Tooth orthodontic correction movement vector detection method, device, medium and equipment

Abstract

The invention discloses a tooth orthodontic correction movement vector detection method and device, a medium and equipment. The method comprises the following steps: acquiring a first model before tooth orthodontic correction and a second model after tooth orthodontic correction of a person with teeth to be corrected; preprocessing the first model and the second model to obtain a preprocessed first model and a preprocessed second model; carrying out registration on the preprocessed first model and the preprocessed second model so as to obtain a tooth orthodontic correction registration model; and based on the tooth orthodontic correction registration model, detecting a movement vector of the person with the teeth to be corrected after orthodontic correction relative to the movement vector before orthodontic correction. According to the invention, the motion vector of the tooth after orthodontic correction relative to the motion vector before orthodontic correction can be accurately detected.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and specifically relates to a method, device, medium and equipment for detecting a tooth orthodontic correction movement vector. Background Art

[0002] The effectiveness of orthodontic correction is extremely important for maintaining oral health, which is related to the correctness of tooth arrangement and the improvement of chewing function. Improper correction force may cause loose teeth and gum problems, while insufficient correction may lead to unbalanced bite and joint discomfort. If the treatment progress does not meet expectations, in addition to prolonging the treatment course, it may also bring psychological pressure to the patient. Therefore, accurately judging the effect comparison before and after tooth correction is a key link in the whole process.

[0003] During the orthodontic process, a certain area of ​​the upper jaw, especially the area within about 3 to 5 mm close to the gingival margin, showed morphological stability. This means that certain specific landmarks change slightly between multiple correction review intervals, providing a stable reference basis for alignment.

[0004] At present, the technology for evaluating the correction effect covers the use of machine learning to automatically mark key areas, combined with image processing technology to analyze in two-dimensional dental model images, and extended to the automatic positioning of three-dimensional models. Another approach is to directly analyze the three-dimensional oral model and identify feature areas through specific algorithms. Although these technologies have improved the processing speed, they are limited by the dependence and accuracy of data annotation, especially in complex areas. Human participation may lead to errors and increase workload. Traditional model matching technologies such as ICP require high initial setting accuracy, which adds to the difficulty of operation.

[0005] In summary, existing technologies have exposed deficiencies in efficiency, accuracy, and automation levels in monitoring the effectiveness of orthodontic treatment, highlighting the need for more efficient and accurate methods for detecting orthodontic effects. Summary of the invention

[0006] In view of the deficiencies in the prior art, the purpose of the present invention is to provide a method, device, medium and equipment for detecting tooth orthodontic correction movement vector, which can accurately detect the movement vector of teeth after orthodontic correction relative to before correction.

[0007] To achieve the above objectives, the present disclosure provides the following technical solutions:

[0008] A method for detecting a tooth orthodontic correction movement vector, the method comprising: obtaining a first model of a tooth subject before orthodontic correction and a second model after orthodontic correction; preprocessing the first model and the second model to obtain the preprocessed first model and the second model; aligning the preprocessed first model and the second model to obtain an orthodontic correction registration model; and detecting the movement vector of the tooth subject after orthodontic correction relative to before orthodontic correction based on the orthodontic correction registration model.

[0009] Optionally, preprocessing the first model and the second model includes the following steps: cropping the first model and the second model to obtain a first cropped model and a second cropped model; correcting the first cropped model and the second cropped model to obtain a first corrected model and a second corrected model; segmenting the first corrected model and the second corrected model to obtain a first model to be registered and a second model to be registered; and aligning the first model to be registered and the second model to be registered to obtain an orthodontic correction registration model.

[0010] The present disclosure also provides a device for detecting orthodontic correction movement vector, the device comprising: an acquisition module, used to acquire a first model of a person whose teeth need to be corrected before orthodontic correction and a second model after orthodontic correction; a preprocessing module, used to preprocess the first model and the second model to obtain the preprocessed first model and the second model; a registration module, used to register the preprocessed first model and the second model to obtain an orthodontic correction registration model; a detection module, used to detect the movement vector of the tooth after orthodontic correction relative to that before orthodontic correction based on the orthodontic correction registration model.

[0011] The present disclosure also provides a tooth orthodontic correction movement vector detection system, which includes the device as described above.

[0012] The present disclosure also provides a device for detecting a tooth orthodontic correction movement vector, the device comprising a processor for executing any of the above methods.

[0013] The present disclosure also provides a tooth orthodontic correction movement vector detection system, and the system applies any of the methods described above.

[0014] The present disclosure also provides a storage medium comprising instructions, which, when executed on a computer, enables the computer to execute any of the above methods.

[0015] The present disclosure also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above methods when executing the program.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] 1. The present invention realizes the fully automatic registration of the models before and after orthodontic correction by adopting a three-step registration algorithm, which can significantly improve the efficiency and accuracy of orthodontic effect evaluation.

[0018] 2. The present invention can accurately detect the movement vector of teeth after orthodontic correction relative to that before correction, thereby ensuring the effect of tooth correction and avoiding potential harm caused by over- or under-correction. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of a tooth orthodontic correction detection method proposed in one embodiment of the present disclosure;

[0020] Figure 2 (a) is a schematic diagram of the structure of the first model;

[0021] Figure 2 (b) is a schematic diagram of the structure of the second model;

[0022] Figure 3 It is a schematic diagram of the coordinate system construction of the tooth model with standard orientation;

[0023] FIG4( a ) is a schematic structural diagram of a first trimming model proposed in another embodiment of the present disclosure;

[0024] FIG4( b ) is a schematic structural diagram of a second trimming model proposed in another embodiment of the present disclosure;

[0025] FIG5( a ) is a schematic diagram of the structure of a first calibration model proposed in another embodiment of the present disclosure;

