An artificial intelligence assisted child pre-crown fitting method

Abstract

The application discloses an artificial intelligence auxiliary-based child pre-crown fitting method and relates to the technical field of child oral cavity repair, which comprises the following steps: collecting local oral cavity scanning three-dimensional data of a sick child and performing intelligent regular processing, automatically segmenting and completing the contour of caries deciduous teeth of the regularized local oral cavity scanning three-dimensional data of the sick child, and generating a complete prepared tooth three-dimensional model; calling a pre-set unified specification pre-crown database, adopting a three-dimensional multi-index fitting evaluation system, matching the complete prepared tooth three-dimensional model, and screening out a fitting pre-crown; performing comparison operation on the fitting pre-crown obtained through screening and the complete prepared tooth three-dimensional model, calculating and outputting the trimming parameters of each region of the pre-crown; correcting the three-dimensional multi-index fitting evaluation system according to the deciduous tooth development characteristics corresponding to the age of the sick child, and outputting a final pre-crown fitting scheme. The application can greatly improve the first-time fitting success rate of the pre-crown, significantly shorten the diagnosis and treatment time, reduce the repeated trial wearing pain of the sick child and reduce long-term repair complications.

Description

Technical Field

[0001] This invention relates to the field of pediatric oral restoration technology, and more specifically to an artificial intelligence-assisted method for fitting preformed crowns for children. Background Technology

[0002] Preformed crowns are commonly used to restore decayed primary teeth in children. The mainstream clinical fitting methods are divided into two categories: manual fitting and general dental AI digital matching. Both methods have inherent limitations in adapting to pediatric treatment scenarios, as detailed below: (1) Traditional manual fitting relies on manual measurement of tooth size with calipers and multiple trial fittings to select preformed crowns. Children have poor cooperation and limited oral field of vision. Manual measurement has large errors, requiring repeated replacement and grinding of crowns. The treatment is time-consuming, and multiple trial fittings can easily cause children to resist and cry. Moreover, it only relies on two-dimensional size judgment and does not take into account the three-dimensional fitting relationship of cervical margin seal, occlusion, and interdental space. After restoration, long-term complications such as high occlusion point, food impaction, marginal leakage and secondary caries are likely to occur.

[0003] (2) Existing general oral AI systems have obvious shortcomings in adaptability: First, they only support the recognition of teeth based on complete full-mouth scan data. Young children are difficult to cooperate with full-mouth scans. When only local point clouds can be collected, the outline of decayed deciduous teeth cannot be completely segmented. Doctors still need to manually repair the defect boundaries, resulting in low automation. Second, the matching logic only compares dimensions and does not establish multi-dimensional three-dimensional adaptation evaluation criteria, resulting in poor matching results. Third, the matching model uses fixed evaluation weights and does not distinguish the differences in jawbone and deciduous tooth development in children aged 3-10. Children of different ages share the same matching standard, resulting in obvious adaptation deviations. Fourth, it only selects pre-formed crown models and cannot automatically quantify the amount of grinding in each part. It still relies on the doctor's experience to manually adjust the grinding, which cannot simplify clinical operations from the root.

[0004] In summary, existing technologies lack an integrated AI adaptation solution that adapts to children's local oral scan data, integrates multi-dimensional three-dimensional adaptation assessment, can automatically predict the amount of grinding and trimming, and combines age-level correction and matching standards. Overcoming the shortcomings of existing technologies, such as poor measurement accuracy, low diagnostic efficiency, and unsatisfactory adaptation effects, is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides an artificial intelligence-assisted method for fitting preformed crowns for children, which solves the problems existing in the background technology.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An AI-assisted method for fitting pre-formed crowns in children includes the following steps: S1. Collect local three-dimensional scan data of the child's oral cavity and perform intelligent normalization processing. Automatically segment and complete the outline of the decayed deciduous teeth from the normalized local three-dimensional scan data of the child's oral cavity to generate a complete three-dimensional model of the prepared teeth. S2. Call the preset unified specification preformed crown database, adopt the three-dimensional multi-index adaptation evaluation system, match the complete prepared tooth three-dimensional model, and screen out the suitable preformed crowns; S3. Compare and calculate the trimming parameters of each area of ​​the preformed crown with the selected adaptive preformed crown and the complete prepared tooth three-dimensional model. S4. Based on the developmental characteristics of deciduous teeth corresponding to the child's age, modify the three-dimensional multi-index fitting assessment system, and output the final pre-formed crown fitting plan based on the modified three-dimensional multi-index fitting assessment system.

[0007] Optionally, in S1, the three-dimensional data of the child's oral cavity scan are intelligently normalized, specifically including the following steps: A dynamic graph convolutional network is used for feature learning and topology reconstruction. Based on oral anatomical features, tooth position classification is performed to distinguish the region of deciduous teeth to be restored, the region of adjacent teeth, and the gingival background region. Coordinate alignment correction is performed based on the standard deciduous tooth anatomical coordinate system to unify spatial posture and coordinate reference. The data were filtered, denoised, and density homogenized to remove scanning noise and invalid background data, resulting in standardized deciduous tooth scanning data.

