A dental model fusion method and a dental model fusion device

Abstract

The present application relates to the technical field of digital modeling, in particular to a tooth model fusion method and a tooth model fusion device. The method comprises: acquiring crown oral scanning data and root image data; generating a tooth grid according to the crown oral scanning data and the root image data to obtain a tooth grid model; performing registration and suture processing on the tooth grid model to obtain an initial tooth model; performing smoothing repair processing on the initial tooth model according to preset smoothing adjustment parameters to obtain a target tooth model; the smoothing adjustment parameters comprise a forward shrinkage parameter and a reverse expansion parameter, the forward shrinkage parameter is used for eliminating local corners of the model, and the reverse expansion parameter is used for controlling the rebound of the model grid boundary. By configuring the forward shrinkage parameter and the reverse expansion parameter, a perfect filter is constructed in the frequency domain, so that the real details of the tooth are effectively protected from being damaged in the process of repairing and iterating the initial tooth model, and the accuracy of tooth model fusion is improved.

Description

Technical Field

[0001] This invention belongs to the field of digital modeling technology, specifically relating to a method and device for fusing tooth models. Background Technology

[0002] In recent years, with the development of digital dental technology, the 3D fusion of dental crown data (which can reproduce the occlusal surface fissures and external features) and root data (which reflects the intraosseous morphology) using CBCT (cone beam computed tomography) has become the absolute foundation for orthodontic, orthognathic, and implant treatments. Currently, the primary method is to use implicit surface fitting algorithms to globally reconstruct the mesh and fuse the crown and root data to obtain the dental model.

[0003] However, in practice, it has been found that adopting the global re-network method requires uniform mathematical smoothing and topological reconstruction of the entire tooth surface. This process will destroy the micron-level high-frequency anatomical details that are extremely important in oral scan data (such as the fine pits and fissures of the occlusal surface and the sharp edges of the cusps), causing the fused crown to become "fleshy" and blurry, losing the original real physiological morphological characteristics, and the accuracy of tooth model fusion is not high.

[0004] Therefore, improving the accuracy of dental model fusion is a pressing technical problem that needs to be solved. Summary of the Invention

[0005] The purpose of this invention is to provide a method and apparatus for fusing tooth models, which aims to improve the accuracy of tooth model fusing.

[0006] To address the aforementioned technical problems, the first aspect of this invention discloses a method for fusing tooth models, the method comprising: Acquire dental crown intraoral scan data and dental root image data; the dental crown intraoral scan data is dental crown data obtained by performing a three-dimensional oral cavity scan of the target tooth in advance, and the dental root image data is dental root data obtained by performing a computed tomography scan of the target tooth in advance. Based on the dental crown intraoral scan data and the dental root image data, a dental mesh is generated to obtain a dental mesh model; The tooth mesh model is registered and stitched to obtain the initial tooth model; The initial tooth model is smoothed and repaired according to preset smoothing adjustment parameters to obtain the target tooth model; The smoothing adjustment parameters include forward shrinkage parameters and reverse expansion parameters. The forward shrinkage parameters are used to eliminate local sharp edges of the model, and the reverse expansion parameters are used to control the springback of the model mesh boundaries. The registration stitching process is used to perform non-destructive local mesh construction and topology modification. The smoothing repair process is used to keep the original bottom-level geometric coordinates of the crown occlusal scan data and the root image data in the target tooth model unchanged.

[0007] As an optional implementation, in the first aspect of the present invention, the step of generating a tooth mesh based on the crown intraoral scan data and the root image data to obtain a tooth mesh model includes: The dental crown oral scan data is separated into individual tooth instances according to a preset instance segmentation network to obtain oral scan tooth instances; The tooth root image data is segmented into individual tooth instances according to a preset instance segmentation network to obtain image tooth instances. The oral scan tooth instance is subjected to topological cleaning, and the closed edge loop of the oral scan tooth instance is extracted based on a preset edge reference statistics method to obtain the target closed edge loop; the topological cleaning is used to remove non-manifold edges and degenerate triangles; The target closed edge ring is smoothed in one dimension to obtain a crown mesh model; the one-dimensional smoothing is used to remove edge burrs. The image tooth instance is subjected to manifold cleaning and noise reduction processing to obtain a tooth root mesh model; The tooth mesh model includes the crown mesh model and the root mesh model.

[0008] As an optional implementation, in the first aspect of the present invention, the registration and stitching process of the tooth mesh model to obtain an initial tooth model includes: Rigid body registration is performed on the tooth mesh model to obtain a tooth registration model; the tooth registration model includes a crown registration model and a root registration model; The tooth root registration model is subjected to crown removal processing to obtain a tooth root crown-removed model; The crown registration model and the root crown removal model are geometrically sutured to obtain the initial tooth model.

[0009] As an optional implementation, in the first aspect of the present invention, the step of performing rigid body registration on the tooth mesh model to obtain a tooth registration model includes: The tooth mesh model is downsampled according to the preset first voxel mesh, and the three-dimensional bounding box of the tooth mesh model is calculated. The target cost matrix is ​​obtained by constructing a cost matrix based on the three-dimensional bounding box and the preset volume inclusion rate threshold. Based on the target cost matrix, the tooth mesh model is subjected to mesh mapping filtering to obtain a tooth filtering model; The tooth filtering model is downsampled according to the preset second voxel mesh, and the robustness of the tooth filtering model is adjusted to obtain the tooth adjustment model. The tooth adjustment model is registered using normal locking at full resolution according to a preset convergence index to obtain a tooth registration model.

[0010] As an optional implementation, in the first aspect of the present invention, the crown removal process performed on the tooth root registration model to obtain a crown-removed tooth root model includes: The tooth root registration model is subjected to registration anomaly filtering based on the preset maximum allowable nearest neighbor distance to obtain a tooth root filtering model; The root filtering model is transformed into an undirected weighted graph model to obtain a root weighted graph model. Based on the preset maximum search diffusion radius, the root weighted graph model is subjected to shortest path traversal for crown removal to obtain the root crown removal model.

[0011] As an optional implementation, in the first aspect of the present invention, the geometric suturing of the crown registration model and the root crown removal model to obtain an initial tooth model includes: The suture mapping start point and suture mapping direction between the crown registration model and the root crown removal model are determined according to the preset global absolute alignment operator; A target bridging mesh is generated based on a predetermined radial expansion compensation amount, the stitching mapping start point, and the stitching mapping direction; The crown registration model and the root crown removal model are stitched together according to the target bridging mesh to obtain the initial tooth model.

[0012] As an optional implementation, in the first aspect of the present invention, the step of smoothing and repairing the initial tooth model according to preset smoothing adjustment parameters to obtain the target tooth model includes: Within a preset polishing iteration cycle, the initial tooth model is smoothed by the positive shrinkage parameter and the reverse expansion parameter to obtain a smooth tooth model. Determine the extrapolated normal vector of the oral scan based on the oral scan data of the crown. Obtain the outward normal vector of the suture surface of the initial tooth model; If the vector product of the external normal vector of the oral scan and the outgoing normal vector is less than zero, then the normal orientation of the smooth tooth model is repaired to obtain the target tooth model.