[0026] FIG5( b ) is a schematic diagram of the structure of a second calibration model proposed in another embodiment of the present disclosure;

[0027] FIG6 (a) is a schematic diagram of the structure of a first segmentation model proposed in another embodiment of the present disclosure;

[0028] FIG6( b ) is a schematic diagram of the structure of a second segmentation model proposed in another embodiment of the present disclosure;

[0029] Figure 7 is a schematic diagram of a U-shaped curved surface of an alveolar bone proposed in another embodiment of the present disclosure;

[0030] FIG8( a ) is a schematic structural diagram of a first model to be registered proposed in another embodiment of the present disclosure;

[0031] FIG8( b ) is a schematic diagram of the structure of a second model to be registered proposed in another embodiment of the present disclosure;

[0032] Fig. 9is a red palate wrinkle wall area registration diagram after registration of digital models of teeth before and after orthodontic correction, as proposed in another embodiment of the present disclosure;

[0033] Fig.10 is the standard deviation of the displacement of each tooth in the first model and the second model proposed in another embodiment of the present disclosure after 5 alignment experiments. DETAILED DESCRIPTION

[0034] The following will refer to the attached Figures 1 to 10 Specific embodiments of the present disclosure are described in detail. Although specific embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0035] It should be noted that certain words are used in the specification and claims to refer to specific components. Those skilled in the art should understand that technicians may use different nouns to refer to the same component. This specification and claims do not use the difference in nouns as a way to distinguish components, but use the difference in the functions of the components as the criterion for distinction. As mentioned throughout the specification and claims, "including" or "comprising" is an open term, so it should be interpreted as "including but not limited to". The subsequent description of the specification is a preferred embodiment of the present disclosure, but the description is based on the general principles of the specification and is not used to limit the scope of the present disclosure. The scope of protection of the present disclosure shall be determined by the attached claims.

[0036] To facilitate understanding of the embodiments of the present disclosure, further explanation will be given below using specific embodiments as examples in conjunction with the accompanying drawings, and each of the accompanying drawings does not constitute a limitation on the embodiments of the present disclosure.

[0037] In an exemplary embodiment, Figure 1 As shown, the present disclosure provides a method for orthodontic correction detection, comprising:

[0038] S100: obtaining a first model of the teeth of the patient before orthodontic correction and a second model of the teeth after orthodontic correction, as shown in FIG. 2( a ) and FIG. 2( b ) respectively;

[0039] S200: preprocessing the first model and the second model to obtain the preprocessed first model and the second model;

[0040] S300: aligning the preprocessed first model and the second model to obtain an orthodontic correction alignment model;

[0041] S400: Detecting the movement vector of the teeth of the person to be corrected after orthodontic correction relative to before orthodontic correction based on the orthodontic correction registration model.

[0042] This embodiment can ensure the consistency of the tooth models before and after correction in spatial position and direction by registering the first model of the tooth of the person to be corrected before orthodontic correction and the second model after orthodontic correction. By accurately aligning the two models, the corresponding positions of the teeth can accurately correspond. If not registered, there may be differences in rotation, translation or scaling between the two models, and direct comparison cannot obtain the real tooth movement information. In addition, through registration, manual intervention can be reduced, errors caused by personal experience or subjective judgment can be avoided, and the tooth movement vector detection results are more objective and reliable. Among them, the first model refers to the three-dimensional digital model of the teeth of the person to be corrected before receiving orthodontic correction, which is obtained by intraoral scanning technology. The model reflects the real state of the teeth of the person to be corrected before the correction, including information such as the position, shape and occlusal relationship of the teeth. The second model refers to the three-dimensional digital model of the teeth of the same person to be corrected after completing the current orthodontic correction, which is obtained by using the same intraoral scanning technology again. The model shows the changes in the position of the teeth of the person to be corrected after the current orthodontic correction.

[0043] In another exemplary embodiment, in step S200, preprocessing the first model and the second model includes the following steps:

[0044] S201: trimming the first model and the second model to obtain a first trimmed model and a second trimmed model;

[0045] S202: calibrating the first trimming model and the second trimming model to obtain a first calibrated model and a second calibrated model;

[0046] S203: Segment the first calibrated model and the second calibrated model to obtain a first model to be registered and a second model to be registered.

[0047] In another exemplary embodiment, in step S201, the first model and the second model are trimmed to obtain a first trimmed model and a second trimmed model, including the following steps:

[0048] S2011: constructing a first minimum directional bounding box of the first model, and constructing a second minimum directional bounding box of the second model;

[0049] In this step, the acquired tooth models (the first model and the second model) are first imported into a three-dimensional modeling or point cloud processing software (such as Open3D software) for preliminary denoising and filtering to ensure the clarity and accuracy of the model data.

[0050] Secondly, use the convex hull algorithm or optimization method (the convex hull algorithm is an algorithm used to calculate the smallest convex polygon or polyhedron of a set of points. This convex polygon or polyhedron contains all the input points, and any point not on the boundary will not be included in the convex polygon or polyhedron. In three-dimensional space, the convex hull is the smallest convex polyhedron that contains all points. The convex hull algorithm can be used to simplify point cloud data, feature extraction, collision detection and other application scenarios. Common convex hull algorithms include Graham scanning algorithm, Jarvis stepping algorithm (incremental algorithm), Monotone Chain algorithm (monotone chain algorithm), etc. Optimization method refers to the process of finding the extreme value (minimum or maximum value) of the objective function. Optimization methods are usually used to solve the problem of finding the best solution under constraints. These methods are widely used in machine learning, data mining, engineering design and other fields. Common optimization methods include gradient descent method, Newton method, conjugate gradient method, quasi-Newton method, etc.) to calculate the minimum oriented bounding box (OBB) of each model. When calculating the minimum oriented bounding box (OBB), the convex hull algorithm or optimization method described above can help determine the optimal direction and size so that the bounding box can fit the target model closely.