[0008] Optionally, in S1, the specific method for generating the complete 3D model of the prepared tooth is as follows: A two-layer hierarchical tooth contour segmentation network was used to perform coarse segmentation on the standardized deciduous tooth scanning data to delineate the deciduous tooth region to be restored; For the area of ​​the primary tooth to be restored, a fine contour extraction was performed to separate the primary tooth body from the adjacent tooth area and the gingival background area; Identify the carious and missing areas within the area of ​​the deciduous tooth to be restored, call the preset standard deciduous tooth anatomical template, and perform point interpolation processing on the carious and missing areas by filling branches through the defect boundary to complete the complete outline of the deciduous tooth; Three-dimensional mesh reconstruction is performed based on the complete outline of the deciduous tooth to generate a complete three-dimensional model of the prepared tooth.

[0009] Optionally, in S2, a pre-defined database of uniform specifications and pre-formed crowns is invoked, specifically as follows: Original 3D data of preformed crowns of different brands and models are collected in advance. Size normalization conversion is performed on various types of preformed crowns, and a multi-brand preformed crown size mapping relationship is established to build a unified specification preformed crown database. Using the dimensional parameters of the complete prepared tooth 3D model as search criteria, the database of preformed crowns with uniform specifications is traversed to retrieve all preformed crown 3D templates that match the size range.

[0010] Optionally, in S2, the specific screening method for adapting pre-formed crowns is as follows: A three-dimensional multi-index adaptation evaluation system was established, which includes three constraint indicators: cervical margin occlusion gap, occlusal collision amount, and adjacent tooth interference amount. The retrieved 3D templates of each preformed crown were spatially aligned with the complete 3D model of the prepared tooth, and the values ​​of cervical margin occlusion gap, occlusal collision, and adjacent tooth interference were calculated for each preformed crown. The individual fit score is calculated based on the weight of each constraint indicator, and the overall matching score of the corresponding pre-formed crown is obtained by integrating all individual fit scores. All preformed crowns were sorted from high to low according to their comprehensive matching score. The top three preformed crowns were selected as the matching preformed crowns, and a 3D superimposed comparison cloud map of the complete prepared tooth 3D model and each matching preformed crown was generated.

[0011] Optionally, in S3, the selected pre-formed crowns are compared and calculated with the complete prepared tooth 3D model. Specifically: Retrieve the matching preformed crown 3D template, and use the cervical margin baseline of the complete prepared tooth 3D model as the alignment constraint to perform rigid global spatial registration between the matching preformed crown 3D template and the complete prepared tooth 3D model to unify their spatial coordinate references; The registered 3D template of the pre-formed crown is defined as the target entity, and the complete 3D model of the prepared tooth is defined as the subtracted entity. Perform a 3D mesh Boolean difference operation on the target entity and the subtracted entity to remove non-overlapping spatial regions and extract the spatial interference regions that overlap between the two.

[0012] Optionally, in S3, the trimming parameters for each region of the preformed crown are calculated and output, specifically: The spatial interference region is preprocessed by mesh patch normalization to unify the mesh vertex density and topology. The preprocessed spatial interference region is then input into the pre-trained trimming prediction branch network. The trimming amount prediction branch network divides the buccal and lingual side, mesial and distal side, and neck edge into three-dimensional mesh boundary anatomical features, and calculates the corresponding grinding thickness value for each side. Establish a one-to-one mapping relationship between the grinding thickness values ​​of each partition and the grid vertex coordinates of the spatial interference area, match differentiated identifier color values ​​for different partitions, and generate visual trimming marks with partition identifiers; The grinding thickness values ​​of each zone, the coordinates of the corresponding grid vertices of each zone, and the visual trimming marks are output synchronously and used as trimming parameters for each area of ​​the preformed crown.

[0013] Optionally, in S4, the specific method for obtaining the final pre-formed crown adapter solution is as follows: Retrieve the preset age-deciduous tooth development characteristics comparison table and match the corresponding deciduous tooth development stage parameters according to the child's age; Based on the parameters of the primary tooth development stage, the calculation thresholds and weight coefficients of the three constraint indicators of cervical margin occlusion gap, occlusal collision amount, and adjacent tooth interference amount in the three-dimensional multi-index adaptation assessment system were adjusted to complete the correction of the assessment system; Input the complete three-dimensional model of the prepared tooth into the revised three-dimensional multi-index fitting evaluation system, re-execute the matching scoring calculation, and update the fitting preformed crown model; If the compatible preformed crown model is updated, re-execute S3 to generate trimming parameters for each area of ​​the preformed crown; The updated preformed crown model is integrated with the trimming parameters of each area of ​​the regenerated preformed crown to generate the final preformed crown adaptation solution.