[0013] A second aspect of the present invention discloses a dental model fusion device, the device comprising: The data acquisition module is used to acquire crown intraoral scan data and root image data; the crown intraoral scan data is obtained by performing a three-dimensional oral cavity scan on the target tooth in advance, and the root image data is obtained by performing a computed tomography scan on the target tooth in advance. The mesh processing module is used to generate a tooth mesh based on the crown intraoral scan data and the root image data, thereby obtaining a tooth mesh model. The registration and stitching module is used to perform registration and stitching on the tooth mesh model to obtain an initial tooth model. The smoothing repair module is used to perform smoothing repair processing on the initial tooth model according to preset smoothing adjustment parameters to obtain the target tooth model; The smoothing adjustment parameters include forward shrinkage parameters and reverse expansion parameters. The forward shrinkage parameters are used to eliminate local sharp edges of the model, and the reverse expansion parameters are used to control the springback of the model mesh boundaries. The registration stitching process is used to perform non-destructive local mesh construction and topology modification. The smoothing repair process is used to keep the original bottom-level geometric coordinates of the crown occlusal scan data and the root image data in the target tooth model unchanged.

[0014] As an optional implementation, in a second aspect of the present invention, the mesh processing module generates a tooth mesh based on the crown intraoral scan data and the root image data to obtain a tooth mesh model in the following specific ways: The dental crown oral scan data is separated into individual tooth instances according to a preset instance segmentation network to obtain oral scan tooth instances; The tooth root image data is segmented into individual tooth instances according to a preset instance segmentation network to obtain image tooth instances. The oral scan tooth instance is subjected to topological cleaning, and the closed edge loop of the oral scan tooth instance is extracted based on a preset edge reference statistics method to obtain the target closed edge loop; the topological cleaning is used to remove non-manifold edges and degenerate triangles; The target closed edge ring is smoothed in one dimension to obtain a crown mesh model; the one-dimensional smoothing is used to remove edge burrs. The image tooth instance is subjected to manifold cleaning and noise reduction processing to obtain a tooth root mesh model; The tooth mesh model includes the crown mesh model and the root mesh model.

[0015] As an optional implementation, in a second aspect of the present invention, the registration and stitching module performs registration and stitching processing on the tooth mesh model to obtain an initial tooth model in the following specific ways: Rigid body registration is performed on the tooth mesh model to obtain a tooth registration model; the tooth registration model includes a crown registration model and a root registration model; The tooth root registration model is subjected to crown removal processing to obtain a tooth root crown-removed model; The crown registration model and the root crown removal model are geometrically sutured to obtain the initial tooth model.

[0016] As an optional implementation, in a second aspect of the present invention, the registration stitching module performs rigid body registration on the tooth mesh model to obtain a tooth registration model in the following specific ways: The tooth mesh model is downsampled according to the preset first voxel mesh, and the three-dimensional bounding box of the tooth mesh model is calculated. The target cost matrix is ​​obtained by constructing a cost matrix based on the three-dimensional bounding box and the preset volume inclusion rate threshold. Based on the target cost matrix, the tooth mesh model is subjected to mesh mapping filtering to obtain a tooth filtering model; The tooth filtering model is downsampled according to the preset second voxel mesh, and the robustness of the tooth filtering model is adjusted to obtain the tooth adjustment model. The tooth adjustment model is registered using normal locking at full resolution according to a preset convergence index to obtain a tooth registration model.

[0017] As an optional implementation, in the second aspect of the present invention, the registration and suturing module performs crown removal processing on the tooth root registration model to obtain a crown-removed tooth root model, specifically in the following ways: The tooth root registration model is subjected to registration anomaly filtering based on the preset maximum allowable nearest neighbor distance to obtain a tooth root filtering model; The root filtering model is transformed into an undirected weighted graph model to obtain a root weighted graph model. Based on the preset maximum search diffusion radius, the root weighted graph model is subjected to shortest path traversal for crown removal to obtain the root crown removal model.

[0018] As an optional implementation, in a second aspect of the present invention, the registration and suturing module performs geometric suturing on the crown registration model and the root crown removal model to obtain an initial tooth model, specifically in the following ways: The suture mapping start point and suture mapping direction between the crown registration model and the root crown removal model are determined according to the preset global absolute alignment operator; A target bridging mesh is generated based on a predetermined radial expansion compensation amount, the stitching mapping start point, and the stitching mapping direction; The crown registration model and the root crown removal model are stitched together according to the target bridging mesh to obtain the initial tooth model.

[0019] As an optional implementation, in a second aspect of the present invention, the smoothing repair module performs smoothing repair processing on the initial tooth model according to preset smoothing adjustment parameters to obtain the target tooth model in the following specific ways: Within a preset polishing iteration cycle, the initial tooth model is smoothed by the positive shrinkage parameter and the reverse expansion parameter to obtain a smooth tooth model. Determine the extrapolated normal vector of the oral scan based on the oral scan data of the crown. Obtain the outward normal vector of the suture surface of the initial tooth model; If the vector product of the external normal vector of the oral scan and the outgoing normal vector is less than zero, then the normal orientation of the smooth tooth model is repaired to obtain the target tooth model.

[0020] A third aspect of the present invention discloses another dental model fusion device, the device comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute a tooth model fusion method disclosed in the first aspect of the present invention.

[0021] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked by a processor, are used to execute a tooth model fusion method disclosed in the first aspect of the present invention.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: After acquiring the oral scan data of the crown and the image data of the root, a tooth mesh is generated based on the data to obtain a tooth mesh model. The tooth mesh model is then registered and stitched to obtain an initial tooth model. The initial tooth model is then smoothed and repaired according to preset forward contraction and reverse expansion parameters to obtain the target tooth model. By configuring the forward contraction and reverse expansion parameters, a perfect filter is constructed in the frequency domain, effectively protecting the real details of the teeth from damage during the iterative repair process of the initial tooth model, thereby improving the accuracy of tooth model fusion. Furthermore, a non-destructive local meshing strategy is employed to lock and inherit the underlying geometric coordinates and original topology of the original oral scan crown and the effective CT tooth root, thus more comprehensively preserving the high-frequency anatomical features of the oral scan end and the real intraosseous morphology of the CT end. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a schematic flowchart of a tooth model fusion method disclosed in an embodiment of the present invention; Figure 2 This is a schematic diagram of a dental crown mesh model according to an embodiment of the present invention; Figure 3 This is a schematic diagram of a tooth root mesh model according to an embodiment of the present invention; Figure 4 This is a schematic diagram of a model of a tooth crown and tooth root in a state to be sutured, according to an embodiment of the present invention; Figure 5 This is a schematic diagram of a target tooth model according to an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a tooth model fusion device disclosed in an embodiment of the present invention; Figure 7 This is a schematic diagram of another tooth model fusion device disclosed in an embodiment of the present invention. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, or product may include a series of steps or units, or may not be limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or processes.

