An artificial intelligence driven bionic reconstruction system and method for jaw bone defects

Abstract

The application discloses an artificial intelligence driven jaw bone defect bionic reconstruction system and method. A three-dimensional segmentation model is used to segment the facial CBCT image of a healthy person to construct a jaw bone dentition database; the segmentation model is used to segment the healthy side and the diseased side of the jaw bone of a patient, and intelligent matching is performed in the database based on the segmented healthy side jaw bone data as a reference to obtain the most similar target jaw bone; then the opposite jaw bone of the target jaw bone is taken as a to-be-applied template, three-dimensional integration is performed with the diseased side defect of the patient to generate a basic virtual jaw bone support; then a planting scheme is generated according to the matched healthy dentition data to obtain a segmented virtual support; trabecular bone structure filling is performed in the segmented virtual support to output a bionic reconstruction customized body; finally, the overhanging structure of the trabecular bone is detected according to the complex topological structure of the trabecular bone, and corresponding printing support is generated. The above can improve the bionic adaptation degree of the reconstructed defect jaw bone, and support entity production such as 3D printing.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-driven biomimetic reconstruction system and method for jawbone defects. Background Technology

[0002] The maxillofacial system is a complex composed of bone, muscles, nerves, and soft tissues. The jawbone, as the core supporting structure, not only forms the bony framework of the face, maintaining the spatial position and individualized aesthetic morphology of soft tissues, but also, through its biomechanical coordination with the temporomandibular joint, masticatory muscles, and dentition, enables physiological functions such as chewing, swallowing, and speech. However, factors such as tumors, trauma, and infection often lead to continuous or penetrating defects in the jawbone, resulting in problems such as masticatory dysfunction, stomatognathic system disorders, and facial deformities. Currently, autologous bone grafting is the gold standard for treating large-area jawbone defects. However, establishing a second surgical site directly increases the surgical trauma and prolongs the operation time. Furthermore, the volume and anatomical morphology of the donor bone tissue are difficult to fully match the complex curved structure of the jawbone, resulting in significant limitations in the accuracy of occlusal reconstruction, the coordination of maxillofacial tissue functions, and aesthetic restoration effects, failing to fully restore the physiological adaptation between the original jawbone and surrounding tissues. Effective repair of large-area jawbone defects has become an important and urgent clinical need in the field of oral and maxillofacial surgery. For reconstruction of large-area jawbone defects, ideally, data matching can be performed in a healthy population jawbone database based on digital technology and artificial intelligence methods to obtain the most similar maxilla / mandible as a reference for reconstruction of the defect area.

[0003] Artificial intelligence-based jawbone defect reconstruction design requires searching for similar matching objects in a database using the defective jawbone, and then using the jawbone shape of the matching object to design the shape of the defect. However, at present, artificial intelligence-based jawbone defect reconstruction design is usually only applicable to localized defects with relatively minor structural damage. For complex, large-area jawbone defects that span multiple regions and dimensions, it is difficult to simultaneously meet the core clinical needs of anatomical morphology restoration, occlusal repair and reconstruction, and physiological function adaptation. Moreover, the designed bionic structure is difficult to integrate with physical manufacturing technologies such as 3D printing. Its intelligent bionic reconstruction still faces many technical challenges: 1) Difficulty in matching defective jawbones: In cases of tumor invasion, severe trauma, etc., the anatomical structure of the defective jawbone is often severely damaged, resulting in the lack of matching benchmarks and unstable reconstruction morphology. Furthermore, existing three-dimensional... Matching algorithms suffer from computational complexity and time-consuming nature, making them difficult to effectively adapt to clinical needs; 2) Reconstructed structures are difficult to biomimetic: The reconstructed jaw medullary cavity lacks biomimetic simulation of the trabecular pore structure and biomechanical properties of the natural jawbone. At present, the technology can use microscopically scanned trabecular structures to fill the medullary cavity, but the filling contents are not compatible with the complex shape of the reconstructed jawbone; 3) Biomimetic structures are difficult to connect with printing equipment: Advanced manufacturing processes such as 3D printing have layer-by-layer accumulation characteristics. Problems such as poor continuity of the porous structure of biomimetic trabeculae and lack of support for suspended structures will make it difficult for advanced manufacturing equipment to produce the complete structure. Summary of the Invention

[0004] To address the aforementioned issues, this invention proposes an AI-driven biomimetic reconstruction system and method for jawbone defects. Innovative solutions are designed for each step of the entire biomimetic reconstruction process for large-area jawbone defects: 1) A sequential matching algorithm for jawbone is constructed, reducing jawbone image data to two-dimensional horizontal slices. Based on these horizontal slices, a comprehensive matching similarity is calculated, significantly reducing the computational effort and time required for three-dimensional matching, achieving predictable matching results; 2) A biomimetic trabecular bone expansion and generation algorithm is constructed. Random central trabecular bone cubes are first generated within the reconstructed jawbone medullary cavity. Guided by the structural features of the six faces of the cube, the network expands in an orderly manner, ensuring the continuity of the overall trabecular bone network and its porous structure, simulating the mechanical conduction characteristics of natural bone; 3) An adaptive trabecular bone support algorithm is constructed. By retrieving the suspended structures of the trabecular bone, a top-down reverse backfilling mechanism is used to generate vertically projected support columns for the determined suspended structures until they contact the underlying solid, thereby constructing a stable printed support system.

[0005] To achieve the above objectives, embodiments of the present invention provide an artificial intelligence-driven biomimetic reconstruction system and method for jawbone defects, including: a database, a jawbone segmentation module, a jawbone matching module, a three-dimensional integration module, an implantation simulation module, a biomimetic trabecular bone generation module, and a trabecular bone adaptive support module. The healthy population jawbone and dentition database uses segmentation algorithms to segment CBCT images of healthy individuals' faces, establishing a structured database containing the three-dimensional morphology of the dentition and jawbone. This provides objective anatomical templates for defect restoration, addressing the issue of insufficient design basis. The jawbone segmentation module processes input patients with jawbone defects using segmentation algorithms, obtaining their defective jawbone and the healthy side jawbone. The jawbone matching module employs a designed sequential matching algorithm to convert three-dimensional matching into two-dimensional images. Based on horizontal slices, it calculates the comprehensive similarity between the healthy side jawbone data and each candidate jawbone data. It then selects the target jawbone data with the highest similarity by considering the comprehensive similarity, age difference, gender, and the number of horizontal slices. The contralateral jawbone of the target jawbone is obtained as the jawbone template to be applied, achieving effective pairing of the defective jawbone with the healthy jawbone template. The jawbone three-dimensional integration module uses point cloud fusion and other algorithms to supplement the jawbone data to be applied. The entire segment is spliced ​​with the existing part of the defective jawbone on the affected side, and the boundary topology is restored. The contact area after spatial fusion is locally smoothed and areas such as the alveolar socket are filled to obtain a basic virtual jawbone framework that conforms to the defect space. The implant simulation module uses the jawbone template to be used as a reference to simultaneously reconstruct the shape of the patient's defective jawbone and plan the implant scheme. It optimizes the jawbone reconstruction design according to the principle of implant bone volume and outputs a segmental virtual framework that integrates form and function. The bionic trabecular bone generation module uses the designed bionic trabecular bone extension generation algorithm to simulate the microstructure of natural bone using a generative model and generate a bionic trabecular bone network that conforms to the direction of mechanical transmission inside the framework. Finally, the trabecular bone adaptive support module uses the designed trabecular bone adaptive support algorithm to automatically search for the suspended areas of the trabecular bone and generate vertical projection support columns, thereby constructing a stable printed support system and forming a customized bionic jawbone reconstruction file.

