Artificial intelligence assisted standard artificial temporomandibular joint automatic placement method

Abstract

The invention discloses an artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method. The method comprises the following steps: training a three-dimensional reconstruction model used for extracting a mandible and zygomatic bone zygomatic arch joint prosthesis placement area; based on an improved 3D U-Net neural network, carrying out segmentation extraction on the interested areas of the glenoid fossa and the mandibular branch, and obtaining a space geometric structure for prosthesis installation; the method comprises the following steps: extracting outer bone surface point cloud data of a zygomatic arch and a mandibular ramus on the basis of a space geometric structure of prosthesis installation, identifying anatomical key points, and generating a standard prosthesis fitting surface through geometric fitting and constrained optimization; determining the relative position relation between the maxillary joint fossa prosthesis and the mandibular branch prosthesis, and determining the installation direction, the hole position definition and the assembly reference point; and determining a contact area, a posture direction and a placement position of the prosthesis, and completing three-dimensional positioning and posture matching of the prosthesis in the patient model as a surgical planning scheme.

Description

Technical Field

[0001] The present invention relates to the field of digital design of oral surgery, and in particular to an artificial intelligence-assisted automatic placement method for a standard artificial temporomandibular joint. Background Art

[0002] Artificial Intelligence (AI) is a technical discipline involving computer theory and systems that can solve certain problems by simulating human intelligence. In recent years, due to the rapid development of AI, its application in dentistry has become increasingly widespread. Most studies use AI to process input data, such as photographs or imaging data, to improve the accuracy and efficiency of oral disease diagnosis, treatment, and prognosis. In the field of medical imaging, the most common segmentation neural networks are U-Net [A], V-Net [B], and 3D U-Net [C], which are improvements on the Fully Convolutional Network (FCN). By adopting an end-to-end symmetrical network structure and using an encoder-decoder network structure to continuously aggregate semantic information, they enable better fusion of shallow and deep features, thereby achieving more refined and accurate segmentation.

[0003] The standard artificial joint prosthesis is a stock product and is currently the only product approved for use in China. It consists of two parts: the maxillary fossa prosthesis and the mandibular ramus prosthesis. When used, the specific placement of the prosthesis must be designed preoperatively to adapt to the different maxillary and mandibular shapes of the patient.

[0004] Computer-assisted surgery (CAS) technology is currently used to guide the bone trimming and positioning of standard artificial joint prostheses. Preoperative design and production of digital guides improve surgical accuracy and safety. However, there are certain limitations: the specific placement angle and position of the prosthesis during design is highly dependent on clinical experience, requiring the designer to make small adjustments and repeated verification with the clinician. This invisibly increases the time cost and efficiency of preoperative design. The average time it takes to manually design a prosthetic installation plan is more than 40 minutes.

[0005] In view of this, the present invention provides an artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method to solve the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide an artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method, establish an artificial intelligence-assisted standard artificial joint design automation algorithm, improve preoperative design efficiency, and evaluate the quality of its automatically planned surgical plan to solve the problems in the above-mentioned background technology.

[0007] In order to achieve the above object, the present invention provides the following technical solutions:

[0008] In a first aspect, the present invention provides an artificial intelligence-assisted method for automatically placing a standard artificial temporomandibular joint, comprising the following steps:

[0009] Training is used to extract the 3D reconstruction model of the mandibular and zygomatic arch joint prosthesis placement area;

[0010] Using an improved 3D U-Net neural network, the regions of interest (ROIs) of the mandibular fossa and mandibular ramus are segmented and extracted to obtain the spatial geometry of the prosthesis installation. The ROIs are the three-dimensional markers of the mandibular bone and zygomatic arch where the prosthesis is in direct contact with the force.

[0011] Based on the spatial geometric structure of the prosthesis installation, the point cloud data of the lateral bone surface of the zygomatic arch and the mandibular ramus is extracted, the anatomical key points are identified, and the standard prosthesis fitting surface is generated through geometric fitting and constraint optimization;

[0012] Determine the relative position relationship between the maxillary fossa prosthesis and the mandibular ramus prosthesis, clarify the installation direction, hole definition and assembly reference point; determine the contact area, posture direction and placement position of the prosthesis, and complete the three-dimensional positioning and posture matching of the prosthesis in the patient model as a surgical planning plan.

