Digitized intelligent orthodontic diagnosis and treatment method
By combining features such as point density and local curvature to automatically calculate the probability of defects and using grid completion technology, the problem of automatic completion of missing areas in point clouds in orthodontics is solved, and the detection accuracy and adaptability are improved.
Patent Information
- Application Number
- CN202510732353.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-03
- Publication Date
- 2025-09-12
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing three-dimensional optical scanning methods in orthodontics result in missing areas in the point cloud due to patient movement and posture deviation. Traditional completion methods are not effective in irregular and soft tissue-covered areas and lack automated judgment capabilities.
By combining self-explanatory features such as point density, local curvature and regional symmetry, the true defect probability of the defective block is automatically calculated, and fully automated completion is achieved using existing grid completion technology.
It improves the accuracy of automatic detection and completion of defective areas, reduces errors, adapts to different scanning equipment and environments, and reduces the workload of manual adjustment.
Smart Images
Figure CN120635356A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of orthodontic diagnosis and treatment, and specifically to a digital orthodontic intelligent diagnosis and treatment method. Background Art
[0002] With the development of digital medicine and intelligent diagnostic technology, the field of orthodontics generally uses 3D optical scanning or intraoral scanners to obtain point cloud data of the patient's oral surface. 3D point clouds provide an accurate basic model for subsequent correction plan design, customized brackets or invisible braces. However, in actual clinical scanning, due to involuntary movements of the patient's tongue and lips, breathing, or deviations in mouth opening posture, point cloud missing areas are often generated in local areas of the oral cavity. Traditional point cloud completion mostly relies on symmetry assumptions or template matching-based methods, such as mirror mapping of the contralateral dentition or standard oral model, geometric interpolation based on mean surface fitting, etc. These methods can well restore the model for regular symmetrical defects or small-scale holes, but are not effective for irregular, asymmetric areas and areas covered by soft tissue, often resulting in geometric distortion or excessive smoothing of the completed area, increasing errors in subsequent diagnosis and treatment plans. On the other hand, point cloud quality assessment and true defect determination mostly rely on manual visual inspection or simple threshold screening based on point cloud sparsity, lacking multi-dimensional geometric feature fusion and difficult to quickly and automatically judge. To this end, this method proposes a digital orthodontic intelligent diagnosis and treatment method that integrates multi-dimensional geometric features to automatically judge the authenticity of defects and efficiently complete them. This method needs to combine self-explanatory features such as point density, local curvature and regional symmetry. Without relying on large-scale annotation and deep models, it automatically calculates the true defect probability of the defect block, and calls the existing one-click mesh completion technology after confirming the true defect, realizing full automation and explainability of the clinical process. Summary of the Invention
[0003] The present invention provides a digital intelligent orthodontic diagnosis and treatment method, which helps solve the problems mentioned in the above background technology.
[0004] The present invention provides the following technical solution: a digital intelligent orthodontic diagnosis and treatment method, comprising: The three-dimensional oral point cloud is obtained by scanning and recorded as ; right Perform denoising and unify the normal of each point in the point cloud, and output the processed point cloud ; Point Cloud Filter out the defective areas and form a defective block set ; Calculate the set of missing blocks The geometric feature vector of each block in ; The geometric feature vector of each block , calculate the true defect probability ; Set the probability threshold for determining whether to rescan to obtain a 3D oral point cloud ; Comparison with actual defect probability With probability threshold ; like , it will prompt you to rescan to obtain the 3D oral point cloud; like , the defective area is filled and a complete three-dimensional oral point cloud is obtained.