[0006] Therefore, embodiments of this application provide an artificial intelligence-driven biomimetic reconstruction system and method for jawbone defects, which improves the adaptability of the generated biomimetic structure.

[0007] In one aspect, this application provides an artificial intelligence-driven biomimetic reconstruction system for jawbone defects.

[0008] This application is achieved through the following technical solution: an artificial intelligence-driven bionic reconstruction system for jawbone defects, comprising: A database is used to collect facial CBCT images of healthy individuals. A pre-trained 3D segmentation model is used to segment the facial CBCT images to obtain maxillary bone data, mandibular bone data, and dentition data, thereby establishing a structured jawbone and dentition database. The jaw segmentation module is used to acquire facial CBCT images of patients with jaw defects, and to perform instance segmentation of the facial CBCT images using a pre-trained 3D segmentation model to obtain data of the affected side jawbone and the healthy side jawbone. The jawbone matching module is used to search for all candidate jawbone data on the same side in the jawbone dentition database based on the healthy side jawbone data, reduce the healthy side jawbone data and candidate jawbone data to horizontal slices, calculate the comprehensive similarity between the healthy side jawbone data and each candidate jawbone data based on the horizontal slices, and calculate the final matching score by weighting the comprehensive similarity, age difference, gender and difference in the number of horizontal slices, and filter out the target jawbone data with the highest similarity based on the final matching score; The three-dimensional integration module is used to search for the contralateral jawbone data of the target jawbone data in the jawbone dentition database, and use it as the jawbone data to be applied. The jawbone data to be applied is then integrated with the data of the affected side of the jaw in three dimensions to generate a basic virtual jawbone framework that conforms to the defect space. The implant simulation module is used to generate an implantation plan based on the dental arch data of the complete segment of the jawbone data to be applied, and to optimize the shape of the basic virtual jawbone framework based on the implantation plan to obtain a segmental virtual framework that can be reconstructed with the defective jawbone. A biomimetic trabecular bone generation module is used to fill the segmental virtual scaffold with trabecular bone structure using a biomimetic trabecular bone extension generation algorithm. The trabecular adaptive support module uses a trabecular adaptive support algorithm to retrieve the suspended structures of the trabecular bone, generate vertically projected support columns for the determined suspended structures, and so on until they contact the underlying solid, thereby constructing a stable printed support system and generating the final biomimetic jawbone reconstruction custom body.

[0009] In a preferred embodiment of this application, the jawbone matching module is further configured to: Both the healthy side jawbone data and the candidate jawbone data were divided into horizontal slice sequences according to the CBCT voxel interval; The horizontal slice sequences of the healthy jawbone data and the horizontal slice sequences of the candidate jawbone data are aligned in position. For each pair of horizontally aligned slices, a similarity score is calculated. Calculate the average similarity score of all horizontal slice pairs to obtain the comprehensive similarity between the healthy side jawbone data and the candidate jawbone data; The final matching score between the healthy jawbone data and the candidate jawbone data is calculated by weighting the overall similarity, age difference, gender, and difference in the number of horizontal slices.

[0010] In a preferred embodiment of this application, the jawbone matching module is further configured to: For each pair of horizontally aligned slices, the peak signal-to-noise ratio, structural similarity, and cosine similarity are calculated respectively. The peak signal-to-noise ratio, structural similarity, and cosine similarity are normalized, and the similarity score of the horizontal slice pairs is calculated.

[0011] In a preferred embodiment of this application, the three-dimensional integration module is further configured to: Using the contralateral jawbone of the target jawbone matched in the database as the jawbone data to be applied, the nearest point is calculated iteratively with the data of the affected jawbone, and the existing part of the defective jawbone on the affected side is aligned to the corresponding area of ​​the jawbone data to be applied. A point cloud fusion algorithm is used to stitch together the completed segment of the jawbone data to be applied with the existing part of the defective jawbone on the affected side, and to perform topological repair on the triangular facets of the junction area so that the boundary of the jawbone to be applied and the boundary of the jawbone on the affected side are spatially fused. Local surface smoothing is applied to the contact area after spatial fusion; The alveolar sockets in the complete jawbone segment were filled and the curved surfaces were smoothed.

[0012] In a preferred example of this application, the planting simulation module can be further configured to specifically be used for: Based on the crown direction and root length of the dentition corresponding to the jawbone data to be applied, the three-dimensional placement position and specifications of the implant are generated. Combined with the bone volume requirements of the implant, the local bone volume of the reconstructed jawbone is adjusted.

[0013] In a preferred example of this application, the biomimetic trabecular bone extension module can be further configured to specifically be used for: Using a pre-trained generative neural network model, an initial central trabecular cube is generated. Based on the structural features of the six faces of the central trabecular cube, adjacent trabecular structures are generated in all directions to form a spatially continuous biomimetic trabecular spatial model with a porous mesh structure. The segmental virtual scaffold is uniformly offset inward to define the medullary cavity region. Under the spatial constraints of the medullary cavity region, the biomimetic trabecular bone spatial model and the segmental virtual scaffold are subjected to Boolean intersection calculation, and the medullary cavity region is filled with trabecular bone structures.

[0014] In a preferred example of this application, the trabecular adaptive support module is specifically used for: Based on interlayer differential detection and historical stack data detection of the suspended structure of the trabecular bone, the suspended area to be supported is obtained; The minimum Euclidean distance between the upper solid region and the suspended region to be supported is calculated to obtain the candidate support mask; The candidate support mask is calculated based on the support strength index and material cost function to obtain the suspended support region; The suspended support area is reverse-projected from top to bottom to generate a vertical projection support column, and a custom-made bionic jawbone reconstruction body is generated based on the vertical projection support column.

[0015] Secondly, this application provides an artificial intelligence-driven biomimetic reconstruction method for jawbone defects.

[0016] This application is achieved through the following technical solution: An artificial intelligence-driven biomimetic reconstruction method for jawbone defects, executed using the artificial intelligence-driven biomimetic reconstruction system for jawbone defects described in the first aspect above, includes: Collect facial CBCT images of healthy individuals, and use a pre-trained 3D segmentation model to perform instance segmentation on the facial CBCT images to obtain maxillary bone data, mandibular bone data, and dentition data, and establish a structured jawbone and dentition database. Acquire facial CBCT images of patients with jaw defects, and use a pre-trained 3D segmentation model to perform instance segmentation on the facial CBCT images to obtain data of the affected side jawbone and the healthy side jawbone. Based on the healthy side jawbone data, all candidate jawbone data on the same side are searched in the jawbone dentition database. The healthy side jawbone data and candidate jawbone data are reduced to horizontal slices. The comprehensive similarity between the healthy side jawbone data and each candidate jawbone data is calculated based on the horizontal slices. The final matching score is calculated by weighting the comprehensive similarity, age difference, gender, and difference in the number of horizontal slices. The target jawbone data with the highest similarity is selected based on the final matching score. Search the target jawbone data in the jawbone dentition database for the contralateral jawbone data, and use it as the jawbone data to be applied. Integrate the jawbone data to be applied with the affected jawbone data in three dimensions to generate a basic virtual jawbone framework that conforms to the defect space. Based on the dental data of the segment to be filled using the jawbone data, an implantation plan is generated. Based on the implantation plan, the shape of the basic virtual jawbone framework is optimized to obtain a segmental virtual framework that can be reconstructed with the defective jawbone. A biomimetic trabecular bone expansion and generation algorithm is used to fill the segmental virtual scaffold with trabecular bone structure. Using a trabecular adaptive support algorithm, the suspended areas of the trabecular bone are retrieved and vertically printed supports are generated to form a customized biomimetic jawbone reconstruction body.