[0027] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0028] In recent years, with the development of digital dental technology, the 3D fusion of dental crown data (which can reproduce the occlusal surface fissures and external features) and root data (which reflects the intraosseous morphology) using CBCT (cone beam computed tomography) has become the absolute foundation for orthodontic, orthognathic, and implant treatments. Currently, the primary method is to use implicit surface fitting algorithms to globally reconstruct the mesh and fuse the crown and root data to obtain the dental model.

[0029] However, in practice, it has been found that adopting the global re-network method requires uniform mathematical smoothing and topological reconstruction of the entire tooth surface. This process will destroy the micron-level high-frequency anatomical details that are extremely important in oral scan data (such as the fine pits and fissures of the occlusal surface and the sharp edges of the cusps), causing the fused crown to become "fleshy" and blurry, losing the original real physiological morphological characteristics, and the accuracy of tooth model fusion is not high.

[0030] Therefore, improving the accuracy of dental model fusion is a pressing technical problem that needs to be solved.

[0031] To address the aforementioned technical problems, this invention discloses a method and apparatus for fusing tooth models, aiming to improve the accuracy of tooth model fusing. These will be described in detail below.

[0032] Example 1 Please see Figure 1 , Figure 1 This is a schematic flowchart of a tooth model fusion method disclosed in an embodiment of the present invention. Figure 1 The method shown can be applied to dental model fusion devices, which can improve the accuracy of dental model fusion. For example... Figure 1 As shown, the tooth model fusion method disclosed in this embodiment of the invention includes, but is not limited to, the following operations: 101. Acquire crown intraoral scan data and root image data; crown intraoral scan data is obtained by performing a three-dimensional oral cavity scan on the target tooth in advance, and root image data is obtained by performing a computed tomography scan on the target tooth in advance. 102. Based on the oral scan data of the crown and the image data of the root, a tooth mesh is generated to obtain a tooth mesh model; 103. Perform registration and stitching on the tooth mesh model to obtain the initial tooth model; 104. Perform smoothing and repair processing on the initial tooth model according to the preset smoothing adjustment parameters to obtain the target tooth model; Among them, the smoothing adjustment parameters include forward shrinkage parameters and reverse expansion parameters. Forward shrinkage parameters are used to eliminate local sharp edges of the model, and reverse expansion parameters are used to control the springback of the model mesh boundary. Registration stitching is used for non-destructive local mesh construction and topology modification, and smoothing repair is used to keep the original bottom-level geometric coordinates of the crown occlusal scan data and root image data in the target tooth model unchanged.

[0033] In this embodiment of the invention, after acquiring the oral scan data of the crown and the image data of the root, a tooth mesh is generated based on the oral scan data of the crown and the image data of the root to obtain a tooth mesh model. The tooth mesh model is then registered and stitched to obtain an initial tooth model. The initial tooth model is then smoothed and repaired according to preset forward contraction parameters and reverse expansion parameters to obtain a target tooth model. By configuring the forward contraction parameters and reverse expansion parameters, a perfect filter is constructed in the frequency domain, thereby effectively protecting the real details of the teeth from being destroyed during the repair iteration process of the initial tooth model, thus improving the accuracy of tooth model fusion. Furthermore, a non-destructive local meshing strategy is adopted to lock and inherit the underlying geometric coordinates and original topology of the original oral scan crown and the effective CT tooth root, thereby more comprehensively preserving the high-frequency anatomical features of the oral scan end and the real intraosseous morphology of the CT end.

[0034] In an optional embodiment, a tooth mesh model is generated based on the crown intraoral scan data and the root image data, including: Based on the preset instance segmentation network, the dental crown oral scan data is separated into individual tooth instances to obtain oral scan tooth instances; Based on a pre-defined instance segmentation network, the root image data is segmented into individual tooth instances to obtain image tooth instances; Topological cleaning is performed on the dental instances scanned from the oral cavity, and closed edge loops of the dental instances are extracted based on a preset edge reference statistics method to obtain the target closed edge loops; the topological cleaning is used to remove non-manifold edges and degenerate triangles. One-dimensional smoothing is applied to the target closed edge ring to obtain the crown mesh model; one-dimensional smoothing is used to remove edge burrs. Manifold cleaning and noise reduction are performed on the image tooth instances to obtain the tooth root mesh model; The tooth mesh model includes the crown mesh model and the root mesh model.

[0035] In this optional embodiment, a pre-trained 3D deep learning instance segmentation network is used to separate and extract individual teeth from root CBCT data and crown intraoral scan data, respectively. A manifold cleaning mechanism is also introduced to remove non-manifold edges and degenerate triangles that may cause subsequent algorithm crashes. During the mesh preprocessing stage, to extract pure intraoral scan topological boundaries, an algorithm based on edge reference count is used to extract one-dimensional closed polygons. A dual Laplacian smoothing process, combining forward contraction for anti-aliasing and reverse expansion for volume preservation, is then used to clean one-dimensional edge burrs, outputting healthy mesh instances with independent tooth position identifiers (IDs).

[0036] Please see Figure 2 , Figure 2 This is a schematic diagram of a dental crown mesh model according to an embodiment of the present invention. Figure 2 (a) is a side view. Figure 2 (b) is a top view.

[0037] Please see Figure 3 , Figure 3 This is a schematic diagram of a tooth root mesh model according to an embodiment of the present invention. Figure 3 (a) is a side view. Figure 3 (b) is a top view.

[0038] In another optional embodiment, the tooth mesh model is registered and stitched to obtain an initial tooth model, including: Rigid body registration is performed on the tooth mesh model to obtain the tooth registration model; the tooth registration model includes the crown registration model and the root registration model; The crown removal process is performed on the registered tooth root model to obtain the crown-removed tooth root model; Geometric suturing was performed on the crown registration model and the root crown removal model to obtain the initial tooth model.

[0039] In another optional embodiment, rigid body registration is performed on the tooth mesh model to obtain a tooth registration model, including: The tooth mesh model is downsampled based on the preset first voxel mesh, and the three-dimensional bounding box of the tooth mesh model is calculated. The target cost matrix is ​​obtained by constructing a cost matrix based on the 3D bounding box and the preset volume inclusion rate threshold. Based on the target cost matrix, the tooth mesh model is subjected to mesh mapping and filtering to obtain the tooth filtered model; The tooth filtering model is downsampled according to the preset second voxel mesh, and the robustness of the tooth filtering model is adjusted to obtain the tooth adjustment model. At full resolution, the tooth adjustment model is registered using normal locking based on a preset convergence index to obtain the tooth registration model.

[0040] Because CBCT root data suffers from severe metal artifacts and scattered noise, and has significant non-rigid distortions in morphology compared to the real oral surface, traditional global ICP registration techniques are prone to falling into local optima traps when calculating global point cloud distances, leading to severe translational or rotational deviations or even reversals.

[0041] In this embodiment of the invention, a 0.5mm voxel mesh is used to downsample the high-density point cloud to accelerate computation, thereby calculating the three-dimensional spatial axis-aligned bounding box (AABB). A stringent interception mechanism is introduced: in addition to ensuring that the length, width, and height of the intersection bounding box are all greater than zero, a unidirectional proportion threshold of 40% of the CT scan volume is explicitly set. If it is lower than this proportion, an infinite cost (10^7) is assigned to the pairing combination. After constructing a cost matrix containing the actual surface geometric error, the Hungarian allocation algorithm is called to achieve the globally optimal match, and a maximum allowable error threshold of 5.0mm is set for safe interception filtering.