[0013] As a preferred technical solution of the first aspect of the present invention, the construction logic of the three-dimensional reconstruction model is:

[0014] Extracting maxillofacial CT image data of the patient, wherein the maxillofacial CT image data is DICOM format image data; the maxillofacial CT image data includes 2D sequence views of cross-section, sagittal and coronal planes;

[0015] Maxillofacial CT image data is 3D reconstructed using 3D reconstruction software to generate a 3D reconstruction model; the 3D reconstruction software is ProPlan CMF;

[0016] The three-dimensional reconstruction model is trained to extract the target areas of the mandible and zygomatic arch for placing the joint prosthesis, and shape modeling is performed in the target areas to obtain shape analysis results.

[0017] As a preferred technical solution of the first aspect of the present invention, the logic for obtaining the shape analysis results is:

[0018] Acquire maxillofacial CT image data, perform threshold segmentation, binarization, and 3D reconstruction on the maxillofacial CT image data, extract the target areas of the mandible and zygomatic arch in the oral and maxillofacial region for placement of joint prostheses, and annotate and mesh the target areas to form a polygonal vertex point set;

[0019] Generalized Prototype Analysis is performed on the grid point sets of multiple samples to eliminate differences in position, scale, and rotation, ensuring that the shape data are aligned in a unified reference frame. Based on the aligned vertex data, the principal component analysis method is used to extract eigenvectors.

[0020] Shape analysis results are extracted based on feature vector combinations and used for automatic prosthesis design and placement tasks.

[0021] As a preferred technical solution of the first aspect of the present invention, the construction logic based on the improved 3D U-Net is:

[0022] Based on the 3D reconstruction model and shape analysis results, the regions of interest of the mandible and zygomatic arch were extracted; the image data of the regions of interest were converted into NIFTI format image data and input into the improved 3D U-Net network;

[0023] We designed an improved 3D U-Net neural network structure, which adopts a symmetrical encoder-decoder architecture, combines multi-layer convolution, max pooling, and deconvolution operations, adds an additional feature fusion layer between the encoder and decoder, and adds a batch normalization (BN) layer after each convolution layer.

[0024] Design an optimized loss function, combine the LR region loss and the LB boundary loss to set the optimized loss function, and form a weighted combination of the two to form the final loss function, which optimizes segmentation accuracy, accelerates training and prevents overfitting;

[0025] Output the key structure labels of the regions of interest of the mandible and zygomatic arch respectively.

[0026] As a preferred technical solution of the first aspect of the present invention, the preoperative planning and design logic of the prosthesis is:

[0027] A standard temporomandibular joint prosthesis model and its corresponding geometric parameters were selected. Based on the results of rigid region of interest and shape analysis, the contact areas between the mandibular and maxillary surfaces and the prosthesis were extracted. The implantation path, direction, and spatial posture were determined by combining the geometric constraint relationships between the prosthesis components. The positioning and alignment of the prosthesis were automatically planned to obtain a prosthesis matching solution.

[0028] As a preferred technical solution of the first aspect of the present invention, the bone surface point cloud modeling and prosthesis contact surface fitting process is:

[0029] Construct a directed bounding box based on the bone surface point cloud data to clarify the directions of the sagittal, coronal and horizontal planes;

[0030] The sagittal symmetry plane of the patient's maxillofacial region is calculated based on the directed bounding box of the bone surface, and the left-right side conversion is realized to identify the anatomical key points of the region of interest, including the most prominent point of the zygomatic arch, the edge of the bony external auditory canal, and the condyle, coracoid process, and mandibular angle of the mandibular region.

[0031] Based on anatomical key points, a point cloud subset containing the prosthesis contact area is selected. The selected point cloud subset is projected onto the bounding box plane, and the point closest to the bone surface in each sub-area is found through meshing. The least squares method is used to fit the point set and solve the contact surface plane equation.