[0017] In a preferred example of this application, the healthy side jawbone data and candidate jawbone data can be further configured to reduce the dimensionality of the data to horizontal slices, calculate the comprehensive similarity between the healthy side jawbone data and each candidate jawbone data based on the horizontal slices, and calculate the final matching score by weighting the comprehensive similarity, age difference, gender, and difference in the number of horizontal slices, including: Both the healthy side jawbone data and the candidate jawbone data were divided into horizontal slice sequences according to the CBCT voxel interval; The horizontal slice sequences of the healthy jawbone data and the horizontal slice sequences of the candidate jawbone data are aligned in position. For each pair of horizontally aligned slices, a similarity score is calculated. Calculate the average similarity score of all horizontal slice pairs to obtain the comprehensive similarity between the healthy side jawbone data and the candidate jawbone data; The final matching score between the healthy jawbone data and the candidate jawbone data is calculated by weighting the overall similarity, age difference, gender, and difference in the number of horizontal slices.

[0018] Thirdly, this application provides a computer-readable storage medium.

[0019] This application is achieved through the following technical solution: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of any of the above-described artificial intelligence-driven biomimetic reconstruction methods for jaw defects.

[0020] Fourthly, this application provides a computer device.

[0021] This application is achieved through the following technical solution: A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above-described artificial intelligence-driven biomimetic reconstruction methods for jaw defects.

[0022] In summary, compared with the prior art, the beneficial effects of the technical solution provided in this application include at least the following: By segmenting facial CBCT images of healthy individuals, a structured jawbone database containing the three-dimensional morphology of the dentition and jawbone is established, providing an objective anatomical template for defect restoration and solving the problem of insufficient design basis. Sequence slice comparison technology is employed to convert three-dimensional matching into two-dimensional image comparison for rapid comparison, significantly improving computational efficiency and matching accuracy. The most similar target jawbone is matched using the patient's healthy side jawbone, and the contralateral jawbone of the target jawbone is obtained as the jawbone template to be applied, achieving effective pairing of the defective jawbone with the healthy template. Based on the jawbone template to be applied, the three-dimensional integrated reconstruction of the patient's defective jawbone shape and implant planning are carried out simultaneously. The jawbone reconstruction design is optimized through bone volume principles, ultimately outputting a virtual jawbone framework that integrates morphology and function. Generative models are used to simulate the microstructure of natural bone, generating a biomimetic trabecular network within the framework that conforms to the direction of mechanical transmission. Using a trabecular adaptive support algorithm, suspended areas of the trabecular bone are retrieved and vertical printing supports are generated, forming a final biomimetic jawbone reconstruction custom body that can be manufactured using advanced technologies such as 3D printing. Attached Figure Description

[0023] Figure 1 A schematic diagram of the structure of an artificial intelligence-driven biomimetic reconstruction system for jaw defects provided in an embodiment of this application; Figure 2 A schematic diagram illustrating the construction of a jawbone dentition database according to an embodiment of this application; Figure 3 A schematic diagram illustrating the jawbone matching module provided in one embodiment of this application for retrieving the most similar jawbone from a jawbone dentition database; Figure 4 A schematic diagram illustrating the three-dimensional integration module and implantation simulation module provided in an embodiment of this application, which optimize bone mass based on matching results through three-dimensional integration and implantation simulation. Figure 5 A schematic diagram illustrating the biomimetic trabecular bone generation module provided in an embodiment of this application for the expansion generation of biomimetic trabecular bone and the fabrication of a digitally customized biomimetic jawbone. Figure 6 This is a flowchart illustrating a trabecular adaptive support module provided in an embodiment of this application. Figure 7 This is a flowchart illustrating an artificial intelligence-driven biomimetic reconstruction method for jawbone defects, provided as an embodiment of this application. Detailed Implementation

[0024] This specific embodiment is merely an explanation of this application and is not intended to limit it. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0025] 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.

[0026] Furthermore, the term "and / or" in this application is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this application, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0027] In this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor are there any restrictions on quantity or execution order.

[0028] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0029] Before providing a further detailed description of the embodiments of this application, the nouns and terms used in the embodiments of this application are explained, and the nouns and terms used in the embodiments of this application shall be interpreted as follows: The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0030] Please see Figure 1 This application provides an exemplary embodiment of an artificial intelligence-driven biomimetic reconstruction system for jawbone defects, the system specifically comprising: The database is used to collect facial CBCT images of healthy individuals. A pre-trained 3D segmentation model is used to segment the facial CBCT images to obtain maxillary bone data, mandibular bone data, and dentition data, thus establishing a structured jawbone and dentition database.

[0031] Specifically, to construct a database of jaw and dentition for healthy individuals, it is necessary to collect facial CBCT images of healthy individuals according to inclusion and exclusion criteria, ensuring that the data covers samples of different ages, sexes, and jaw anatomical features, and excluding individuals with severe inflammation, jaw deformities, tumors, or a history of trauma. Data annotation software is used to annotate the maxilla, mandible, and teeth of a certain number of healthy individuals' facial CBCT images. The facial CBCT images of these healthy individuals, along with their corresponding annotation information, are input into a 3D instance segmentation network. The 3D instance segmentation network is trained, and its parameters are optimized until a preset segmentation performance or a preset number of training iterations is achieved, ensuring the segmentation results match the true anatomical structure and preserving clean jaw and dentition data. At this point, the model parameters are fixed, and it is used as the 3D segmentation model. In this application, the 3D instance segmentation network can be a 3D nnU-Net. Segmentation performance can be evaluated using metrics such as Intersection over Union (IoU) and Dice coefficient.

[0032] See Figure 2 As shown, after obtaining the pre-trained 3D segmentation model, the facial CBCT images of healthy individuals are input into the 3D segmentation model to obtain the corresponding 3D structural data of the maxilla, mandible, and dentition. The data are then classified and stored according to standardized fields, such as gender, age, dentition status, and jaw volume, to construct a structured database of the jaw and dentition of healthy individuals.

[0033] The jawbone segmentation module is used to acquire facial CBCT images of patients with jawbone defects. It uses a pre-trained 3D segmentation model to segment the facial CBCT images into instances, obtaining data of the affected side jawbone and the healthy side jawbone.

[0034] For patients with large-area jawbone defects, a pre-trained 3D segmentation model is used to segment facial CBCT images into instances, obtaining data of the affected side and the healthy side of the jawbone.