[0005] Optionally, the pair Perform denoising and unify the normal of each point in the point cloud, and output the processed point cloud ,include: For each point ,Calculate point cloud Other points in The Euclidean distance of The Represented in point cloud The points, whose coordinates are three-dimensional vectors ; Set the number of neighbor points ; Point Cloud Medium For other points outside the The Euclidean distance is sorted from small to large; In the sorting order, select Points form a set, marked as points The k-nearest neighbor set ; Get Point The k-nearest neighbor set For each point in , calculate the point The average neighborhood distance ; Traversing the point cloud At each point in the , forming a neighborhood set; Calculate the mean in a set of neighborhoods and standard deviation ; Set the distance outlier removal threshold : in: , represents the empirical coefficient, which is used to relax the judgment criteria; If there is a point of , then determine the point It is a noise point and is removed; Traversing the point cloud For each point in the image, remove all noise points and output the processed point cloud .
[0006] Optionally, the acquisition point The k-nearest neighbor set For each point in , calculate the point The average neighborhood distance , specifically: The calculation point The average neighborhood distance : in: Represents the set of k nearest neighbors The Point coordinates; represents the Euclidean distance; The mean value in the calculated neighborhood set and standard deviation , specifically: in: Indicates the total number of elements in the neighborhood set.
[0007] Optionally, the pair Perform denoising and unify the normal of each point in the point cloud, and output the processed point cloud , also includes: Point Construct the local covariance matrix for the center: Where: the subscript T represents the matrix transpose; Represents the neighborhood centroid: beg The characteristic decomposition of , the minimum eigenvalue The corresponding eigenvector is used as the normal ; Get the scanner's viewing angle vector when scanning and acquiring 3D oral point clouds ; calculate and compare with 0; when When , flip the normal, that is ; Otherwise, keep the normal unchanged.
[0008] Optionally, the point cloud Filter out the defective areas and form a defective block set ,include: For point clouds Each point in , with radius For the range, get all the points in the range to form a set ; For each point Perform point density field calculations: in: Indicates a point The density of , which represents the number of points per unit area; Indicates radius is the set of points in the neighborhood; Indicates the number of points in the set; represents the neighborhood area; For each point , perform curvature estimation: in: Indicates the surface curvature, the smaller it is, the smoother it is; Indicated by point Construct the eigenvalues of the local covariance matrix for the center.
[0009] Optionally, the point cloud Filter out the defective areas and form a defective block set , also includes: Get point cloud Each point in Density and surface curvature ; Point Cloud Point in The total number of ; Mean of the global statistical density and standard deviation :
[0010] Similarly, the mean of the global statistical surface curvature is and standard deviation ; To set the filter threshold:
[0011] For each point ,like and , it is determined to be a possible defect point; Get all points that are judged as possible defect points to form a possible defect point set ; Use Euclidean distance clustering to extract defective blocks: Set the neighborhood radius , minimum number of neighbors ; Initialize Collection All points in the state are "not visited"; S2, the machine selects an "unvisited" point , marked as "visited", query its -Neighborhood: in: Representing a collection Other points in Indicates the Euclidean distance, and the subscript 2 indicates that the L2 norm is used; S3, if , then mark If it is noise, it will be discarded; otherwise, a new cluster is started. ,Will and all points in its neighborhood are added , and repeat steps S2 and S3 for each new point in the neighborhood until there are no new extensible points; Repeat S2 and S3 to traverse all "unvisited" points until all points are marked; Output all the A collection of ; Get all the missing blocks to form a missing block set .
[0012] Optionally, the calculation of the defective block set The geometric feature vector of each block in ,include: For each piece : Calculating missing blocks The average density of all points in ; Calculating missing blocks The average surface curvature of all points in ; Calculate 3D features: in: Indicates that After symmetry, the average distance between the original block and the original block is obtained by fitting the plane symmetry, specifically: Get Block Points contained in The number of ; Block , calculate its center of mass :
[0013] Constructing the covariance matrix : right Perform eigendecomposition and obtain three eigenvalues and orthogonal eigenvectors: ; The smallest eigenvalue corresponding to As the plane normal vector ; perpendicular to the plane normal vector And through the point A plane is defined as the fitting plane ; For any block Point in , calculate its Mirror point : in: Represents the directed distance from a point to a plane; For any block Point in , calculate the distance between it and the mirror point :
[0014] Take the average of all points to get the symmetric error : in: Expressed as the distance between each point and its mirror image point.