[0051] Secondly, output the calculated bounding box parameters, including but not limited to the center point, size, rotation angle, etc. of the bounding box, and use visualization tools (such as Open3D software, MeshLab, Blender, 3D Slicer, CloudCompare, etc.) to check whether the bounding box reasonably covers the main part of the tooth model, especially near the occlusal surface. The specific steps include:

[0052] 1. Import the first or second model into any of the above processing software. Make sure the model data has been loaded correctly and can be viewed and operated in the software.

[0053] 2. Display OBB: Calculate and display OBB in the software. Most professional software provides the function of directly generating OBB, which can usually be achieved through menu options or related command line instructions.

[0054] 3. Adjust the viewing angle: Adjust the viewing angle to observe the relationship between the model and the OBB from different directions. Pay special attention to the occlusal area to ensure that this key area is properly covered by the OBB. If you find that the occlusal surface is not completely covered by the OBB, you can try to fine-tune the position of the OBB or expand its size until the occlusal surface and the surrounding important structures are reasonably included. Specifically, you can use the transformation tools provided by the software to translate, rotate or scale the OBB so that the OBB can better adapt to the actual shape of the model.

[0055] 4. Verify details: Zoom in on the local view and carefully check the occlusal surface and other key areas to ensure that these locations are inside the OBB.

[0056] Different from the axis-aligned bounding box, the OBB can better represent the size and orientation of the model and fit the object more closely.

[0057] S2012: identifying a first occlusal surface of the first model based on the first minimum oriented bounding box, and identifying a second occlusal surface of the second model based on the second minimum oriented bounding box;

[0058] In this step, the eight groups of vertex coordinates of the OBB are first analyzed (the eight groups of data output by the minimum directed bounding box OBB represent the eight vertex coordinates of the bounding box, which are divided into six surface information of the bounding box according to the arrangement order). According to the vertex distribution, the two largest surface areas related to the occlusal surface of the teeth are determined as candidate occlusal surfaces, namely, the E surface and the F surface as shown in Figure 4 (a) and Figure 4 (b).

[0059] Secondly, the identified first occlusal surface and the identified second occlusal surface are clearly marked on the first model and the second model respectively, so as to provide clear instructions for subsequent cutting steps.

[0060] S2013: Perform trimming along the first occlusal surface and its opposite surface toward the center of the first minimum directed bounding box to obtain a first trimmed model as shown in FIG. 4( a ), and perform trimming along the second occlusal surface and its opposite surface toward the center of the second minimum directed bounding box to obtain a second trimmed model as shown in FIG. 4( b ).

[0061] In this step, the clipping path from the center of the occlusal surface to the center of the minimum directed bounding box is first determined to ensure that the clipping plane is perpendicular to the occlusal surface, while considering the opposite surface of the occlusal surface to maintain the symmetry and integrity of the clipping.

[0062] Secondly, the software tools are used to perform trimming along the planned path to remove the redundant parts of the model away from the center of the occlusal surface, including non-critical structures such as gums and soft tissues, and focus on retaining important areas of the tooth body and alveolar bone.

[0063] The above trimming process can be further expressed by the following formula:

[0064]

[0065]

[0066]

[0067] Among them, ratio represents the cropping ratio of the first model or the second model. Represents the center vector coordinates of the first model or the second model bounding box, Represents the top vector coordinate of the bounding box of the first model or the second model, Represents the bottom vector coordinate of the bounding box of the first model or the second model, Indicates the volume reduction ratio calculated based on the top clipping direction. represents the volume reduction ratio calculated according to the bottom trimming direction, P represents the total volume of the first model or the second model before trimming, and P′ represents the volume of the first model or the second model after trimming.

[0068] In summary, this embodiment can accurately identify the occlusal surface of the teeth by constructing a minimum directed bounding box, ensure that the clipping process is highly targeted, focus on the main structural parts of the teeth, remove unnecessary background areas, thereby reducing the amount of data, and accelerating the operation speed of subsequent processing steps such as correction, segmentation and registration, thereby improving the efficiency of the overall detection process. Furthermore, the above-mentioned clipping method can determine the position and direction of the bounding box according to the specific shape and direction of the tooth model by constructing a minimum directed bounding box. This clipping method does not require any directional adjustment of the model in advance, and can directly calculate a bounding box that fits the model closely. Once the bounding box is determined, the occlusal surface of the teeth can be identified based on the bounding box, and clipping can be performed along the occlusal surface. Because OBB adjusts its direction according to the characteristics of the model itself, even if the model is in any posture, OBB can still accurately capture the key features of the model, thereby ensuring the consistency of the clipping results. Based on the above reasons, the clipping method adopted in this embodiment can be unaffected by the original orientation of the model. Even if the model is in any posture, it can be accurately performed based on the minimum directed bounding box, ensuring rotation invariance, that is, no matter how the model is rotated, the clipping result can remain consistent. Other clipping methods, such as semantic-based clipping methods, can theoretically provide better clipping effects, but this method requires the model to be in a standard posture to achieve the best effect, which means that the tooth model needs to be correctly aligned to a predefined standard frame before processing, which is difficult to achieve in the orthodontic environment with diverse and complex tooth models, increasing the complexity of the preliminary processing. In addition, in addition to the teeth themselves, the tooth model also includes complex structures such as gums and soft tissues, which will undergo morphological changes during the orthodontic process (the gums and soft tissues may undergo random changes during the orthodontic process, such as swelling, retraction or morphological changes, which are dynamic and difficult to predict, and will increase the difficulty of model registration). The semantic clipping method cannot effectively distinguish and remove these time-varying tissues, thereby introducing interference in subsequent registration and motion vector detection, affecting the accuracy of the measurement. However, the present invention will not cause morphological changes during the orthodontic process because: the present invention removes non-critical structures such as gums and soft tissues by cutting the tooth model, and only retains hard tissue structures such as the tooth body and alveolar bone. The position of such hard tissue structures is relatively stable during correction, and even if there is movement, it can be predicted and measured, thereby avoiding interference caused by morphological changes in the gums or soft tissues.