[0035] The jawbone matching module is used to search for all candidate jawbone data on the same side in the jawbone dentition database based on the healthy side jawbone data, reduce the healthy side jawbone data and candidate jawbone data to horizontal slices, calculate the comprehensive similarity between the healthy side jawbone data and each candidate jawbone data based on the horizontal slices, and calculate the final matching score by weighting the comprehensive similarity, age difference, gender and difference in the number of horizontal slices, and filter out the target jawbone data with the highest similarity based on the final matching score.

[0036] Considering the significant loss of data in the affected side of the jawbone, making it difficult to use as a matching subject, this application uses the data of the healthy side of the jawbone opposite the affected side as the matching subject for database retrieval and matching. For example, when there is a large area of ​​defect in the mandible, the defective mandible and the intact maxilla are segmented, with the intact maxilla being the object to be matched.

[0037] The three-dimensional integration module is used to search for the contralateral jawbone data of the target jawbone data in the jawbone dentition database, and use it as the jawbone data to be applied. The jawbone data to be applied is then integrated with the jawbone data of the affected side in three dimensions to generate a basic virtual jawbone framework that conforms to the defect space.

[0038] The implant simulation module is used to generate an implantation plan based on the dental arch data of the complete segment of the jawbone data to be applied, and to optimize the shape of the basic virtual jawbone framework based on the implantation plan to obtain a segmental virtual framework that can be reconstructed with the defective jawbone.

[0039] A biomimetic trabecular bone generation module is used to fill the segmental virtual scaffold with trabecular bone structure using a biomimetic trabecular bone extension generation algorithm.

[0040] The trabecular adaptive support module uses a trabecular adaptive support algorithm to retrieve the suspended structures of the trabecular bone, generate vertically projected support columns for the determined suspended structures, and so on until they contact the underlying solid, thereby constructing a stable printed support system and generating the final biomimetic jawbone reconstruction custom body.

[0041] By segmenting facial CBCT images of healthy individuals, a structured jawbone database containing the three-dimensional morphology of the dentition and jawbone is established, providing an objective anatomical template for defect restoration and addressing the problem of insufficient design basis. Sequential slice comparison technology is employed to convert three-dimensional matching into two-dimensional images for rapid comparison. Using the patient's healthy jawbone as a reference, and considering factors such as age, gender, and the number of slices, the most similar target jawbone in the database is determined. The contralateral jawbone of the target jawbone is then obtained as the jawbone template to be used, achieving effective pairing of the defective jawbone with the healthy jawbone template. Based on the jawbone template to be used, the three-dimensional integrated reconstruction of the patient's defective jawbone shape and implant planning are performed simultaneously. The jawbone reconstruction design is optimized based on bone volume principles, ultimately outputting a virtual jawbone framework that integrates morphology and function. Generative models are used to simulate the microstructure of natural bone, generating a biomimetic trabecular network within the framework that conforms to the direction of mechanical conduction. An adaptive trabecular support algorithm is used to retrieve suspended trabecular areas and generate vertical printing supports, forming a final biomimetic jawbone reconstruction custom body that can be manufactured using advanced technologies such as 3D printing.

[0042] See Figure 3 As shown, in some preferred embodiments, the jawbone matching module is specifically used for: Both the healthy side jawbone data and the candidate jawbone data were divided into horizontal slice sequences according to the CBCT voxel interval; Align the horizontal slice sequences of the healthy jawbone data with the horizontal slice sequences of the candidate jawbone data. For each pair of horizontally aligned slices, a similarity score is calculated. Calculate the average similarity score of all horizontal slice pairs to obtain the comprehensive similarity between the healthy jawbone data and the candidate jawbone data; The final matching score between the healthy jawbone data and the candidate jawbone data is calculated by weighting the overall similarity, age difference, gender, and difference in the number of horizontal slices.

[0043] It should be noted that the maxillary bone defect is used as an example in this application, but this is not a limitation on the technical solution of this application. This application is also applicable to the case of mandibular bone defect.

[0044] For example, the healthy side jawbone data The CBCT voxel intervals were used to divide the sequence into a group of horizontal slices. Retrieve a candidate jawbone data from the jawbone dentition database according to its number. Similarly, it was divided into a group of horizontal slice sequences according to the CBCT voxel intervals. ; Horizontal slice sequence and horizontal slice sequence Perform position alignment; calculate horizontal slices separately. and horizontal slices Similarity score Calculate horizontal slices and horizontal slices Similarity score Calculate horizontal slices and horizontal slices Similarity score Similarly, similarity scores are calculated between all aligned horizontal slices. Then calculate the average similarity score of all horizontal slices. As data of the healthy side of the jaw. With candidate jawbone data The overall similarity between them.

[0045] By reducing the dimensionality of three-dimensional jawbone data and using a sequential matching algorithm to perform matching at the two-dimensional level, the amount of computation can be greatly reduced and the matching speed can be improved.

[0046] For each pair of horizontally aligned slices, a similarity score is calculated, including: For each pair of horizontally aligned slices, the peak signal-to-noise ratio, structural similarity, and cosine similarity are calculated respectively. Peak signal-to-noise ratio, structural similarity, and cosine similarity are normalized, and the similarity score of horizontal slice pairs is calculated based on the normalized indices.

[0047] To calculate horizontal slices and horizontal slices Similarity score Taking a specific example, we will calculate the horizontal slice. and horizontal slices The peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and cosine similarity are used to evaluate horizontal slices. and horizontal slices Similarity in jaw shape, features, and structure.

[0048] The formula for calculating the peak signal-to-noise ratio can be expressed as: , , Where MAX represents the maximum gray value of a CBCT slice, and MSE represents the mean square of the difference in gray values ​​between corresponding pixels in two images.

[0049] Structural similarity is used to assess the similarity of anatomical structures, comprehensively evaluating three dimensions: brightness, contrast, and structure. The numerical range is 0 to 1, with values ​​closer to 1 indicating greater structural similarity. The formula for calculating structural similarity can be expressed as: , in, and These represent the mean gray values ​​of image X and Y, respectively. The mean gray value reflects the brightness and corresponds to the average density of bone tissue. and These represent the grayscale standard deviations of images X and Y, respectively. The grayscale standard deviation reflects the contrast and corresponds to the grayscale differences in the jawbone. This represents the covariance of images X and Y, with consistent covariance structure.

[0050] Cosine similarity focuses on key anatomical features of the jawbone, such as the contour of the alveolar ridge and the angle of the mandibular angle, supplementing the limitations of structural similarity in matching local features. Cosine similarity ranges from 0 to 1, with values ​​closer to 1 indicating greater similarity of feature vectors. The calculation first extracts the anatomical feature vectors from the slices, then calculates the cosine angle between the vectors. , in, and The feature vector of the slice to be matched / / database slice.

[0051] Since the three indicators have different value ranges, the three indicators are first normalized to the 0~1 range, and the weighted sum of the normalized indicators is used to obtain the comprehensive similarity S.

[0052] The final matching score is obtained by combining age difference, gender, difference in the number of horizontal slices, and overall similarity of the jawbone using a weighted summation method. The weights of these factors are set according to clinical requirements; for example, if external similarity is emphasized, the overall similarity is set to 0.7, and the weights for age difference, gender, and difference in the number of slices are each 0.1. Further, the allowable age range T and the allowable difference in the number of slices Y are defined. The age difference (0-T) is assigned a value of 1-0, and the difference in the number of slices is assigned a value of D. A gender value G is assigned, for example, 1 for the same sex and 0.5 for the opposite sex. All factors must first be converted to normalized values ​​in the 0-1 range for calculation, as shown in the following formula: The final match score can be expressed as: , Where S, A, G, and D represent the overall similarity value, age assignment, gender assignment, and slice number difference assignment, respectively. , , and These are the weights corresponding to the four factors.