[0014] As can be seen from the above technical solution, compared with the prior art, the present invention discloses an artificial intelligence-assisted method for fitting preformed crowns for children, which has the following beneficial effects: This invention first utilizes local oral cavity scanning 3D data to automatically reconstruct a 3D model of primary teeth. Then, it employs a multi-size preformed crown database and a 3D multi-index fitting evaluation system to complete initial model selection. Next, it uses 3D Boolean operations combined with an AI branch network to precisely quantify the grinding and trimming parameters for each area. Finally, it outputs a fitting plan based on the child's primary tooth development characteristics and matching criteria. The entire process does not rely on complete full-mouth scanning data, significantly reducing the cooperation threshold for young children. The layered and decoupled matching architecture can quickly output candidate crowns for dentists' reference and can also finely calibrate the fitting standards based on age and physiological characteristics. Simultaneously, it automatically outputs visual trimming marks and grinding values, reducing errors caused by manual measurement, repeated fittings, and experience-based adjustments. This effectively shortens treatment time, reduces the probability of long-term restorative complications such as occlusal high points and marginal leakage, and improves the success rate of first-time fitting of preformed crowns for children's primary teeth. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 The flowchart illustrates the artificial intelligence-assisted preformed crown fitting method for children provided by this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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.

[0018] This invention discloses an artificial intelligence-assisted method for fitting preformed crowns in children, such as... Figure 1 As shown, it includes the following steps: S1. Collect local three-dimensional scan data of the child's oral cavity and perform intelligent normalization processing. Automatically segment and complete the outline of the decayed deciduous teeth from the normalized local three-dimensional scan data of the child's oral cavity to generate a complete three-dimensional model of the prepared teeth. S2. Call the preset unified specification preformed crown database, adopt the three-dimensional multi-index adaptation evaluation system, match the complete prepared tooth three-dimensional model, and screen out the suitable preformed crowns; S3. Compare and calculate the trimming parameters of each area of ​​the preformed crown with the selected adaptive preformed crown and the complete prepared tooth three-dimensional model. S4. Based on the developmental characteristics of deciduous teeth corresponding to the child's age, modify the three-dimensional multi-index fitting assessment system, and output the final pre-formed crown fitting plan based on the modified three-dimensional multi-index fitting assessment system.

[0019] Reference Figure 1 The process described herein uses 3D local oral scan data of children as input. It utilizes a multi-layer AI network to automatically model decayed deciduous teeth. A standardized cross-brand preformed crown database and a 3D multi-index quantitative evaluation mechanism are used for initial selection of preformed crowns. Then, 3D mesh geometric calculations and an AI prediction network automatically output precise trimming parameters for each area. Finally, the system dynamically corrects and matches the evaluation criteria based on the physiological characteristics of deciduous tooth development corresponding to the child's age, integrating them into a complete clinical adaptation plan. This technical solution minimizes manual intervention throughout the process and takes into account the differences in deciduous tooth development among children of different ages. The following detailed explanations of the implementation logic, design advantages, and specific computational details of this solution are provided step-by-step.

[0020] First, existing pediatric oral scans can only acquire local point clouds. The raw scan data is mixed with noise from the gums, adjacent teeth, and saliva, resulting in uneven point cloud density and random scanning posture shifts. Conventional processing methods cannot automatically distinguish between teeth and soft tissue, requiring manual selection of target deciduous teeth, leading to low automation. Furthermore, the lack of a unified spatial reference causes coordinate misalignment in subsequent segmentation and matching operations, significantly reducing model reconstruction accuracy. To address the technical shortcomings of cluttered local scan data, inconsistent spatial postures, and the inability to automatically distinguish between teeth and background, S1 in this embodiment includes a corresponding normalization processing flow. Specifically, intelligent normalization processing is performed on the 3D local scan data of the child's oral cavity, including the following steps: A dynamic graph convolutional network is used for feature learning and topology reconstruction. Based on oral anatomical features, tooth position classification is performed to distinguish the region of deciduous teeth to be restored, the region of adjacent teeth, and the gingival background region. Coordinate alignment correction is performed based on the standard deciduous tooth anatomical coordinate system to unify spatial posture and coordinate reference. The data were filtered, denoised, and density homogenized to remove scanning noise and invalid background data, resulting in standardized deciduous tooth scanning data.

[0021] Regarding the dynamic graph convolutional network, this embodiment uses the DGCNN dynamic graph convolutional network as the backbone for point cloud feature extraction. This network contains four EdgeConv feature extraction layers. Each layer dynamically constructs a K-nearest neighbor topology graph in the feature space, updates the adjacency relationship between points in real time, extracts local anatomical geometric features through a multilayer perceptron, aggregates neighborhood features using max pooling, and outputs semantic segmentation labels for three categories: tooth body, gingiva, and adjacent teeth by relying on a hybrid loss function. It can autonomously identify the boundaries of deciduous tooth crowns and gingival mucosa without the need for manual labeling of region boundaries.