[0032] Geometric constraints for bone-prosthesis contact were set, including the distance constraint between the posterior edge of the glenoid fossa prosthesis and the bony external auditory canal. The top surface of the glenoid fossa prosthesis was parallel to the orbitoauricular plane and aligned as closely as possible with the lateral surface of the zygomatic arch. Based on the geometric constraints, the lateral section of the zygomatic arch and the upper edge of the orbitoauricular plane were generated. The prosthesis rotation matrix was determined by the intersection of the two planes, corresponding to the placement angle of the glenoid fossa prosthesis in three-dimensional space.

[0033] As a preferred technical solution of the first aspect of the present invention, the relative posture registration of the prosthesis and the automated surgical planning include:

[0034] Set up a standardized coordinate system for the prosthesis. Set up a local coordinate system for the maxillary fossa prosthesis and the mandibular ramus prosthesis separately: the contact surface is the XY plane, the Z axis points into the bone, and the center point of the key hole or anatomical feature point is the coordinate origin; ensure that the coordinate axis is aligned with the prosthesis geometry;

[0035] Determine the spatial corresponding points between the prostheses. The lowest point of the concave surface of the maxillary fossa prosthesis should coincide with the highest point of the convex surface of the mandibular ramus prosthesis. Convert them to the same world coordinate system and set the simultaneous common point constraint.

[0036] The spatial posture constraint is introduced, and three plane posture constraints are constructed based on the fact that the upper edge plane of the maxillary prosthesis should be parallel to the orbitoauricular plane and the prosthesis contact surface should be parallel to the lateral surface of the bone.

[0037] Based on the simultaneous common point constraints and three plane posture constraints, a six-variable linear equation system is formed, and the rotation matrix and translation vector of the upper and lower prostheses are obtained based on the six-variable linear equation system;

[0038] Apply rotation and translation parameters to the STL data of the standard prosthesis model to automatically and accurately place the prosthesis at the corresponding position of the patient's bone model; output the STL data as a preoperative digital planning file, which can be used for surgical navigation, guide design or 3D printing.

[0039] As a preferred technical solution of the first aspect of the present invention, the design logic of the maxillary fossa prosthesis is:

[0040] The prosthesis contact area was extracted from the point cloud of the right zygomatic arch area in the 3D model. The range of the screening area was determined as the initial analysis object. The directed bounding box of the point cloud in this area was recalculated to extract the main axis and boundary information.

[0041] The boundary information of the contact area between the prosthesis and the bone surface is framed to obtain the lateral section;

[0042] The lateral section was divided into several small squares, and the anatomical key points closest to the bone surface were found in each square. All these points were collected as representative points of the zygomatic arch bone surface, and the section equation was fitted using the least squares method as the lateral fitting surface.

[0043] In the same way, generate the upper edge fitting surface in the upper boundary direction;

[0044] The intersection of the lateral fitting surface and the upper edge fitting surface is used as the prosthesis placement direction, and the spatial posture of the prosthesis is determined based on the prosthesis placement direction. The final output is the rotation matrix of the prosthesis relative to the patient's three-dimensional model, which is used for precise prosthesis placement.

[0045] As a preferred technical solution of the first aspect of the present invention, the mandibular ramus prosthesis placement logic is:

[0046] The lateral surface of the mandible is screened from the point cloud of the region of interest, the directed bounding box is calculated, the condyle, coracoid process and mandibular angle are determined, the mandibular lateral section is generated and the bone-prosthesis fitting surface is optimized;

[0047] Determine the placement angle through the plane and line direction and the rotation matrix, determine the standardized prosthesis coordinate system based on the prosthesis position under geometric constraints, and set the contact surface and hole center;

[0048] By using the common point constraint (the lowest point of the concave surface of the glenoid fossa coincides with the highest point of the convex surface of the mandibular ramus) and the plane constraint, a linear equation system is established to solve the translation matrix.

[0049] In a second aspect, the present invention provides a computer program product stored on a computer-readable medium, comprising a computer-readable program, which, when executed on an electronic device, provides a user input interface to implement the method of the first aspect.

[0050] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0051] This patented invention establishes an AI-assisted standard artificial joint design algorithm, enabling automated placement of standard artificial joint prostheses before surgery. This reduces the time it takes to manually design a prosthesis from 40 minutes to 30 seconds, significantly improving preoperative planning efficiency. The quality of the resulting surgical plan is no different from that of a manual design. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0053] Figure 1This is a flowchart of the oral and maxillofacial key tissue segmentation based on the three-dimensional reconstruction model and improved 3D U-Net of the present invention.

