Oral and maxillofacial cone beam CT image tooth three-dimensional reconstruction method and system thereof

Abstract

The application belongs to the technical field of image processing, and specifically provides a method and system for tooth three-dimensional reconstruction of a cone beam CT image of an oral and maxillofacial region. The method mainly performs gray separation and morphological denoising on a CBCT voxel matrix to define a tooth core and an environment matrix; a probability field and a gradient sequence are generated through density attenuation mapping to detect extreme value inflection points to extract a cyst bag point cloud and thickness; after tolerance checking and reconstruction of a grid, hierarchical splicing is performed to realize high-precision tooth reconstruction containing a cyst bag feature. The application realizes high-fidelity three-dimensional shape restoration of an embedded and impacted tooth wrapped by a complex cyst bag, effectively eliminates geometric deviation at the junction of blurred tissues, and thus provides a reliable spatial position basis for preoperative risk accurate assessment, guide plate high-precision design and safe planning of a surgical path.

Description

Technical Field

[0001] This application belongs to the field of image processing technology, and specifically relates to a method and system for three-dimensional reconstruction of teeth from cone-beam computed tomography (CBCT) images of the oral and maxillofacial region. Background Technology

[0002] In oral and maxillofacial medical image processing, a widely used standard procedure is to acquire voxel grayscale information generated by cone-beam computed tomography (CBCT) scans and then use a fixed grayscale threshold segmentation algorithm to achieve three-dimensional reconstruction and target segmentation of the target tooth and surrounding anatomical structures.

[0003] However, impacted teeth and their surrounding anatomical tissues are generally tightly wrapped by a dentin-containing pocket. The density difference between the soft tissue filling the pocket and structures such as the dental pulp and periodontal ligament is very small. Furthermore, due to the objective limitations of the resolution and partial volume effect of CBCT equipment, the pocket boundary and the adjacent trabeculae, tooth roots, and mandibular nerve canal often appear as an intertwined and blurred transition in the images. Under these conditions, it becomes very difficult to accurately separate the tissue interface from these blurred transition areas with extremely similar densities. This can directly lead to pseudo-fusion or geometric distortion at the edges of the reconstructed tooth model, resulting in spatial positional deviations in preoperative risk assessment, clinical guide design, or surgical path planning.

[0004] Therefore, it is necessary to propose a method and system for three-dimensional reconstruction of teeth from cone-beam computed tomography (CBCT) images of the oral and maxillofacial region to solve the above problems. Summary of the Invention

[0005] This application provides a method and system for three-dimensional reconstruction of teeth using cone-beam computed tomography (CBCT) images of the oral and maxillofacial region. This method effectively solves the problem in existing technologies where the density of the impacted tooth pocket and the surrounding tissue are extremely close and the boundaries are blurred and intertwined. This makes it difficult for traditional segmentation algorithms to accurately separate the tissue interface, which easily leads to false fusion and geometric distortion at the edges of the reconstructed model. This method achieves high-fidelity three-dimensional morphological restoration of impacted teeth under complex pockets, effectively eliminating geometric deviations at blurred tissue boundaries. As a result, it provides a reliable spatial location basis for accurate preoperative risk assessment, high-precision guide design, and safe surgical path planning.

[0006] To achieve the above objectives, this application adopts the following technical solution:

[0007] In a first aspect, this application provides a method for three-dimensional reconstruction of teeth from cone-beam computed tomography (CBCT) images of the oral and maxillofacial region, including:

[0008] A three-dimensional region of interest (ROI) voxel matrix containing the target anatomical structure is obtained. Gray-level separation is performed on the ROI voxel matrix to extract the initial screening highlighted connected regions and local background space. Morphological denoising and entity definition are performed on the initial screening highlighted connected regions and local background space to obtain a high-density tooth core and environmental reference matrix. Peripheral density attenuation mapping is performed based on the high-density tooth core and environmental reference matrix to extract the spatial topological probability distribution field representing the tissue transition boundary and the gradient direction vector sequence pointing to the outer edge. Extreme inflection point detection at the soft-hard tissue interface is performed on the spatial topological probability distribution field and gradient direction vector sequence to obtain the pocket envelope surface point cloud and physiological thickness range. Tolerance verification and surface reconstruction are performed based on the pocket envelope surface point cloud, physiological thickness range, and spatial topological probability distribution field to generate a pocket layered mesh. The pocket layered mesh and the spatial coordinates of the high-density tooth core are stitched together to generate a three-dimensional tooth reconstruction model containing pocket layered features.

[0009] Secondly, this application provides a three-dimensional reconstruction system for teeth from cone-beam computed tomography (CBCT) images of the oral and maxillofacial region, comprising:

[0010] Voxel extraction module: Used to obtain a three-dimensional region of interest voxel matrix containing the target anatomical structure, perform grayscale separation on the three-dimensional region of interest voxel matrix, and extract the initial screening of bright connected components and local background space.

[0011] Morphological denoising module: used to perform morphological denoising and entity definition on the initially screened bright connected components and local background space to obtain high-density tooth core and environmental reference matrix.

[0012] Density mapping module: used to perform peripheral density attenuation mapping based on the high-density tooth core and environmental reference matrix, and extract the spatial topological probability distribution field and gradient direction vector sequence pointing to the outer edge that characterize the tissue transition boundary.

[0013] Inflection point detection module: used to perform extreme inflection point detection at the junction of soft and hard tissues on the spatial topological probability distribution field and gradient direction vector sequence, to obtain the point cloud of the capsule envelope surface and the physiological thickness range.

[0014] Surface Reconstruction Module: Used to perform tolerance verification and surface reconstruction based on the point cloud of the capsule envelope surface, physiological thickness range and spatial topological probability distribution field, and generate capsule layered mesh.

[0015] Model stitching module: Used to stitch together the spatial coordinates of the pocket layered mesh and the high-density tooth core to generate a 3D reconstructed tooth model containing pocket layered features.

[0016] Thirdly, this application provides a readable storage medium, comprising: computer program instructions stored in the readable storage medium, wherein the computer program instructions are read and executed by a processor to perform the steps of a method for three-dimensional reconstruction of teeth from cone-beam CT images of the oral and maxillofacial region.