[0022] The standard deciduous tooth anatomical coordinate system uses the long axis of the deciduous tooth as the Z-axis, the mesiodistal direction as the X-axis, and the buccal-lingual direction as the Y-axis, with the midpoint of the cervical margin as the origin. All local scan point clouds are uniformly mapped to this coordinate system, which can eliminate posture deviations caused by scanning angle and head rotation. Denoising processing uses statistical filtering to remove isolated outliers, and density homogenization is achieved by resampling the farthest point to unify the vertex density of the point cloud, ensuring the consistency of subsequent network input data.

[0023] Based on the above design, this embodiment can automatically complete the classification, coordinate correction and data cleaning of local oral scan data without the need for doctors to manually remove background noise and adjust the scanning posture. The unified anatomical coordinate system provides a unified calculation benchmark for subsequent deciduous tooth contour segmentation and preformed crown matching, reducing coordinate errors in subsequent modeling and matching calculations from the source and greatly improving the integrity of carious deciduous tooth reconstruction.

[0024] In children, decayed primary teeth often exhibit tooth structure defects and localized point cloud gaps. General segmentation networks can only segment complete tooth structures, failing to automatically fill in defective areas. This necessitates manual contour repair by the dentist, a time-consuming process whose accuracy depends heavily on experience. Furthermore, conventional single-level segmentation easily leads to adhesion between adjacent teeth and the gingiva, resulting in blurred segmentation boundaries and an inability to generate a closed, complete primary tooth mesh model. To achieve automatic filling of defective areas in decayed primary teeth and precise segmentation of tooth structure boundaries, S1 in this embodiment further incorporates a two-layer, hierarchical tooth contour segmentation and defect interpolation completion process. Specifically, the generation of a complete 3D model of the prepared tooth includes the following steps: A two-layer hierarchical tooth contour segmentation network was used to perform coarse segmentation on the standardized deciduous tooth scanning data to delineate the deciduous tooth region to be restored; For the area of ​​the primary tooth to be restored, a fine contour extraction was performed to separate the primary tooth body from the adjacent tooth area and the gingival background area; Identify the carious and missing areas within the area of ​​the deciduous tooth to be restored, call the preset standard deciduous tooth anatomical template, and perform point interpolation processing on the carious and missing areas by filling branches through the defect boundary to complete the complete outline of the deciduous tooth; Three-dimensional mesh reconstruction is performed based on the complete outline of the deciduous tooth to generate a complete three-dimensional model of the prepared tooth.

[0025] The dual-layer hierarchical tooth contour segmentation network is divided into a coarse segmentation branch and a fine boundary segmentation branch: the coarse segmentation branch completes the global region division and quickly locks the approximate range of a single deciduous tooth to be restored; the fine segmentation branch introduces a boundary-aware attention module to enhance the feature extraction of the tooth and gingiva and the junction area of ​​adjacent teeth, and eliminate the problem of contour adhesion.

[0026] The defect boundary completion branch employs a cubic spline point interpolation algorithm. Using the anatomical curvature of the standard deciduous tooth anatomical template as a priori constraint, it interpolates and fills in missing vertices along the normal direction of the defect boundary, automatically closing the contour gaps formed by caries. The standard deciduous tooth anatomical template is a standardized three-dimensional mesh of complete deciduous teeth of different sizes. The standard buccal, lingual, mesial, distal, and cervical margin anatomical morphologies of each tooth position are pre-entered. The interpolation completion process strictly conforms to the physiological and anatomical morphology of deciduous teeth, avoiding any deformed contours that do not conform to oral physiology.

[0027] The 3D mesh reconstruction uses a triangular patch uniform meshing algorithm to generate a closed, hole-free STL 3D mesh model from the completed continuous contour.

[0028] In the above design, this embodiment achieves precise separation of the tooth body through two-level layered segmentation of coarse and fine thickness. It automatically repairs carious defects by interpolation based on the standard deciduous tooth anatomical template. The entire process does not require the doctor to manually repair the defect boundary. It can generate a complete and physiologically anatomically consistent three-dimensional model of the prepared tooth for carious deciduous teeth that are only scanned locally, which greatly reduces the amount of manual intervention and improves the anatomical fit accuracy of the deciduous tooth reconstruction model.

[0029] Secondly, there are numerous brands of pre-formed crowns on the market, and the size standards are not uniform. The three-dimensional shapes of the same model from different manufacturers vary significantly. Traditional matching databases only include pre-formed crowns from a single brand, making it impossible to search for compatible models across brands. Physicians need to manually switch databases for comparison, which is cumbersome. Furthermore, database searches rely solely on single diameter size filtering, resulting in a broad search range and low matching efficiency. To address the technical shortcomings of incompatible multi-brand pre-formed crown data and low search matching efficiency, S2 of this embodiment sets up a multi-brand normalized database construction and size search retrieval process. Specifically, calling a preset unified specification pre-formed crown database includes the following steps: Original 3D data of preformed crowns of different brands and models are collected in advance. Size normalization conversion is performed on various types of preformed crowns, and a multi-brand preformed crown size mapping relationship is established to build a unified specification preformed crown database. Using the dimensional parameters of the complete prepared tooth 3D model as search criteria, the database of preformed crowns with uniform specifications is traversed to retrieve all preformed crown 3D templates that match the size range.