[0042] In the subsequent rigid alignment, an affine matrix safety check is introduced (by calculating the magnitude of the translation vector and the trace of the rotation matrix to limit anomalous drift). Registration employs 0.2mm voxel downsampling to extract high-frequency details, and executes a multi-scale convergence threshold sequence (1.0mm, 0.5mm, 0.25mm, and 0.15mm sequentially), while simultaneously dynamically lowering the robust Tukey Loss control threshold parameter. Finally, at full resolution, extremely high-precision normal-locked registration is achieved with an extremely small threshold of 0.1mm and stringent convergence metrics (error variation between two consecutive iterations less than 10^-6, maximum 50 iterations).

[0043] In this optional embodiment, downsampling and 3D AABB collision extremum calculation are employed, and failure mappings are filtered using a minimum volume capacity of 40%, significantly saving matching computational power and eliminating mismatches. A threshold sequence reduced from 1.0mm to 0.1mm is combined with robust attenuation weights to immunize against metal artifacts and provide... Extremely high level of geometric locking precision.

[0044] In another alternative embodiment, for applications that allow for some manual intervention or do not require high local precision, the bounding box matrix can be automatically assigned instead of being used in the initial registration stage. Instead, the physician can manually select multiple sets of homologous anatomical feature points (such as specific cusps and fissures) on the intraoral scan and CBCT to provide the initial rigid body transformation matrix, and then perform ICP iteration.

[0045] In another optional embodiment, the root registration model is subjected to crown removal processing to obtain a root crown-removed model, including: The tooth root registration model is filtered for registration anomalies based on the preset maximum allowable nearest neighbor distance to obtain the tooth root filtering model; The root filtering model is transformed into an undirected weighted graph model to obtain a root weighted graph model. Based on the preset maximum search diffusion radius, the root weighted graph model is subjected to shortest path traversal for crown removal to obtain the root crown removal model.

[0046] Traditional techniques heavily rely on manual extraction of the cervical margin line by dentists to deconstruct the tooth structure, or on simple cross-sectional geometric plane cutting to remove the crown of the CBCT. The cutting transition is stiff and unnatural, and manual drawing is extremely inefficient. Conventional geometric plane cutting produces a flat concave effect like "cutting an apple" in cross-section, resulting in obvious step-like faults in the cervical region of the fused model. It cannot truly reflect the natural convex and full shape of the junction between dentin and enamel.

[0047] In this embodiment of the invention, instead of traditional geometric plane cutting, when mapping the intraoral scan edge to the CT surface as an anchor point, the maximum allowable distance between the nearest neighbors in the KD-Tree is set to 2.0 mm to filter out registration anomalies or damaged surfaces. The CT surface is then converted into an undirected weighted graph model containing edge connectivity weights, and the Dijkstra shortest path algorithm is used for pathfinding. To prevent the cutting line from "running across the entire image" due to CT perforations, a dynamic maximum search diffusion radius is introduced, with this limit set to 5.0 times the straight-line distance between two points (and the lower limit not less than 5.0 mm). After coronal resection, the ClusterConnected Triangles connected domain clustering algorithm is used to accurately extract and retain the largest single-volume CT root fragment.

[0048] In this alternative embodiment, by introducing a maximum diffusion radius limited to 5 times the actual physical straight-line distance, the undulations of the real surface mesh are tightly engaged, thereby completely eliminating the industrial problem of spatial discontinuity jumps in the cutting path.

[0049] It should be noted that, in addition to using graph theory-based shortest path search, the "Region Growing Algorithm Based on Curvature Tensor and Normal Divergence" or the "Graph Cut Algorithm Based on Markov Random Field" can also be used to achieve automated topological stripping of CT pathological coronal regions.

[0050] Please see Figure 4 , Figure 4 This is a schematic diagram of a tooth crown and tooth root in a state to be sutured, according to an embodiment of the present invention. Figure 4 (a) is a side view. Figure 4 (b) is a top view. Figure 2 and Figure 3 Based on this, after high-precision registration and adaptive graph theory crown removal, the spatial alignment between the dental crown and the CT root, and the physical gap to be sutured, are as follows: Figure 4 As shown.

[0051] In another alternative embodiment, for applications that allow for some manual intervention or do not require high local precision, instead of using graph theory (Dijkstra) pathfinding mapping during boundary extraction and crown removal, a three-dimensional interactive interface is provided, allowing operators to manually select and draw the cervical line on the CBCT mesh using a pen or punctuation tool, and then perform surface removal based on this line.

[0052] In another optional embodiment, the crown registration model and the root crown removal model are geometrically sutured to obtain an initial tooth model, including: The starting point and direction of the suture mapping between the crown registration model and the root crown removal model are determined based on the preset global absolute alignment operator. The target bridging mesh is generated based on the predetermined radial expansion compensation amount, stitching mapping start point, and stitching mapping direction; The crown registration model and the root crown removal model are stitched together based on the target bridging mesh to obtain the initial tooth model.

[0053] Traditional techniques for joining tooth crowns and roots typically employ simple spatial linear interpolation to transition and connect boundary vertices. However, at complex tooth edges, simple linear interpolation and blind stitching of uneven starting points can easily lead to serious topological errors such as twisted and self-intersecting meshes on the bridging surface, and normal flipping (which appears as black holes in the rendering).

[0054] In this embodiment of the invention, the arc length of the same number of vertices is uniformly resampled based on the upper and lower boundary rings extracted from the model. To avoid the "twisted" effect caused by direct stitching, a global exhaustive absolute alignment operator is introduced (a step size of 2 can be set when traversing offsets to accelerate Euclidean distance calculation) to find the topological mapping starting point and direction with the minimum overall connection cost. During bridging mesh generation, an expansion stitching mechanism based on real cross-sections is pioneered. First, the joint spatial centroid of the two endpoint sets is calculated as the growth base point. An S-shaped easing function is set. Calculate the weighted interpolation. For each grid generation layer, calculate the target desired radius. Compare this with the actual bias shrinkage radius of the direct linear interpolation point. Subtracting the two values ​​yields the radial expansion compensation amount. Finally, this compensation amount is accumulated along the outward normal direction of the center of gravity, perfectly eliminating the "apple-cutting" concavity and restoring the true natural physiological fullness.

[0055] In this optional embodiment, utilizing The solution using bidirectional operators with varying complexity overcomes the limitations of single-feature alignment; it fits the true physiological expectation radius using sigmoid weights and applies them. The compensation scalar restores the fullness of the cervical protrusion.

[0056] It should be noted that, in generating the transition zone, in addition to using the radial expansion supplementation algorithm, implicit surface function fitting algorithms such as Poisson Surface Reconstruction can also be applied to fill the gaps between the upper and lower mesh segments.