[0017] The beneficial effects of this application are:

[0018] This application employs a method of extracting a high-density tooth core and an environmental reference matrix, and generating a probability distribution field and gradient vector sequence through peripheral density attenuation mapping. Based on this, extreme inflection points are detected to obtain the point cloud and thickness of the capsule. Finally, after tolerance verification, the mesh is reconstructed and stitched with the core to generate a reconstructed model. This effectively solves the problem in existing technologies where the density of the impacted tooth capsule and the surrounding tissue are extremely close and the boundaries are blurred and intertwined, making it difficult for traditional segmentation algorithms to accurately separate the tissue interface. This easily leads to false fusion and geometric distortion at the edges of the reconstructed model. This method achieves high-fidelity three-dimensional morphological restoration of impacted teeth under complex capsules, effectively eliminating geometric deviations at blurred tissue boundaries. As a result, it provides a reliable spatial location basis for accurate preoperative risk assessment, high-precision guide design, and safe surgical path planning.

[0019] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description and the accompanying drawings. Attached Figure Description

[0020] Figure 1 A flowchart illustrating the method for three-dimensional reconstruction of teeth from cone-beam computed tomography images of the oral and maxillofacial region according to this application is shown.

[0021] Figure 2 A schematic diagram of the process for generating the initial topological probability field and the initial direction vector array in this application is shown;

[0022] Figure 3 A schematic diagram of the process for generating the initial thickness sequence and the initial pocket point cloud in this application is shown;

[0023] Figure 4 A schematic diagram of the modules of the oral and maxillofacial cone-beam CT image tooth three-dimensional reconstruction system of this application is shown. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] In the clinical application of cone-beam computed tomography (CBCT) in the oral and maxillofacial region, three-dimensional reconstruction of impacted teeth is crucial for preoperative risk assessment. However, because impacted teeth are usually enclosed in dentin pockets, the density difference between the soft tissues filling the pockets and structures such as the pulp and periodontal ligament is minimal, limiting image contrast. Due to the inherent resolution and partial volume effect of CBCT equipment, the pocket boundaries often appear as blurred transitions with adjacent trabeculae, roots, and the mandibular canal on images. Traditional segmentation methods based on fixed grayscale thresholds struggle to accurately separate these interfaces of similar density, easily leading to pseudo-fusion or geometric distortion at the edges of the reconstructed tooth model, resulting in spatial deviations in clinical guide design or surgical path planning. This approach aims to accurately segment and reconstruct the three-dimensional structural model of the tooth and its associated pockets by processing voxel data.

[0026] In some embodiments, such as Figure 1 As shown, this application provides a method for three-dimensional reconstruction of teeth from cone-beam computed tomography (CBCT) images of the oral and maxillofacial region, including:

[0027] S1. Obtain the three-dimensional region of interest voxel matrix containing the target anatomical structure, perform grayscale separation on the three-dimensional region of interest voxel matrix, and extract the initial screening of bright connected components and local background space.

[0028] Cone-beam technique in the maxillofacial region During image processing, the three-dimensional image data is first acquired after preliminary reconstruction by the scanning device. This three-dimensional image data consists of voxel points arranged in a grid in three-dimensional space. Each voxel point contains specific three-dimensional spatial coordinate information and gray value reflecting the tissue density at that location, thus forming a three-dimensional region of interest voxel matrix.

[0029] To distinguish between high-density hard tissue and low-density soft tissue, existing image processing operations such as grayscale thresholding and connected component labeling are performed on the voxel matrix of the 3D region of interest. A set of voxels with high-brightness features is selected based on defined grayscale conditions, and connected voxels are grouped into initial high-brightness connected components using spatial adjacency relationships. Simultaneously, all other 3D spatial coordinates not belonging to these connected components are categorized and extracted to form a local background space.

[0030] For example, maxillofacial tomographic scan data is a three-dimensional region of interest voxel matrix. Through grayscale separation, the crowns and dense bone that present extremely high brightness can be initially separated out, while the areas containing dull pocket soft tissue, nerve canals and loose bone are classified as background space.

[0031] S2. Perform morphological denoising and entity definition on the initially screened bright connected components and local background space to obtain the high-density tooth core and environmental reference matrix.

[0032] Due to unavoidable system noise and local volumetric effects during the scanning and imaging process, burrs or artifacts may appear at the edges of the extracted connected regions. Therefore, existing three-dimensional mathematical morphological filtering operations are performed on the initially screened bright connected regions to remove unstable voxel structures at the edges, retaining the core structure with extremely high density as the high-density tooth core. Based on the defined core structure, a certain spatial range is extended outward to extract the specific region surrounding the core in the local background space, forming an environmental reference matrix for subsequent density analysis.

[0033] For example, when dealing with impacted teeth, the initially extracted bright areas may adhere to the surrounding alveolar bone due to image blurring. Through noise reduction and delimitation operations, this blurry layer of adhesion can be peeled off, exposing the internal core that is absolutely the hardest part of the tooth. A three-dimensional annular observation area that can cover the entire tooth-containing pocket can then be delineated outward from this core.

[0034] S3. Based on the high-density tooth core and the environmental reference matrix, perform peripheral density attenuation mapping to extract the spatial topological probability distribution field characterizing the tissue transition boundary and the gradient direction vector sequence pointing to the outer edge.

[0035] After defining the core structure of the tooth, it is necessary to analyze the physical laws governing the transition from high-density hard tissue to low-density soft tissue in the periphery. By traversing the three-dimensional spatial coordinates in the environmental reference matrix, a distance transformation algorithm is used to calculate the spatial distance characteristics between the core structure and the high-density tooth structure, and a three-dimensional spatial difference algorithm is used to calculate the direction and amplitude characteristics of its grayscale changes.

[0036] By performing nonlinear mathematical fusion on spatial distance features and grayscale change features, the probability value of each voxel belonging to the tissue boundary edge is calculated, and the direction vector of the fastest grayscale decrease is recorded. This generates a spatial topological probability distribution field containing all probability values, and a gradient direction vector sequence containing all direction vectors.

[0037] For example, when probing the boundary of a dental pocket, the area where the fluid inside the pocket transitions to the external alveolar bone will exhibit a specific density gradient change. Through the above feature calculation and fusion, this gray-scale gradient process, which is difficult to discern with the naked eye, can be transformed into an intuitive probability numerical distribution. At the same time, the normal direction of the pocket's outward expansion can be indicated by the vector sequence.

[0038] S4. Perform extreme inflection point detection at the soft-hard tissue interface on the spatial topological probability distribution field and gradient direction vector sequence to obtain the point cloud of the capsule envelope surface and the physiological thickness range.

[0039] The physical properties of the transition boundary between soft and hard tissues surrounding the tooth manifest as local extrema in a probability distribution field. Following the spatial direction indicated by the gradient direction vector sequence, a local extrema search algorithm is executed in the spatial topological probability distribution field to locate discrete spatial coordinate points representing the physical separation interface between soft and hard tissues. These discrete coordinate points are combined to construct a point cloud of the capsule envelope surface describing the capsule's geometry. Simultaneously, statistical analysis of the distances between these discrete coordinate points and the internal core is performed to calculate the maximum extraocular thickness and minimum apposition thickness of the capsule structure in space, defining a reasonable physiological thickness range.

