Method and device for selecting matching bone flap for jaw defect repair and generating guide plate

Abstract

The application provides a jaw defect repair bone flap selection matching and guide plate generation method and device, which comprises the following steps: inputting the recipient bone image data and the donor bone image data into corresponding bone segmentation models respectively to obtain a recipient bone three-dimensional model and a donor bone three-dimensional model; constructing an optimization problem and solving it according to the recipient bone three-dimensional model, the recipient osteotomy plane model of the defect boundary and the three-dimensional space curve of the recipient post-reconstruction profile to obtain a target chord length list; searching the donor bone three-dimensional model according to the target chord length list based on the space curve of the available bone harvesting area of the donor to obtain the start and end point coordinates of a plurality of candidate bone flap segments; calculating an adjustment transformation matrix of the donor bone to the recipient bone according to the start and end point coordinates of the plurality of candidate bone flap segments on the recipient bone and the start and end point coordinates of the plurality of candidate bone flap segments on the donor bone; and generating a planning guide plate according to the inverse matrix of the adjustment transformation matrix, so that the image input to the guide plate output is automatically generated in a one-stop mode, and the efficiency is high.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence diagnostic technology, and in particular to a method and apparatus for selecting and matching bone flaps and generating guide plates for jawbone defect repair. Background Technology

[0002] Before digital surgical planning, the first step in autologous bone grafting is to obtain a precise three-dimensional anatomical model from the patient's computed tomography (CT) images. However, current techniques for obtaining three-dimensional models and surgical planning present the following problems: The generation of 3D models mainly relies on doctors or technicians manually or semi-automatically outlining (segmenting) CT 2D slice images in software. This process is extremely time-consuming; processing the iliac bone and jawbone of a single case typically takes several hours. At the same time, the segmentation results are greatly affected by the operator's subjectivity, resulting in significant inter-operator variability, which makes it difficult to guarantee the accuracy and consistency of the model.

[0003] Even after obtaining the model, traditional surgical planning still relies heavily on the surgeon's experience and manual operation. During the operation, the surgeon directly cuts bone fragments and repeatedly grinds them to match the recipient area. This process lacks precision control, has low surgical efficiency, poor bone flap fit to the recipient area, and the surgical outcome is highly dependent on the surgeon's experience, lacking standardized procedures.

[0004] In summary, existing technologies suffer from problems such as heavy reliance on manual interaction and low efficiency in model acquisition and surgical planning. Summary of the Invention

[0005] This invention provides a method and apparatus for selecting and matching bone flaps and generating guide plates for jawbone defect repair, which solves the defects of existing technologies that rely heavily on manual interaction and are inefficient, and realizes automated and highly efficient selection and matching of bone flaps and generating guide plates for jawbone defect repair.

[0006] This invention provides a method for selecting and matching bone flaps and generating guide plates for jawbone defect repair, comprising the following steps: Recipient bone image data is input into a pre-constructed recipient bone segmentation model to obtain a three-dimensional recipient bone model; donor bone image data is input into a pre-constructed donor bone segmentation model to obtain a three-dimensional donor bone model; wherein, the recipient bone segmentation model is based on a deep neural network and is trained using recipient bone image samples with labeled contours, and the donor bone segmentation model is based on a deep neural network and is trained using donor bone image samples with labeled contours. Based on the three-dimensional model of the receptor skeleton, the predefined defect boundary receptor osteotomy plane model, and the three-dimensional spatial curve of the receptor reconstructed contour, an optimization problem is constructed and solved to obtain a list of target chord lengths. Based on the spatial curve of the predefined available bone harvesting area of ​​the donor, the three-dimensional model of the donor skeleton is searched according to the target chord length list to obtain the start and end point coordinates of multiple candidate bone flap segments. Based on the coordinates of the origin and termination points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the origin and termination points in the donor bone, calculate the adjustment transformation matrix from the donor bone to the recipient bone. The planning guide is generated based on the inverse of the adjustment transformation matrix.

[0007] According to the present invention, a method for selecting and matching bone flaps and generating guide plates for jawbone defect repair is provided. Based on the three-dimensional model of the recipient skeleton, the predefined recipient osteotomy plane model of the defect boundary, and the three-dimensional spatial curve of the reconstructed recipient contour, an optimization problem is constructed and solved to obtain a list of target chord lengths, including: The intersection of the three-dimensional spatial curve of the reconstructed receptor contour and the plane model of the receptor osteotomy is calculated as the starting point and ending point of reconstruction. The coordinates of several dividing points on the three-dimensional space curve are used as optimization variables. The preset geometric deviation index between the straight line segments connected by the dividing points and the curve segments of the corresponding three-dimensional space curve is used as the objective function. The optimal dividing point coordinates are obtained by solving the optimization algorithm. Based on the coordinates of the reconstruction start point, the reconstruction end point, and the optimal segmentation point, the target chord lengths corresponding to the bone flap segments are generated, and a list of target chord lengths is obtained.

[0008] According to the present invention, a method for selecting and matching bone flaps and generating guide plates for jawbone defect repair is provided. Based on a predefined spatial curve of the available bone harvesting area of ​​the donor, the method searches the three-dimensional model of the donor skeleton according to the target chord length list to obtain multiple candidate bone flap segments that satisfy the target chord length list, including: S1: Based on the spatial curve of the available bone harvesting area of ​​the donor, and according to a target chord length in the target chord length list, the three-dimensional model of the donor bone is searched to obtain candidate bone flap segments that meet the target chord length, which are then used as newly planned candidate bone flap segments. S2: Perform spatial interference detection on the newly planned candidate bone flap segments and the planned candidate bone flap segments in the set of planned candidate bone flap segments; S3: If the newly planned candidate bone flap segment and the already planned candidate bone flap segment are completely separated in space, retain the newly planned candidate bone flap segment and add the newly planned candidate bone flap segment to the set of already planned candidate bone flap segments. S4: Repeat steps S1-S3 until the target chord length search in the target chord length list is completed; S5: Output the start and end coordinates of multiple candidate bone flap segments in the planned set of candidate bone flap segments.