[0015] Optionally, the geometric feature vector of each block , calculate the true defect probability ,include: Set the scoring function : in: , , represents the empirical weighting coefficient, which is set to 0.4, 0.3, and 0.3 respectively; , used to normalize the symmetric error; represents the standard deviation of the global statistical density; represents the standard deviation of the global statistical surface curvature; Calculating the true defect probability : in: and Represents the mean and standard deviation of the scores precomputed in multiple high-quality samples.
[0016] The present invention has the following beneficial effects: 1. A statistical filtering algorithm effectively removes isolated noise points by calculating the Euclidean distance between each point and a fixed number of k nearest neighbors and applying a global threshold. This method is more adaptive than traditional methods that simply remove points based on the mean distance because it uses a dynamic threshold weighted by the standard deviation of the mean, which can account for uneven point cloud density. Furthermore, the normal estimation and orientation consistency steps utilize local PCA to construct a centralized covariance matrix for the neighborhood point set and calculate the eigenvector corresponding to the minimum eigenvalue as the normal, avoiding the directional bias introduced by a non-centered matrix. The normal is compared with the scanner's view vector and flipped when necessary to ensure that all normals are uniformly oriented toward the scan source. This process simultaneously achieves noise reduction and stable normal acquisition without requiring a large number of pre-set parameters, providing accurate geometric information for subsequent defect detection and completion. Compared to existing methods that rely solely on global filtering or empirical thresholding, this method can dynamically adapt and maintain consistency under high noise conditions and varying scanner parameters. This step reduces the error rate in subsequent processing and significantly reduces the workload of manual threshold adjustment, enabling rapid deployment of the algorithm in various clinical scenarios.
[0017] 2. The local point density and curvature indicators calculated based on the number of points in the radius neighborhood and the neighborhood covariance characteristics can fully reflect the local geometric sparseness and flatness of the point cloud; the mean and standard deviation of the density and curvature distribution are calculated through global statistics, and dynamic thresholds are set. and the lower limit of empirical curvature to achieve efficient screening of "possible defect points"; combined with the dual conditions of low density and low curvature, it eliminates misjudgments caused by normal tooth depressions or high curvature areas; compared with traditional single density-based or single curvature-based threshold screening, the dual threshold joint strategy of this algorithm effectively reduces false detections and missed detections; then the selected candidate points are clustered by density reachability, and further clustered based on the ε neighborhood and Parameters distinguish between noise, boundary points and core points; the clustering method can adapt to defect point clusters with different density distributions, avoiding the drawbacks of one-size-fits-all Euclidean distance or unstable parameters; the entire detection process not only ensures the degree of automation of the algorithm, but also significantly improves the adaptability to different defect forms; this step does not rely on external annotation data or traditional template libraries, and adapts to a variety of oral local structure change characteristics, providing a reliable regional block division basis for subsequent authenticity judgment.
[0018] 3. Construct a three-dimensional geometric feature vector for each defective block , integrating three complementary features of point density, local curvature and symmetry error; using weighted linear combination score function , and normalize the scores to probabilities by This process not only retains the interpretable linear feature combination, but also uses the pre-calculated statistical samples to and , ensuring the stability of the probability mapping; this method can be directly used in new equipment and new environments without the need for large-scale labeled data; the multi-dimensional weighting strategy allows users to emphasize single features according to different scenarios, such as increasing the symmetry weight to enhance the recognition of asymmetric artifacts; combined with the probability threshold judgment mechanism, it can automatically screen out low-confidence artifacts caused by tongue and lip movements, and accurately identify the real defect area; this step significantly improves the robustness and flexibility of defect discrimination, reduces the error of human experience judgment, and provides a reliable decision-making basis for the next step of completion.