[0040] For example, when tracing the outer wall of the cyst, the probe is made along the outward ray direction of the tooth. The point where the probability value fluctuates drastically and reaches a minimum value is usually the interface between the soft tissue of the cyst and the surrounding jawbone. Collecting the three-dimensional points on these interfaces forms a discrete point cloud network surrounding the impacted tooth, and the thickness distribution range of the cyst is obtained simultaneously, providing a size reference for screening pathological cysts.

[0041] S5. Based on the point cloud of the capsule envelope surface, the physiological thickness range, and the spatial topological probability distribution field, tolerance verification and surface reconstruction are performed to generate a capsule layered mesh.

[0042] To eliminate erroneously extracted free noise during extreme value detection, the thickness dimensions of each point in the capsule envelope surface point cloud are numerically constrained and compared with the physiological thickness range. After removing abnormal discrete points exceeding the reasonable thickness range, the effective point set conforming to anatomical rules is retained. Using a 3D surface reconstruction algorithm, the effective point set is connected according to spatial topological relationships to form a 3D surface model composed of continuous polygons, generating a capsule layered mesh reflecting the outer boundary of the capsule.

[0043] For example, before generating the surgical navigation model, the system may mistake the distant neural canal boundary for the bag boundary. Since the neural canal is extremely far away, its thickness will inevitably exceed the normal physiological thickness range. After removing the abnormal point through tolerance verification, the remaining accurate point cloud is used to construct a smooth and closed bag mesh epidermis, which can effectively prevent interference in the design of the surgical guide.

[0044] S6. Based on the spatial coordinates of the pocket layered mesh and the high-density tooth core, a three-dimensional reconstruction model of the tooth containing pocket layered features is generated by stitching them together.

[0045] The three-dimensional spatial coordinates of the high-density tooth core, representing the dense hard tissue inside the tooth, and the three-dimensional spatial coordinates of the layered mesh of the capsule, representing the outer soft tissue covering layer, are placed in the same three-dimensional Cartesian coordinate system for spatial alignment. Using Boolean union operations in the field of three-dimensional computer-aided design, the spatial volume information of the two is logically integrated to finally generate a three-dimensional reconstruction model of the tooth that includes both the internal crown and root entities and the external layered structure containing the capsule.

[0046] For example, when the final clinical plan is output, what doctors see on the workstation screen will no longer be a single bone block model, but a three-dimensional entity with a bright and dense inner layer of tooth structure and an outer layer of translucent pockets. This provides an intuitive digital twin model for accurately assessing the resistance to extraction of impacted teeth and avoiding damage to adjacent tissues.

[0047] In some embodiments, grayscale separation is performed on the voxel matrix of the three-dimensional region of interest to extract the initially screened bright connected components and local background space, including:

[0048] S11. Extract voxels with gray values ​​greater than the preset hard tissue gray value threshold from the voxel matrix of the three-dimensional region of interest to generate a preliminary set of high-brightness voxels.

[0049] The hard tissue grayscale threshold is the grayscale boundary value used to distinguish between high-density bone tissue and low-density soft tissue. A large number of real clinical extracted tooth specimens and their corresponding oral and maxillofacial cone beams can be collected in advance. Image data is analyzed using statistical analysis software to extract grayscale distribution curves of known enamel and dentin regions in the images. The values ​​corresponding to the peaks of the grayscale distribution are calculated, and the arithmetic mean is used as the threshold.

[0050] The grayscale value of each voxel in the voxel matrix of the 3D region of interest is read one by one, and compared with a preset hard tissue grayscale threshold. The 3D spatial coordinates and corresponding voxels with grayscale values ​​strictly greater than the threshold are retained. All voxels that meet the condition are aggregated and combined to generate an initial set of high-brightness voxels.

[0051] For example, the crown surface of an impacted tooth is usually covered with a very thin soft tissue pocket, while the enamel inside the crown appears to have extremely high grayscale brightness in the image due to its high degree of calcification. Through the above extraction operation, the bright enamel and part of the dense jawbone can be initially separated from the surrounding dark soft tissue pocket and the cancellous part of the alveolar bone, thus providing initial high-brightness voxel data support for locking the core of the tooth.

[0052] S12. Perform three-dimensional connected domain volume calculation on the initial screening high-brightness voxel set, and extract the three-dimensional connected voxel cluster with the largest volume to define the initial screening high-brightness connected domain.

[0053] Within the target anatomical region, the main structure of the tooth to be reconstructed typically occupies the largest continuous high-density volume.

[0054] For any voxel in the initial screening highlight voxel set, retrieve its 26 immediately adjacent voxel positions in 3D space. If adjacent positions contain voxels belonging to the same set, mark them as the same connected region. Continue marking all contacting voxels until all voxels are marked, thus dividing the initial screening highlight voxel set into several independent connected voxel clusters. Count the total number of voxels contained within each connected voxel cluster; the total number of voxels represents the 3D volume of the connected region. Sort the volume values ​​of all connected voxel clusters and extract the 3D connected voxel cluster with the largest volume value, defining it as the initial screening highlight connected region.

[0055] For example, scattered artifact fragments from metal brackets can form multiple isolated bright spots, while the crown and root of the target impacted tooth as a whole can form a single, continuous bright area. By extracting the largest volume cluster, the surrounding discrete trabecular bone fragments or metal artifacts can be effectively filtered out.

[0056] S13. Remove the three-dimensional spatial coordinates occupied by the initially screened highlighted connected components from the three-dimensional region of interest voxel matrix, and obtain all the remaining three-dimensional spatial coordinates to form the local background space.

[0057] Using the full set of 3D spatial coordinates contained in the voxel matrix of the 3D region of interest as the base set, the 3D spatial coordinates belonging to the initially highlighted connected regions are used as the subtraction set. Coordinates within the subtraction set are completely removed from the base set, retaining the remaining 3D spatial coordinates. All these remaining 3D spatial coordinates are then integrated to form a local background space. This local background space represents the anatomical environment surrounding the high-density tooth core, filled with soft tissue, cystic fluid, and low-density bone.

[0058] For example, after extracting the main body of the impacted tooth, the coordinate culling operation can hollow out the tooth body in three-dimensional space. The remaining space includes the dentin pocket tightly surrounding the tooth, the periodontal ligament space, and the adjacent nerve canal trend area, providing a precise search range for extracting the transition boundary in these relatively dark areas.