[0009] According to the present invention, a method for selecting and matching bone flaps and generating guide plates for jawbone defect repair includes calculating an adjustment transformation matrix from donor bone to recipient bone based on the origin and termination coordinates of multiple candidate bone flap segments in the recipient bone and the origin and termination coordinates of the donor bone. Based on the coordinates of the origin and termination points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the origin and termination points in the donor bone, the Protodyakonov analysis algorithm is used to calculate and obtain the coarse registration matrix. Based on the coordinates of the starting and ending points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the starting and ending points in the donor bone, a fine adjustment matrix is ​​obtained by performing a rotational iterative search based on the chord axis of the bone flap using the fit cost function. The adjustment transformation matrix is ​​obtained based on the coarse registration matrix and the fine adjustment matrix.

[0010] According to the present invention, a method for selecting and matching bone flaps and generating guide plates for jawbone defect repair is provided, wherein a planned guide plate is generated based on the inverse matrix of the adjustment transformation matrix, including: Based on the inverse of the adjustment transformation matrix, the recipient osteotomy plane is mapped onto the donor bone three-dimensional model; A guide module with a guide groove of preset thickness is generated at each mapped planar position; The guide module generates a conforming surface; The bonding surface is thickened to obtain a negative bonding substrate; Based on the negative-shape bonding substrate and the guiding module, a planning guide plate is constructed.

[0011] The present invention provides a method for selecting and matching bone flaps and generating guide plates for jawbone defect repair. Based on a deep neural network, the method trains a recipient bone segmentation model using recipient bone image samples with labeled contours, including: Acquire receptor bone image samples; The receptor skeleton image samples are resampled, normalized, and augmented to obtain receptor skeleton segmentation training data. The receptor skeleton segmentation training data is outlined and labeled to obtain receptor skeleton segmentation training samples. The first base model is constructed based on the nnU-Net neural network; Based on the first base model, the receptor skeleton segmentation training sample is used for training until the first preset termination condition is reached, thus obtaining the receptor skeleton segmentation model.

[0012] According to the present invention, a method for selecting and matching bone flaps and generating guide plates for jawbone defect repair is provided. Based on a deep neural network, the method trains a donor bone segmentation model using donor bone image samples with labeled contour lines, including: Obtain donor bone image samples; The donor bone image samples are resampled, normalized, and augmented to obtain donor bone segmentation training data. The donor bone segmentation training data is outlined and labeled to obtain donor bone segmentation training samples; A second base model is constructed based on the nnU-Net neural network; Based on the second base model, the donor bone segmentation model is trained using donor bone segmentation training samples until the second preset termination condition is met.

[0013] The present invention also provides a bone flap selection and matching and guide plate generation device for jawbone defect repair, comprising the following modules: The three-dimensional model module is used to input recipient bone image data into a pre-constructed recipient bone segmentation model to obtain a three-dimensional model of the recipient bone; and to input donor bone image data into a pre-constructed donor bone segmentation model to obtain a three-dimensional model of the donor bone; wherein, the recipient bone segmentation model is based on a deep neural network and is trained using recipient bone image samples with labeled contours, and the donor bone segmentation model is based on a deep neural network and is trained using donor bone image samples with labeled contours. The target chord length module is used to construct and solve an optimization problem based on the three-dimensional model of the recipient skeleton, the predefined defect boundary recipient osteotomy plane model, and the three-dimensional spatial curve of the reconstructed recipient contour, to obtain a list of target chord lengths. The candidate bone flap module is used to search the three-dimensional model of the donor bone based on the spatial curve of the predefined available bone harvesting area of ​​the donor and according to the target chord length list to obtain the start and end point coordinates of multiple candidate bone flap segments. The adjustment transformation module is used to calculate the adjustment transformation matrix from donor bone to recipient bone based on the coordinates of the start and end points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the start and end points of the donor bone. The planning guide module is used to generate a planning guide based on the inverse matrix of the adjustment transformation matrix.

[0014] The present invention also provides a planning guide plate, wherein the planning guide plate assembly is generated by the bone flap selection and matching and guide plate generation method for jaw defect repair as described above.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the bone flap selection and matching and guide plate generation method for jaw defect repair as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the bone flap selection and matching and guide plate generation method for jaw defect repair as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the bone flap selection and matching and guide plate generation method for jaw defect repair as described above.

[0018] The present invention provides a method and apparatus for selecting and matching bone flaps and generating guide plates for jawbone defect repair. By training recipient and donor bone segmentation models, it automatically and accurately segments the corresponding bones and constructs corresponding three-dimensional models. The determination of segmentation points is transformed into a mathematical optimization problem, and an optimization algorithm is used to solve for the optimal segmentation scheme, resulting in a list of target chord lengths. On the donor bone three-dimensional model, an algorithm-driven automated search is performed based on the spatial curve of the available bone harvesting area and the target chord lengths to obtain the start and end coordinates of multiple candidate bone flap segments. Based on the start and end coordinates of the multiple candidate bone flap segments in the recipient and donor bones, an adjustment transformation matrix from donor to recipient bone is obtained. Based on the inverse of the adjustment transformation matrix, a planning guide plate is generated, achieving a "one-stop" automated generation from image input to guide plate output, greatly reducing manpower and increasing efficiency. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this 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 some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is one of the flowcharts illustrating the method for selecting and matching bone flaps and generating guide plates for jawbone defect repair provided by the present invention. Figure 2 This is the second flowchart of the method for selecting and matching bone flaps and generating guide plates for jawbone defect repair provided by the present invention. Figure 3 This is a schematic diagram of the structure of the bone flap selection and matching and guide plate generation device for jawbone defect repair provided by the present invention; Figure 4This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0022] The following is combined with Figures 1-2 The present invention describes a method for selecting and matching bone flaps and generating guide plates for jawbone defect repair. Figure 1 This is one of the flowcharts illustrating the method for selecting and matching bone flaps and generating guide plates for jawbone defect repair provided by the present invention, such as... Figure 1 As shown, the method includes: Step 110: Input the recipient bone image data into the pre-constructed recipient bone segmentation model to obtain a three-dimensional recipient bone model; input the donor bone image data into the pre-constructed donor bone segmentation model to obtain a three-dimensional donor bone model; wherein, the recipient bone segmentation model is based on a deep neural network and is trained using recipient bone image samples with outlined contours, and the donor bone segmentation model is based on a deep neural network and is trained using donor bone image samples with outlined contours.

[0023] The recipient bone imaging data includes tomographic images of the recipient bone, and the acquisition methods include, but are not limited to, CT (computed tomography), MRI (magnetic resonance imaging), ultrasound tomography, PET / SPECT (nuclear medicine tomography), OCT (optical coherence tomography), and DTS (digital tomography fusion), etc. This invention does not limit these methods.