[0019] 4. By Query to obtain k nearest neighboring points; then calculate the relationship between these neighboring points and point The Euclidean distance of each point is calculated and the arithmetic mean is calculated. This operation is highly adaptive and efficient compared to fixed radius or random sampling methods. First, the use of a fixed k-nearest neighbor ensures that each point has a stable number of neighbors, thereby avoiding instability caused by too many or too few neighbors in areas with uneven point cloud density. Second, averaging the distance from each point to its neighboring points helps smooth local density fluctuations and reduce the influence of outliers. The average neighborhood distance obtained in this way not only accurately reflects the local point cloud distribution state, but also provides a reliable basis for subsequent noise removal based on statistical thresholds. At the same time, the dynamic threshold judgment mechanism formed by the mean plus standard deviation can automatically adjust the point cloud noise level of different scanning devices and different parts, thereby improving the robustness and generalization ability of the algorithm.
[0020] 5. Construct a dynamic threshold based on the calculated neighborhood average distance, combined with the global mean and standard deviation The statistical filtering mechanism can adapt to different point cloud density distributions and take into account both overall and local differences. In actual clinical scans, there are complex environments such as soft tissue and tongue occlusion inside the oral cavity, and the point cloud density is extremely uneven. Traditional fixed threshold or empirical value elimination often leads to over-elimination or residual noise. This method uses a threshold based on sample statistical characteristics to finely distinguish between noise points and edge points. Specifically, isolated floating points will be accurately eliminated because their distance from the neighborhood is much greater than the global distribution. Normal concave points in dense areas will not be mistakenly deleted due to the limitations of the mean and fluctuation range. This step does not require prior knowledge and can be completed self-consistently based on the current scan data. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] Figure 1 It is a basic flow chart of the present invention. DETAILED DESCRIPTION
[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.
[0023] Example, see Figure 1 , a digital orthodontic intelligent diagnosis and treatment method, including: The three-dimensional oral point cloud is obtained by scanning and recorded as ; right Perform denoising and unify the normal of each point in the point cloud, and output the processed point cloud A statistical filtering algorithm effectively removes isolated noise points by calculating the Euclidean distance between each point and a fixed number of k nearest neighbors and applying a global threshold. This method is more adaptive than traditional simple elimination based on the mean distance because it uses a dynamic threshold weighted by the standard deviation of the mean to account for scenarios with uneven point cloud density. Furthermore, the normal estimation and direction consistency steps utilize local PCA to construct a centralized covariance matrix for the neighborhood point set and calculate the eigenvector corresponding to the minimum eigenvalue as the normal, avoiding the directional bias introduced by the non-centered matrix. The normal is compared with the scanner's view vector and flipped when necessary to ensure that all normals are uniformly oriented toward the scan source. This process simultaneously achieves noise reduction and stable normal acquisition without requiring a large number of pre-set parameters, providing accurate geometric information for subsequent defect detection and completion. Compared with existing methods that rely solely on global filtering or empirical thresholds, this method can dynamically adapt and maintain consistency under high noise levels and varying scanning device parameters. This step reduces the error rate in subsequent processing and significantly reduces the workload of manual threshold adjustment, enabling rapid deployment of the algorithm in various clinical scenarios.
[0024] Point Cloud Filter out the defective areas and form a defective block set The local point density and curvature indicators calculated based on the number of points in the radius neighborhood and the neighborhood covariance characteristics can fully reflect the local geometric sparsity and flatness of the point cloud; the mean and standard deviation of the density and curvature distribution are calculated through global statistics, and dynamic thresholds are set. and the lower limit of empirical curvature to achieve efficient screening of "possible defect points"; combined with the dual conditions of low density and low curvature, it eliminates misjudgments caused by normal tooth depressions or high curvature areas; compared with traditional single density-based or single curvature-based threshold screening, the dual threshold joint strategy of this algorithm effectively reduces false detections and missed detections; then the selected candidate points are clustered by density reachability, and further clustered based on the ε neighborhood and Parameters distinguish between noise, boundary points and core points; the clustering method can adapt to defect point clusters with different density distributions, avoiding the drawbacks of one-size-fits-all Euclidean distance or unstable parameters; the entire detection process not only ensures the degree of automation of the algorithm, but also significantly improves the adaptability to different defect forms; this step does not rely on external annotation data or traditional template libraries, and adapts to a variety of oral local structure change characteristics, providing a reliable regional block division basis for subsequent authenticity judgment.