[0053] In some preferred embodiments, the three-dimensional integration module is specifically used for: Using the contralateral jawbone of the target jawbone matched in the database as the jawbone data to be applied, the nearest point is calculated iteratively with the data of the affected jawbone, and the existing part of the defective jawbone on the affected side is aligned to the corresponding area of ​​the jawbone data to be applied. A point cloud fusion algorithm is used to stitch together the completed segment of the jawbone data to be applied with the existing part of the defective jawbone on the affected side, and to perform topological repair on the triangular facets of the junction area so that the boundary of the jawbone to be applied and the boundary of the jawbone on the affected side are spatially fused. Local surface smoothing is applied to the contact area after spatial fusion; The alveolar sockets in the complete jawbone segment were filled and the curved surfaces were smoothed.

[0054] See Figure 4 As shown, in actual implementation, the jawbone data and dentition data of the healthy individual with the highest similarity in the jawbone dentition database are retrieved and three-dimensionally registered with the existing data of the affected side of the jawbone defect of the patient. Based on the dentition arrangement of the healthy individual, the implantation plan is specified using restoration-oriented intelligent implant simulation technology, including the three-dimensional placement and specifications of the implants. Based on this, the shape of the restored jawbone is automatically fine-tuned, and finally a virtual jawbone framework that takes into account both anatomical shape and occlusal reconstruction needs is constructed.

[0055] The process involves using the matched jawbone data as a baseline, performing iterative nearest-point calculations with the affected side's defective jawbone, and aligning the existing portion of the defective jawbone to the corresponding region of the jawbone to be applied. This alignment can be achieved by adjusting the scaling factor according to the size of both jawbones. After alignment, a point cloud fusion algorithm is used to stitch the completed segment of the jawbone data to be applied with the existing portion of the defective jawbone on the affected side. Topological repair is then performed on the triangular facets at the stitching point to eliminate overlaps and / or gaps, thereby spatially fusing the completed segment of the jawbone data to be applied with the patient's defect boundary.

[0056] B-spline curves or moving least squares method are used to optimize the surface of the contact area to eliminate splicing gaps and sharp edges.

[0057] The alveolar sockets in the complete jawbone segment were filled and the curved surfaces were smoothed.

[0058] In some preferred embodiments, the planting simulation module is specifically used for: Based on the crown direction and root length of the dentition corresponding to the jawbone data to be applied, the three-dimensional placement position and specifications of the implant are generated. Combined with the biomechanical principles of the implant, the local bone volume of the reconstructed jawbone is adjusted.

[0059] In some preferred embodiments, the biomimetic trabecular bone generation module is specifically used for: Using a pre-trained generative neural network model, an initial central trabecular cube is generated. Based on the structural features of the six faces of the central trabecular cube, adjacent trabecular structures are generated in all directions to form a spatially continuous biomimetic trabecular spatial model with a porous mesh structure. The segmental virtual scaffold is uniformly offset inward to define the medullary cavity region. Under the spatial constraints of the medullary cavity region, the biomimetic trabecular bone spatial model and the segmental virtual scaffold are subjected to Boolean intersection calculation, and the medullary cavity region is filled with trabecular bone structures.

[0060] See Figure 5 As shown, in actual implementation, clinical allogeneic bone blocks are first obtained, and Micro-CT scanning imaging is used to cut the trabeculae into standardized cubic units (such as 5mm×5mm×5mm) in batches to construct a trabeculae structure sample library.

[0061] Generative neural networks, such as 3D GAN (Generative Adversarial Network), are constructed using trabecular bone cube imaging as training data to learn the physiological structural characteristics of the trabecular bone sample library, such as porosity, trabecular diameter, and spatial distribution patterns, in order to generate biomimetic trabecular bone structures. Using a trained generative neural network, a random central trabecular bone cube is first generated. Guided by the structural features of the six faces of the cube, the network expands in an orderly manner, ensuring the continuity of the overall trabecular bone network and its porous structure. This simulates the mechanical conduction characteristics of natural bone, ultimately forming a biomimetic trabecular bone spatial model of a certain scale. The generated biomimetic trabecular bone spatial model is then integrated with a virtual scaffold: First, a personalized cortical bone thickness is formed by uniformly offsetting the outer contour of the virtual scaffold inward, constituting the inner boundary of the cortical bone (i.e., the outer boundary of the medullary cavity region). The area between the inner and outer boundaries is defined as the cortical bone layer (preserving the original dense structure), and the area inside the inner boundary is the medullary cavity region (which needs to be filled with trabecular bone). Under the spatial constraints of the medullary cavity region of the virtual scaffold, the biomimetic trabecular bone spatial model and the segmental virtual scaffold are calculated using Boolean intersection, thereby filling the medullary cavity region of the virtual scaffold with trabecular bone structure, ultimately forming a customized biomimetic jawbone reconstruction.

[0062] The consistency between the generated results and natural bone is verified by using trabecular bone structural parameters (such as porosity and trabecular diameter); the uniformity of trabecular bone filling inside the scaffold and its integration with the outer wall of the scaffold are obtained as assessed by professional physicians.

[0063] See Figure 6 As shown, Figure 6 This is a schematic diagram illustrating the process of a trabecular adaptive support module, provided in one embodiment of this application, detecting the suspended area of ​​the trabecular bone and generating a printed support. By printing a support for the trabecular bone structure, the design results of a virtual jawbone scaffold containing biomimetic trabecular bone can be stored in a format compatible with 3D printing or other advanced manufacturing technologies (such as STL), the scaffold material parameters can be labeled (such as elastic modulus matching natural bone), and biomaterials with appropriate biocompatibility and strength can be used to complete the physical fabrication by connecting to clinical 3D printing equipment.

[0064] See Figure 7 As shown, the first exemplary embodiment of this application provides an artificial intelligence-driven biomimetic reconstruction method for jawbone defects, the specific steps of which include: Step S1: Collect facial CBCT images of healthy individuals, use a pre-trained 3D segmentation model to perform instance segmentation on the facial CBCT images to obtain maxillary bone data, mandibular bone data and dentition data, and establish a structured jawbone and dentition database. Specifically, to construct a database of jaw and dentition for healthy individuals, it is necessary to collect facial CBCT images of healthy individuals according to inclusion and exclusion criteria, ensuring that the data covers samples of different ages, sexes, and jaw anatomical features, and excluding individuals with severe inflammation, jaw deformities, tumors, or a history of trauma. Data annotation software is used to annotate the maxilla, mandible, and teeth of a certain number of healthy individuals' facial CBCT images. The facial CBCT images of these healthy individuals, along with their corresponding annotation information, are input into a 3D instance segmentation network. The 3D instance segmentation network is trained, and its parameters are optimized until a preset segmentation performance or a preset number of training iterations is achieved, ensuring the segmentation results match the true anatomical structure and preserving clean jaw and dentition data. At this point, the model parameters are fixed, and it is used as the 3D segmentation model. In this application, the 3D instance segmentation network can be a 3D nnU-Net. Segmentation performance can be evaluated using metrics such as Intersection over Union (IoU) and Dice coefficient.