[0054] Figure 2 This is the segmentation and visualization diagram of key oral and maxillofacial tissues of the present invention.

[0055] Figure 3 This is the overall flow chart of the surgical planning of the present invention.

[0056] Figure 4 This is a flow chart for the placement of the maxillary fossa prosthesis of the present invention.

[0057] Figure 5 The figure is a flow chart of the placement of the mandibular ramus prosthesis of the present invention.

[0058] Figure 6 Schematic diagram of the contact points of the two parts of the prosthesis of the present invention. DETAILED DESCRIPTION

[0059] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of this disclosure will be more comprehensive and complete, and will fully convey the concepts of the example embodiments to those skilled in the art. The accompanying drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. Identical reference numerals in the figures indicate identical or similar parts, and thus repeated descriptions thereof will be omitted.

[0060] In addition, the described features, structures or characteristics can be combined in any suitable manner in one or more example embodiments. In the following description, many specific details are provided to provide a full understanding of the example embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure can be practiced while omitting one or more of the specific details, or methods, components, steps, etc. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0061] Artificial temporomandibular joint replacement is an important means of treating end-stage joint diseases. The installation of a standard temporomandibular joint prosthesis usually requires the removal of a large amount of bone from the patient's zygomatic arch and the outer side of the mandibular ramus to ensure the correct installation of the prosthesis. If the prosthesis is not placed properly, it is very likely to damage the external auditory canal, alveolar nerve or skull base. Therefore, preoperative planning of the placement of a standardized prosthesis is very important. The process of computer three-dimensional design relies heavily on clinical experience. The designer needs to repeatedly check and confirm the plan with the doctor, which requires design and production time. This patent uses the automation, strong learning ability and high flexibility of artificial intelligence to optimize and solve the limitations of current preoperative design, which can improve the formulation of personalized treatment plans for standard artificial joint prostheses and the efficiency of preoperative design.

[0062] Example 1

[0063] like Figure 1 As shown, the present invention provides an artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method, comprising the following steps:

[0064] Step S1: training a three-dimensional reconstruction model for extracting the mandibular and zygomatic arch joint prosthesis placement area;

[0065] Specifically, the construction logic of the 3D reconstruction model is as follows:

[0066] Extracting maxillofacial CT image data of the patient, wherein the maxillofacial CT image data is DICOM format image data; the maxillofacial CT image data includes 2D sequence views of cross-section, sagittal and coronal planes;

[0067] Maxillofacial CT image data is 3D reconstructed using 3D reconstruction software to generate a 3D reconstruction model; the 3D reconstruction software is ProPlan CMF;

[0068] The 3D reconstruction model is trained to extract the target areas of the mandible and zygomatic arch for placement of the joint prosthesis, and shape modeling is performed in the target areas to obtain shape analysis results;

[0069] Specifically, the target area includes the areas of interest corresponding to the mandible and zygomatic arch. Based on each target area, the three-dimensional reconstructed model supports interactive browsing of multi-plane views (MPR); through the segmentation function equipped in the software, the maxillary fossa prosthesis and the mandibular ramus prosthesis are separately visualized through shape analysis results, thereby simplifying the visual information of the joint area, which is helpful for the next step of prosthesis placement.

[0070] Furthermore, the target area morphology of the mandible and zygomatic arch for placing the joint prosthesis was extracted from the overall three-dimensional reconstructed model of the maxillofacial region, and the average shape of the maxillary fossa prosthesis and the mandibular ramus prosthesis was obtained. The geometric information was described in the form of a vertex point set of a polygonal mesh. 100 maxillofacial CT image data (including labeled data) were used as training data. The data were aligned using the generalized Prokaryotic Analysis (GPA) method, and the principal component analysis (PCA) method was used to extract the main shape variation patterns, and the shape analysis results were output.

[0071] Align data: use Generalized Platts Analysis (GPA) to eliminate position / rotation differences between samples;

[0072] Dimensionality reduction modeling: Principal component analysis (PCA) was used to extract the main shape variation patterns and construct the average mandibular / zygomatic arch shape.