[0030] The dimensional normalization conversion uses the cervical margin circumference of deciduous teeth, mesiodistal diameter of the crown, and buccal-lingual diameter as unified reference parameters. It maps the dimensional parameters of each brand to a unified measurement standard, establishes a cross-brand dimensional comparison table, and realizes the correspondence between the dimensions of different brand models. The database stores the triangular mesh 3D template, reference dimensional parameters, cervical margin surface parameters, and occlusal surface morphology parameters of each preformed crown. The KD-Tree spatial index structure is used to accelerate the dimensional range retrieval. During the retrieval, the 3D dimensions of the cervical margin, mesiodistal diameter, and buccal-lingual diameter of the prepared tooth are extracted. A dimensional tolerance range of ±0.8mm is set, and all preformed crown templates falling within the tolerance range are quickly filtered out, while models with excessive dimensional deviations are filtered out to reduce the amount of subsequent multi-index matching calculations.

[0031] Based on the above design, this embodiment can achieve standardized and unified storage of pre-formed crown data from multiple brands, and quickly narrow down the candidate range by relying on three-dimensional size range retrieval. It eliminates the need for physicians to manually switch between multiple databases to compare models, improves the retrieval speed of pre-formed crown candidate templates, and eliminates the retrieval omission problem caused by inconsistent size standards among different brands.

[0032] Current digital matching relies solely on two-dimensional size comparisons, neglecting three key three-dimensional clinical constraints: cervical margin seal, occlusal high point, and interference with adjacent teeth. This leads to preformed crowns prone to problems such as cervical margin leakage, occlusal interference, and food impaction on adjacent surfaces. Furthermore, the lack of a quantitative comprehensive scoring mechanism, relying solely on rough size selection, fails to provide physicians with an intuitive reference for fitting priority. To address the technical problems of one-dimensional matching evaluation being inadequate and lacking three-dimensional quantitative standards for clinical fitting, S2 in this embodiment establishes a three-dimensional multi-index weighted scoring and screening process. Specifically, the screening process for fitting preformed crowns includes: A three-dimensional multi-index adaptation evaluation system was established, which includes three constraint indicators: cervical margin occlusion gap, occlusal collision amount, and adjacent tooth interference amount. The retrieved 3D templates of each preformed crown were spatially aligned with the complete 3D model of the prepared tooth, and the values ​​of cervical margin occlusion gap, occlusal collision, and adjacent tooth interference were calculated for each preformed crown. The individual fit score is calculated based on the weight of each constraint indicator, and the overall matching score of the corresponding pre-formed crown is obtained by integrating all individual fit scores. All preformed crowns were sorted from high to low according to their comprehensive matching score. The top three preformed crowns were selected as the matching preformed crowns, and a 3D superimposed comparison cloud map of the complete prepared tooth 3D model and each matching preformed crown was generated.

[0033] The three-dimensional multi-index fitting evaluation system incorporates standardized calculation logic: the cervical margin occlusion gap is calculated point-by-point along the cervical margin ring of the prepared tooth, showing the vertical distance between the inner wall of the preformed crown and the tooth surface, with an ideal gap range of 0.05~0.2mm; the occlusal collision volume is the volume of the overlapping interference area of ​​the occlusal surfaces, with a larger value indicating a more severe occlusal high point; the adjacent tooth interference volume is the overlapping area between the mesial and distal walls of the preformed crown and the adjacent tooth mesh, directly reflecting the risk of food impaction. Each of the three indicators is assigned an independent weight: cervical margin occlusion weight 0.5, occlusal collision weight 0.3, and adjacent tooth interference weight 0.2. Individual scores are linearly converted according to the ideal range of the indicators, with a total score of 0~100 points, where a higher score indicates a better fit. The three-dimensional overlay comparison cloud map uses gradient colors to mark gaps and interference areas, with red markings for high interference and large gap risk areas, visually displaying the location of fitting defects.

[0034] In the above design, this embodiment quantifies the fitting effect from three clinical dimensions: cervical margin seal, occlusion, and interproximal gap. It automatically sorts the best candidate preformed crowns through weighted comprehensive scoring, and outputs a visual overlay cloud map to intuitively display fitting defects. This avoids clinical fitting defects caused by single-size matching and provides physicians with a quantitative and visual initial selection reference.