[0069] Furthermore, through the above steps, since the OBB is dynamically adjusted according to the specific shape and orientation of the model, rather than just a bounding box with a fixed orientation, it is possible to ensure that the minimum oriented bounding box (OBB) can closely fit the contour of the tooth model while taking into account the orientation of the model. Specifically, the OBB is calculated based on the actual geometry of the target model. Unlike the axis-aligned bounding box (AABB), the OBB allows rotation to find an optimal orientation so that the bounding box can wrap the entire model as compactly as possible, which means that the OBB can not only adapt to the overall size of the model, but also consider the aspect ratio and tilt angle of the model. In addition, the calculation process of the OBB includes determining the optimal rotation angle of the bounding box, which allows it to naturally take into account the orientation of the model. For example, when processing a tooth model, the OBB will automatically adjust its orientation to better match the growth direction of the tooth or the position of the occlusal surface. Such adaptive capabilities help improve the accuracy of subsequent processing steps (such as registration).

[0070] The benefit of this is that it can adapt to tooth models in different orientations, thus ensuring the consistency and effectiveness of the cutting process. In addition, this method is rotationally invariant, meaning that the cutting results remain consistent regardless of how the model is rotated, thus improving the efficiency and accuracy of subsequent processing steps.

[0071] In another embodiment, in step S202, calibrating the first trimming model and the second trimming model to obtain a first calibrated model and a second calibrated model includes the following steps:

[0072] S2021: Introduction of tooth models with standard orientations;

[0073] The tooth model with standard orientation is usually constructed based on standard anatomical posture, for example, according to the standard tooth position arrangement of FDI (International Dental Federation), ensuring that the long axis direction of the tooth, the occlusal surface and the coordinate axis are aligned. Such a model can be constructed by 3D scanning a known standard tooth model or using professional software.

[0074] The coordinate system of the introduced tooth model with standard orientation (denoted as model P) is set as the world coordinate system (such as Figure 3 As shown, a coordinate axis that satisfies the right-hand rule is defined for the tooth model with a standard direction. The point cloud center O of the model is located at the origin of the coordinate system. The positive direction of the Z axis is perpendicular to and points to the occlusal surface. The positive direction of the Y axis passes through the center of the suture formed by the left and right incisors and points to the center of the line connecting the midpoints of the left and right first molars. The coordinate center is set to .

[0075] S2022: Rotate the occlusal surface of the first trimming model as shown in FIG. 4 (a) based on the coordinate system where the tooth model with the standard orientation is located to obtain the first correction model as shown in FIG. 5 (a); rotate the occlusal surface of the second trimming model as shown in FIG. 4 (b) based on the coordinate system where the tooth model with the standard orientation is located to obtain the second correction model as shown in FIG. 5 (b).

[0076] In this embodiment, the first trimmed model shown in FIG4 (a) is also used as an example to illustrate the correction process. Specifically, since most teeth grow perpendicular to the occlusal surface, the number of normal vectors on the Z axis of the point cloud that are positive accounts for the majority. Therefore, the first trimmed model is rotated twice around its own coordinate axis so that the centers of its E and F surfaces are aligned with the center of the minimum directional bounding box. The connecting lines of are aligned with the positive direction of the Z axis of the model P, and the number of Z-axis normal vectors of all point clouds in the first clipping model that are greater than 0 is counted. The plane with the largest number of normal vectors greater than 0 is recorded as the top surface of the model, and the other side is recorded as the bottom surface of the model. The top surface of the first clipping model is rotated to the positive direction of the Z axis to complete the correction of the first clipping model.

[0077] This embodiment introduces a tooth model with a standard orientation to ensure that the first clipping model and the second clipping model can be registered in a unified coordinate frame, thereby significantly improving the accuracy of the registration between the two models and reducing the registration error caused by differences in model posture.

[0078] In another embodiment, in step S203, segmenting the first calibration model and the second calibration model to obtain a first model to be registered and a second model to be registered includes the following steps:

[0079] S2031: performing tooth instance segmentation on the first calibration model and the second calibration model respectively to obtain a tooth point cloud and an alveolar bone point cloud;

[0080] In this step, it is first necessary to perform tooth instance segmentation on the first correction model and the second correction model based on the Tgnet tooth instance segmentation algorithm to obtain the first segmentation model shown in FIG. 6 (a) and the second segmentation model shown in FIG. 6 (b).

[0081] Specifically, the Tgnet tooth instance segmentation algorithm includes a tooth segmentation network and a tooth grouping network, wherein the tooth segmentation network takes the complete first correction model or the second correction model as input, and fuses the two parts of the point cloud that have undergone the farthest point sampling and edge-aware sampling to obtain sampling points. Then, the tooth grouping network based on PointTransformer as the backbone network is used to extract and output the features of the sampling points to identify and distinguish different tooth individuals.

[0082] Furthermore, the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm was used to cluster the tooth point cloud, and each independent tooth was automatically identified and segmented based on the density distribution of the point cloud. For each cluster, the FDI tooth position representation method was used to assign a unique number to each tooth point cloud according to the anatomical position and morphological characteristics of the tooth, which facilitated subsequent tracking and alignment.

[0083] The above description clusters the tooth point cloud, and the specific calculation process is as follows:

[0084] 1. Mark all points in the tooth point cloud as unvisited;

[0085] 2. Loop through each point: if the point has been visited, continue to the next point; if the point has not been visited, retrieve its r neighborhood; if the number of points in the neighborhood is less than M, mark it as a noise point.