[0024] In practice, the recipient bone can be any bone that needs to be reconstructed. This invention uses the jawbone as the recipient bone for illustration, but this does not imply any limitation on the invention.

[0025] The donor bone imaging data includes tomographic images of the donor bone, and the acquisition methods include, but are not limited to, CT (computed tomography), MRI (magnetic resonance imaging), ultrasound tomography, PET / SPECT (nuclear medicine tomography), OCT (optical coherence tomography), and DTS (digital tomography fusion), etc. This invention does not limit these methods.

[0026] In practice, the donor bone can be any bone suitable for autologous transplantation and reconstruction of the recipient bone. This invention uses the iliac bone as the donor bone, but this does not imply any limitation on the invention.

[0027] Specifically, during training, the recipient bone segmentation model is trained using recipient bone image samples containing the recipient bone, while the donor bone segmentation model is trained using donor bone image samples containing the donor bone. For example, in one specific embodiment, the jawbone is used as the recipient bone and the iliac bone as the donor bone. The recipient bone segmentation model (jawbone segmentation model) is trained using head and neck CT scans as recipient bone image samples, while the donor bone segmentation model (iliac bone segmentation model) is trained using abdominal and pelvic spiral CT scans as donor bone image samples.

[0028] It should be noted that the image samples used to train the skeletal segmentation model are image samples with outlines.

[0029] Furthermore, the present invention does not impose restrictions on the base model during the training process of the recipient bone segmentation model and the donor bone segmentation model, and the model can be selected according to the circumstances.

[0030] During the inference process, the recipient bone image data and the donor bone image data are input into their respective recipient bone segmentation models and donor bone segmentation models to obtain the three-dimensional models of the recipient bone and the donor bone.

[0031] Furthermore, to ensure the smoothness and accuracy of the model output, the inference process employs a sliding window strategy with Gaussian Importance Weighting to seamlessly predict the volume of the entire image (e.g., CT), avoiding artifacts at image patch stitching and ensuring the smoothness and accuracy of the output probability map.

[0032] It should be noted that the skeletal segmentation model outputs a 3D mask of the input image data. A 3D model can be obtained from the 3D mask. The specific conversion method can be selected according to the actual situation, and this invention will not elaborate on it.

[0033] In one specific embodiment, a threshold (e.g., 0.5) is set on the output probability map to generate a binarized 3D mask. Then, the Marching Cubes algorithm is used to convert this 3D mask into a triangular mesh surface model (STL format). Finally, the model is optimized using Laplacian smoothing and operations such as removing small connected components to obtain a high-fidelity 3D model for practical operation.

[0034] Furthermore, before the recipient's and donor's bone image data are input into the corresponding model, preprocessing can be performed, including resampling and data normalization.

[0035] Step 120: Based on the three-dimensional model of the receptor skeleton, the predefined defect boundary receptor osteotomy plane model, and the three-dimensional spatial curve of the reconstructed receptor contour, construct and solve an optimization problem to obtain a list of target chord lengths.

[0036] In this step, the inputs include the generated three-dimensional model of the recipient skeleton, the predefined osteotomy plane model of the defect boundary (recipient), and a three-dimensional spatial curve representing the ideal reconstructed contour of the recipient.

[0037] Based on the above input, the boundary is determined, and segmentation optimization is performed. Specifically, the coordinates of several segmentation points on the reconstruction curve are used as optimization variables to construct an objective function. The minimum value of this objective function is obtained through iterative calculation, thereby obtaining a set of optimal segmentation point coordinates. Based on the start, end, and intermediate segmentation points, the target chord length required for each bone flap segment is obtained, and a list of target chord lengths is then compiled.

[0038] Step 130: Based on the predefined spatial curve of the available bone harvesting area of ​​the donor, and according to the target chord length list, search the three-dimensional model of the donor bone to obtain the start and end point coordinates of multiple candidate bone flap segments.

[0039] In this step, the inputs include a 3D model of the donor skeleton, a predefined spatial curve of the available bone harvesting area for the donor, and a list of target chord lengths.

[0040] Based on the above input, an automated search is performed along the spatial curve of the available bone harvesting area of ​​the donor to obtain the coordinates of the start and end points of multiple candidate bone flap segments.

[0041] Step 140: Calculate the adjustment transformation matrix from donor bone to recipient bone based on the coordinates of the starting and ending points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the starting and ending points of the donor bone.

[0042] Based on the coordinates of the start and end points of multiple candidate bone flap segments in the recipient bone and the start and end points of the donor bone, the donor bone flap is virtually placed and precisely fitted to obtain the adjustment transformation matrix from the donor bone to the recipient bone.

[0043] Step 150: Generate the planning guide based on the inverse matrix of the adjustment transformation matrix.

[0044] Adjusting the inverse of the transformation matrix can accurately map any object in the recipient bone region coordinate system back to its corresponding position in the donor bone region original coordinate system.

[0045] During the generation of the planning guide, the osteotomy plane of the recipient bone region is mapped back to the surface of the three-dimensional model of the donor bone region by adjusting the inverse matrix of the transformation matrix. Based on the mapping result, a solid structure is generated through Boolean operations to obtain the planning guide.

[0046] The present invention provides a method for selecting and matching bone flaps and generating guide plates for jawbone defect repair. This method automatically and accurately segments corresponding bones and constructs corresponding three-dimensional models by training recipient and donor bone segmentation models. The determination of segmentation points is transformed into a mathematical optimization problem, and an optimization algorithm is used to solve for the optimal segmentation scheme, resulting in a list of target chord lengths. On the donor bone three-dimensional model, an algorithm-driven automated search is performed based on the spatial curve of the available bone harvesting area and the target chord lengths to obtain the start and end coordinates of multiple candidate bone flap segments. Based on the start and end coordinates of the multiple candidate bone flap segments in the recipient and donor bones, an adjustment transformation matrix from donor to recipient bone is obtained. The inverse of the adjustment transformation matrix is ​​used to generate a planned guide plate, achieving a "one-stop" automated generation from image input to guide plate output, greatly reducing manpower and increasing efficiency.

[0047] The following provides a further explanation of step 120. In some embodiments, based on the three-dimensional model of the recipient skeleton, the predefined recipient osteotomy plane model of the defect boundary, and the three-dimensional spatial curve of the reconstructed recipient contour, an optimization problem is constructed and solved to obtain a list of target chord lengths, including: Step 121: Calculate the intersection point of the three-dimensional spatial curve of the reconstructed contour of the receptor and the plane model of the receptor osteotomy, and use it as the starting point and ending point of reconstruction.