[0065] After obtaining the pre-trained 3D segmentation model, the facial CBCT images of healthy individuals are input into the 3D segmentation model to obtain the corresponding 3D structural data of the maxilla, mandible, and dentition. The data are then classified and stored according to standardized fields such as gender, age, dentition status, and jaw volume to construct a structured database of the jaw and dentition of healthy individuals.

[0066] Step S2: Obtain facial CBCT images of patients with jaw defects, and use a pre-trained 3D segmentation model to perform instance segmentation on the facial CBCT images to obtain data of the affected side jawbone and the healthy side jawbone.

[0067] For patients with large-area jawbone defects, a pre-trained 3D segmentation model is used to segment facial CBCT images into instances, obtaining data of the affected side and the healthy side of the jawbone.

[0068] Step S3: Using the healthy side jawbone data as the matching basis, search for all candidate jawbone data on the same side in the jawbone dentition database. Reduce the dimension of the healthy side jawbone data and the candidate jawbone data to horizontal slices. Calculate the comprehensive similarity between the healthy side jawbone data and each candidate jawbone data based on the horizontal slices. Combine the comprehensive similarity, age difference, gender, and difference in the number of horizontal slices to calculate the final matching score. Based on the final matching score, select the target jawbone data with the highest similarity.

[0069] Considering the significant loss of data in the affected side of the jawbone, making it difficult to use as a matching subject, this application uses the data of the healthy side of the jawbone opposite the affected side as the matching subject for database retrieval and matching. For example, when there is a large area of ​​defect in the mandible, the defective mandible and the intact maxilla are segmented, with the intact maxilla being the object to be matched.

[0070] Step S4: Locate the contralateral jawbone data of the target jawbone data in the jawbone dentition database as the jawbone data to be used. Integrate the jawbone data to be used with the affected jawbone data in three dimensions to generate a basic virtual jawbone framework that conforms to the defect space.

[0071] Step S5: Based on the data of the dental arch of the segment to be filled in using the jawbone data, generate the implantation plan, optimize the shape of the basic virtual jawbone framework based on the implantation plan, and obtain a segmental virtual framework that can be reconstructed with the defective jawbone.

[0072] Step S6: Use a biomimetic trabecular bone extension generation algorithm to fill the segmental virtual scaffold with trabecular bone structure.

[0073] Step S7: Using the trabecular adaptive support algorithm, the suspended areas of the trabecular bone are retrieved and vertical printing supports are generated to form a customized biomimetic jawbone reconstruction body.

[0074] In some embodiments, the healthy side jawbone data and candidate jawbone data are reduced in dimensionality to horizontal slices, and the comprehensive similarity between the healthy side jawbone data and each candidate jawbone data is calculated based on the horizontal slices, including: Both the healthy side jawbone data and the candidate jawbone data were divided into horizontal slice sequences according to the CBCT voxel interval; The horizontal slice sequences of the healthy side jawbone data and the horizontal slice sequences of the candidate jawbone data are aligned with the midline slice as the starting point. For each pair of horizontally aligned slices, a similarity score is calculated. The average similarity score of all horizontal slice pairs is calculated to obtain the comprehensive similarity between the healthy jawbone data and the candidate jawbone data.

[0075] The final matching score between the healthy jawbone data and the candidate jawbone data is calculated by weighting the overall similarity, age difference, gender, and difference in the number of horizontal slices.

[0076] In some embodiments, the jawbone data to be applied is integrated with the jawbone data of the affected side in three dimensions, including: The most similar target jawbone is matched with the healthy side jawbone of the patient, and the contralateral jawbone of the target jawbone is obtained as the template to be applied. The closest point is calculated iteratively with the data of the affected side jawbone, and the existing part of the defective jawbone on the affected side is aligned to the corresponding area of ​​the jawbone data to be applied. A point cloud fusion algorithm is used to stitch together the completed segment of the jawbone data to be applied with the existing part of the defective jawbone on the affected side, and to perform topological repair on the triangular facets of the junction area so that the boundary between the jawbone to be applied and the jawbone on the affected side can be spatially fused. Local surface smoothing is applied to the contact area after spatial fusion; The alveolar sockets in the complete jawbone segment were filled and the curved surfaces were smoothed.

[0077] In some embodiments, based on the dentition data corresponding to the jawbone data to be applied, an implantation plan is generated, and the shape of the basic virtual jawbone framework is optimized based on the implantation plan to obtain a segmental virtual framework capable of reconstructing the defective jawbone, including: Based on the crown direction and root length of the dentition corresponding to the jawbone data to be applied, the three-dimensional placement position and specifications of the implant are generated. Combined with the biomechanical principles of the implant, the local bone volume of the jawbone is adjusted.

[0078] Using a biomimetic trabecular bone expansion and generation algorithm, trabecular bone structures are filled into a segmental virtual scaffold to generate the final customized biomimetic jawbone reconstruction, including: Using a pre-trained generative neural network model, an initial central trabecular cube is generated. Based on the structural features of the six faces of the central trabecular cube, adjacent trabecular structures are generated in all directions, forming a spatially continuous biomimetic trabecular spatial model with a porous mesh structure. The segmental virtual scaffold contour is uniformly offset inward to determine the medullary cavity region. Under the spatial constraints of the medullary cavity region, the biomimetic trabecular space model and the segmental virtual scaffold are calculated using Boolean intersection, and the trabecular structure is filled into the medullary cavity region.

[0079] In practice, clinical allogeneic bone blocks are first obtained, and Micro-CT scanning imaging is used to cut the trabeculae into standardized cubic units (such as 5mm×5mm×5mm) in batches to build a trabeculae structure sample library.

[0080] Generative neural networks, such as 3D GAN (Generative Adversarial Network), are constructed using trabecular bone cube imaging as training data to learn the physiological structural characteristics of the trabecular bone sample library, such as porosity, trabecular diameter, and spatial distribution patterns, in order to generate biomimetic trabecular bone structures. Using a trained generative neural network, a random central trabecular bone cube is first generated. Guided by the structural features of the six faces of the cube, the network expands in an orderly manner, ensuring the continuity of the overall trabecular bone network and its porous structure. This simulates the mechanical conduction characteristics of natural bone, ultimately forming a biomimetic trabecular bone spatial model of a certain scale. The generated biomimetic trabecular bone spatial model is then integrated with a virtual scaffold: First, a personalized cortical bone thickness is formed by uniformly offsetting the outer contour of the virtual scaffold inward, constituting the inner boundary of the cortical bone (i.e., the outer boundary of the medullary cavity region). The area between the inner and outer boundaries is defined as the cortical bone layer (preserving the original dense structure), and the area inside the inner boundary is the medullary cavity region (which needs to be filled with trabecular bone). Under the spatial constraints of the medullary cavity region of the virtual scaffold, the biomimetic trabecular bone spatial model and the segmental virtual scaffold are calculated using Boolean intersection, thereby filling the medullary cavity region of the virtual scaffold with trabecular bone structure, ultimately forming a customized biomimetic jawbone reconstruction.