[0086] If the number of points in the neighborhood r is greater than or equal to M, a new cluster is created and the cluster is continuously expanded through density reachability relations.

[0087] 3. Expand the cluster: For each point in the new cluster, if there are unvisited points in the neighborhood, add these points to the cluster; if the number of points in the neighborhood of these points also meets the requirements of M, these points are also regarded as core points and continue to expand.

[0088] 4. Repeat the above steps until all points have been visited.

[0089] S2032: distinguishing the tooth point cloud from the alveolar bone point cloud, and retaining the alveolar bone point cloud;

[0090] In this step, after completing the tooth instance segmentation, the alveolar bone and tooth point clouds need to be separated, that is, the alveolar bone point cloud and the tooth point cloud need to be distinguished from the overall point cloud, and the tooth point cloud needs to be removed, and only the alveolar bone point cloud is retained. At this time, the retained alveolar bone model is the model to be registered. Specifically, the alveolar bone point cloud can be finely cut using the minimum oriented bounding box (OBB) as described above. This strategy aims to remove irregular interference and ensure that only the U-shaped surface of the alveolar bone (such as Figure 7 As shown in the figure, more realistic and interference-free alveolar bone morphology data can be obtained, providing an accurate basis for subsequent registration work.

[0091] S2033: Optimizing the alveolar bone point cloud.

[0092] In this step, a point cloud filtering technique (such as statistical filtering or morphological filtering) is first applied to remove isolated points, cluttered points or noise in the alveolar bone point cloud to improve the purity of the point cloud.

[0093] Secondly, surface reconstruction algorithms (such as Poisson reconstruction and spherical harmonic function expansion) are used to smooth the alveolar bone point cloud while maintaining its original morphological features, reducing jagged edges, and making the alveolar bone surface smoother and more continuous to facilitate subsequent registration.

[0094] Finally, the alveolar bone point cloud is locally optimized, such as feature-based local refinement, to further improve the model's detail expression ability and accuracy, ensuring that the alveolar bone model can accurately reflect the anatomical structure and provide a more accurate basis for alignment.

[0095] After the above steps, the first model to be registered as shown in FIG. 8( a ) and the second model to be registered as shown in FIG. 8( b ) can be obtained.

[0096] In another embodiment, in step S300, registering the first model to be registered and the second model to be registered to obtain an orthodontic registration model includes the following steps:

[0097] S301: performing coordinate system-based registration on the first model to be registered and the second model to be registered to obtain a first registration model, specifically comprising:

[0098] 1. Standardized data:

[0099] The average value of the coordinates of each point in the point cloud of the first model to be registered and the second model to be registered is adjusted to 0, and the standard deviation is adjusted to 1 to ensure fairness in comparison between features.

[0100] 2. Calculate the covariance matrix Σ of the point cloud of the first model to be registered and the second model to be registered, which can be calculated by the following formula:

[0101]

[0102] in, Represents the point cloud with sequence number i The coordinates of the first model to be registered and the second model to be registered are combined, N represents the number of point clouds, T represents transposition, Represents the mean vector of the coordinates of all points in the three axis directions of the entire point cloud, and n represents the number of point clouds in the first model to be registered or the second model to be registered.

[0103] 3. Solve the eigenvalues ​​and eigenvectors of the covariance matrix Σ, which can be calculated by the following formula:

[0104]

[0105] in, represents the eigenvalue, represents a feature vector.

[0106] The eigenvector determines the direction of the new coordinate axis, and the eigenvalue determines the importance of the coordinate axis (that is, the variance of the data in that direction).

[0107] 4. Select the principal components.

[0108] The eigenvalue Sort from large to small, and select the eigenvectors corresponding to the first k largest eigenvalues ​​as the principal components. These eigenvectors constitute the basis of the new coordinate system.

[0109] The center of the U-shaped surface point cloud of the alveolar bone is set as the origin so that its model satisfies the Cartesian coordinate system, that is, the three axes of the coordinate system are perpendicular to each other and satisfy the right-hand rule.

[0110] Calculate the angle between the Y axis of the first model to be registered S and the Y axis of the second model to be registered T , using the rotation transformation formula to get the rotation matrix , the calculation formula is as follows:

[0111]

[0112] Based on the above formula, a model based on principal component analysis matching is finally obtained, that is, a first registration model in which the coordinate axes of the first model to be registered and the second model to be registered are aligned.

[0113] S302: performing global coarse registration on the first registration model to obtain a second registration model;

[0114] In this step, the RANSAC (Random Sample Consensus) global rough matching algorithm is used to perform global rough registration on the first registration model. The algorithm can find an accurate registration transformation model even in the presence of a large number of outliers. It randomly selects a set of samples from the data and uses these samples to fit the model, and judges the quality of the model by evaluating the degree of fit between other data points and the model. The following are the specific implementation steps for global rough registration based on the RANSAC algorithm:

[0115] 1. Randomly select a small group of point clouds from the first registration model as a sample set, where the small group of point clouds corresponds to the features of the target object or scene to be registered.

[0116] 2. Use the selected sample point set to fit and calculate the transformation matrix. Common transformation matrices include rigid body transformation (such as translation, rotation) and affine transformation.

[0117] 3. Apply the transformation matrix to other point cloud data to obtain the transformation model, and calculate the error metric between them and the model through the difference between corresponding points. Usually the nearest neighbor distance or reprojection error is used as the error metric.

[0118] 4. For each data point, calculate its fitting error with the fitted model. If the error is less than a predefined threshold (i.e., the inlier threshold), it is considered an inlier, otherwise it is considered an outlier.

[0119] 5. Count the number of inliers and compare it with the pre-defined inlier threshold. If the number of inliers exceeds the threshold, the model is considered to fit the data well.