[0048] In other words, the intersection of the contour curve and the osteotomy plane is calculated and used as the starting and ending points of the reconstruction.

[0049] Step 122: Using the coordinates of several segmentation points on the three-dimensional space curve as optimization variables, and using the preset geometric deviation index between the straight line segments connected by all segmentation points and the corresponding curve segments of the three-dimensional space curve as the objective function, the optimal segmentation point coordinates are obtained by solving the optimization algorithm.

[0050] Specifically, a stochastic optimization algorithm (such as differential evolution algorithm) is used, with the coordinates of several segmentation points on the reconstructed curve (i.e., the three-dimensional curve) as optimization variables, and the preset geometric deviation index between all straight line segments connected by the segmentation points and their corresponding original curve segments as the objective function to be minimized. The minimum value of the objective function is solved by iterative calculation, thereby obtaining a set of optimal segmentation point coordinates.

[0051] In one specific embodiment, the preset geometric deviation index is the maximum Hausdorff distance.

[0052] Step 123: Based on the coordinates of the reconstruction start point, the reconstruction end point, and the optimal segmentation point, generate the target chord length corresponding to the bone flap segment, and then obtain a list of target chord lengths.

[0053] Output the 3D coordinates of all key points (start, end, and intermediate segmentation points), as well as the target chord length required for each bone flap segment, to obtain a list of target chord lengths.

[0054] This invention transforms the determination of segmentation points into a mathematical optimization problem, thereby enabling automatic segmentation. The time required to acquire a 3D model is reduced from hours of manual labor to an average of less than 5 minutes of automated computer processing, resulting in a tenfold increase in the efficiency of the entire planning process.

[0055] The following provides further explanation of step 130. For example... Figure 2 As shown, in some embodiments, based on the spatial curve of the predefined donor bone harvesting area, the three-dimensional model of the donor skeleton is searched according to the target chord length list to obtain multiple candidate bone flap segments that satisfy the target chord length list, including: S1: Based on the spatial curve of the available bone harvesting area of ​​the donor, and according to a target chord length in the target chord length list, the three-dimensional model of the donor bone is searched to obtain candidate bone flap segments that meet the target chord length, which are then used as newly planned candidate bone flap segments.

[0056] An algorithmic search is performed along the bone harvesting path curve of the donor bone (i.e., the spatial curve of the available bone harvesting area of ​​the donor) to find candidate bone flap segments that meet the first target chord length, which are then used as newly planned candidate bone flap segments.

[0057] S2: Perform spatial interference detection on the newly planned candidate bone flap segments and the planned candidate bone flap segments in the set of planned candidate bone flap segments.

[0058] When locating subsequent bone flap segments, spatial interference detection is required to ensure that the newly planned bone flap is completely separated from all previously planned bone flaps in space.

[0059] In specific spatial interference detection, a 3D convex hull can be generated for each planned virtual bone flap. Specifically, for a newly planned candidate bone flap segment, a 3D convex hull is generated, and spatial interference detection is performed by traversing every existing 3D convex hull in the global list. If the 3D convex hull of the newly planned candidate bone flap segment interferes (does not separate) with any existing 3D convex hull, the current position or shape of the newly planned candidate bone flap segment may need to be adjusted. Otherwise, the planning is successful.

[0060] S3: If the newly planned candidate bone flap segment and the already planned candidate bone flap segment are completely separated in space, retain the newly planned candidate bone flap segment and add the newly planned candidate bone flap segment to the already planned candidate bone flap segment set.

[0061] S4: Repeat steps S1-S3 until the target chord length search in the target chord length list is completed.

[0062] S5: Output the start and end coordinates of multiple candidate bone flap segments in the planned set of candidate bone flap segments.

[0063] The start and end coordinates of the multiple candidate bone flap segments output are the precise start and end coordinates of all planned bone flaps on the donor bone surface.

[0064] Step 140 is further described below. In some embodiments, based on the origin and termination coordinates of the plurality of candidate bone flap segments in the recipient bone and the origin and termination coordinates in the donor bone, a modulation transformation matrix from donor bone to recipient bone is calculated, including: Step 141: Based on the coordinates of the start and end points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the start and end points in the donor bone, the Protodyakonov analysis algorithm is used to calculate and obtain the coarse registration matrix.

[0065] For each bone flap, using its origin and destination coordinates in the donor area and the target origin and destination coordinates in the recipient area, a rigid transformation matrix is ​​calculated using the Procrustes Analysis algorithm as a coarse registration matrix to quickly align the bone flap from the donor area to the recipient area.

[0066] Step 142: Based on the coordinates of the starting and ending points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the starting and ending points in the donor bone, a fine adjustment matrix is ​​obtained by performing a rotational iterative search based on the chord axis of the bone flap using the fit cost function.

[0067] Specifically, the osteotomy section of the bone flap and the osteotomy section of the recipient's bone defect are separated first, these two surfaces that will soon come into contact with each other.

[0068] Then, a fitting cost function is constructed, which is a weighted sum of a: the average nearest distance between the vertices of the two contact surfaces and b: the angular deviation penalty term of the average normal vectors of the two contact surfaces.

[0069] Using the chord axis of the bone flap as the center of rotation, iterative rotation is performed within a preset range to find the optimal rotation angle that minimizes the cost function mentioned above, and a fine-tuning transformation matrix (i.e., fine-tuning matrix) is generated.

[0070] Step 143: Obtain the adjustment transformation matrix based on the coarse registration matrix and the fine adjustment matrix.

[0071] Multiplying the coarse registration matrix by the fine adjustment matrix yields the adjustment transformation matrix.

[0072] The embodiments of the present invention first achieve rapid initial alignment through rigid transformation; then, based on a "fitting cost function" that includes multiple factors such as distance and normal vector, fine angle optimization is performed to achieve the best matching of the contact surfaces.

[0073] The following provides a further explanation of step 150. In some embodiments, generating a planning guide based on the inverse of the adjustment transformation matrix includes: Step 151: Map the recipient osteotomy plane to the donor bone three-dimensional model according to the inverse of the adjustment transformation matrix.