[0081] The consistency between the generated results and natural bone is verified by using trabecular bone structural parameters (such as porosity and trabecular diameter); the uniformity of trabecular bone filling inside the scaffold and its integration with the outer wall of the scaffold are obtained as assessed by professional physicians.

[0082] In one feasible implementation, the trabecular adaptive support generation algorithm is designed based on a morphological support generation strategy using sliced ​​image flow, addressing the complex topology of trabecular bone and the interlayer accumulation characteristics during 3D printing. The core of the algorithm lies in a dual-stage cantilever detection mechanism and a reverse backfilling construction strategy: First, cantilever regions are identified by combining interlayer differential detection with multi-layer historical support determination, eliminating noise interference; second, the minimum Euclidean distance between the cantilever region and adjacent entities is evaluated, and sites that can be supported at close range are selected accordingly; finally, a top-down reverse backfilling mechanism is used to generate vertically projected support columns for the identified cantilever structure until they contact the underlying entity, thereby constructing a stable printed support system.

[0083] Specifically, the first step is data preparation: obtaining three-dimensional slice data of the trabecular bone structure to be printed, and binarizing it into a series of two-dimensional bitmap sequences arranged along the Z-axis { I 1 ，I 2 ，...，I N},in I k This represents the slice at layer k. I k ( x, y )=1 represents a skeletal entity. I k ( x, y =0 represents pores or background.

[0084] Support generation algorithm construction: Design an adaptive support algorithm based on morphological image processing. The core steps include: First, suspended regions are detected: a dual-judgment logic combining inter-layer difference and historical stack is adopted.

[0085] This includes: First, differential detection: comparing the current layer. I curr With the previous layer I curr-1 Extract regions that appear only in the current layer and whose corresponding positions in the previous layer are empty; then perform connected component analysis on the candidate regions, remove isolated noise regions with areas smaller than a preset threshold, and retain connected regions that meet the area requirements as differential floating results; Second, rigorous historical verification: A historical stack containing the most recent m layers (e.g., m=5) of slices is established. A pixel is only judged as "true dangling" if it is a solid in the current layer and a non-solid in all corresponding layers of the historical stack. This effectively avoids misjudgments caused by noise in single-layer slices.

[0086] Specifically: Suspended region determination model: Based on a comprehensive evaluation of interlayer variations and historical accumulation, the suspended mask is defined as... F This indicator reflects whether the current layer of the skeletal structure lacks physical support. The calculation formula is as follows: ; ; ; in, I curr This is the binary image of the current layer. I curr-k This is the binary image of the k-th layer preceding the current layer. C(·) represents the connected component area filtering operator, used to remove isolated noise regions with an area smaller than a threshold. m is the number of historical backtracking layers. F final This is the final determined suspended area to be supported.

[0087] Furthermore, support site determination is performed: for the detected suspended island regions, the minimum spatial distance to the upper-layer solid region is evaluated using Euclidean distance transformation, and candidate support masks are generated by combining morphological dilation operations. Within a preset number of dilation iterations, the support strength index and material cost function of each candidate mask are calculated, and the support site and scheme with lower cost are selected under the premise of satisfying the minimum support strength constraint.

[0088] This includes: First, assessing the support strength: using Euclidean distance transformation to evaluate the spatial distance from the suspended point to the nearest entity, thereby quantifying the necessity and connection strength of the support. The upper-level entity set is defined as...S prev For any point within the suspended region p , its arrival S prev Shortest distance D( p The calculation is as follows: ; Support connection strength ( p The calculation is as follows: ; in, Represents Euclidean distance. To prevent the use of tiny constants with a denominator of zero. D( p The smaller the value, the closer the suspended point is to the upper layer of the solid structure. When the distance exceeds the set threshold, a vertical support column will be generated through subsequent reverse backfilling.

[0089] Second, the material cost function: In order to balance the printing success rate and material consumption, a cost function J is constructed for the expanded temporary support mask. B temp Conduct an assessment: in, W vol It is a volumetric weight, reflecting the cost of material consumption; W sparse The sparsity weights reflect the complexity and cost of the printed structure. The specific values ​​of the two weights can be set according to the resolution of the printing equipment, material properties, or experimental experience. N comp This represents the number of connected components. The algorithm selects the mask that satisfies the support strength constraint and has the lowest cost from a finite number of expansion candidates.

[0090] Finally, reverse backfilling is performed to construct the support: a top-down reverse projection is performed on the identified suspended areas and support points. Starting from the current layer, the support shape is copied and backfilled layer by layer downwards, stopping when an existing solid structure is detected at that location, thus forming continuous vertical support columns.

[0091] The design results of the virtual jawbone scaffold containing biomimetic trabeculae are stored in a format compatible with 3D printing or other advanced manufacturing technologies (such as STL), the scaffold material parameters are labeled (such as elastic modulus matching natural bone), and biomaterials with appropriate biocompatibility and strength are used to complete the physical fabrication by connecting to clinical 3D printing equipment.

[0092] The specific limitations of the AI-driven biomimetic reconstruction method for jaw defects provided in this embodiment can be found in the embodiment of an AI-driven biomimetic reconstruction system for jaw defects described above, and will not be repeated here. Each module in the aforementioned AI-driven biomimetic reconstruction system for jaw defects can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0093] This application provides a computer device that may include a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it causes the processor to perform the steps of an artificial intelligence-driven biomimetic reconstruction method for jaw defects, as described in any of the above embodiments.

[0094] The working process, working details, and technical effects of the computer device provided in this embodiment can be found in the embodiment of an artificial intelligence-driven bionic reconstruction method for jaw defects described above, and will not be repeated here.

[0095] This application provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the steps of an artificial intelligence-driven bionic reconstruction method for jaw defects as described in any of the above embodiments. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0096] The working process, working details, and technical effects of the computer-readable storage medium provided in this embodiment can be found in the embodiment of a biomimetic reconstruction method for a defective jawbone described above, and will not be repeated here.

[0097] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0098] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0099] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system described in this application can be divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. An artificial intelligence-driven biomimetic reconstruction system for jawbone defects, characterized in that, The system includes: A database is used to collect facial CBCT images of healthy individuals. A pre-trained 3D segmentation model is used to segment the facial CBCT images to obtain maxillary bone data, mandibular bone data, and dentition data, thereby establishing a structured jawbone and dentition database. The jaw segmentation module is used to acquire facial CBCT images of patients with jaw defects, and to perform instance segmentation of the facial CBCT images using a pre-trained 3D segmentation model to obtain data of the affected side jawbone and the healthy side jawbone. The jawbone matching module is used to search for all candidate jawbone data on the same side in the jawbone dentition database based on the healthy side jawbone data, reduce the healthy side jawbone data and candidate jawbone data to horizontal slices, calculate the comprehensive similarity between the healthy side jawbone data and each candidate jawbone data based on the horizontal slices, and calculate the final matching score by weighting the comprehensive similarity, age difference, gender and difference in the number of horizontal slices, and filter out the target jawbone data with the highest similarity based on the final matching score; The three-dimensional integration module is used to search for the contralateral jawbone data of the target jawbone data in the jawbone dentition database, and use it as the jawbone data to be applied. The jawbone data to be applied is then integrated with the data of the affected side of the jaw in three dimensions to generate a basic virtual jawbone framework that conforms to the defect space. The implant simulation module is used to generate an implantation plan based on the dental arch data of the complete segment of the jawbone data to be applied, and to optimize the shape of the basic virtual jawbone framework based on the implantation plan to obtain a segmental virtual framework that can be reconstructed with the defective jawbone. A biomimetic trabecular bone generation module is used to fill the segmental virtual scaffold with trabecular bone structure using a biomimetic trabecular bone extension generation algorithm. The trabecular adaptive support module is used to retrieve the suspended structure of the trabecular bone using the trabecular adaptive support algorithm, generate vertical projection support columns, and generate a customized biomimetic jawbone reconstruction body based on the vertical projection support columns.