[0025] Calculate the set of missing blocks The geometric feature vector of each block in ; The geometric feature vector of each block , calculate the true defect probability ; Set the probability threshold for determining whether to rescan to obtain a 3D oral point cloud ; By constructing a three-dimensional geometric feature vector for each defective block , integrating three complementary features of point density, local curvature and symmetry error; using weighted linear combination score function , and normalize the scores to probabilities by This process not only retains the interpretable linear feature combination, but also uses the pre-calculated statistical samples to and , ensuring the stability of the probability mapping; this method can be directly used in new equipment and new environments without the need for large-scale labeled data; the multi-dimensional weighting strategy allows users to emphasize single features according to different scenarios, such as increasing the symmetry weight to enhance the recognition of asymmetric artifacts; combined with the probability threshold judgment mechanism, it can automatically screen out low-confidence artifacts caused by tongue and lip movements, and accurately identify the real defect area; this step significantly improves the robustness and flexibility of defect discrimination, reduces the error of human experience judgment, and provides a reliable decision-making basis for the next step of completion.
[0026] Comparison with actual defect probability With probability threshold ; like , it means that the defective block is a scanning error and needs to be re-scanned, prompting you to re-scan to obtain the 3D oral point cloud; like , it means that the defective area is a real defect, and the defective area is completed to obtain a complete three-dimensional oral point cloud. The defective area is completed by existing technology, for example, for each block to be completed , enter the "Fill Holes" function in MeshLab to automatically detect the hole boundaries and generate triangles, and then call the "Laplacian Smooth" or "Taubin Smooth" filter with one click to obtain a closed and smooth completed mesh model.
[0027] The pair Perform denoising and unify the normal of each point in the point cloud, and output the processed point cloud ,include: For each point ,Calculate point cloud Other points in The Euclidean distance of The Represented in point cloud The points, whose coordinates are three-dimensional vectors ; Set the number of neighbor points ; Point Cloud Medium For other points outside the The Euclidean distance is sorted from small to large; In the sorting order, select Points form a set, marked as points The k-nearest neighbor set ; Get Point The k-nearest neighbor set For each point in , calculate the point The average neighborhood distance ; By each point Query to obtain k nearest neighboring points; then calculate the relationship between these neighboring points and point The Euclidean distance of each point is calculated and the arithmetic mean is calculated. This operation is highly adaptive and efficient compared to fixed radius or random sampling methods. First, the use of a fixed k-nearest neighbor ensures that each point has a stable number of neighbors, thereby avoiding instability caused by too many or too few neighbors in areas with uneven point cloud density. Second, averaging the distance from each point to its neighboring points helps smooth local density fluctuations and reduce the influence of outliers. The average neighborhood distance obtained in this way not only accurately reflects the local point cloud distribution state, but also provides a reliable basis for subsequent noise removal based on statistical thresholds. At the same time, the dynamic threshold judgment mechanism formed by the mean plus standard deviation can automatically adjust the point cloud noise level of different scanning devices and different parts, thereby improving the robustness and generalization ability of the algorithm.
[0028] Traversing the point cloud At each point in the , forming a neighborhood set; Calculate the mean in a set of neighborhoods and standard deviation ; Set the distance outlier removal threshold : in: , represents the empirical coefficient, which is used to relax the judgment criteria; If there is a point of , then determine the point It is a noise point and is removed; Traversing the point cloud For each point in the image, remove all noise points and output the processed point cloud The dynamic threshold is constructed based on the calculated neighborhood average distance and combined with the global mean and standard deviation. The statistical filtering mechanism can adapt to different point cloud density distributions and take into account both overall and local differences. In actual clinical scans, there are complex environments such as soft tissue and tongue occlusion inside the oral cavity, and the point cloud density is extremely uneven. Traditional fixed threshold or empirical value elimination often leads to over-elimination or residual noise. This method uses a threshold based on sample statistical characteristics to finely distinguish between noise points and edge points. Specifically, isolated floating points will be accurately eliminated because their distance from the neighborhood is much greater than the global distribution. Normal concave points in dense areas will not be mistakenly deleted due to the limitations of the mean and fluctuation range. This step does not require prior knowledge and can be completed self-consistently based on the current scan data.