[0074] Calculate the inverse of the adjustment transformation matrix. This inverse matrix accurately maps any object in the recipient bone region coordinate system back to its corresponding position in the original donor bone region coordinate system. Map all osteotomy planes in the recipient bone region back to the surface of the donor bone 3D model using the inverse of the adjustment transformation matrix.

[0075] Step 152: Generate a guide module with a guide groove of preset thickness at each mapped planar position.

[0076] The preset thickness can be the standard thickness.

[0077] Step 153: Generate a fitting surface based on the guide module.

[0078] Specifically, all guide modules are Boolean-unified, and their intersection with the donor skeleton is calculated to form a fitting surface.

[0079] Step 154: Thicken the bonding surface to obtain a negative bonding substrate.

[0080] The bonding surface is uniformly thickened (e.g., 3 mm) to generate a negative bonding substrate that precisely conjugates with the donor bone surface.

[0081] Step 155: Construct a planning guide plate based on the negative-shape bonding substrate and the guiding module.

[0082] The bonding base and guide module are Boolean-unified, and then the guide groove and preset screw holes are subtracted to finally obtain a solid planning guide plate that can be directly used for 3D printing.

[0083] The highly conforming bone contact surface (gap <1mm) provided by this invention offers an optimal mechanical and biological environment for bone healing, significantly improving the success rate and long-term stability of transplantation procedures. Through algorithm-driven planning and guide plate guidance, the deviation between the reconstructed bone shape and the preoperative virtual plan can be controlled within 2mm, achieving highly precise anatomical repositioning.

[0084] The receptor skeleton segmentation model is further described below. In some embodiments, the receptor skeleton segmentation model is trained using receptor skeleton image samples with outlined contours, based on a deep neural network, including: Step 101: Obtain a sample of the recipient's bone structure.

[0085] Collect large-scale imaging data, including recipient bones, covering different ages, genders, and pathological states, as recipient bone imaging samples.

[0086] In one specific embodiment, image data including recipient skeletons from more than 500 cases are collected as recipient skeleton image samples.

[0087] Step 102: Resample, normalize, and augment the receptor bone image samples to obtain receptor bone segmentation training data.

[0088] The data resampling process involves resampling all receptor bone image samples with different original resolutions to an isotropic target voxel spacing (e.g., 1.5 x 1.5 x 1.5 mm) using third-order spline interpolation. This aims to eliminate resolution differences caused by different scanning devices and provide the network with scale-consistent input.

[0089] Data normalization involves processing the voxel intensity values ​​for each recipient bone image sample. In some embodiments, head and neck CT images are used as recipient bone image samples. In this case, data normalization involves: first, clipping based on a typical range of CT values ​​(e.g., [-1000, 2000] HU) to remove outliers; then, performing foreground-based Z-score normalization on the clipped data, i.e., subtracting the mean of the foreground voxel intensities and dividing by its standard deviation. This allows the model to focus on structural information without being affected by the window width and level settings of the CT equipment.

[0090] Data augmentation includes employing a range of powerful online data augmentation techniques during model training to improve the model's generalization ability and prevent overfitting. These techniques include random 3D rotation, random scaling, elastic deformation, random addition of Gaussian noise, and gamma correction (brightness and contrast adjustment).

[0091] Step 103: Contour delineation and annotation are performed on the receptor skeleton segmentation training data to obtain receptor skeleton segmentation training samples.

[0092] Outline drawing and annotation can be done using automatic annotation tools or manually by technicians; this invention does not impose any restrictions on this.

[0093] In one specific embodiment, a predetermined number of senior imaging experts use a double-blind method to precisely manually delineate the contours of the receptor skeleton in each receptor skeleton segmentation training data, and cross-validation is used to ensure the consistency of the annotation, ultimately forming a high-precision 3D mask as the "gold standard" for training.

[0094] Step 104: Construct the first base model based on the nnU-Net neural network.

[0095] The reason for choosing the nnU-Net (no-new-Net, automated medical image segmentation framework) framework as the core segmentation algorithm is as follows: 1) nnU-Net can handle the high heterogeneity of medical images. Different hospitals, different scanning equipment, and different scanning protocols can lead to huge differences in image resolution, slice thickness, signal-to-noise ratio, window width, and window level. Fixed network models are difficult to adapt to this heterogeneity.

[0096] 2) nnU-Net possesses adaptive configuration capabilities. The core innovation of nnU-Net lies in its "adaptive" capability. Before training begins, it automatically analyzes the "data fingerprint" of the entire training set, including statistical information such as voxel spacing, image size, and orientation. Based on this information, it automatically makes key decisions, including: Optimal network architecture selection: Automatically determines whether to use a 2D network, a 3D full-resolution network, or a 3D cascaded network for processing large images (first low-resolution global segmentation, then high-resolution local fine segmentation).

[0097] Data resampling strategy: Automatically determine the optimal target voxel spacing and select an appropriate interpolation method.

[0098] Normalization scheme: Automatically select the voxel intensity normalization method that best suits the dataset.

[0099] Training parameter optimization: Automatically configure the optimal batch size and patch size to adapt to hardware limitations (such as GPU memory).

[0100] 3) nnU-Net exhibits robustness and state-of-the-art performance. nnU-Net has consistently achieved top results in multiple internationally open medical image segmentation challenges, demonstrating its ability as an "out-of-the-box" solution to provide state-of-the-art performance for various segmentation tasks. Therefore, choosing nnU-Net is not merely about selecting an algorithm, but about choosing an automated, complete technical process capable of building customized optimal models, thereby maximizing the accuracy and generalization ability of iliac bone segmentation.

[0101] Step 105: Based on the first pedestal model, train using the receptor skeleton segmentation training samples until the first preset termination condition is met to obtain the receptor skeleton segmentation model.

[0102] The first preset termination condition can be the maximum number of training rounds or the model performance evaluation index reaching a threshold; this invention does not impose any restrictions on this.

[0103] During training, a combined loss function can be used, which is a weighted sum of Dice loss and cross-entropy loss, to simultaneously optimize the model's region overlap accuracy and pixel-level classification accuracy.

[0104] In one specific embodiment, the entire dataset (500 cases) is randomly divided into a training set (400 cases), a validation set (50 cases), and an independent test set (50 cases) in an 8:1:1 ratio for model training, hyperparameter tuning, and objective evaluation of final performance. The model is trained on a high-performance GPU server until the Dice coefficient of the validation set no longer increases for several consecutive epochs, at which point training stops, the best-performing model weights are saved, and the receptor skeleton segmentation model is output.