[0035] However, after obtaining the preformed crown template through initial screening, it is impossible to visually distinguish the interference area between the preformed crown and the prepared tooth. Traditionally, relying on manual observation of the 3D model to determine the grinding area results in significant subjective errors and an inability to accurately locate the buccal, lingual, mesial, distal, and cervical margin areas requiring grinding. Furthermore, the lack of a standardized spatial registration process leads to inconsistencies in the coordinate systems of the preformed crown and the prepared tooth, resulting in a large number of redundant and invalid meshes in the extraction of the interference area. To address the technical problems of non-standard spatial alignment between the preformed crown and the prepared tooth and the inability to automatically and accurately extract the interference area, S3 in this embodiment sets up a rigid registration and 3D mesh Boolean difference set operation process. Specifically, the selected suitable preformed crown is compared and calculated with the complete prepared tooth 3D model, as follows: Retrieve the matching preformed crown 3D template, and use the cervical margin baseline of the complete prepared tooth 3D model as the alignment constraint to perform rigid global spatial registration between the matching preformed crown 3D template and the complete prepared tooth 3D model to unify their spatial coordinate references; The registered 3D template of the pre-formed crown is defined as the target entity, and the complete 3D model of the prepared tooth is defined as the subtracted entity. Perform a 3D mesh Boolean difference operation on the target entity and the subtracted entity to remove non-overlapping spatial regions and extract the spatial interference regions that overlap between the two.

[0036] The rigid global spatial registration employs the ICP iterative nearest point algorithm, with the alignment benchmark being the closed curve of the complete cervical margin of the prepared tooth. The iterative convergence error of fixed translation and rotation parameters is less than 0.01 mm, ensuring complete fit between the preformed crown and the anatomical position of the prepared tooth. The 3D mesh Boolean difference operation is based on a triangular facet intersection detection algorithm. It traverses all triangular faces in both meshes, identifies overlapping areas, removes non-overlapping blank meshes, and retains only the set of interfering faces as the spatial interference region. This interference region strictly corresponds to the excess portion of the preformed crown that needs to be ground away, without any irrelevant redundant meshes interfering with subsequent grinding amount prediction.

[0037] In the above design, this embodiment completes standardized rigid registration based on the anatomical benchmark of the neck margin. It automatically extracts all interference mesh regions through Boolean difference set operation, accurately locates the parts of the preformed crown that need to be polished, completely replaces manual visual judgment of the interference region, greatly improves the accuracy of interference region extraction, and provides clean computational input data for subsequent AI prediction of polishing thickness.

[0038] Traditional digital solutions only screen preformed crown models and cannot automatically quantify the grinding thickness of each area. All grinding amounts rely on manual judgment by the physician's clinical experience. Insufficient grinding can lead to occlusal high points and poor marginal fit, while excessive grinding can cause leakage at the neck margin. Furthermore, the lack of visual markers for different zones means physicians have no intuitive numerical reference during grinding, resulting in low grinding efficiency. To address the technical shortcomings of not being able to automatically quantify the grinding amount in each area of ​​the preformed crown and lacking visual grinding guidance, S3 in this embodiment incorporates mesh preprocessing and trimming amount prediction branch networks. Specifically, the trimming parameters for each area of ​​the preformed crown are calculated and output, including the following steps: The spatial interference region is preprocessed by mesh patch normalization to unify the mesh vertex density and topology. The preprocessed spatial interference region is then input into the pre-trained trimming prediction branch network. The trimming amount prediction branch network divides the buccal and lingual side, mesial and distal side, and neck edge into three-dimensional mesh boundary anatomical features, and calculates the corresponding grinding thickness value for each side. Establish a one-to-one mapping relationship between the grinding thickness values ​​of each partition and the grid vertex coordinates of the spatial interference area, match differentiated identifier color values ​​for different partitions, and generate visual trimming marks with partition identifiers; The grinding thickness values ​​of each zone, the coordinates of the corresponding grid vertices of each zone, and the visual trimming marks are output synchronously and used as trimming parameters for each area of ​​the preformed crown.

[0039] The preprocessing of mesh normalization involves resampling the mesh in the interference region, standardizing the number of triangular patches per square millimeter, eliminating mesh density differences caused by scanning and Boolean operations, and ensuring uniformity of network input data specifications. The trimming amount prediction branch network adopts a Mesh R-CNN mesh regression architecture. The encoder extracts the three-dimensional geometric features of the interference region, and the multi-branch decoder outputs the regression values ​​of the grinding thickness at the vertices of the buccal and lingual sides, mesial and distal sides, and cervical margin. The network training dataset includes tens of thousands of clinical grinding samples of preformed crowns for pediatric deciduous teeth. The loss function uses vertex coordinate L2 regression loss to ensure that the grinding thickness values ​​closely match actual clinical grinding standards. During zoning, the deciduous teeth are automatically distinguished based on their anatomical orientation: buccal and lingual sides are marked in blue, mesial and distal sides in yellow, and cervical margins in red. Each mesh vertex is bound to a corresponding grinding thickness value, and the visual markers can be directly imported into 3D printing and digital grinding equipment.

[0040] In the above design, this embodiment relies on AI branch network to automatically and accurately predict the grinding thickness by region, and simultaneously output numerical parameters and visual markers of the regions. It eliminates the need for physicians to rely on experience to subjectively judge the amount of grinding, standardizes and quantifies the trimming needs of each region, greatly shortens the clinical grinding operation time, and reduces the adaptation deviation caused by manual grinding.