[0057] In another alternative embodiment, for applications where some manual intervention is permissible or where local precision requirements are not high, the upper and lower boundaries can be directly connected using existing conventional Hole Filling algorithms (such as linear interpolation stitching) during the transition zone stitching stage, and basic Laplacian smoothing can be applied. Although this approach cannot avoid local shrinkage and mesh kinking, it can serve as a simple alternative for low-precision requirements.

[0058] In another optional embodiment, the initial tooth model is smoothed and repaired according to preset smoothing adjustment parameters to obtain the target tooth model, including: Within a preset polishing iteration cycle, the initial tooth model is smoothed by using forward shrinkage parameters and reverse expansion parameters to obtain a smooth tooth model. Determine the extrapolated normal vector of the oral scan based on the oral scan data of the crown; Obtain the outward normal vector of the suture surface of the initial tooth model; If the vector product of the external normal vector and the outward normal vector is less than zero, then the normal orientation of the smooth tooth model is restored to obtain the target tooth model.

[0059] Traditional techniques for handling geometric abrupt changes and step noise in spliced ​​meshes typically employ conventional Laplacian smoothing operators. However, inherent mathematical flaws in conventional Laplacian smoothing cause the solid mesh to shift towards a local mean (i.e., inward). After multiple iterations, this can lead to catastrophic shrinkage, altering the true physiological dimensions of teeth. In orthodontics, this can even introduce errors of a few tenths of a millimeter, which is unacceptable and fails to meet the extremely high precision requirements of subsequent orthodontic biomechanical simulations or implant guide design.

[0060] In this embodiment of the invention, the Laplacian smoothing operator, which leads to irreversible shrinkage, is no longer used for the abrupt changes in the machining polylines of the bridging mesh. A two-parameter Taubin relaxation controller with four layers of diffusion into the neighborhood is deployed. Positive shrinkage parameters are alternately applied within a set optimal polishing iteration cycle of 40 times. To eliminate local sharp edges, a reverse expansion parameter is then applied. Forced mesh boundary bounce. Before final output, the centroid ray method, which easily causes incorrect flipping of faces in groove regions, is no longer used. Instead, the outer normal of the original mesh edge starting point is used as the truth criterion, and a vector dot product is performed between it and the outgoing normal of the newly generated bridging surface. If the dot product is less than zero, the transition area faces are instructed to uniformly swap the underlying vertex indices, correcting the normal orientation in one go and eliminating downstream rendering black holes.

[0061] In this optional embodiment, specific combinations of mutual exclusion factors are configured. A perfect low-pass filter was constructed in the frequency domain, which maintained its low-frequency intrinsic gain without shrinking after dozens of deep iterations; and all reflective holes and leaks were repaired by resetting the winding order through the dot product verification of the native boundary extension normal.

[0062] It should be noted that, in addition to the Taubin two-parameter smoothing operator in this embodiment of the invention, the HC-Laplacian Smoothing algorithm can also be used in engineering. This algorithm calculates the coordinate difference between the initial mesh and the mesh after conventional Laplacian smoothing, and adds the lost volume back to the vertex positions in an inverse proportion, which can be used as another alternative algorithm for anti-shrinkage smoothing.

[0063] Please see Figure 5 , Figure 5 This is a schematic diagram of a target tooth model according to an embodiment of the present invention. Figure 5 (a) is a side view. Figure 5 (b) is a top view. Figure 4 Based on this, a high-fidelity fusion model is generated by radial expansion suturing and dual-parameter volume-preserving smoothing. The seam exhibits natural physiological fullness with no volume shrinkage.

[0064] The tooth model fusion method in this invention employs a non-destructive local meshing strategy, limiting the generation and topology modification of mesh vertices to the physical tomographic gaps after crown removal. By directly locking and inheriting the underlying geometric coordinates and native topology of the original oral scan crown and effective CT root, it completely abandons conventional global implicit surface reconstruction algorithms. The final output watertight solid model retains 100% lossless micron-level high-frequency anatomical features at the oral scan end and the true intraosseous morphology at the CT end.

[0065] Example 2 Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a tooth model fusion device disclosed in an embodiment of the present invention. Figure 6 The apparatus shown can be used to perform the method described in Embodiment 1, and this apparatus can improve the accuracy of dental model fusion. Figure 6 As shown, the tooth model fusion device disclosed in this invention includes, but is not limited to: The data acquisition module 201 is used to acquire crown intraoral scan data and root image data. The crown intraoral scan data is obtained by performing a three-dimensional oral cavity scan on the target tooth in advance, and the root image data is obtained by performing a computed tomography scan on the target tooth in advance. Mesh processing module 202 is used to generate tooth meshes based on crown intraoral scan data and root image data to obtain a tooth mesh model; The registration and stitching module 203 is used to perform registration and stitching on the tooth mesh model to obtain the initial tooth model. The smoothing repair module 204 is used to perform smoothing repair processing on the initial tooth model according to the preset smoothing adjustment parameters to obtain the target tooth model. Among them, the smoothing adjustment parameters include forward shrinkage parameters and reverse expansion parameters. Forward shrinkage parameters are used to eliminate local sharp edges of the model, and reverse expansion parameters are used to control the springback of the model mesh boundary. Registration stitching is used for non-destructive local mesh construction and topology modification, and smoothing repair is used to keep the original bottom-level geometric coordinates of the crown occlusal scan data and root image data in the target tooth model unchanged.

[0066] In this embodiment of the invention, after acquiring the oral scan data of the crown and the image data of the root, a tooth mesh is generated based on the oral scan data of the crown and the image data of the root to obtain a tooth mesh model. The tooth mesh model is then registered and stitched to obtain an initial tooth model. The initial tooth model is then smoothed and repaired according to preset forward contraction parameters and reverse expansion parameters to obtain a target tooth model. By configuring the forward contraction parameters and reverse expansion parameters, a perfect filter is constructed in the frequency domain, thereby effectively protecting the real details of the teeth from being destroyed during the repair iteration process of the initial tooth model, thus improving the accuracy of tooth model fusion. Furthermore, a non-destructive local meshing strategy is adopted to lock and inherit the underlying geometric coordinates and original topology of the original oral scan crown and the effective CT tooth root, thereby more comprehensively preserving the high-frequency anatomical features of the oral scan end and the real intraosseous morphology of the CT end.

[0067] In an optional embodiment, the mesh processing module 202 generates a tooth mesh based on the crown intraoral scan data and the root image data, and the specific method for obtaining the tooth mesh model includes: Based on the preset instance segmentation network, the dental crown oral scan data is separated into individual tooth instances to obtain oral scan tooth instances; Based on a pre-defined instance segmentation network, the root image data is segmented into individual tooth instances to obtain image tooth instances; Topological cleaning is performed on the dental instances scanned from the oral cavity, and closed edge loops of the dental instances are extracted based on a preset edge reference statistics method to obtain the target closed edge loops; the topological cleaning is used to remove non-manifold edges and degenerate triangles. One-dimensional smoothing is applied to the target closed edge ring to obtain the crown mesh model; one-dimensional smoothing is used to remove edge burrs. Manifold cleaning and noise reduction are performed on the image tooth instances to obtain the tooth root mesh model; The tooth mesh model includes the crown mesh model and the root mesh model.