[0105] The donor bone segmentation model is further described below. In some embodiments, the donor bone segmentation model is trained using donor bone image samples with labeled contours, based on a deep neural network, including: Step 111: Obtain donor bone image samples.

[0106] We collected a large-scale collection of donor bone images covering different ages, sexes, and pathological conditions as donor bone image samples.

[0107] In one specific embodiment, spiral CT images (DICOM format) of the abdominal and pelvic cavities of 500 patients with ethical approval were collected as donor bone image samples.

[0108] Step 112: Resample, normalize, and augment the donor bone image samples to obtain donor bone segmentation training data.

[0109] Resampling, data normalization, and data augmentation are similar to step 102, and will not be described in detail here.

[0110] Step 113: Contour delineation and annotation are performed on the donor bone segmentation training data to obtain donor bone segmentation training samples.

[0111] Outline drawing and annotation can be done using automatic annotation tools or manually by technicians; this invention does not impose any restrictions on this.

[0112] In one specific embodiment, a predetermined number of senior imaging experts use a double-blind method to precisely manually delineate the contours of the donor bones in each donor bone segmentation training data, and cross-validation is used to ensure the consistency of the annotations, ultimately forming a high-precision 3D mask as the "gold standard" for training.

[0113] Step 114: Construct a second pedestal model based on the nnU-Net neural network.

[0114] The reason for choosing nnU-Net as the base model (denoted as the second base model) for the donor bone segmentation model in this application is the same as the reason for choosing nnU-Net as the base model for the recipient bone segmentation model, and this application will not elaborate on this.

[0115] Step 115: Based on the second base model, train using donor bone segmentation training samples until the second preset termination condition is met to obtain the donor bone segmentation model.

[0116] The second preset termination condition can be the maximum number of training rounds or the model performance evaluation index reaching a threshold; this invention does not impose any restrictions on this.

[0117] During training, a combined loss function can be used, which is a weighted sum of Dice loss and cross-entropy loss, to simultaneously optimize the model's region overlap accuracy and pixel-level classification accuracy.

[0118] In one specific embodiment, the entire dataset (500 cases) is randomly divided into a training set (400 cases), a validation set (50 cases), and an independent test set (50 cases) in an 8:1:1 ratio for model training, hyperparameter tuning, and objective evaluation of final performance. The model is trained on a high-performance GPU server until the Dice coefficient of the validation set no longer increases for several consecutive epochs, at which point training stops, the best-performing model weights are saved, and the receptor skeleton segmentation model is output.

[0119] Furthermore, based on the above embodiments, during the training process, all DICOM sequences are uniformly converted to NIFTI format for easier processing.

[0120] This invention does not employ a fixed deep learning network structure, but instead utilizes the nnU-Net adaptive framework. This framework can automatically analyze the data "fingerprint" based on the unique attributes of the input dataset (such as anisotropy, resolution, window width, and window level), and optimize the network topology (such as 2D U-Net, 3D full-resolution U-Net, or 3D U-Netcascade), training parameters, and data preprocessing and augmentation strategies. This allows for the construction of an optimal customized model for a specific segmentation task (such as iliac bone segmentation), ensuring high robustness and high accuracy in segmentation.

[0121] In one specific embodiment, experiments show that deep learning models trained on large-scale data eliminate the subjectivity of manual segmentation and operator differences. The segmentation results have an average Dice similarity coefficient of over 0.97 with the expert "gold standard," providing a highly consistent and accurate anatomical basis for all subsequent planning steps.

[0122] In one specific embodiment, a simulation test was conducted on the bone flap selection and matching and guide plate generation method for jaw defect repair provided by the present invention. The simulation test showed that when different operators used this system to plan the same case, the final generated bone flap length difference was less than 0.5 mm, and the guide plate osteotomy plane angle difference was less than 1°, ensuring the high quality and consistency of the surgical plan.

[0123] The following describes the bone flap selection and matching and guide plate generation device for jaw defect repair provided by the present invention. The bone flap selection and matching and guide plate generation device for jaw defect repair described below can be referred to in correspondence with the bone flap selection and matching and guide plate generation method for jaw defect repair described above. Figure 3 This is a schematic diagram of the structure of the bone flap selection and matching and guide plate generation device for jawbone defect repair provided by the present invention, as shown below. Figure 3 As shown, the device includes the following modules: The 3D model module 310 is used to input recipient bone image data into a pre-constructed recipient bone segmentation model to obtain a recipient bone 3D model; and to input donor bone image data into a pre-constructed donor bone segmentation model to obtain a donor bone 3D model; wherein, the recipient bone segmentation model is based on a deep neural network and is trained using recipient bone image samples with labeled contours, and the donor bone segmentation model is based on a deep neural network and is trained using donor bone image samples with labeled contours.

[0124] The target chord length module 320 is used to construct and solve an optimization problem based on the three-dimensional model of the recipient skeleton, the predefined defect boundary of the recipient osteotomy plane model, and the three-dimensional spatial curve of the reconstructed contour of the recipient, to obtain a list of target chord lengths.

[0125] The candidate bone flap module 330 is used to search the three-dimensional model of the donor bone based on the spatial curve of the predefined available bone harvesting area of ​​the donor and according to the target chord length list to obtain the start and end coordinates of multiple candidate bone flap segments. The adjustment transformation module 340 is used to calculate the adjustment transformation matrix from donor bone to recipient bone based on the coordinates of the start and end points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the start and end points of the donor bone.

[0126] The planning guide module 350 is used to generate a planning guide based on the inverse matrix of the adjustment transformation matrix.

[0127] According to the present invention, a bone flap selection and matching and guide plate generation device for jawbone defect repair constructs and solves an optimization problem based on the three-dimensional model of the recipient skeleton, the predefined recipient osteotomy plane model of the defect boundary, and the three-dimensional spatial curve of the reconstructed recipient contour, to obtain a list of target chord lengths, including: The intersection of the three-dimensional spatial curve of the reconstructed receptor contour and the plane model of the receptor osteotomy is calculated as the starting point and ending point of reconstruction. The coordinates of several dividing points on the three-dimensional space curve are used as optimization variables. The preset geometric deviation index between the straight line segments connected by the dividing points and the curve segments of the corresponding three-dimensional space curve is used as the objective function. The optimal dividing point coordinates are obtained by solving the optimization algorithm. Based on the coordinates of the reconstruction start point, the reconstruction end point, and the optimal segmentation point, the target chord lengths corresponding to the bone flap segments are generated, and a list of target chord lengths is obtained.