[0059] In some embodiments, morphological denoising and entity definition are performed on the initially screened bright connected components and local background space to obtain a high-density tooth core and an environmental reference matrix, including:

[0060] S21. Perform a three-dimensional morphological erosion operation with a fixed kernel radius on the initially screened bright connected regions to generate a high-density tooth core.

[0061] Preset cube structuring elements, such as voxels voxels A voxel is used as the erosion nucleus, a cubic structural element. This nucleus is slid across a 3D grid space. If all voxels within the nucleus's coverage area belong to the initially screened bright connected regions, the center point is retained; otherwise, it is discarded. This process strips away blurred voxels at the boundaries, removes fine protrusion noise, and ensures that the resulting high-density tooth core represents the densest internal region of the tooth.

[0062] S22. Extract the coordinates of the outermost boundary voxels of the high-density tooth core and generate the coordinate set of the outer surface of the core.

[0063] Traverse all voxel points in the high-density tooth core. For each point, check whether there are voxels in its 26 neighborhoods that do not belong to the core. If so, determine that the point is a boundary point and record its three-dimensional spatial coordinates.

[0064] S23. Using the core outer surface coordinate set as the starting coordinate, perform unidirectional spatial expansion extraction in the local background space according to a fixed three-dimensional voxel step size, and define the extracted three-dimensional voxel set as the environment reference matrix.

[0065] An expansion algorithm is used to search outwards from the coordinate set of the core's outer surface, with the step size set to cover the maximum physiological range that the capsule may exist. Each unit voxel extracts voxels from the local background space within the search range, forming a strip-shaped 3D array around the core of the tooth, providing a precise feature calculation region for subsequent probability mapping.

[0066] In some embodiments, peripheral density attenuation mapping is performed based on the high-density tooth core and the environmental reference matrix to extract the spatial topological probability distribution field characterizing the tissue transition boundary and the gradient direction vector sequence pointing to the outer edge, including:

[0067] S31. Extract the three-dimensional spatial coordinates and grayscale values ​​of each voxel in the environmental reference matrix, calculate the minimum Euclidean distance from the coordinates of each voxel to the outermost boundary of the high-density tooth core, and generate a spatial distance matrix.

[0068] For any voxel point in the environmental reference matrix, search for the point closest to it in the coordinate set of the outer surface of the core. Calculate the geometric interval between the two points using the Euclidean distance formula, and store this interval value as the distance feature value of the voxel point, thereby constructing a data matrix describing the distance of the voxel from the core.

[0069] S32. Extract each voxel from the environment reference matrix along... , , The first-order difference of gray levels of the coordinate axes is used to generate a three-dimensional gray-level gradient field by combining the first-order difference of gray levels according to the three-dimensional spatial coordinates.

[0070] Central difference operator or The operator calculates the partial derivative of the gray value in three-dimensional space. Specifically, it can calculate the partial derivative of the gray value at that point with respect to its relative position in three-dimensional space. , , The grayscale difference between adjacent points in the positive and negative directions of the three coordinate axes reflects the degree and direction of density change at that point.

[0071] S33. Perform nonlinear feature fusion and probability decay calculation based on the spatial distance matrix and the three-dimensional gray-level gradient field to generate the initial topological probability field and the initial direction vector array.

[0072] Since the density at the junction of the tooth cell and the tooth pocket exhibits a non-linear decay trend with distance, and the partial volume effect causes the gradient information distribution to become uneven, it is necessary to use fusion calculation to quantify the probability that each voxel belongs to the edge of the tooth tissue.

[0073] S34. Extract each three-dimensional spatial vector from the initial direction vector array, divide the three-dimensional spatial vector by its own Euclidean modulus, and generate a gradient direction vector sequence.

[0074] Normalization is performed on each three-dimensional vector in the initial direction vector array. The square root of the sum of the squares of the three-dimensional components of the vector is calculated as the modulus. If the modulus is not zero, the component is divided by the modulus. The aim is to preserve the main trend direction of density change at each point and eliminate the absolute numerical difference of gradient modulus, thereby locking the geometric normal path pointing to the periphery of the tooth body.

[0075] S35. Extract the three-dimensional spatial coordinates of voxels with a gray value of zero in the environment reference matrix, extract the initial probabilities with the same three-dimensional spatial coordinates in the initial topological probability field and assign them to zero, and combine all the assigned probabilities to generate a spatial topological probability distribution field.

[0076] In clinical practice In an image, a gray value of zero usually represents an air region or an invalid background with extremely low density. This cleaning step can reduce the probability of noise in the distribution field that does not belong to biological tissue to zero, thereby improving the signal-to-noise ratio of the probability distribution field.

[0077] In some embodiments, such as Figure 2 As shown, nonlinear feature fusion and probability decay calculations are performed based on the spatial distance matrix and the three-dimensional grayscale gradient field to generate an initial topological probability field and an initial direction vector array, including:

[0078] S331. Extract the three-dimensional first-order difference vector of each voxel in the three-dimensional gray-level gradient field to obtain the initial direction vector array.

[0079] S332. Calculate the first... in the initial direction vector array The gradient magnitude is obtained by taking the Euclidean modulus of the three-dimensional first-order difference vector of the individual voxels. Extract the first element from the spatial distance matrix. Minimum Euclidean distance of voxels .

[0080] The unique index number of each voxel unit in the traversal environment reference matrix is ​​used to calculate the first voxel using the modulus formula. The length of each vector is used to quantify the intensity of grayscale abrupt changes at that point.

[0081] S333. Based on gradient magnitude and minimum Euclidean distance Calculate the first Fusion eigenvalues ​​of individual elements , .

[0082] By performing logarithmic compression on the gradient information, we can prevent high gradient values ​​caused by local metal artifacts or noise from excessively interfering with the overall structure. At the same time, by introducing distance information as a spatial constraint, the fusion result can reflect the coupling relationship between position and gray level.

[0083] S334. Based on gradient magnitude and fusion eigenvalues Calculate the first initial probability of individual elements , The initial topological probability field is generated by combining the initial probabilities of all voxels.

[0084] By utilizing the negative exponential decay property of the natural exponential function, the fused complex eigenvalues ​​are nonlinearly mapped to... Within the closed interval, the abstract numerical values ​​are transformed into the probability of a voxel belonging to the tissue boundary edge. Finally, the initial topological probability field is generated by combining the initial probabilities calculated from all voxels.