[0068] In this optional embodiment, a pre-trained 3D deep learning instance segmentation network is used to separate and extract individual teeth from root CBCT data and crown intraoral scan data, respectively. A manifold cleaning mechanism is also introduced to remove non-manifold edges and degenerate triangles that may cause subsequent algorithm crashes. During the mesh preprocessing stage, to extract pure intraoral scan topological boundaries, an algorithm based on edge reference count is used to extract one-dimensional closed polygons. A dual Laplacian smoothing process, combining forward contraction for anti-aliasing and reverse expansion for volume preservation, is then used to clean one-dimensional edge burrs, outputting healthy mesh instances with independent tooth position identifiers (IDs).

[0069] Please see Figure 2 , Figure 2 This is a schematic diagram of a dental crown mesh model according to an embodiment of the present invention. Figure 2 (a) is a side view. Figure 2 (b) is a top view.

[0070] Please see Figure 3 , Figure 3 This is a schematic diagram of a tooth root mesh model according to an embodiment of the present invention. Figure 3 (a) is a side view. Figure 3 (b) is a top view.

[0071] In another optional embodiment, the registration and stitching module 203 performs registration and stitching processing on the tooth mesh model to obtain the initial tooth model in the following specific ways: Rigid body registration is performed on the tooth mesh model to obtain the tooth registration model; the tooth registration model includes the crown registration model and the root registration model; The crown removal process is performed on the registered tooth root model to obtain the crown-removed tooth root model; Geometric suturing was performed on the crown registration model and the root crown removal model to obtain the initial tooth model.

[0072] In yet another optional embodiment, the registration stitching module 203 performs rigid body registration on the tooth mesh model to obtain the tooth registration model in the following specific ways: The tooth mesh model is downsampled based on the preset first voxel mesh, and the three-dimensional bounding box of the tooth mesh model is calculated. The target cost matrix is ​​obtained by constructing a cost matrix based on the 3D bounding box and the preset volume inclusion rate threshold. Based on the target cost matrix, the tooth mesh model is subjected to mesh mapping and filtering to obtain the tooth filtered model; The tooth filtering model is downsampled according to the preset second voxel mesh, and the robustness of the tooth filtering model is adjusted to obtain the tooth adjustment model. At full resolution, the tooth adjustment model is registered using normal locking based on a preset convergence index to obtain the tooth registration model.

[0073] Because CBCT root data suffers from severe metal artifacts and scattered noise, and has significant non-rigid distortions in morphology compared to the real oral surface, traditional global ICP registration techniques are prone to falling into local optima traps when calculating global point cloud distances, leading to severe translational or rotational deviations or even reversals.

[0074] In this embodiment of the invention, a 0.5mm voxel mesh is used to downsample the high-density point cloud to accelerate computation, thereby calculating the three-dimensional spatial axis-aligned bounding box (AABB). A stringent interception mechanism is introduced: in addition to ensuring that the length, width, and height of the intersection bounding box are all greater than zero, a unidirectional proportion threshold of 40% of the CT scan volume is explicitly set. If it is lower than this proportion, an infinite cost (10^7) is assigned to the pairing combination. After constructing a cost matrix containing the actual surface geometric error, the Hungarian allocation algorithm is called to achieve the globally optimal match, and a maximum allowable error threshold of 5.0mm is set for safe interception filtering.

[0075] In the subsequent rigid alignment, an affine matrix safety check is introduced (by calculating the magnitude of the translation vector and the trace of the rotation matrix to limit anomalous drift). Registration employs 0.2mm voxel downsampling to extract high-frequency details, and executes a multi-scale convergence threshold sequence (1.0mm, 0.5mm, 0.25mm, and 0.15mm sequentially), while simultaneously dynamically lowering the robust Tukey Loss control threshold parameter. Finally, at full resolution, extremely high-precision normal-locked registration is achieved with an extremely small threshold of 0.1mm and stringent convergence metrics (error variation between two consecutive iterations less than 10^-6, maximum 50 iterations).

[0076] In this optional embodiment, downsampling and 3D AABB collision extremum calculation are employed, and failure mappings are filtered using a minimum volume capacity of 40%, significantly saving matching computational power and eliminating mismatches. A threshold sequence reduced from 1.0mm to 0.1mm is combined with robust attenuation weights to immunize against metal artifacts and provide... Extremely high level of geometric locking precision.

[0077] In another alternative embodiment, for applications that allow for some manual intervention or do not require high local precision, the bounding box matrix can be automatically assigned instead of being used in the initial registration stage. Instead, the physician can manually select multiple sets of homologous anatomical feature points (such as specific cusps and fissures) on the intraoral scan and CBCT to provide the initial rigid body transformation matrix, and then perform ICP iteration.

[0078] In yet another optional embodiment, the registration and suturing module 203 performs crown removal processing on the tooth root registration model to obtain the tooth root crown-removed model. The specific methods include: The tooth root registration model is filtered for registration anomalies based on the preset maximum allowable nearest neighbor distance to obtain the tooth root filtering model; The root filtering model is transformed into an undirected weighted graph model to obtain a root weighted graph model. Based on the preset maximum search diffusion radius, the root weighted graph model is subjected to shortest path traversal for crown removal to obtain the root crown removal model.

[0079] Traditional techniques heavily rely on manual extraction of the cervical margin line by dentists to deconstruct the tooth structure, or on simple cross-sectional geometric plane cutting to remove the crown of the CBCT. The cutting transition is stiff and unnatural, and manual drawing is extremely inefficient. Conventional geometric plane cutting produces a flat concave effect like "cutting an apple" in cross-section, resulting in obvious step-like faults in the cervical region of the fused model. It cannot truly reflect the natural convex and full shape of the junction between dentin and enamel.

[0080] In this embodiment of the invention, instead of traditional geometric plane cutting, when mapping the intraoral scan edge to the CT surface as an anchor point, the maximum allowable distance between the nearest neighbors in the KD-Tree is set to 2.0 mm to filter out registration anomalies or damaged surfaces. The CT surface is then converted into an undirected weighted graph model containing edge connectivity weights, and the Dijkstra shortest path algorithm is used for pathfinding. To prevent the cutting line from "running across the entire image" due to CT perforations, a dynamic maximum search diffusion radius is introduced, with this limit set to 5.0 times the straight-line distance between two points (and the lower limit not less than 5.0 mm). After coronal resection, the ClusterConnected Triangles connected domain clustering algorithm is used to accurately extract and retain the largest single-volume CT root fragment.

[0081] In this alternative embodiment, by introducing a maximum diffusion radius limited to 5 times the actual physical straight-line distance, the undulations of the real surface mesh are tightly engaged, thereby completely eliminating the industrial problem of spatial discontinuity jumps in the cutting path.

[0082] It should be noted that, in addition to using graph theory-based shortest path search, the "Region Growing Algorithm Based on Curvature Tensor and Normal Divergence" or the "Graph Cut Algorithm Based on Markov Random Field" can also be used to achieve automated topological stripping of CT pathological coronal regions.