[0041] Furthermore, existing AI matching systems use a fixed set of indicator weights and thresholds, failing to differentiate between the developmental differences in deciduous teeth and jawbones in children aged 3-10. Younger children have unresorbed deciduous tooth roots and ample space for jawbone development, so stringent matching standards may exclude suitable models. Older children experience significant desorption of deciduous tooth roots, making lenient standards prone to occlusal abnormalities, resulting in significant mismatches with a uniform standard. Additionally, updates to the initial screening model do not synchronously update the grinding and trimming parameters, leading to insufficient system completeness. To address the technical issues of using a single matching standard for children of different ages and the inability to synchronously iterate trimming parameters after model updates, S4 in this embodiment incorporates age development parameter correction, iterative recalculation of trimming parameters, and integration of the complete solution process. Specifically, the final pre-formed crown fitting solution is obtained through the following steps: Retrieve the preset age-deciduous tooth development characteristics comparison table and match the corresponding deciduous tooth development stage parameters according to the child's age; Based on the parameters of the primary tooth development stage, the calculation thresholds and weight coefficients of the three constraint indicators of cervical margin occlusion gap, occlusal collision amount, and adjacent tooth interference amount in the three-dimensional multi-index adaptation assessment system were adjusted to complete the correction of the assessment system; Input the complete three-dimensional model of the prepared tooth into the revised three-dimensional multi-index fitting evaluation system, re-execute the matching scoring calculation, and update the fitting preformed crown model; If the compatible preformed crown model is updated, re-execute S3 to generate trimming parameters for each area of ​​the preformed crown; The updated preformed crown model is integrated with the trimming parameters of each area of ​​the regenerated preformed crown to generate the final preformed crown adaptation solution.

[0042] The age-deciduous tooth development characteristics comparison table divides deciduous tooth development into four stages: 3-4 years, 5-6 years, 7-8 years, and 9-10 years. Each stage has independent indicator thresholds and weights: for 3-4 year olds, the upper limit of cervical margin gap is relaxed, and the weight of occlusal collision penalty is reduced; for 9-10 year olds in the pre-tooth-changing stage, occlusion is tightened, and the interproximal interference threshold is adjusted to match the physiological characteristics of older children with root resorption and mature jawbone development. When the revised evaluation system re-matches to obtain a new preformed crown model, the system fully utilizes the entire process of S3 registration, Boolean operations, and trimming prediction to recalculate exclusive grinding and trimming parameters for the new model, ensuring a one-to-one correspondence between the model and the trimming parameters. The final adaptation scheme integrates the optimal preformed crown model, grinding thickness of each zone, visual trimming marks, and a 3D overlay comparison cloud map, forming a complete digital solution that can be directly used for clinical diagnosis and treatment.

[0043] In the above design, this embodiment dynamically corrects the matching evaluation criteria according to the age-stratified development of children's deciduous teeth, adapts to the physiological development differences of children of different ages, and greatly reduces the matching deviation of children of different ages; at the same time, it establishes a closed-loop logic for synchronous recalculation and adjustment of model updates to ensure that the final output scheme model and polishing parameters are completely matched, providing doctors with an integrated and complete digital restoration scheme that is adapted to the age and physiological characteristics of the children.

[0044] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0045] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for fitting pre-formed crowns for children based on artificial intelligence assistance, characterized in that, Includes the following steps: S1. Collect local three-dimensional scan data of the child's oral cavity and perform intelligent normalization processing. Automatically segment and complete the outline of the decayed deciduous teeth from the normalized local three-dimensional scan data of the child's oral cavity to generate a complete three-dimensional model of the prepared teeth. S2. Call the preset unified specification preformed crown database, adopt the three-dimensional multi-index adaptation evaluation system, match the complete prepared tooth three-dimensional model, and screen out the suitable preformed crowns; S3. Compare and calculate the trimming parameters of each area of ​​the preformed crown with the selected matching preformed crown and the complete prepared tooth three-dimensional model. S4. Based on the developmental characteristics of deciduous teeth corresponding to the child's age, modify the three-dimensional multi-index fitting assessment system, and output the final pre-formed crown fitting plan based on the modified three-dimensional multi-index fitting assessment system.

2. The method for fitting preformed crowns for children based on artificial intelligence assistance according to claim 1, characterized in that, In S1, the three-dimensional data of the child's oral cavity scan are intelligently processed, specifically including the following steps: A dynamic graph convolutional network is used for feature learning and topology reconstruction. Based on oral anatomical features, tooth position classification is performed to distinguish the region of deciduous teeth to be restored, the region of adjacent teeth, and the gingival background region. Coordinate alignment correction is performed based on the standard deciduous tooth anatomical coordinate system to unify spatial posture and coordinate reference. The data were filtered, denoised, and density homogenized to remove scanning noise and invalid background data, resulting in standardized deciduous tooth scanning data.