[0073] More specifically, the logic for obtaining the shape analysis results is as follows:

[0074] Acquire maxillofacial CT image data, perform threshold segmentation, binarization, and 3D reconstruction on the maxillofacial CT image data, extract the target areas of the mandible and zygomatic arch in the oral and maxillofacial region for placement of joint prostheses, and annotate and mesh the target areas to form a polygonal vertex point set;

[0075] Maxillofacial CT image data was preprocessed using threshold segmentation. A threshold interval was selected, and the initial DICOM image was binarized and isosurfaces extracted to visualize the bone tissue in the oral and maxillofacial region. The mandibular ramus and zygomatic joint area were then precisely annotated, and an eight-vertex hexahedral mesh was extracted as the basic data for shape analysis.

[0076] Generalized Proteus analysis is performed on the grid point sets of multiple samples to eliminate the differences in position, scale and rotation, ensure that the shape data are aligned in a unified reference system, and extract the eigenvectors based on the aligned vertex data using the principal component analysis method.

[0077] In order to eliminate the shape deviation caused by differences in position, size and rotation between different samples, the generalized Procrustes Analysis (GPA) method was used to align the data. On this basis, the principal component analysis (PCA) method was used to extract the main change patterns of the shape. Specifically, PCA is achieved by calculating the covariance matrix of the data and solving its eigenvalues ​​and eigenvectors. The eigenvectors represent the main variation directions of the shape data (ie, principal components), while the eigenvalues ​​quantify the degree of variation in these directions. Based on the size of the eigenvalues, the data can be reduced in dimensionality, and several principal components with the largest weights can be retained to construct a three-dimensional reconstruction model. Taking the mandible as an example, the constructed three-dimensional reconstruction model is as follows Figure 1 shown.

[0078] Extract shape analysis results based on feature vector combinations for automatic prosthesis design and placement tasks;

[0079] In the constructed 3D reconstruction model, each shape can be approximately reconstructed using a linear combination of the mean shape and the principal components. This approach not only significantly improves computational efficiency but also effectively captures key shape features, providing a reliable theoretical foundation and technical support for subsequent prosthesis design and shape analysis.

[0080] Step S2: Segment and extract the regions of interest of the glenoid fossa and mandibular ramus based on the improved 3D U-Net neural network to obtain the spatial geometric structure of the prosthesis installation;

[0081] During prosthesis installation, the three-dimensional areas of the mandible and zygomatic arch where the prosthesis is in direct contact with the prosthesis are marked as regions of interest, and the other areas are marked as non-regions of interest.

[0082] Specifically, the construction logic based on the improved 3D U-Net is:

[0083] Based on the 3D reconstruction model and shape analysis results, the regions of interest of the mandible and zygomatic arch were extracted; the image data of the regions of interest were converted into NIFTI format image data and input into the improved 3D U-Net network;

[0084] We designed an improved 3D U-Net neural network structure, which adopts a symmetrical encoder-decoder architecture, combines multi-layer convolution, max pooling, and deconvolution operations, adds an additional feature fusion layer between the encoder and decoder, and adds a batch normalization (BN) layer after each convolution layer.

[0085] Design and optimize loss function, combined with regional loss L R and boundary loss L BSet the optimized loss function and combine the two weightedly to form the final loss function to optimize segmentation accuracy, accelerate training and prevent overfitting;

[0086] Output the key structure labels of the regions of interest of the mandible and zygomatic arch respectively.

[0087] A 3D reconstructed model is constructed using key structure labels and displayed in the user interface in cross-sectional, sagittal, coronal, and 3D views, providing intuitive observation of the temporomandibular joint area structure and preoperative intervention reference.

[0088] More specifically, based on the shape extraction results of the 3D reconstructed model, CT cross-sectional slices encompassing the temporomandibular joint area and the mandibular ramus (including the condyle) were extracted as regions of interest (ROI). Subsequently, the image slices in the ROI area were converted into NIFTI format for input into an improved 3D U-Net neural network. This segmentation network was used to train the dataset, thereby achieving accurate segmentation of the bone tissue anatomical structure in the temporomandibular joint area. This method is expected to significantly improve segmentation precision and accuracy, and promote the automation of the segmentation process.