[0085] In some embodiments, extreme inflection point detection is performed at the soft-hard tissue interface on the spatial topological probability distribution field and gradient direction vector sequence to obtain the point cloud of the capsule envelope surface and the physiological thickness range, including:

[0086] S41. Extract the probabilities and 3D spatial coordinates of all voxels in the spatial topological probability distribution field, and combine them to generate a global topological dataset. The global topological dataset reflects the probability tendency of each coordinate position in 3D space to belong to the organizational transition boundary.

[0087] S42. Based on the global topology dataset, gradient direction vector sequence, and high-density tooth core, perform spatial minimum detection and thickness power scaling mapping to generate an initial thickness sequence and an initial pocket point cloud.

[0088] Along the normalized normal vector path, the minimum point where the value is reversed is found within the probability distribution space. This is used to extract the three-dimensional spatial discrete coordinate sequence representing the outermost physical interface of the dentate pocket. At the same time, nonlinear mapping calculations are performed in combination with the physical distance from each point to the core to evaluate and restore the true simulated soft tissue thickness at each extreme point. Initial pocket point cloud recording the three-dimensional coordinate point set and initial thickness sequence recording the thickness attribute values ​​are generated respectively.

[0089] S43. Extract discrete 3D spatial coordinates from the initial bag point cloud, perform Delaunay triangulation to generate a spatial triangular mesh, extract the 3D vertex coordinate set of all triangles in the spatial triangular mesh, and obtain the bag envelope surface point cloud.

[0090] The discrete 3D spatial coordinates of the initial pocket point cloud are used as the mesh vertices input to the Delaunay triangulation algorithm. Through mathematical programming and empty circle property verification, a network of interconnected and non-overlapping tetrahedral or spatial triangles is constructed in 3D space. The core property of this algorithm guarantees that the circumcircle or circumsphere of any triangle in the mesh absolutely does not contain any other free vertices from the mesh point set, thus minimizing the generation of elongated and deformed triangles and achieving optimal compactness of the topological connections between vertices and surface smoothness.

[0091] After the mesh construction process is completed, the set of coordinates of all three-dimensional vertices contained in the connected spatial triangle surfaces is extracted. This set is defined as the point cloud of the capsule envelope surface that characterizes the geometry of the capsule's outer membrane.

[0092] For example, when dealing with scattered capsule boundary detection points, these points are like suspended grains of sand in three-dimensional space. Through Delaunay triangulation calculation, the algorithm will automatically determine which grains of sand are closest and form the most reasonable plane. Then, a virtual film made up of countless tiny triangles is used to cover and connect these grains of sand, forming a continuous point cloud model with clear surface properties.

[0093] S44. Extract all thickness mapping values ​​from the initial thickness sequence, sort them in descending order, extract the maximum thickness value at the first position and the minimum thickness value at the last position, and combine them to form the physiological thickness range.

[0094] Using quicksort or mergesort algorithms, all simulated thickness values ​​recorded in the initial thickness sequence are read into memory and rearranged in absolute descending order from the largest to the smallest value.

[0095] The first data point in the sorted array sequence, representing the maximum physiological limit thickness of the bag-enclosed tissue in the entire 3D reconstructed area, is read. Simultaneously, the last data point in the array sequence, representing the minimum physiological limit thickness of the bag-enclosed tissue at the adhesion site, is read. These two extreme thickness values ​​are used as the upper and lower control values ​​of the thickness range, respectively. Their combination constitutes the baseline scale range for subsequent geometric tolerance filtering, i.e., the physiological thickness range.

[0096] In some embodiments, such as Figure 3 As shown, based on the global topology dataset, gradient direction vector sequence, and high-density tooth core, spatial minima detection and thickness power scaling mapping are performed to generate an initial thickness sequence and an initial pocket point cloud, including:

[0097] S421. Calculate the minimum Euclidean distance from each 3D spatial coordinate in the global topology dataset to the outermost boundary of the high-density tooth core to obtain the basic physical thickness. The probability fluctuation is obtained by extracting the probabilities of three adjacent voxels in the global topological dataset from the pointing vector directions with the same three-dimensional spatial coordinates in the gradient direction vector sequence and calculating the sum of the absolute values ​​of the differences between each pair of voxels. The three-dimensional spatial coordinates of the accumulated sum when it reaches a local minimum are used as the initial pocket point cloud.

[0098] Specifically, Represents the integer index number for iterating through discrete points sampled continuously in three-dimensional space. For the th... Using a sampling point, the basic physical thickness is obtained by executing the shortest spatial distance calculation logic described above. Then, following the direction of the unit vector corresponding to that point in the gradient direction vector sequence, the probability values ​​of three adjacent voxels in three-dimensional space are extracted along the normal forward, current, and backward respectively. The previous probability value is subtracted from the current probability value and the absolute value is taken, and the current probability value is subtracted from the next probability value and the absolute value is taken. Then, these two absolute values ​​are summed by algebraic addition to calculate the probability fluctuation describing the curvature of the local curve. .

[0099] Traverse the entire vector search path, when the probability fluctuation... When the value reaches a local minimum on the curve, it indicates that the first derivative of the probability distribution curve changes most smoothly and stably at this point. This minimum phenomenon reflects the physiological critical point at which the soft tissue inside the capsule gradually transitions to the alveolar bone tissue outside and eventually undergoes physical separation.

[0100] Record all the three-dimensional spatial coordinates that make the probability fluctuation reach a minimum value, and combine all the recorded discrete coordinates to form an initial pocket point cloud representing the boundary position.

[0101] For example, when tracing a vector line pointing toward the jawbone, the probability value initially changes drastically. Then, at the moment it passes through the cyst fluid and contacts the bone wall, the probability change tends to stagnate. At this point, the sum of the absolute values ​​of the probability differences between adjacent voxels falls to the bottom, and the system accurately captures the three-dimensional coordinates corresponding to this bottom as the true edge point of the cyst.

[0102] S422. Extract the probability that the points in the global topology dataset have the same 3D spatial coordinates as those in the initial pocket point cloud. According to probability fluctuation Sum of probabilities Calculate the first Power scaling factor at each spatial index location , .

[0103] Since the minimum point of the probability curve extracted by a single step is extremely susceptible to displacement due to local random speckle noise during image imaging, an adaptive scaling factor is constructed by performing a linear multiplication combination of the absolute probability level feature of this point and the probability fluctuation feature representing the stability of the local region. This scaling factor will be specifically used in subsequent steps to perform anatomical-level dynamic elastic compensation and correction for rigid physical geometric distances.

[0104] S423. Based on the basic physical thickness and power scaling factor Calculate the first Thickness mapping value at each spatial index position , The initial thickness sequence is generated by combining all thickness mapping values.

