the present application, the device comprising:

[0100] Figure 4 As an implementation of the methods shown in the above figures, the present invention discloses an embodiment of a device for generating a gingival filling network. Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0101] like Figure 4As shown, an embodiment of the present invention provides a device for generating a gingival completion network, the device comprising: a data acquisition module 401 , a real value acquisition module 402 , a predicted value acquisition module 403 and a model training module 404 . Among them, the data acquisition module 401 is used to acquire oral scan data; wherein the oral scan data is grid data obtained after scanning the inside of the oral cavity of the target object; the real value acquisition module 402 is used to preprocess the oral scan data and obtain the real signed distance field value and the real curvature value corresponding to the target query point; wherein the target query point is obtained by sampling the query point of the gingival completion data, and the gingival completion data is used to characterize the state of the oral scan data after the gingival completion processing; the predicted value acquisition module 403 is used to obtain the predicted signed distance field value and the predicted curvature value corresponding to the target query point by using the initial gingival completion network calculation according to the voxel data of the oral scan data; the model training module 404 is used to adjust the parameters of the initial gingival completion network according to the predicted signed distance field value, predicted curvature value, real signed distance field value and real curvature value of the target query point to obtain the trained gingival completion network.

[0102] In this embodiment, the specific processing of the data acquisition module 401, the real value acquisition module 402, the predicted value acquisition module 403 and the model training module 404 in the device for generating the gingival completion network and the technical effects thereof can be referred to respectively. Figure 1 This corresponds to steps S101 to S104 in the embodiment.

[0103] In some embodiments, the initial gingival completion network includes a multi-scale sampling module, a feature aggregation module, a first decoding module and a second decoding module; and a predicted value acquisition module 403, including: a first unit, used to process the voxel data of the oral scan data through the multi-scale sampling module to obtain the splicing features of the target query point; a second unit, used to input the splicing features of the target query point into the feature aggregation module to obtain the curvature features of each vertex in the vertex set; wherein each vertex is obtained by vertex sampling of the completed gingival data; a third unit, used to input the splicing features of the target query point into the first decoding module to obtain the predicted signed distance field value corresponding to the target query point; and input the curvature features of each vertex into the second decoding module to obtain the predicted curvature value.

[0104] In some embodiments, the first unit is specifically used to: input voxel data of oral scanning data into a multi-scale sampling module to obtain shape features; perform trilinear interpolation on the shape features to obtain multi-scale features of a target query point; splice the multi-scale features of the target query point with the neighborhood features of the target query point to obtain the splicing features of the target query point; wherein the neighborhood features of the target query point are features obtained by processing data within a preset range of the gum line of the distance-completing gum data.

[0105] In some embodiments, the model training module 404 is specifically used to: input the predicted symbol distance field value, predicted curvature value, true symbol distance field value and true curvature value of the target query point into a preset loss function respectively, to obtain a loss value corresponding to the distance field value and a loss value corresponding to the curvature value; wherein the preset loss function is a loss function represented by an implicit function; and adjust the parameters of the initial gingival completion network according to the loss value corresponding to the distance field value and the loss value corresponding to the curvature value to obtain a trained gingival completion network.

[0106] In some embodiments, the formula of the preset loss function is:

[0107] L=L SDF +L BCE +L CUR

[0108] Among them, L is the loss function expressed by implicit function; is the predicted symbol distance field value, d is the true symbol distance field value; L BCE For values ​​in the range y∈{0,1} Use binary cross entropy for supervision; is the predicted curvature value, c is the true curvature value, and x is the loss value corresponding to the curvature value.