2. The artificial intelligence-driven biomimetic reconstruction system for jawbone defects according to claim 1, characterized in that, The jawbone matching module is specifically used for: Both the healthy side jawbone data and the candidate jawbone data were divided into horizontal slice sequences according to the CBCT voxel interval; The horizontal slice sequences of the healthy jawbone data and the horizontal slice sequences of the candidate jawbone data are aligned in position. For each pair of horizontally aligned slices, a similarity score is calculated. Calculate the average similarity score of all horizontal slice pairs to obtain the comprehensive similarity between the healthy side jawbone data and the candidate jawbone data; The final matching score between the healthy jawbone data and the candidate jawbone data is calculated by weighting the overall similarity, age difference, gender, and difference in the number of horizontal slices.

3. The artificial intelligence-driven biomimetic reconstruction system for jawbone defects according to claim 2, characterized in that, The jawbone matching module is specifically used for: For each pair of horizontally aligned slices, the peak signal-to-noise ratio, structural similarity, and cosine similarity are calculated respectively. The peak signal-to-noise ratio, structural similarity, and cosine similarity are normalized, and the similarity score of the horizontal slice pairs is calculated.

4. The artificial intelligence-driven biomimetic reconstruction system for jawbone defects according to claim 1, characterized in that, The three-dimensional integration module is specifically used for: Using the contralateral jawbone of the target jawbone matched in the database as the jawbone data to be applied, the nearest point is calculated iteratively with the data of the affected jawbone, and the existing part of the defective jawbone on the affected side is aligned to the corresponding area of ​​the jawbone data to be applied. A point cloud fusion algorithm is used to stitch together the completed segment of the jawbone data to be applied with the existing part of the defective jawbone on the affected side, and to perform topological repair on the triangular facets of the junction area so that the boundary of the jawbone to be applied and the boundary of the jawbone on the affected side are spatially fused. Local surface smoothing is applied to the contact area after spatial fusion; The alveolar sockets in the complete jawbone segment were filled and the curved surfaces were smoothed.

5. The artificial intelligence-driven biomimetic reconstruction system for jawbone defects according to claim 1, characterized in that, The planting simulation module is specifically used for: Based on the crown direction and root length of the dentition corresponding to the jawbone data to be applied, the three-dimensional placement position and specifications of the implant are generated. Combined with the bone volume requirements of the implant, the local bone volume of the reconstructed jawbone is adjusted.

6. The artificial intelligence-driven biomimetic reconstruction system for jawbone defects according to claim 1, characterized in that, The biomimetic trabecular bone generation module is specifically used for: Using a pre-trained generative neural network model, an initial central trabecular cube is generated. Based on the structural features of the six faces of the central trabecular cube, adjacent trabecular structures are generated in all directions to form a spatially continuous biomimetic trabecular spatial model with a porous mesh structure. The segmental virtual scaffold is uniformly offset inward to define the medullary cavity region. Under the spatial constraints of the medullary cavity region, the biomimetic trabecular bone spatial model and the segmental virtual scaffold are subjected to Boolean intersection calculation, and the medullary cavity region is filled with trabecular bone structures.

7. The artificial intelligence-driven biomimetic reconstruction system for jawbone defects according to claim 1, characterized in that, The trabecular adaptive support module is specifically used for: Based on interlayer differential detection and historical stack data detection of the suspended structure of the trabecular bone, the suspended area to be supported is obtained; The minimum Euclidean distance between the upper solid region and the suspended region to be supported is calculated to obtain the candidate support mask. The candidate support mask is calculated based on the support strength index and material cost function to obtain the suspended support region; The suspended support area is reverse-projected from top to bottom to generate a vertical projection support column, and a bionic jawbone reconstruction custom body is generated based on the vertical projection support column.

8. An artificial intelligence-driven biomimetic reconstruction method for jawbone defects, characterized in that, The method includes: Collect facial CBCT images of healthy individuals, and use a pre-trained 3D segmentation model to perform instance segmentation on the facial CBCT images to obtain maxillary bone data, mandibular bone data, and dentition data, and establish a structured jawbone and dentition database. Acquire facial CBCT images of patients with jaw defects, and use a pre-trained 3D segmentation model to perform instance segmentation on the facial CBCT images to obtain data of the affected side jawbone and the healthy side jawbone. Based on the healthy side jawbone data, all candidate jawbone data on the same side are searched in the jawbone dentition database. The healthy side jawbone data and candidate jawbone data are reduced to horizontal slices. The comprehensive similarity between the healthy side jawbone data and each candidate jawbone data is calculated based on the horizontal slices. The final matching score is calculated by weighting the comprehensive similarity, age difference, gender, and difference in the number of horizontal slices. The target jawbone data with the highest similarity is selected based on the final matching score. Search the target jawbone data in the jawbone dentition database for the contralateral jawbone data, and use it as the jawbone data to be applied. Integrate the jawbone data to be applied with the affected jawbone data in three dimensions to generate a basic virtual jawbone framework that conforms to the defect space. Based on the dental data of the segment to be filled using the jawbone data, an implantation plan is generated. Based on the implantation plan, the shape of the basic virtual jawbone framework is optimized to obtain a segmental virtual framework that can be reconstructed with the defective jawbone. A biomimetic trabecular bone expansion and generation algorithm is used to fill the segmental virtual scaffold with trabecular bone structure. Using a trabecular adaptive support algorithm, the suspended areas of the trabecular bone are retrieved and vertically printed supports are generated to form a customized biomimetic jawbone reconstruction body.

9. The artificial intelligence-driven biomimetic reconstruction method for jawbone defects according to claim 8, characterized in that, The healthy side jawbone data and candidate jawbone data are reduced in dimensionality to horizontal slices. The comprehensive similarity between the healthy side jawbone data and each candidate jawbone data is calculated based on these horizontal slices. The final matching score is then calculated by weighting the comprehensive similarity, age difference, gender, and difference in the number of horizontal slices, including: Both the healthy side jawbone data and the candidate jawbone data were divided into horizontal slice sequences according to the CBCT voxel interval; The horizontal slice sequences of the healthy jawbone data and the horizontal slice sequences of the candidate jawbone data are aligned in position. For each pair of horizontally aligned slices, a similarity score is calculated. Calculate the average similarity score of all horizontal slice pairs to obtain the comprehensive similarity between the healthy side jawbone data and the candidate jawbone data; The final matching score between the healthy jawbone data and the candidate jawbone data is calculated by weighting the overall similarity, age difference, gender, and difference in the number of horizontal slices.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the artificial intelligence-driven biomimetic reconstruction method for jaw defects according to any one of claims 8 to 9.

Images