[0128] According to the present invention, a bone flap selection and matching device and guide plate generation device for jawbone defect repair, based on a predefined spatial curve of the available bone harvesting area of ​​the donor, searches the three-dimensional model of the donor skeleton according to the target chord length list to obtain multiple candidate bone flap segments that satisfy the target chord length list, including: S1: Based on the spatial curve of the available bone harvesting area of ​​the donor, and according to a target chord length in the target chord length list, the three-dimensional model of the donor bone is searched to obtain candidate bone flap segments that meet the target chord length, which are then used as newly planned candidate bone flap segments. S2: Perform spatial interference detection on the newly planned candidate bone flap segments and the planned candidate bone flap segments in the set of planned candidate bone flap segments; S3: If the newly planned candidate bone flap segment and the already planned candidate bone flap segment are completely separated in space, retain the newly planned candidate bone flap segment and add the newly planned candidate bone flap segment to the set of already planned candidate bone flap segments. S4: Repeat steps S1-S3 until the target chord length search in the target chord length list is completed; S5: Output the start and end coordinates of multiple candidate bone flap segments in the planned set of candidate bone flap segments.

[0129] According to the present invention, a bone flap selection and matching and guide plate generation device for jawbone defect repair calculates an adjustment transformation matrix from donor bone to recipient bone based on the origin and end point coordinates of multiple candidate bone flap segments in the recipient bone and the origin and end point coordinates of the donor bone, including: Based on the coordinates of the origin and termination points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the origin and termination points in the donor bone, the Protodyakonov analysis algorithm is used to calculate and obtain the coarse registration matrix. Based on the coordinates of the starting and ending points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the starting and ending points in the donor bone, a fine adjustment matrix is ​​obtained by performing a rotational iterative search based on the chord axis of the bone flap using the fit cost function. The adjustment transformation matrix is ​​obtained based on the coarse registration matrix and the fine adjustment matrix.

[0130] According to the present invention, a bone flap selection and matching and guide plate generation device for jawbone defect repair generates a planned guide plate based on the inverse matrix of the adjustment transformation matrix, comprising: Based on the inverse of the adjustment transformation matrix, the recipient osteotomy plane is mapped onto the donor bone three-dimensional model; A guide module with a guide groove of preset thickness is generated at each mapped planar position; The guide module generates a conforming surface; The bonding surface is thickened to obtain a negative bonding substrate; Based on the negative-shape bonding substrate and the guiding module, a planning guide plate is constructed.

[0131] According to the present invention, a bone flap selection and matching and guide plate generation device for jawbone defect repair is provided. Based on a deep neural network, the device trains the recipient bone segmentation model using recipient bone image samples with labeled contours, including: Acquire receptor bone image samples; The receptor skeleton image samples are resampled, normalized, and augmented to obtain receptor skeleton segmentation training data. The receptor skeleton segmentation training data is outlined and labeled to obtain receptor skeleton segmentation training samples. The first base model is constructed based on the nnU-Net neural network; Based on the first base model, the receptor skeleton segmentation training sample is used for training until the first preset termination condition is reached, thus obtaining the receptor skeleton segmentation model.

[0132] According to the present invention, a bone flap selection and matching and guide plate generation device for jawbone defect repair is provided. Based on a deep neural network, the device trains a donor bone segmentation model using donor bone image samples with labeled contour lines, including: Obtain donor bone image samples; The donor bone image samples are resampled, normalized, and augmented to obtain donor bone segmentation training data. The donor bone segmentation training data is outlined and labeled to obtain donor bone segmentation training samples; A second base model is constructed based on the nnU-Net neural network; Based on the second base model, the donor bone segmentation model is trained using donor bone segmentation training samples until the second preset termination condition is met.

[0133] The bone flap selection and matching and guide plate generation device for jawbone defect repair provided by this invention automatically and accurately segments the corresponding bones and constructs corresponding three-dimensional models by training the recipient bone segmentation model and the donor bone segmentation model. The determination of the segmentation points is transformed into a mathematical optimization problem, and the optimal segmentation scheme is solved using an optimization algorithm to obtain a list of target chord lengths. On the donor bone three-dimensional model, an algorithm-driven automated search is performed based on the spatial curve of the available bone harvesting area of ​​the donor and the target chord lengths to obtain the start and end point coordinates of multiple candidate bone flap segments. According to the start and end point coordinates of multiple candidate bone flap segments in the recipient bone and the donor bone, the adjustment transformation matrix from the donor bone to the recipient bone is obtained. According to the inverse matrix of the adjustment transformation matrix, a planning guide plate is generated, realizing "one-stop" automated generation from image input to guide plate output, which greatly liberates manpower and has high efficiency.

[0134] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include a processor 410, a communication interface 420, a memory 430, and a communication bus 440. The processor 410, communication interface 420, and memory 430 communicate with each other via the communication bus 440. The processor 410 can call logical instructions from the memory 430 to execute a method for selecting and matching bone flaps and generating guide plates for jawbone defect repair.

[0135] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0136] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the bone flap selection and matching and guide plate generation methods for jaw defect repair provided by the above methods.

[0137] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the methods for selecting and matching bone flaps and generating guide plates for jaw defect repair provided by the above methods.

[0138] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0139] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for selecting and matching bone flaps and generating guide plates for jawbone defect repair, characterized in that, include: The receptor skeleton image data is input into a pre-built receptor skeleton segmentation model to obtain a three-dimensional receptor skeleton model. The donor bone image data is input into a pre-constructed donor bone segmentation model to obtain a three-dimensional donor bone model; wherein, the recipient bone segmentation model is based on a deep neural network and is trained using recipient bone image samples with outlines. Based on the three-dimensional model of the receptor skeleton, the predefined defect boundary receptor osteotomy plane model, and the three-dimensional spatial curve of the receptor reconstructed contour, an optimization problem is constructed and solved to obtain a list of target chord lengths. Based on the spatial curve of the predefined available bone harvesting area of ​​the donor, the three-dimensional model of the donor skeleton is searched according to the target chord length list to obtain the start and end point coordinates of multiple candidate bone flap segments. Based on the coordinates of the origin and termination points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the origin and termination points in the donor bone, calculate the adjustment transformation matrix from the donor bone to the recipient bone. The planning guide is generated based on the inverse of the adjustment transformation matrix.