[0120] 6. Repeat the above steps multiple times (the number of iterations is defined in advance) and select the model with the largest number of inliers as the best model.

[0121] 7. Use the best model to estimate the transformation relationship between point clouds, so as to achieve coarse registration of point clouds, that is, to achieve feature alignment.

[0122] It should be noted that the key idea of ​​the RANSAC algorithm is to find the best model from data containing noise and outliers through random sampling and iteration. By repeating this process, RANSAC can find an accurate registration transformation model even in the presence of a large number of outliers. For the tooth model, this step of registration can fit the U-shaped surface of the alveolar bone, extract the feature points of the surface model and align it.

[0123] S303: Perform fine registration on the second registration model to obtain Figure 7 Orthodontic registration model shown.

[0124] In the present disclosure, the global coarse registration method based on the RANSAC algorithm can align the overall direction of the alveolar bone model, and the algorithm is very accurate in the registration of the feature points of the model, but there is still a situation where the local detail registration is not accurate enough. Therefore, it is necessary to further limit and narrow the search range of the registration point cloud and perform precise registration on the second registration model. This embodiment uses the point-to-plane ICP (Iterative Closest Point) registration algorithm. ICP is a commonly used point cloud registration method for aligning a cluster of point clouds to another cluster of reference point clouds. The specific implementation steps are as follows:

[0125] 1. Initialization: Given the reference point cloud of the second registration model, generally, the source point cloud is the point cloud that needs to be registered, and the reference point cloud is a known or registered point cloud.

[0126] 2. Nearest neighbor search: For each point in the source point cloud, find the corresponding nearest neighbor point in the reference point cloud through nearest neighbor search. The purpose of this step is to establish the correspondence between point pairs.

[0127] 3. Initial transformation estimation: Using the correspondence between point pairs, an initial transformation matrix T is estimated. The example is as follows:

[0128]

[0129] in, is a 3×3 rotation matrix, Is a 3×1 translation vector. The last row of the matrix is ​​usually [0,0,0,1], so that the transformation matrix can be applied to points in the homogeneous coordinate system.

[0130] A common approach is to use the least squares method to fit the transformation matrix so that the error in the direction of the normal vector of the corresponding point in the source point cloud is minimized.

[0131] 4. Iterative optimization: The transformation matrix is ​​continuously optimized through iterations to make the registered source point cloud closer to the reference point cloud. For each iterative step, the points in the source point cloud are transformed to the coordinate system of the reference point cloud according to the current transformation matrix. For each point in the source point cloud, find its corresponding nearest neighbor in the reference point cloud. Based on the correspondence between point pairs, a new transformation matrix is ​​estimated by the least squares method to minimize the error in the direction of the normal vector of the corresponding point in the reference point cloud. According to the new transformation matrix, the position of the source point cloud is updated.

[0132] 5. Convergence determination: According to certain convergence criteria (such as the change in the transformation matrix or the change in the error), determine whether the algorithm has reached a convergence state. If convergence has not been reached, return to step 4 to continue iterating; otherwise, the algorithm ends.

[0133] 6. Output results: The final transformation matrix is ​​the result of point-to-surface ICP registration.

[0134] The point-to-plane ICP algorithm can be expressed as the following equation:

[0135]

[0136] Among them, E(R,t) represents the energy function, which is used to measure the quality of registration. The lower the value, the better the registration effect. represents the i-th point in the source point cloud, represents the rotation matrix, represents the translation vector, Representation and serial number The sequence number in the set of matched points, Indicates that the target point cloud is concentrated and The nearest point on the surface, express The normal vector of the point on the target surface, (R pi+t) represents the position of the ith point after rotation R and translation t, and N represents the total number of point clouds.

[0137] The point-to-plane ICP method continuously optimizes the registration results by minimizing the distance or error between point pairs. Compared with the point-to-point ICP method, the point-to-plane ICP method can better utilize the normal information of the point cloud in the tooth model, thereby improving the accuracy and stability of the registration. Based on the model registration of the above three steps, the registration model before and after orthodontic correction is as follows Fig. 9 As shown in Figure 9, the palatine fold is one of the criteria for judging the overlap quality. As can be seen from Figure 9, this registration method makes the palatine fold area of ​​the upper jaw roughly overlap, which has a certain objectivity in the evaluation of the therapeutic effect.

[0138] In summary, the present disclosure achieves preliminary alignment of the models by registering the first model and the second model based on the coordinate system, thus ensuring a standardized starting point for subsequent processing.

[0139] First, by standardizing the data and principal component analysis (PCA), a specific step was implemented to make the model's feature alignment more accurate, effectively reducing the complexity of the subsequent registration steps and improving the level of automation;

[0140] Secondly, the present disclosure implements a specific step of performing global coarse registration by adopting the RANSAC (Random Sample Consensus) algorithm, so that an accurate registration transformation model can still be found in the presence of a large amount of noise and abnormal data points. In the present disclosure, RANSAC randomly selects a set of sample points, fits the transformation matrix, and repeatedly iterates to select the model with the most inliers (i.e., the best fit to the model), thereby achieving robust registration of key features such as the U-shaped surface of the alveolar bone. This improves the model's ability to resist interference from abnormal data and ensures the robustness of the registration results;

[0141] Finally, by using the point-to-plane ICP (Iterative Closest Point) algorithm for precise registration, a specific step was achieved, that is, by minimizing the error of the point cloud data in the normal vector direction, the transformation matrix was continuously optimized iteratively until the model reached the best alignment state. The ICP algorithm is particularly suitable for processing local details. It uses the normal information of the point cloud in the tooth model to improve the accuracy and stability of the registration, so that the fine structures such as the palatal fold area of ​​the upper jaw can be well overlapped, ensuring the objectivity and accuracy of the efficacy evaluation.