3. The method for fitting preformed crowns for children based on artificial intelligence assistance according to claim 2, characterized in that, In S1, the specific method for generating a complete 3D model of the prepared tooth is as follows: A two-layer hierarchical tooth contour segmentation network was used to perform coarse segmentation on the standardized deciduous tooth scanning data to delineate the deciduous tooth region to be restored; For the area of ​​the primary tooth to be restored, a fine contour extraction was performed to separate the primary tooth body from the adjacent tooth area and the gingival background area; Identify the carious and missing areas within the area of ​​the deciduous tooth to be restored, call the preset standard deciduous tooth anatomical template, and perform point interpolation processing on the carious and missing areas by filling branches through the defect boundary to complete the complete outline of the deciduous tooth; Three-dimensional mesh reconstruction is performed based on the complete outline of the deciduous tooth to generate a complete three-dimensional model of the prepared tooth.

4. The method for fitting preformed crowns for children based on artificial intelligence assistance according to claim 1, characterized in that, In S2, the preset unified specification pre-formed crown database is called, specifically as follows: Original 3D data of preformed crowns of different brands and models are collected in advance. Size normalization conversion is performed on various types of preformed crowns, and a multi-brand preformed crown size mapping relationship is established to build a unified specification preformed crown database. Using the dimensional parameters of the complete prepared tooth 3D model as search criteria, the database of preformed crowns with uniform specifications is traversed to retrieve all preformed crown 3D templates that match the size range.

5. The method for fitting preformed crowns for children based on artificial intelligence assistance according to claim 4, characterized in that, In S2, the specific screening method for adapting pre-formed crowns is as follows: A three-dimensional multi-index adaptation evaluation system was established, which includes three constraint indicators: cervical margin occlusion gap, occlusal collision amount, and adjacent tooth interference amount. The retrieved 3D templates of each preformed crown were spatially aligned with the complete 3D model of the prepared tooth, and the values ​​of cervical margin occlusion gap, occlusal collision, and adjacent tooth interference were calculated for each preformed crown. The individual fit score is calculated based on the weight of each constraint indicator, and the overall matching score of the corresponding pre-formed crown is obtained by integrating all individual fit scores. All preformed crowns were sorted from high to low according to their comprehensive matching score. The top three preformed crowns were selected as the matching preformed crowns, and a 3D superimposed comparison cloud map of the complete prepared tooth 3D model and each matching preformed crown was generated.

6. The method for fitting preformed crowns for children based on artificial intelligence assistance according to claim 1, characterized in that, In S3, the selected pre-formed crowns are compared and calculated with the complete prepared tooth 3D model. Specifically: Retrieve the matching preformed crown 3D template, and use the cervical margin baseline of the complete prepared tooth 3D model as the alignment constraint to perform rigid global spatial registration between the matching preformed crown 3D template and the complete prepared tooth 3D model to unify their spatial coordinate references; The registered 3D template of the pre-formed crown is defined as the target entity, and the complete 3D model of the prepared tooth is defined as the subtracted entity. Perform a 3D mesh Boolean difference operation on the target entity and the subtracted entity to remove non-overlapping spatial regions and extract the spatial interference regions that overlap between the two.

7. A method for fitting preformed crowns for children based on artificial intelligence assistance according to claim 6, characterized in that, In S3, the trimming parameters for each region of the preformed crown are calculated and output, specifically: The spatial interference region is preprocessed by mesh patch normalization to unify the mesh vertex density and topology. The preprocessed spatial interference region is then input into the pre-trained trimming prediction branch network. The trimming amount prediction branch network divides the buccal and lingual sides, mesial and distal sides, and neck edge sides according to the anatomical features of the three-dimensional mesh boundary, and fits and calculates the corresponding grinding thickness value for each side. Establish a one-to-one mapping relationship between the grinding thickness values ​​of each partition and the grid vertex coordinates of the spatial interference area, match differentiated identifier color values ​​for different partitions, and generate visual trimming marks with partition identifiers; The grinding thickness values ​​of each zone, the coordinates of the corresponding grid vertices of each zone, and the visual trimming marks are output synchronously and used as trimming parameters for each area of ​​the preformed crown.

8. The method for fitting preformed crowns for children based on artificial intelligence assistance according to claim 1, characterized in that, In S4, the specific method for obtaining the final pre-formed crown adapter solution is as follows: Retrieve the preset age-deciduous tooth development characteristics comparison table and match the corresponding deciduous tooth development stage parameters according to the child's age; Based on the parameters of the primary tooth development stage, the calculation thresholds and weight coefficients of the three constraint indicators of cervical margin occlusion gap, occlusal collision amount, and adjacent tooth interference amount in the three-dimensional multi-index adaptation assessment system were adjusted to complete the correction of the assessment system; Input the complete three-dimensional model of the prepared tooth into the revised three-dimensional multi-index fitting evaluation system, re-execute the matching scoring calculation, and update the fitting preformed crown model; If the compatible preformed crown model is updated, re-execute S3 to generate trimming parameters for each area of ​​the preformed crown; The updated preformed crown model is integrated with the trimming parameters of each area of ​​the regenerated preformed crown to generate the final preformed crown adaptation solution.

Images