2. The method for selecting and matching bone flaps and generating guide plates for jawbone defect repair according to claim 1, characterized in that, Based on the three-dimensional model of the receptor skeleton, the predefined defect boundary plane model of the receptor osteotomy, and the three-dimensional spatial curve of the reconstructed receptor contour, an optimization problem is constructed and solved to obtain a list of target chord lengths, including: The intersection of the three-dimensional spatial curve of the reconstructed receptor contour and the plane model of the receptor osteotomy is calculated as the starting point and ending point of reconstruction. The coordinates of several dividing points on the three-dimensional space curve are used as optimization variables. The preset geometric deviation index between the straight line segments connected by the dividing points and the curve segments of the corresponding three-dimensional space curve is used as the objective function. The optimal dividing point coordinates are obtained by solving the optimization algorithm. Based on the coordinates of the reconstruction start point, the reconstruction end point, and the optimal segmentation point, the target chord lengths corresponding to the bone flap segments are generated, and a list of target chord lengths is obtained.

3. The method for selecting and matching bone flaps and generating guide plates for jawbone defect repair according to claim 1, characterized in that, Based on the spatial curve of the predefined donor bone harvesting area, and according to the target chord length list, the three-dimensional model of the donor skeleton is searched to obtain multiple candidate bone flap segments that satisfy the target chord length list, including: S1: Based on the spatial curve of the available bone harvesting area of ​​the donor, and according to a target chord length in the target chord length list, the three-dimensional model of the donor bone is searched to obtain candidate bone flap segments that meet the target chord length, which are then used as newly planned candidate bone flap segments. S2: Perform spatial interference detection on the newly planned candidate bone flap segments and the planned candidate bone flap segments in the set of planned candidate bone flap segments; S3: If the newly planned candidate bone flap segment and the already planned candidate bone flap segment are completely separated in space, retain the newly planned candidate bone flap segment and add the newly planned candidate bone flap segment to the set of already planned candidate bone flap segments. S4: Repeat steps S1-S3 until the target chord length search in the target chord length list is completed; S5: Output the start and end coordinates of multiple candidate bone flap segments in the planned set of candidate bone flap segments.

4. The method for selecting and matching bone flaps and generating guide plates for jawbone defect repair according to claim 1, characterized in that, Based on the origin and termination coordinates of the multiple candidate bone flap segments in the recipient bone and the origin and termination coordinates in the donor bone, the adjustment transformation matrix from donor bone to recipient bone is calculated, including: Based on the coordinates of the origin and termination points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the origin and termination points in the donor bone, the Protodyakonov analysis algorithm is used to calculate and obtain the coarse registration matrix. Based on the coordinates of the starting and ending points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the starting and ending points in the donor bone, a fine adjustment matrix is ​​obtained by performing a rotational iterative search based on the chord axis of the bone flap using the fit cost function. The adjustment transformation matrix is ​​obtained based on the coarse registration matrix and the fine adjustment matrix.

5. The method for selecting and matching bone flaps and generating guide plates for jawbone defect repair according to claim 1, characterized in that, Based on the inverse of the adjustment transformation matrix, a planning guide is generated, including: Based on the inverse of the adjustment transformation matrix, the recipient osteotomy plane is mapped onto the donor bone three-dimensional model; A guide module with a guide groove of preset thickness is generated at each mapped planar position; The guide module generates a conforming surface; The bonding surface is thickened to obtain a negative bonding substrate; Based on the negative-shape bonding substrate and the guiding module, a planning guide plate is constructed.

6. The method for selecting and matching bone flaps and generating guide plates for jawbone defect repair according to claim 1, characterized in that, The receptor skeleton segmentation model is obtained by training on receptor skeleton image samples with labeled contours based on a deep neural network, including: Acquire receptor bone image samples; The receptor skeleton image samples are resampled, normalized, and augmented to obtain receptor skeleton segmentation training data. The receptor skeleton segmentation training data is outlined and labeled to obtain receptor skeleton segmentation training samples. The first base model is constructed based on the nnU-Net neural network; Based on the first base model, the receptor skeleton segmentation training sample is used for training until the first preset termination condition is reached, thus obtaining the receptor skeleton segmentation model.

7. The method for selecting and matching bone flaps and generating guide plates for jawbone defect repair according to claim 1, characterized in that, The donor bone segmentation model is trained using deep neural networks and labeled donor bone image samples with outlined contours, including: Obtain donor bone image samples; The donor bone image samples are resampled, normalized, and augmented to obtain donor bone segmentation training data. The donor bone segmentation training data is outlined and labeled to obtain donor bone segmentation training samples; A second base model is constructed based on the nnU-Net neural network; Based on the second base model, the donor bone segmentation model is trained using donor bone segmentation training samples until the second preset termination condition is met.

8. A device for selecting and matching bone flaps and generating guide plates for jawbone defect repair, characterized in that, include: The three-dimensional model module is used to input recipient bone image data into a pre-constructed recipient bone segmentation model to obtain a three-dimensional model of the recipient bone; and to input donor bone image data into a pre-constructed donor bone segmentation model to obtain a three-dimensional model of the donor bone; wherein, the recipient bone segmentation model is based on a deep neural network and is trained using recipient bone image samples with labeled contours, and the donor bone segmentation model is based on a deep neural network and is trained using donor bone image samples with labeled contours. The target chord length module is used to construct and solve an optimization problem based on the three-dimensional model of the recipient skeleton, the predefined defect boundary recipient osteotomy plane model, and the three-dimensional spatial curve of the reconstructed recipient contour, to obtain a list of target chord lengths. The candidate bone flap module is used to search the three-dimensional model of the donor bone based on the spatial curve of the predefined available bone harvesting area of ​​the donor and according to the target chord length list to obtain the start and end point coordinates of multiple candidate bone flap segments. The adjustment transformation module is used to calculate the adjustment transformation matrix from donor bone to recipient bone based on the coordinates of the start and end points of the multiple candidate bone flap segments in the recipient bone and the coordinates of the start and end points of the donor bone. The planning guide module is used to generate a planning guide based on the inverse matrix of the adjustment transformation matrix.

9. A planning guide plate, characterized in that, The planning guide plate is generated by the bone flap selection and matching and guide plate generation method for jaw defect repair as described in any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method for selecting and matching bone flaps and generating guide plates for jawbone defect repair as described in any one of claims 1 to 7.

Images