[0105] Cone bundles of oral and maxillofacial region Due to limitations in device pixel size and the significant influence of partial volume effects, the thickness of extremely thin soft tissues such as the dentin pocket wall often appears as a visual illusion of contraction or expansion on tomographic images. By using a power quadratic term to perform nonlinear amplification or reduction adjustment on a basic distance with physical length dimensions, the geometrically rigid straight-line distance can be nonlinearly remapped to a more realistic physiological thickness distribution that better reflects the actual encapsulation state of soft tissue in a real oral cavity environment. All calculated thickness mapping values ​​are arranged and combined according to the index order of discrete points to ultimately generate an initial thickness sequence for subsequent tolerance verification.

[0106] For example, the actual physical thickness of a certain sac is However, due to the volumetric effect causing image blurring, the extracted basic physical thickness is only [amount missing]. At this point, by performing nonlinear amplification through a power scaling factor that integrates probability and volatility information, the [data / property] can be [absorbed / suppressed]. The calculated value is dynamically compensated and restored to near its original value. The thickness mapping value is obtained, thereby greatly improving the dimensional accuracy of the model reconstruction.

[0107] In some embodiments, tolerance verification and surface reconstruction are performed based on the point cloud of the capsule envelope surface, the physiological thickness range, and the spatial topological probability distribution field to generate a capsule layered mesh, including:

[0108] S51. Extract the three-dimensional spatial coordinates of each discrete point in the point cloud of the capsule envelope surface, calculate the minimum Euclidean distance from each voxel coordinate to the outermost boundary of the high-density tooth core, and generate a measured thickness feature sequence with physical length dimensions.

[0109] Using the spatial shortest distance search calculation method, for each scattered coordinate point in the point cloud set, a global search is performed on all boundary coordinates of the dense solid surface of the tooth core, and the Euclidean distance of the shortest straight line connecting the two is calculated. This straight line Euclidean distance objectively reflects the direct physical span between the soft tissue surface boundary extracted by the algorithm in space and the internal hard tooth solid. The absolute distance values ​​calculated from all discrete coordinate points are arranged and combined in sequence to generate a measured thickness feature sequence containing only physical length information.

[0110] For example, the distance from the first point in the envelope surface point cloud to the tooth surface is calculated to be... The second point is The 500th point mutation is These specific physical measurement values, arranged sequentially, constitute this characteristic sequence.

[0111] S52. Compare the thickness values ​​in the measured thickness feature sequence with the upper and lower limits of the physiological thickness range, remove discrete points whose thickness values ​​exceed the physiological thickness range, and generate a cleaned set of boundary points.

[0112] Each specific thickness value in the measured thickness feature sequence is sequentially traversed, and each value is compared with the maximum and minimum thickness control values ​​of the pre-established physiological thickness range. If the thickness value is determined to be greater than the maximum thickness control value or less than the minimum thickness control value, the corresponding three-dimensional discrete point is determined to be a mis-extracted free noise point, distal bone tissue, or abnormal anatomical boundary point. The coordinate indices of these points that are determined to be abnormal are completely deleted from the basic data structure that records the three-dimensional spatial coordinates. All discrete point coordinates that have not been deleted and have been retained and fully meet the physiological range tolerance constraints are extracted to generate a high-purity boundary point set after noise cleaning.

[0113] For example, the thickness measurement value at the 500th point is The upper limit of the physiological thickness range is ,Should The scattered points were clearly caused by the algorithm mistaking the distant mandibular canal bone wall for the bag boundary. After comparison, the point was accurately removed, ensuring the anatomical purity of the boundary point set.

[0114] S53. Extract the gradient vectors in the gradient direction vector sequence that have the same three-dimensional spatial coordinates as the cleaned boundary point set as normal constraints, perform three-dimensional Poisson surface reconstruction calculation on the cleaned boundary point set, and generate a bag-type layered mesh.

[0115] Using a mature algorithm for Poisson surface reconstruction, coordinate matching is performed in a pre-generated gradient direction vector sequence based on the three-dimensional spatial coordinates of each cleaned and retained point in the boundary point set to find the three-dimensional unit vector corresponding to the physical location. The matched unit vector is then directly assigned to the discrete coordinate point as a spatial geometric normal vector constraint condition that controls the surface orientation.

[0116] The Poisson surface reconstruction algorithm utilizes these normal constraint point sets, which not only have position coordinates but also have a clear orientation direction, to construct and solve a large set of partial differential Laplace equations in three-dimensional space. It mathematically integrates the discrete point cloud and normal vector field into a continuous implicit geometric indicator function. The isosurface where the value of this implicit indicator function is equal to zero in space is the reconstructed continuous three-dimensional surface.

[0117] By extracting standard triangular polygonal mesh surfaces from implicit isosurfaces, a layered pocket mesh that can completely and watertightly encapsulate the tooth core and has a highly smooth and continuous surface is ultimately generated.

[0118] In some embodiments, a three-dimensional reconstructed tooth model containing pocket layering features is generated by stitching together the spatial coordinates of the pocket layered mesh and the high-density tooth core, including:

[0119] S61. Extract the voxel space coordinates of the high-density tooth core and generate the internal entity coordinate set.

[0120] The algorithm iterates through the three-dimensional data storage matrix representing the dense core of the tooth, which contains the hard, dense internal tissue. It then rapidly reads and aggregates the three-dimensional spatial coordinates of all discrete voxels marked by the algorithm as belonging to the hard tissue, generating a set of internal entity coordinates containing a large dataset of points. This coordinate set geometrically reflects the dense crown and root hard solid structures within the output three-dimensional reconstruction model.

[0121] S62. Map the 3D vertex coordinates of the bag-like layered mesh to the 3D Cartesian coordinate system containing the coordinate set of the internal entities, generating a hierarchical nested coordinate matrix that includes the correspondence between the internal and external spaces.

[0122] Since the coordinates of the mesh model generated by the point cloud reconstruction algorithm may be in a local computational reference frame, a matrix multiplication spatial transformation involving rotation and translation components is performed to losslessly transform the coordinates of all surface vertices of the layered mesh into a global absolute 3D Cartesian coordinate system completely consistent with the coordinate set of the internal entities. After the transformation and alignment are completed, a multi-level associative data structure matrix is ​​generated, namely a hierarchical nested coordinate matrix. This matrix not only records the dense lattice coordinate spatial arrangement of the inner hard entities, but also records the absolute spatial geometric position of the outer soft tissue mesh shell, mathematically establishing the complex topological nested dependency relationship between the interface between the inner hard tissue and the outer soft tissue.

[0123] S63. Perform a three-dimensional spatial Boolean union operation on the hierarchical nested coordinate matrix to synthesize a three-dimensional reconstructed tooth model containing the layered features of the pockets.