[0083] Please see Figure 4 , Figure 4 This is a schematic diagram of a tooth crown and tooth root in a state to be sutured, according to an embodiment of the present invention. Figure 4 (a) is a side view. Figure 4 (b) is a top view. Figure 2 and Figure 3 Based on this, after high-precision registration and adaptive graph theory crown removal, the spatial alignment between the dental crown and the CT root, and the physical gap to be sutured, are as follows: Figure 4 As shown.

[0084] In another alternative embodiment, for applications that allow for some manual intervention or do not require high local precision, instead of using graph theory (Dijkstra) pathfinding mapping during boundary extraction and crown removal, a three-dimensional interactive interface is provided, allowing operators to manually select and draw the cervical line on the CBCT mesh using a pen or punctuation tool, and then perform surface removal based on this line.

[0085] In yet another optional embodiment, the registration and suturing module 203 performs geometric suturing on the crown registration model and the root crown removal model to obtain the initial tooth model in the following specific ways: The starting point and direction of the suture mapping between the crown registration model and the root crown removal model are determined based on the preset global absolute alignment operator. The target bridging mesh is generated based on the predetermined radial expansion compensation amount, stitching mapping start point, and stitching mapping direction; The crown registration model and the root crown removal model are stitched together based on the target bridging mesh to obtain the initial tooth model.

[0086] Traditional techniques for joining tooth crowns and roots typically employ simple spatial linear interpolation to transition and connect boundary vertices. However, at complex tooth edges, simple linear interpolation and blind stitching of uneven starting points can easily lead to serious topological errors such as twisted and self-intersecting meshes on the bridging surface, and normal flipping (which appears as black holes in the rendering).

[0087] In this embodiment of the invention, the arc length of the same number of vertices is uniformly resampled based on the upper and lower boundary rings extracted from the model. To avoid the "twisted" effect caused by direct stitching, a global exhaustive absolute alignment operator is introduced (a step size of 2 can be set when traversing offsets to accelerate Euclidean distance calculation) to find the topological mapping starting point and direction with the minimum overall connection cost. During bridging mesh generation, an expansion stitching mechanism based on real cross-sections is pioneered. First, the joint spatial centroid of the two endpoint sets is calculated as the growth base point. An S-shaped easing function is set. Calculate the weighted interpolation. For each grid generation layer, calculate the target desired radius. Compare this with the actual bias shrinkage radius of the direct linear interpolation point. Subtracting the two values ​​yields the radial expansion compensation amount. Finally, this compensation amount is accumulated along the outward normal direction of the center of gravity, perfectly eliminating the "apple-cutting" concavity and restoring the true natural physiological fullness.

[0088] In this optional embodiment, utilizing The solution using bidirectional operators with varying complexity overcomes the limitations of single-feature alignment; it fits the true physiological expectation radius using sigmoid weights and applies them. The compensation scalar restores the fullness of the cervical protrusion.

[0089] It should be noted that, in generating the transition zone, in addition to using the radial expansion supplementation algorithm, implicit surface function fitting algorithms such as Poisson Surface Reconstruction can also be applied to fill the gaps between the upper and lower mesh segments.

[0090] In another alternative embodiment, for applications where some manual intervention is permissible or where local precision requirements are not high, the upper and lower boundaries can be directly connected using existing conventional Hole Filling algorithms (such as linear interpolation stitching) during the transition zone stitching stage, and basic Laplacian smoothing can be applied. Although this approach cannot avoid local shrinkage and mesh kinking, it can serve as a simple alternative for low-precision requirements.

[0091] In another optional embodiment, the smoothing repair module 204 performs smoothing repair processing on the initial tooth model according to preset smoothing adjustment parameters to obtain the target tooth model in the following specific ways: Within a preset polishing iteration cycle, the initial tooth model is smoothed by using forward shrinkage parameters and reverse expansion parameters to obtain a smooth tooth model. Determine the extrapolated normal vector of the oral scan based on the oral scan data of the crown; Obtain the outward normal vector of the suture surface of the initial tooth model; If the vector product of the external normal vector and the outward normal vector is less than zero, then the normal orientation of the smooth tooth model is restored to obtain the target tooth model.

[0092] Traditional techniques for handling geometric abrupt changes and step noise in spliced ​​meshes typically employ conventional Laplacian smoothing operators. However, inherent mathematical flaws in conventional Laplacian smoothing cause the solid mesh to shift towards a local mean (i.e., inward). After multiple iterations, this can lead to catastrophic shrinkage, altering the true physiological dimensions of teeth. In orthodontics, this can even introduce errors of a few tenths of a millimeter, which is unacceptable and fails to meet the extremely high precision requirements of subsequent orthodontic biomechanical simulations or implant guide design.

[0093] In this embodiment of the invention, the Laplacian smoothing operator, which leads to irreversible shrinkage, is no longer used for the abrupt changes in the machining polylines of the bridging mesh. A two-parameter Taubin relaxation controller with four layers of diffusion into the neighborhood is deployed. Positive shrinkage parameters are alternately applied within a set optimal polishing iteration cycle of 40 times. To eliminate local sharp edges, a reverse expansion parameter is then applied. Forced mesh boundary bounce. Before final output, the centroid ray method, which easily causes incorrect flipping of faces in groove regions, is no longer used. Instead, the outer normal of the original mesh edge starting point is used as the truth criterion, and a vector dot product is performed between it and the outgoing normal of the newly generated bridging surface. If the dot product is less than zero, the transition area faces are instructed to uniformly swap the underlying vertex indices, correcting the normal orientation in one go and eliminating downstream rendering black holes.

[0094] In this optional embodiment, specific combinations of mutual exclusion factors are configured. A perfect low-pass filter was constructed in the frequency domain, which maintained its low-frequency intrinsic gain without shrinking after dozens of deep iterations; and all reflective holes and leaks were repaired by resetting the winding order through the dot product verification of the native boundary extension normal.

[0095] It should be noted that, in addition to the Taubin two-parameter smoothing operator in this embodiment of the invention, the HC-Laplacian Smoothing algorithm can also be used in engineering. This algorithm calculates the coordinate difference between the initial mesh and the mesh after conventional Laplacian smoothing, and adds the lost volume back to the vertex positions in an inverse proportion, which can be used as another alternative algorithm for anti-shrinkage smoothing.

[0096] Please see Figure 5 , Figure 5 This is a schematic diagram of a target tooth model according to an embodiment of the present invention. Figure 5 (a) is a side view. Figure 5 (b) is a top view. Figure 4 Based on this, a high-fidelity fusion model is generated by radial expansion suturing and dual-parameter volume-preserving smoothing. The seam exhibits natural physiological fullness with no volume shrinkage.

[0097] The tooth model fusion method in this invention employs a non-destructive local meshing strategy, limiting the generation and topology modification of mesh vertices to the physical tomographic gaps after crown removal. By directly locking and inheriting the underlying geometric coordinates and native topology of the original oral scan crown and effective CT root, it completely abandons conventional global implicit surface reconstruction algorithms. The final output watertight solid model retains 100% lossless micron-level high-frequency anatomical features at the oral scan end and the true intraosseous morphology at the CT end.