[0109] Embodiment 4

[0110] Embodiment 4 of the present application provides a device for generating a gingival filling network. Figure 5 This is a schematic diagram of a structure of a device for generating a gingival filling network disclosed in Example 4 of the present application, the device comprising:

[0111] Figure 5 As an implementation of the methods shown in the above figures, the present invention discloses an embodiment of a device for filling gums. Figure 1 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0112] like Figure 5As shown, an embodiment of the present invention provides a device for completing gums, the device comprising: an acquisition module 501, a processing module 502, a prediction module 503 and a completion module 504. The acquisition module 501 is used to acquire oral scan data to be predicted; the processing module 502 is used to pre-process the oral scan data to be predicted to obtain voxel data of predicted query points and crown data to be predicted; the prediction module 503 is used to input the voxel data into a pre-trained gum completion network to obtain the signed distance field value corresponding to the predicted query point; the completion module 504 is used to obtain the prediction result data according to the signed distance field value corresponding to the predicted query point; wherein the prediction result data is used to characterize the state of the gums after completion.

[0113] In the present embodiment, the specific processing of the acquisition module 501, the processing module 502, the prediction module 503 and the completion module 504 and the technical effects thereof can be referred to respectively. Figure 2 This corresponds to steps S201 to S204 in the embodiment.

[0114] In some embodiments, the device further includes: a merging module for merging the prediction result data with the crown data to be predicted to obtain complete oral scan data; wherein the merging is used to ensure that the shape and size of the crown of the complete oral scan data are consistent with the crown data to be predicted.

[0115] The gum filling device of this embodiment can realize the corresponding gum filling methods in the aforementioned multiple method embodiments and has the beneficial effects of the corresponding method embodiments, which will not be described in detail here.

[0116] So far, specific embodiments of the present application have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions recorded in the claims can be performed in a different order and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing can be advantageous.

[0117] The present application is described with reference to the flowcharts and / or block diagrams of the methods according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1A device that provides the functions specified in a block or multiple blocks.

[0118] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0119] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods and devices. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0120] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0121] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for generating a gingival filling network, characterized in that: The method comprises: Acquire oral scan data; wherein the oral scan data is grid data obtained after scanning the inside of the oral cavity of the target object; Preprocessing the oral scan data, and obtaining a true signed distance field value and a true curvature value corresponding to a target query point; wherein the target query point is obtained by sampling the query point of the gingival completion data, and the gingival completion data is used to characterize the state of the oral scan data after the gingival completion processing; According to the voxel data of the oral scan data, using the initial gingival completion network to calculate and obtain the predicted signed distance field value and the predicted curvature value corresponding to the target query point; The initial gingival completion network is adjusted according to the predicted signed distance field value, predicted curvature value, true signed distance field value and true curvature value of the target query point to obtain a trained gingival completion network.

2. The method according to claim 1, characterized in that The initial gum completion network includes a multi-scale sampling module, a feature aggregation module, a first decoding module and a second decoding module; and the predicted signed distance field value and the predicted curvature value corresponding to the target query point are calculated using the initial gum completion network according to the voxel data of the oral scan data, including: Processing the voxel data of the oral scan data by the multi-scale sampling module to obtain the splicing features of the target query point; Inputting the splicing features of the target query point into the feature aggregation module to obtain the curvature features of each vertex; wherein each vertex is obtained by performing vertex sampling on the completed gingival data; The splicing feature of the target query point is input into the first decoding module to obtain the predicted signed distance field value corresponding to the target query point; and the curvature feature of each vertex is input into the second decoding module to obtain the predicted curvature value.

3. The method according to claim 2, characterized in that The step of processing the voxel data of the oral scan data by the multi-scale sampling module to obtain the splicing features of the target query point includes: Inputting the voxel data of the oral scan data into the multi-scale sampling module to obtain shape features; Performing trilinear interpolation on the shape feature to obtain a multi-scale feature of the target query point; The multi-scale features of the target query point and the neighborhood features of the target query point are spliced ​​to obtain the spliced ​​features of the target query point; wherein the neighborhood features of the target query point are features obtained by processing data within a preset range of the gum line from the completed gum data.