[0029] The acquisition point The k-nearest neighbor set For each point in , calculate the point The average neighborhood distance , specifically: The calculation point The average neighborhood distance :
[0030] in: Represents the set of k nearest neighbors The Point coordinates; represents the Euclidean distance; The mean value in the calculated neighborhood set and standard deviation , specifically:
[0031]
[0032] in: Indicates the total number of elements in the neighborhood set.
[0033] The pair Perform denoising and unify the normal of each point in the point cloud, and output the processed point cloud , also includes: Point Construct the local covariance matrix for the center:
[0034] Where: the subscript T represents the matrix transpose; Represents the neighborhood centroid:
[0035] beg The characteristic decomposition of , the minimum eigenvalue The corresponding eigenvector is used as the normal The eigendecomposition is performed by the PCA algorithm calling a linear algebra library (such as Eigen, PCL or SVD / eigenSolver in Open3D) to perform standard eigenvalue decomposition and output three eigenvalues; Get the scanner's viewing angle vector when scanning and acquiring 3D oral point clouds ; Specifically, it is the reverse vector pointing from the optical center of the scanner to the center of the scanned object; calculate and compare with 0; when When the normal is turned, it is ensured to face the scanner so that the normal can be guided in the subsequent defect detection and completion. ; Otherwise, keep the normal unchanged.
[0036] Point cloud Filter out the defective areas and form a defective block set ,include: For point clouds Each point in , with radius For the range, get all the points in the range to form a set ; For each point Perform point density field calculations:
[0037] in: Indicates a point The density of , which represents the number of points per unit area; Indicates radius is the set of points in the neighborhood; Indicates the number of points in the set; represents the neighborhood area; For each point , perform curvature estimation:
[0038] in: Indicates the surface curvature, the smaller it is, the smoother it is; Indicated by point Construct the eigenvalues of the local covariance matrix for the center.
[0039] Point cloud Filter out the defective areas and form a defective block set , also includes: Get point cloud Each point in Density and surface curvature ; Point Cloud Point in The total number of ; Mean of the global statistical density and standard deviation :
[0040]
[0041] Similarly, the mean of the global statistical surface curvature is and standard deviation ; To set the filter threshold:
[0042]
[0043] The curvature threshold is set to 0.02 here, which is based on the statistical distribution obtained from a large number of dental models and oral soft and hard tissue scans. The minimum curvature of the tooth surface, especially the defect edge, is generally greater than 0.02; For each point ,like and , it is determined to be a possible defect point; Get all points that are judged as possible defect points to form a possible defect point set ; Use Euclidean distance clustering to extract defective blocks: Set the neighborhood radius , can be fine-tuned according to the scanning accuracy; the minimum number of neighbors ,to avoid extremely small isolated clusters being mistaken for defective blocks; Initialize Collection All points in the state are "not visited"; S2, the machine selects an "unvisited" point , marked as "visited", query its -Neighborhood:
[0044] in: Representing a collection Other points in Indicates the Euclidean distance, and the subscript 2 indicates that the L2 norm is used; S3, if , then mark If it is noise, it will be discarded; otherwise, a new cluster is started. ,Will and all points in its neighborhood are added , and repeat steps S2 and S3 for each new point in the neighborhood until there are no new extensible points; Repeat S2 and S3 to traverse all "unvisited" points until all points are marked; Output all the A collection of ; Get all the missing blocks to form a missing block set .