[0124] By invoking the Boolean set operation method in 3D digital geometric modeling technology, and based on the absolute volume of 3D space under a unified coordinate system, the outer 3D volume model formed by the layered mesh representing the outer soft tissue is logically geometrically joined with the inner core 3D volume model occupied by the coordinate set of the inner entity representing the inner hard tissue.

[0125] The computer's Boolean operation process automatically calculates and processes the intersection interference, spatial overlap, and region merging between the mesh surface and the internal solid, ensuring that the geometric structures with different organizational properties are seamlessly joined at the physical boundary edges without normal misalignment.

[0126] The final integrated and calculated three-dimensional general format data object is the three-dimensional reconstruction model of the tooth. This reconstruction model has high fidelity and high density of the hard tissue structure of the tooth body as well as the layered details of the soft tissue pockets attached to the outside of the body.

[0127] For example, when the reconstructed model after Boolean merging is imported into surgical navigation software or 3D printer slicing software, what will be identified is a complete anatomical digital twin complex with a high-density filling inside (corresponding to teeth) and an outer layer of a translucent shell with a specified thickness (corresponding to pouch soft tissue). This allows doctors to intuitively and accurately avoid accidental spatial interference and mechanical damage to key nerves and bone tissue when designing implant guides or tooth extraction bone removal paths.

[0128] In some embodiments, such as Figure 4 As shown, this application provides a cone-shaped bundle for the oral and maxillofacial region. The image-based 3D tooth reconstruction system includes:

[0129] Voxel extraction module: Used to obtain a three-dimensional region of interest voxel matrix containing the target anatomical structure, perform grayscale separation on the three-dimensional region of interest voxel matrix, and extract the initial screening of bright connected components and local background space.

[0130] Morphological denoising module: used to perform morphological denoising and entity definition on the initially screened bright connected components and local background space to obtain high-density tooth core and environmental reference matrix.

[0131] Density mapping module: used to perform peripheral density attenuation mapping based on the high-density tooth core and environmental reference matrix, and extract the spatial topological probability distribution field and gradient direction vector sequence pointing to the outer edge that characterize the tissue transition boundary.

[0132] Inflection point detection module: used to perform extreme inflection point detection at the junction of soft and hard tissues on the spatial topological probability distribution field and gradient direction vector sequence, to obtain the point cloud of the capsule envelope surface and the physiological thickness range.

[0133] Surface Reconstruction Module: Used to perform tolerance verification and surface reconstruction based on the point cloud of the capsule envelope surface, physiological thickness range and spatial topological probability distribution field, and generate capsule layered mesh.

[0134] Model stitching module: Used to stitch together the spatial coordinates of the pocket layered mesh and the high-density tooth core to generate a 3D reconstructed tooth model containing pocket layered features.

[0135] In some embodiments, this application provides a readable storage medium, comprising: computer program instructions stored in the readable storage medium, wherein the computer program instructions are read and executed by a processor to execute a cone-beam transducer for the oral and maxillofacial region. The steps of a three-dimensional reconstruction method for tooth structure from images.

[0136] It should be noted that, in this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0137] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.

[0138] Although this application 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; and these 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 this application.

Claims

1. A method for three-dimensional reconstruction of teeth from cone-beam computed tomography (CBCT) images of the oral and maxillofacial region, characterized in that, include: Obtain a three-dimensional region of interest voxel matrix containing the target anatomical structure, perform grayscale separation on the three-dimensional region of interest voxel matrix, and extract the initial screening of bright connected components and local background space; Morphological denoising and entity definition are performed on the initially screened bright connected components and local background space to obtain the high-density tooth core and environmental reference matrix; Based on the high-density tooth core and the environmental reference matrix, peripheral density attenuation mapping is performed to extract the spatial topological probability distribution field characterizing the tissue transition boundary and the gradient direction vector sequence pointing to the outer edge. The extreme inflection point detection at the soft-hard tissue interface is performed on the spatial topological probability distribution field and gradient direction vector sequence to obtain the point cloud of the bag envelope surface and the physiological thickness range. Based on the point cloud of the capsule envelope surface, the physiological thickness range and the spatial topological probability distribution field, tolerance verification and surface reconstruction are performed to generate a capsule layered mesh. By stitching together the layered mesh of the capsule and the spatial coordinates of the high-density tooth core, a three-dimensional reconstruction model of the tooth containing the layered features of the capsule is generated.

2. The method based on claim 1, characterized in that, Gray-scale separation is performed on the voxel matrix of the 3D region of interest to extract the initially screened bright connected components and local background space, including: Voxels with gray values ​​greater than the preset hard tissue gray value threshold are extracted from the three-dimensional region of interest voxel matrix to generate a preliminary set of high-brightness voxels. Perform three-dimensional connected component volume calculation on the initial high-brightness voxel set, and extract the three-dimensional connected voxel cluster with the largest volume to define the initial high-brightness connected component; The three-dimensional spatial coordinates occupied by the initially screened highlighted connected components are removed from the three-dimensional region of interest voxel matrix, and the remaining three-dimensional spatial coordinates are used to form the local background space.

3. The method based on claim 1, characterized in that, Morphological denoising and entity definition are performed on the initially screened bright connected components and local background space to obtain a high-density tooth core and an environmental reference matrix, including: A three-dimensional morphological erosion operation with a fixed kernel radius is performed on the initially screened bright connected regions to generate a high-density tooth core; Extract the outermost boundary voxel coordinates of the high-density tooth core to generate a core outer surface coordinate set; Using the coordinate set of the core outer surface as the starting coordinate, unidirectional spatial expansion extraction is performed in the local background space according to a fixed three-dimensional voxel step size, and the extracted three-dimensional voxel set is defined as the environment reference matrix.

4. The method based on claim 1, characterized in that, Based on the high-density tooth core and environmental reference matrix, peripheral density attenuation mapping is performed to extract the spatial topological probability distribution field characterizing the tissue transition boundary and the gradient direction vector sequence pointing to the outer edge, including: Extract the three-dimensional spatial coordinates and grayscale values ​​of each voxel in the environmental reference matrix, calculate the minimum Euclidean distance from the coordinates of each voxel to the outermost boundary of the high-density tooth core, and generate a spatial distance matrix. Extract the first-order gray-level difference of each voxel in the environmental reference matrix along the X, Y, and Z coordinate axes, and combine the first-order gray-level difference according to the three-dimensional spatial coordinates to generate a three-dimensional gray-level gradient field. Based on the spatial distance matrix and the three-dimensional gray-level gradient field, nonlinear feature fusion and probability decay calculation are performed to generate an initial topological probability field and an initial direction vector array. Extract each three-dimensional spatial vector from the initial direction vector array, divide the three-dimensional spatial vector by its own Euclidean modulus, and generate a gradient direction vector sequence. Extract the three-dimensional spatial coordinates of voxels with a gray value of zero in the environment reference matrix, extract the initial probabilities with the same three-dimensional spatial coordinates in the initial topological probability field and assign them a value of zero, and combine all the assigned probabilities to generate a spatial topological probability distribution field.