[0142] The present disclosure creatively proposes the implementation of the specific steps of three-step registration by combining PCA, RANSAC and ICP algorithms, wherein PCA is used to extract the main directions and features of the tooth model. Through PCA, the main directions of the point cloud data can be found, making the model feature alignment more accurate, which is conducive to the efficient implementation of the subsequent registration steps; the RANSAC algorithm can effectively process the noise points that may exist in the model, improve the model's anti-interference ability to abnormal data, thereby achieving robust registration of key features such as the U-shaped surface of the alveolar bone; the ICP algorithm uses the normal information of the point cloud in the tooth model to accurately align the local details of the tooth model, such as the palatal fold area of ​​the upper jaw and other fine structures, thereby improving the accuracy and stability of the registration. The present disclosure uses a three-step registration algorithm, firstly, it can realize the fully automatic registration of the model before and after orthodontic correction, and significantly improve the efficiency of orthodontic effect evaluation. Secondly, by combining PCA, RANSAC and ICP algorithms, not only the accuracy of the registration is improved, but also the stability and practicality of the entire detection process are improved while ensuring high precision. Third, by automating the registration process, the need for manual intervention can be reduced, the risk of human error can be lowered, and processing speed can be increased.

[0143] It should be emphasized that although PCA, RANSAC and ICP algorithms are all existing technologies, the three are not a simple combination. Among them, PCA is generally used as a dimensionality reduction tool in image processing and data analysis, but in the present disclosure, it is innovatively used in orthodontic model registration, and is used for standardized data and alignment coordinate systems in orthodontic correction; RANSAC is generally used to estimate linear model parameters, such as line segments, planes, etc., but in the present disclosure, it is innovatively used in orthodontic model registration, and it is used to robustly estimate the transformation matrix to achieve global coarse registration in orthodontic correction, which is a specific innovative application for tooth models; ICP is a commonly used algorithm in point cloud registration, but in the present disclosure, the point-to-surface ICP method is used and innovatively used in orthodontics, especially using the normal information of the point cloud in the tooth model to improve the registration accuracy and stability, which is an innovative application in orthodontic model registration. In summary, although PCA, RANSAC and ICP are relatively mature and widely used algorithms in their respective fields, combining them for the registration of orthodontic models constitutes an innovative technical solution and plays a new technical role and effect in orthodontic correction. It can effectively improve the accuracy and robustness of registration during orthodontic correction, reduce manual intervention, and improve the stability and practicality of the entire detection process. Therefore, combining these three methods for the registration of orthodontic models is an unconventional and innovative technical means.

[0144] In another embodiment, in step S400, calculating the movement vector of the tooth after orthodontic correction relative to that before orthodontic correction based on the orthodontic correction registration model comprises the following steps:

[0145] S401: extracting all tooth point cloud data with corresponding numbers before and after correction from the tooth orthodontic correction registration model according to the tooth numbers;

[0146] S402: Calculating the center of gravity offset of each pair of teeth with the same number in the tooth point cloud data, wherein the offset includes the direction and distance information of the tooth movement;

[0147] In this step, for each pair of teeth with the same number, their center of gravity offset needs to be calculated. The center of gravity offset can be expressed as a vector, which contains the direction and distance information of the tooth movement. Assuming there is a pair of point cloud data of teeth, the center of gravity offset can be calculated by the following steps:

[0148] First, calculate the center of gravity. For the tooth point cloud data P before correction and the tooth point cloud data Q after correction, first calculate their respective center of gravity C P and C Q , assuming that the point cloud P consists of n points, then the centroid C P It can be expressed as:

[0149]

[0150] Similarly, for point cloud Q, center of gravity C Q It can be expressed as:

[0151]

[0152] Where m is the number of points in the point cloud Q, p i and q j are single points in point clouds P and Q respectively.

[0153] The center of gravity offset It can be expressed as:

[0154] =C Q -C P

[0155] This step can quantify the direction and distance of tooth movement during the correction process by calculating the center of gravity offset of each pair of teeth with the same number. The center of gravity offset includes not only distance information but also direction information, which is helpful for understanding the movement path of the teeth. Since the center of gravity is a stable reference point, the accuracy of the calculation can be improved by using the concept of the center of gravity to calculate the offset.

[0156] S403: Obtaining the movement vector of each tooth during the correction process based on the center of gravity offset of each pair of teeth with the same number, wherein the length of the movement vector corresponds to the total distance of tooth displacement, and the direction points to the direction of tooth movement.

[0157] In this step, for each pair of teeth with the same number, the center of gravity offset That is the moving vector V, which can be specifically expressed as:

[0158] V= =CQ-CP

[0159] This step can intuitively display the movement of each tooth by converting the center of gravity offset into a movement vector, which is convenient for clinicians and patients to understand the correction process. In addition, the movement vector provides a quantitative indicator to measure the actual movement of each tooth during the correction process. By measuring the length of the movement vector, you can know how far the tooth has moved, which helps to determine whether the correction force is sufficient and whether the expected amount of movement has occurred. The direction of the movement vector provides the specific direction of the tooth movement, which is very important for confirming whether the correction force is acting in the expected direction. For example, if you want a tooth to move outward, and the direction of the movement vector shows that the tooth has indeed moved outward, then this indicates that the correction measure is effective.

[0160] In order to verify the rotation invariance of this method, this method was rotated around any coordinate axis, and the displacement standard deviation S of the same tooth number of multiple patients with teeth to be corrected under the same rotation was calculated. td , and its calculation formula is as follows:

[0161]

[0162]

[0163] in, represents the displacement of the tth tooth of the patient with the jth tooth to be corrected, Indicates the point cloud position of the tth tooth of the jth patient to be corrected in the i-th experiment in the source point cloud, represents the point cloud position of the tth tooth of the jth patient to be corrected in the i-th experiment in the target point cloud, M represents the number of teeth, n represents the number of corresponding point sets of P and Q, || || 2 represents the L-2 norm.