[0045] The calculation defective block set The geometric feature vector of each block in ,include: For each piece : Calculating missing blocks The average density of all points in ; Calculating missing blocks The average surface curvature of all points in ; Calculate 3D features:
[0046] in: Indicates that After symmetry, the average distance between the original block and the original block is used to measure its symmetry. It is obtained by fitting the plane symmetry, specifically: Get Block Points contained in The number of ; Block , calculate its center of mass :
[0047] Constructing the covariance matrix :
[0048] right Perform eigendecomposition and obtain three eigenvalues and orthogonal eigenvectors: ; The smallest eigenvalue corresponding to As the plane normal vector ; perpendicular to the plane normal vector And through the point A plane is defined as the fitting plane ; For any block Point in , calculate its Mirror point :
[0049] in: Represents the directed distance from a point to a plane; For any block Point in , calculate the distance between it and the mirror point :
[0050] Take the average of all points to get the symmetric error :
[0051] in: Expressed as the distance between each point and its mirror image point.
[0052] The geometric feature vector of each block is , calculate the true defect probability ,include: Set the scoring function :
[0053] in: , , represents the empirical weighting coefficient, which is set to 0.4, 0.3, and 0.3 respectively; , used to normalize the symmetric error; represents the standard deviation of the global statistical density; Represents the standard deviation of the global statistical surface curvature; the empirical weighting coefficient can be adjusted as needed. If you want to emphasize the density difference more, you can increase If you pay more attention to the curvature, you can increase If symmetry is more important, you can increase ; Calculating the true defect probability :
[0054] in: and The mean and standard deviation of the scores pre-calculated in multiple high-quality samples are obtained by collecting a number of 10-20 complete scans of 3D oral point clouds without motion artifacts and calculating the score value corresponding to each "true defect block" , calculate the global mean and standard deviation of these scores, which is and .
[0055] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.
[0056] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.
Claims
1. A digital intelligent orthodontic diagnosis and treatment method, characterized in that: include: The three-dimensional oral point cloud is obtained by scanning and recorded as ; right Perform denoising and unify the normal of each point in the point cloud, and output the processed point cloud ; Point Cloud Filter out the defective areas and form a defective block set ; Calculate the set of missing blocks The geometric feature vector of each block in ; The geometric feature vector of each block , calculate the true defect probability ; Set the probability threshold for determining whether to rescan to obtain a 3D oral point cloud ; Comparison with actual defect probability With probability threshold ; like , it will prompt you to rescan to obtain the 3D oral point cloud; like , the defective area is filled and a complete three-dimensional oral point cloud is obtained.
2. A digital orthodontic intelligent diagnosis and treatment method according to claim 1, characterized in that: The pair Perform denoising and unify the normal of each point in the point cloud, and output the processed point cloud ,include: For each point ,Calculate point cloud Other points in The Euclidean distance of The Represented in point cloud The points, whose coordinates are three-dimensional vectors ; Set the number of neighbor points ; Point Cloud Medium For other points outside the The Euclidean distance is sorted from small to large; In the sorting order, select Points form a set, marked as points The k-nearest neighbor set ; Get Point The k-nearest neighbor set For each point in , calculate the point The average neighborhood distance ; Traversing the point cloud At each point in the , forming a neighborhood set; Calculate the mean in a set of neighborhoods and standard deviation ; Set the distance outlier removal threshold : in: , represents the empirical coefficient, which is used to relax the judgment criteria; If there is a point of , then determine the point It is a noise point and is removed; Traversing the point cloud For each point in the image, remove all noise points and output the processed point cloud .
3. A digital orthodontic intelligent diagnosis and treatment method according to claim 2, characterized in that: The acquisition point The k-nearest neighbor set For each point in , calculate the point The average neighborhood distance , specifically: The calculation point The average neighborhood distance : in: Represents the set of k nearest neighbors The Point coordinates; represents the Euclidean distance; The mean value in the calculated neighborhood set and standard deviation , specifically: in: Indicates the total number of elements in the neighborhood set.