5. The method based on claim 4, characterized in that, Based on the spatial distance matrix and the three-dimensional grayscale gradient field, nonlinear feature fusion and probability decay calculations are performed to generate an initial topological probability field and an initial direction vector array, including: Extract the three-dimensional first-order difference vector of each voxel in the three-dimensional gray-scale gradient field to obtain the initial direction vector array; Calculate the Euclidean magnitude of the three-dimensional first-order difference vector of the i-th voxel in the initial direction vector array to obtain the gradient magnitude. Extract the first element from the spatial distance matrix. Minimum Euclidean distance of voxels ; Based on gradient magnitude and minimum Euclidean distance Calculate the first Fusion eigenvalues ​​of individual elements , ; According to the gradient magnitude and fusion eigenvalues Calculate the first initial probability of individual elements , The initial topological probability field is generated by combining the initial probabilities of all voxels.

6. The method based on claim 1, characterized in that, The extreme inflection point detection at the soft-hard tissue interface is performed on the spatial topological probability distribution field and gradient direction vector sequence to obtain the point cloud of the capsule envelope surface and the physiological thickness range, including: Extract the probabilities and three-dimensional spatial coordinates of all voxels in the spatial topological probability distribution field, and combine them to generate a global topological dataset; Based on the global topology dataset, gradient direction vector sequence, and high-density tooth core, spatial minimum detection and thickness power scaling mapping are performed to generate an initial thickness sequence and an initial pocket point cloud. Extract the discrete three-dimensional spatial coordinates of the initial bag point cloud and perform Delaunay triangulation calculation to generate a spatial triangular mesh. Extract the three-dimensional vertex coordinate set of all triangles in the spatial triangular mesh to obtain the bag envelope surface point cloud. All thickness mapping values ​​in the initial thickness sequence are extracted, sorted in descending order, and the maximum thickness value at the first position and the minimum thickness value at the last position are extracted and combined to form the physiological thickness range.

7. The method based on claim 6, characterized in that, Based on the global topology dataset, gradient direction vector sequence, and high-density tooth core, spatial minimum detection and thickness power scaling mapping are performed to generate an initial thickness sequence and an initial pocket point cloud, including: Calculate the minimum Euclidean distance from each three-dimensional spatial coordinate in the global topology dataset to the outermost boundary of the high-density tooth core to obtain the basic physical thickness. Along the direction of the gradient direction vector sequence, extract the probabilities of three adjacent voxels in the global topological dataset and calculate the sum of the absolute values ​​of the differences between each pair of voxels to obtain the probability fluctuation. The three-dimensional spatial coordinates of the accumulated sum when it reaches a local minimum are used as the initial pocket point cloud; Extract the probability from the global topology dataset that has the same three-dimensional spatial coordinates as the initial pocket point cloud. According to the probability fluctuation amount Sum of probabilities Calculate the first Power scaling factor at each spatial index location , ; Based on the basic physical thickness and power scaling factor Calculate the first Thickness mapping value at each spatial index position , The initial thickness sequence is generated by combining all thickness mapping values.

8. The method based on claim 1, characterized in that, Based on the point cloud of the capsule envelope surface, the physiological thickness range, and the spatial topological probability distribution field, tolerance verification and surface reconstruction are performed to generate a layered mesh of the capsule, including: Extract the three-dimensional spatial coordinates of each discrete point in the point cloud of the capsule envelope surface, calculate the minimum Euclidean distance from the three-dimensional spatial coordinates to the three-dimensional spatial coordinates of the outermost boundary voxel of the high-density tooth core, and generate a measured thickness feature sequence with physical length dimensions. The thickness values ​​in the measured thickness feature sequence are compared with the upper and lower limits of the physiological thickness range. Discrete points whose thickness values ​​exceed the physiological thickness range are removed to generate a cleaned set of boundary points. The gradient vectors with the same three-dimensional spatial coordinates as the cleaned boundary point set are extracted from the gradient direction vector sequence and used as normal constraints. Three-dimensional Poisson surface reconstruction calculations are then performed on the cleaned boundary point set to generate a bag-shaped layered mesh.

9. The method based on claim 1, characterized in that, By stitching together the spatial coordinates of the pocket-layered mesh and the high-density tooth core, a three-dimensional reconstructed tooth model containing pocket-layered features is generated, including: Extract the voxel space coordinates of the high-density tooth core to generate an internal entity coordinate set; The three-dimensional vertex coordinates of the layered mesh of the bag are mapped to the three-dimensional Cartesian coordinate system where the internal entity coordinate set is located, generating a hierarchical nested coordinate matrix that contains the correspondence between the inner and outer spaces; Perform a three-dimensional spatial Boolean union operation on the nested coordinate matrix to synthesize a three-dimensional reconstructed tooth model containing pocket layer features.

10. A three-dimensional reconstruction system for teeth from cone-beam computed tomography (CBCT) images of the oral and maxillofacial region, characterized in that, include: Voxel extraction module: used to obtain a three-dimensional region of interest voxel matrix containing the target anatomical structure, perform grayscale separation on the three-dimensional region of interest voxel matrix, and extract the initial screening of bright connected components and local background space; Morphological denoising module: used to perform morphological denoising and entity definition on the initially screened bright connected components and local background space to obtain high-density tooth core and environmental reference matrix; Density mapping module: used to perform peripheral density attenuation mapping based on the high-density tooth core and environmental reference matrix, and extract the spatial topological probability distribution field characterizing the tissue transition boundary and the gradient direction vector sequence pointing to the outer edge; Inflection point detection module: used to perform extreme inflection point detection at the soft and hard tissue interface on the spatial topological probability distribution field and gradient direction vector sequence, to obtain the point cloud of the capsule envelope surface and the physiological thickness range; Surface reconstruction module: used to perform tolerance verification and surface reconstruction based on the point cloud of the capsule envelope surface, physiological thickness range and spatial topological probability distribution field, and generate capsule layered mesh; Model stitching module: used to stitch together the spatial coordinates of the pocket layered mesh and the high-density tooth core to generate a three-dimensional tooth reconstruction model containing pocket layered features.

Images