[0164] The calculation result of the above formula is as follows Fig.10 As shown, Fig.10The first model before and after treatment of all patients with teeth to be corrected is rotated along any axis and any angle for 5 times, and the displacement of each tooth with the same number on the first and second precise registration models after registration by the above method is calculated, and the standard deviation of the displacement of the same tooth in 5 experiments is calculated. Fig.10 It can be seen that the displacement of each tooth has a very low standard deviation, indicating that even after multiple rotation transformations, the model still has a high registration accuracy and a certain degree of rotation invariance.

[0165] In summary, the present disclosure can understand whether the tooth arrangement reaches the expected goal and whether the application of the correction force produces the expected biomechanical response by detecting the movement vector before and after tooth correction.

[0166] Finally, the present disclosure also obtains a tooth orthodontic correction alignment model by the ICP method alone, and uses this as a comparative example to perform motion vector detection comparison with the tooth orthodontic correction alignment model obtained by the fusion of PCA, RANSAC and ICP based on the innovative application of the present disclosure. The comparison results are shown in Table 1:

[0167] Table 1

[0168]

[0169] As can be seen from Table 1, after multiple rotational alignments, the method proposed in the present disclosure has lower root mean square error (RMSE) on all three coordinate axes compared to the direct use of the ICP alignment method (RMSE is a method of measuring the difference between an estimated value and a true value. The smaller the value, the closer the estimated value is to the true value, i.e., the higher the accuracy). This indicates that the method based on PCA, RANSAC and ICP fusion proposed in the present disclosure has significantly higher alignment accuracy in orthodontic correction. The higher the alignment accuracy, the more accurately it can reflect the movement of teeth during orthodontic correction, thereby providing more accurate data support for the detection of tooth movement. Therefore, the present disclosure also has higher accuracy in tooth movement detection by adopting the three-step alignment method.

[0170] In another embodiment, the present disclosure further provides a tooth orthodontic correction detection device, comprising:

[0171] An acquisition module, used for acquiring a first model of the teeth of the person to be corrected before orthodontic correction and a second model of the teeth after orthodontic correction;

[0172] A preprocessing module, used for preprocessing the first model and the second model to obtain the preprocessed first model and the second model;

[0173] A registration module, used for registering the preprocessed first model and the second model to obtain an orthodontic registration model;

[0174] The detection module is used to detect the movement vector of the teeth after orthodontic correction relative to that before orthodontic correction based on the orthodontic correction registration model.

[0175] In another exemplary embodiment, the present disclosure further provides a system for detecting a tooth orthodontic correction movement vector, the system comprising the device as described above.

[0176] In another exemplary embodiment, the present disclosure further provides a device for detecting a tooth orthodontic correction movement vector, the device comprising a processor for executing any of the methods described above.

[0177] In another exemplary embodiment, the present disclosure further provides a system for detecting orthodontic correction movement vectors, wherein the system applies any of the above methods.

[0178] In another exemplary embodiment, the present disclosure further provides a storage medium comprising instructions, which, when executed on a computer, enables the computer to execute any of the above methods.

[0179] In another exemplary embodiment, the present disclosure further provides an electronic device, the electronic device comprising:

[0180] A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein:

[0181] When the processor executes the program, any of the above methods is implemented.

[0182] The above embodiments are only for illustrating the technical concept and features of the present disclosure, and their purpose is to enable people familiar with the technology to understand the content of the present disclosure and implement it accordingly, and they cannot be used to limit the protection scope of the present disclosure. Any equivalent changes or modifications made according to the spirit of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A method for detecting a tooth orthodontic correction movement vector, characterized in that: The method comprises: Obtain a first model of the teeth of the patient before orthodontic correction and a second model of the teeth after orthodontic correction; Preprocessing the first model and the second model to obtain the preprocessed first model and the second model; Aligning the preprocessed first model and the second model to obtain an orthodontic correction alignment model; The movement vector of the teeth of the person to be corrected after orthodontic correction relative to before orthodontic correction is detected based on the orthodontic correction registration model.

2. The method according to claim 1, characterized in that Preferably, preprocessing the first model and the second model comprises the following steps: trimming the first model and the second model to obtain a first trimmed model and a second trimmed model; Correcting the first trimming model and the second trimming model to obtain a first corrected model and a second corrected model; The first corrected model and the second corrected model are segmented to obtain a first model to be registered and a second model to be registered.

3. A tooth orthodontic correction movement vector detection device, characterized in that: The device comprises: An acquisition module, used for acquiring a first model of the teeth of the person to be corrected before orthodontic correction and a second model of the teeth after orthodontic correction; A preprocessing module, used for preprocessing the first model and the second model to obtain the preprocessed first model and the second model; A registration module, used for registering the preprocessed first model and the second model to obtain an orthodontic registration model; The detection module is used to detect the movement vector of the teeth after orthodontic correction relative to that before orthodontic correction based on the orthodontic correction registration model.

4. A tooth orthodontic correction movement vector detection system, characterized in that: The system comprises the apparatus of claim 3.

5. A tooth orthodontic correction movement vector detection device, characterized in that: The device comprises a processor, configured to execute the method according to any one of claims 1 to 2.

6. A tooth orthodontic correction movement vector detection system, characterized in that: The system applies the method according to any one of claims 1 to 2.

7. A storage medium, characterized in that: The method comprises instructions, which, when executed on a computer, enable the computer to execute the method according to any one of claims 1 to 2.

8. An electronic device, characterized in that: The electronic device comprises: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 2 is implemented.

Images