4. A digital orthodontic intelligent diagnosis and treatment method according to claim 1, characterized in that: The pair Perform denoising and unify the normal of each point in the point cloud, and output the processed point cloud , also includes: Point Construct the local covariance matrix for the center: Where: the subscript T represents the matrix transpose; Represents the neighborhood centroid: beg The characteristic decomposition of , the minimum eigenvalue The corresponding eigenvector is used as the normal ; Get the scanner's viewing angle vector when scanning and acquiring 3D oral point clouds ; calculate and compare with 0; when When , flip the normal, that is ; Otherwise, keep the normal unchanged.
5. A digital orthodontic intelligent diagnosis and treatment method according to claim 4, characterized in that: Point cloud Filter out the defective areas and form a defective block set ,include: For point clouds Each point in , with radius For the range, get all the points in the range to form a set ; For each point Perform point density field calculations: in: Indicates a point The density of , which represents the number of points per unit area; Indicates radius is the set of points in the neighborhood; Indicates the number of points in the set; represents the neighborhood area; For each point , perform curvature estimation: in: Indicates the surface curvature, the smaller it is, the smoother it is; Indicated by point Construct the eigenvalues of the local covariance matrix for the center.
6. A digital orthodontic intelligent diagnosis and treatment method according to claim 5, characterized in that: Point cloud Filter out the defective areas and form a defective block set , also includes: Get point cloud Each point in Density and surface curvature ; Point Cloud Points in The total number of ; Mean of the global statistical density and standard deviation : Similarly, the mean of the global statistical surface curvature is and standard deviation ; To set the filter threshold: For each point ,like and , it is determined to be a possible defect point; Get all points that are judged as possible defect points to form a possible defect point set ; Use Euclidean distance clustering to extract defective blocks: Set the neighborhood radius , minimum number of neighbors ; Initialize Collection All points in the state are "not visited"; S2, the machine selects an "unvisited" point , marked as "visited", query its -Neighborhood: in: Representing a collection Other points in Indicates the Euclidean distance, and the subscript 2 indicates that the L2 norm is used; S3, if , then mark If it is noise, it will be discarded; otherwise, a new cluster is started. ,Will All points in its neighborhood are added , and repeat steps S2 and S3 for each new point in the neighborhood until there are no new extensible points; Repeat S2 and S3 to traverse all "unvisited" points until all points are marked; Output all the A collection of ; Get all the missing blocks to form a missing block set .
7. A digital orthodontic intelligent diagnosis and treatment method according to claim 1, characterized in that: The calculation defective block set The geometric feature vector of each block in ,include: For each piece : Calculating missing blocks The average density of all points in ; Calculating missing blocks The average surface curvature of all points in ; Calculate 3D features: in: Indicates that After symmetry, the average distance between the original block and the original block is obtained by fitting the plane symmetry, specifically: Get Block Points contained in The number of ; Block , calculate its center of mass : Constructing the covariance matrix : right Perform eigendecomposition and obtain three eigenvalues and orthogonal eigenvectors: ; The smallest eigenvalue corresponding to As the plane normal vector ; perpendicular to the plane normal vector And through the point A plane is defined as the fitting plane ; For any block Points in , calculate its Mirror point : in: Represents the directed distance from a point to a plane; For any block Points in , calculate the distance between it and the mirror point : Take the average of all points to get the symmetric error : in: Expressed as the distance between each point and its mirror image point.
8. A digital orthodontic intelligent diagnosis and treatment method according to claim 7, characterized in that: The geometric feature vector of each block is , calculate the true defect probability ,include: Set the scoring function : in: , , represents the empirical weighting coefficient, which is set to 0.4, 0.3, and 0.3 respectively; , used to normalize the symmetric error; represents the standard deviation of the global statistical density; represents the standard deviation of the global statistical surface curvature; Calculating the true defect probability : in: and Represents the mean and standard deviation of the scores precomputed in multiple high-quality samples.