4. The method according to any one of claims 1 to 3, characterized in that: The initial gingival completion network is adjusted according to the predicted signed distance field value, the predicted curvature value, the true signed distance field value and the true curvature value of the target query point to obtain a trained gingival completion network, including: Inputting the predicted signed distance field value, predicted curvature value, true signed distance field value and true curvature value of the target query point into a preset loss function respectively, and obtaining a loss value corresponding to the distance field value and a loss value corresponding to the curvature value; wherein the preset loss function is a loss function represented by an implicit function; The initial gum completion network is adjusted according to the loss value corresponding to the distance field value and the loss value corresponding to the curvature value to obtain a trained gum completion network.

5. The method according to claim 4, characterized in that The formula of the preset loss function is: L=L SDF +L BCE +L CUR Among them, L is the loss function expressed by implicit function; is the predicted symbol distance field value, d is the true symbol distance field value; L BCE For values ​​in the range y∈{0,1} Use binary cross entropy for supervision; is the predicted curvature value, c is the true curvature value, and x is the loss value corresponding to the curvature value.

6. A method for gum restoration, characterized in that: The method comprises: Acquire oral scan data to be predicted; Preprocessing the oral scan data to be predicted to obtain voxel data of the prediction query point and the crown data to be predicted; Inputting the voxel data into a gingival completion network generated by the method according to any one of claims 1 to 5 to obtain a signed distance field value corresponding to the predicted query point; The prediction result data is obtained according to the signed distance field value corresponding to the prediction query point; wherein the prediction result data is used to characterize the state of the gums after completion.

7. The method according to claim 6, characterized in that The method further comprises: The prediction result data is combined with the crown data to be predicted to obtain complete oral scan data; wherein the combination is used to ensure that the crown shape and size of the complete oral scan data are consistent with the crown data to be predicted.

8. A device for generating a gingival filling network, characterized in that: The device comprises: A data acquisition module, used to acquire oral scan data; wherein the oral scan data is grid data obtained after scanning the inside of the oral cavity of the target object; A real value acquisition module, used for preprocessing the oral scan data and acquiring a real signed distance field value and a real curvature value corresponding to a target query point; wherein the target query point is obtained by sampling the query point of the gum completion data, and the gum completion data is used to characterize the state of the oral scan data after the gum completion processing; A prediction value acquisition module, used to obtain the predicted signed distance field value and the predicted curvature value corresponding to the target query point by using the initial gingival completion network calculation according to the voxel data of the oral scan data; The model training module is used to adjust the parameters of the initial gingival completion network according to the predicted signed distance field value, predicted curvature value, true signed distance field value and true curvature value of the target query point to obtain a trained gingival completion network.

9. The device according to claim 8, characterized in that The initial gingival completion network includes a multi-scale sampling module, a feature aggregation module, a first decoding module and a second decoding module; and the prediction value acquisition module includes: A first unit is used to process the voxel data of the oral scan data through the multi-scale sampling module to obtain the splicing features of the target query point; The second unit is used to input the splicing features of the target query point into the feature aggregation module to obtain the curvature features of each vertex; wherein each vertex is obtained by performing vertex sampling on the completed gingival data; The third unit is used to input the splicing feature of the target query point into the first decoding module to obtain the predicted signed distance field value corresponding to the target query point; and input the curvature feature of each vertex into the second decoding module to obtain the predicted curvature value.

10. A device for filling gums, characterized in that: The device comprises: An acquisition module, used for acquiring oral scan data to be predicted; A processing module, used for preprocessing the oral scan data to be predicted, and obtaining voxel data of the prediction query point and the crown data to be predicted; A prediction module, configured to input the voxel data into a gingival completion network generated by the method according to any one of claims 1 to 5, to obtain a signed distance field value corresponding to the predicted query point; The completion module is used to obtain prediction result data according to the signed distance field value corresponding to the prediction query point; wherein the prediction result data is used to characterize the state of the gums after completion.

Images