[0089] The improved 3D U-Net we used is a fully convolutional neural network with symmetrical encoding and decoding processes. It uses multi-layer convolution operations to extract features from shallow layers to deep layers and perform reverse restoration. During the encoding process, 3DU-Net uses downsampling, with the input image slice size set to 16×256×256, and uses the ReLU function as the activation function. Each layer of the network consists of a convolutional layer with a kernel size of 3×3×3 and a stride of 2. Each convolutional layer is followed by a batch normalization (BN) layer to normalize the features, accelerating network convergence and preventing overfitting. Each layer is terminated by a 3D max pooling layer with a kernel size of 2×2×2 and a stride of 2, halving the size of the feature map and doubling the number of channels.

[0090] During decoding, the 3D U-Net employs upsampling and deconvolution using the same convolutional layers used for downsampling. Each convolutional layer is followed by a batch normalization layer to restore the input image slice size. Finally, a convolutional layer with a kernel size of 1×1×1 and a stride of 1 is used to output the image segmentation result. Furthermore, because the U-Net network aggregates semantic information at the expense of spatial information, to minimize the loss of spatial feature information, a feature fusion layer is established between the symmetrical convolutional modules in the encoding and decoding steps to better integrate contextual shape information.

[0091] The ROI region segmentation process based on the 3D reconstruction model combined with the 3D U-Net deep neural network is as follows Figure 2 This method combines the shape prior knowledge of the 3D reconstruction model with the powerful feature extraction capability of 3D U-Net to achieve high-precision segmentation of bone tissue in the temporomandibular joint area, providing a basis for the design of the prosthesis placement in the next step.

[0092] Because the regional boundary is more important in the segmentation of oral and maxillofacial anatomical tissues, the loss function Loss of the segmentation process adopts the regional loss L R and boundary loss L B Combined evaluation criteria. R and L B Combined with the corresponding weights, the loss function in the segmentation process is superimposed and minimized through repeated iterations. Finally, the segmentation results based on the improved 3D U-Net neural network are output to generate labels for each key tissue. Accurate segmentation of key oral and maxillofacial anatomical tissues such as the maxillary bone, mandibular left and right zygomatic bones, and left and right maxillary sinuses is achieved. By visualizing these key tissues in three dimensions and displaying the cross-sectional, sagittal, and coronal views of the oral and maxillofacial CT images and the 3D view of the temporomandibular joint area ROI in the user interface, such as Figure 2 shown.

[0093] After completing the above-mentioned TMJ area ROI extraction, key tissue segmentation and three-dimensional visualization based on maxillofacial CT images, the next step of prosthesis placement can be started.

[0094] To further explain, the logic of prosthesis preoperative planning and design is:

[0095] A standard temporomandibular joint prosthesis model and its corresponding geometric parameters were selected. Based on the results of rigid region of interest and shape analysis, the contact areas between the mandibular and maxillary surfaces and the prosthesis were extracted. The implantation path, direction, and spatial posture were determined by combining the geometric constraint relationships between the prosthesis components. The positioning and alignment of the prosthesis were automatically planned to obtain a prosthesis matching solution that took into account both functional and structural adaptability.

[0096] As an example, the most commonly used standard temporomandibular joint prosthesis in clinical practice is the Biomet temporomandibular joint replacement system, a stock device that consists of two components: a glenoid fossa and a mandibular ramus. The mandibular ramus component is composed of a chromium-cobalt alloy with a titanium coating to facilitate osseointegration. It is available in three lengths: small (45 mm), medium (50 mm), and large (55 mm), and in three basic styles: standard, offset, and narrow. The glenoid fossa component is composed of ultra-high molecular weight polyethylene and lacks osseointegration with the zygomatic bone but is fixed to the lateral border of the zygomatic arch with bone screws. It is available in three sizes: small, medium, and large. The mandibular ramus component is fixed to the mandible with 2.7 mm screws, and the glenoid fossa component is fixed to the lateral border of the zygomatic arch with 2.0 mm screws. Preoperative planning in this study utilized the most commonly used prosthetic models in clinical practice, including a medium narrow mandibular ramus prosthesis and a small maxillary glenoid fossa prosthesis. The design concept of this study is to locate and divide the region of interest (ROI), extract the planes of the bone surfaces of the maxillary and mandibular parts that match the prosthesis, and then determine the position and posture of the prosthesis implant based on the geometric constraint relationship between the two components of the prosthesis. The overall process of preoperative planning is as follows: Figure 3 shown.