[0098] Example 3 Please see Figure 7 , Figure 7 This is a schematic diagram of another tooth model fusion device disclosed in an embodiment of the present invention. Figure 7 The apparatus shown can be used to perform the method described in Embodiment 1, and this apparatus can improve the accuracy of dental model fusion. Figure 7 As shown, the tooth model fusion device disclosed in this invention includes, but is not limited to: Memory 301 storing executable program code; Processor 302 coupled to memory 301; The processor 302 calls the executable program code stored in the memory 301 to execute some or all of the steps in the tooth model fusion method described in Embodiment 1 of the present invention.

[0099] Example 4 This invention discloses a computer storage medium storing computer instructions. When the computer instructions are invoked by a processor, they are used to execute some or all of the steps in the tooth model fusion method described in Embodiment 1 of this invention.

[0100] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules. They may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0101] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0102] Finally, it should be noted that the technical content disclosed in the embodiments of the present invention is only a preferred embodiment of the present invention and is only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for fusing tooth models, characterized in that, The method includes: Acquire dental crown intraoral scan data and dental root image data; the dental crown intraoral scan data is dental crown data obtained by performing a three-dimensional oral cavity scan of the target tooth in advance, and the dental root image data is dental root data obtained by performing a computed tomography scan of the target tooth in advance. Based on the dental crown intraoral scan data and the dental root image data, a dental mesh is generated to obtain a dental mesh model; The tooth mesh model is registered and stitched to obtain the initial tooth model; The initial tooth model is smoothed and repaired according to preset smoothing adjustment parameters to obtain the target tooth model; The smoothing adjustment parameters include forward shrinkage parameters and reverse expansion parameters. The forward shrinkage parameters are used to eliminate local sharp edges of the model, and the reverse expansion parameters are used to control the springback of the model mesh boundaries. The registration stitching process is used to perform non-destructive local mesh construction and topology modification. The smoothing repair process is used to keep the original bottom-level geometric coordinates of the crown occlusal scan data and the root image data in the target tooth model unchanged.

2. The tooth model fusion method according to claim 1, characterized in that, The step of generating a tooth mesh model based on the crown intraoral scan data and the root image data includes: The dental crown oral scan data is separated into individual tooth instances according to a preset instance segmentation network to obtain oral scan tooth instances; The tooth root image data is segmented into individual tooth instances according to a preset instance segmentation network to obtain image tooth instances. The oral scan tooth instance is subjected to topological cleaning, and the closed edge loop of the oral scan tooth instance is extracted based on a preset edge reference statistics method to obtain the target closed edge loop; the topological cleaning is used to remove non-manifold edges and degenerate triangles; The target closed edge ring is smoothed in one dimension to obtain a crown mesh model; the one-dimensional smoothing is used to remove edge burrs. The image tooth instance is subjected to manifold cleaning and noise reduction processing to obtain a tooth root mesh model; The tooth mesh model includes the crown mesh model and the root mesh model.

3. The tooth model fusion method according to claim 2, characterized in that, The process of registering and stitching the tooth mesh model to obtain the initial tooth model includes: Rigid body registration is performed on the tooth mesh model to obtain a tooth registration model; the tooth registration model includes a crown registration model and a root registration model; The tooth root registration model is subjected to crown removal processing to obtain a tooth root crown-removed model; The crown registration model and the root crown removal model are geometrically sutured to obtain the initial tooth model.

4. The tooth model fusion method according to claim 3, characterized in that, The process of performing rigid body registration on the tooth mesh model to obtain a tooth registration model includes: The tooth mesh model is downsampled according to the preset first voxel mesh, and the three-dimensional bounding box of the tooth mesh model is calculated. The target cost matrix is ​​obtained by constructing a cost matrix based on the three-dimensional bounding box and the preset volume inclusion rate threshold. Based on the target cost matrix, the tooth mesh model is subjected to mesh mapping filtering to obtain a tooth filtering model; The tooth filtering model is downsampled according to the preset second voxel mesh, and the robustness of the tooth filtering model is adjusted to obtain the tooth adjustment model. The tooth adjustment model is registered using normal locking at full resolution according to a preset convergence index to obtain a tooth registration model.

5. The tooth model fusion method according to claim 3, characterized in that, The process of removing the crown from the registered tooth root model to obtain a crown-removed tooth root model includes: The tooth root registration model is subjected to registration anomaly filtering based on the preset maximum allowable nearest neighbor distance to obtain a tooth root filtering model; The root filtering model is transformed into an undirected weighted graph model to obtain a root weighted graph model. Based on the preset maximum search diffusion radius, the root weighted graph model is subjected to shortest path traversal for crown removal to obtain the root crown removal model.

6. The tooth model fusion method according to claim 3, characterized in that, The process of geometrically stitching the crown registration model and the root crown removal model to obtain an initial tooth model includes: The suture mapping start point and suture mapping direction between the crown registration model and the root crown removal model are determined according to the preset global absolute alignment operator; A target bridging mesh is generated based on a predetermined radial expansion compensation amount, the stitching mapping start point, and the stitching mapping direction; The crown registration model and the root crown removal model are stitched together according to the target bridging mesh to obtain the initial tooth model.

7. A method for fusing a tooth model according to any one of claims 1 to 6, characterized in that, The step of smoothing and repairing the initial tooth model according to preset smoothing adjustment parameters to obtain the target tooth model includes: Within a preset polishing iteration cycle, the initial tooth model is smoothed by the positive shrinkage parameter and the reverse expansion parameter to obtain a smooth tooth model. Determine the extrapolated normal vector of the oral scan based on the oral scan data of the crown. Obtain the outward normal vector of the suture surface of the initial tooth model; If the vector product of the external normal vector of the oral scan and the outgoing normal vector is less than zero, then the normal orientation of the smooth tooth model is repaired to obtain the target tooth model.

8. A dental model fusion device, characterized in that, The device includes: The data acquisition module is used to acquire crown intraoral scan data and root image data; the crown intraoral scan data is obtained by performing a three-dimensional oral cavity scan on the target tooth in advance, and the root image data is obtained by performing a computed tomography scan on the target tooth in advance. The mesh processing module is used to generate a tooth mesh based on the crown intraoral scan data and the root image data, thereby obtaining a tooth mesh model. The registration and stitching module is used to perform registration and stitching on the tooth mesh model to obtain an initial tooth model. The smoothing repair module is used to perform smoothing repair processing on the initial tooth model according to preset smoothing adjustment parameters to obtain the target tooth model; The smoothing adjustment parameters include forward shrinkage parameters and reverse expansion parameters. The forward shrinkage parameters are used to eliminate local sharp edges of the model, and the reverse expansion parameters are used to control the springback of the model mesh boundaries. The registration stitching process is used to perform non-destructive local mesh construction and topology modification. The smoothing repair process is used to keep the original bottom-level geometric coordinates of the crown occlusal scan data and the root image data in the target tooth model unchanged.

9. A dental model fusion device, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute a tooth model fusion method according to any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked by a processor, are used to execute a tooth model fusion method according to any one of claims 1 to 7.

Images