[0097] Step S3: Extract the lateral bone surface point cloud data of the zygomatic arch and mandibular ramus based on the spatial geometric structure of the prosthesis installation, identify the anatomical key points, and generate the standard prosthesis fitting surface through geometric fitting and constraint optimization;

[0098] Specifically, the process of bone surface point cloud modeling and prosthesis contact surface fitting is as follows:

[0099] Construct a directed bounding box based on the bone surface point cloud data to clarify the directions of the sagittal, coronal and horizontal planes;

[0100] The sagittal symmetry plane of the patient's maxillofacial region is calculated based on the directed bounding box of the bone surface, and the left-right side conversion is realized to identify the anatomical key points of the region of interest, including the most prominent point of the zygomatic arch, the edge of the bony external auditory canal, and the condyle, coracoid process, and mandibular angle of the mandibular region.

[0101] Based on anatomical key points, a point cloud subset containing the prosthesis contact area is selected. The selected point cloud subset is projected onto the bounding box plane, and the point closest to the bone surface in each sub-area is found through meshing. The least squares method is used to fit the point set and solve the contact surface plane equation.

[0102] Geometric constraints for bone-prosthesis contact were set. The distance between the posterior edge of the glenoid fossa prosthesis and the bony external auditory canal was greater than 2 mm. The top surface of the glenoid fossa prosthesis was parallel to the orbitoauricular plane and aligned as closely as possible with the lateral surface of the zygomatic arch. Based on the geometric constraints, the lateral zygomatic arch section and the upper edge of the orbitoauricular plane were generated. The prosthesis rotation matrix was determined by the intersection of the two planes, corresponding to the placement angle of the glenoid fossa prosthesis in three-dimensional space.

[0103] Specifically, bone reconstruction is performed based on the patient's maxillofacial CT data, generating a stereolithography (STL) model that describes the surface geometry of the patient's maxillofacial bones using a triangular mesh. The STL model M of the patient's maxillofacial region is then filled with a point cloud. Principal component analysis (PCA) is then used to calculate the oriented bounding box (OBB) of the point cloud, generating six bounding box planes. The main steps of this method are as follows:

[0104] Fill the STL model of the patient's maxillofacial area with point cloud to obtain a point cloud dataset containing n three-dimensional points, denoted as P = {p1, p2, ... p n}, where each point p i It can be expressed as p i =(x i ,y i ,z i ) T Then the centroid of the point cloud can be expressed as Center the point cloud to get the dataset P C , where each point is p Ci =p i -p C , then the covariance matrix of the data set is The three eigenvectors of the matrix are the three main axes of the directed bounding box OBB, which represent the vertical axis, sagittal axis and coronal axis of the patient model, respectively, and are denoted as e v 、e s and e c By projecting the point cloud onto each principal axis, the upper boundary b of each principal axis is calculated. max =(x max ,y max ,z max ) T and the lower boundary b min =(x min ,y min ,z min ) T , the OBB of the model can be calculated.

[0105] Because the anatomical structure of the human maxillofacial region is bilaterally symmetrical, the design of left and right temporomandibular joint prostheses can be made equivalent through symmetric transformation. According to the definition of OBB, the normal vector of the sagittal symmetry plane of the model is the coronal axis e c , then the equation of the sagittal symmetry plane can be expressed as e c ·(pp C )=0. For any point p on the left side of the model l , and its corresponding symmetric point p rThe coordinates can be calculated using the following formula:

[0106] p r =p l -2e c ·(p l -p C )·e c

[0107] Through the above steps, the bounding box and sagittal symmetry plane of the patient's maxillofacial model were calculated, laying the foundation for subsequent bone surface extraction. Due to symmetry, the design of the left TMJ prosthesis can be transferred to the right side using Equation 1. Therefore, the following discussion will only focus on the design of the right TMJ prosthesis.

[0108] Step S4: Determine the relative position relationship between the maxillary fossa prosthesis and the mandibular ramus prosthesis, clarify the installation direction, hole position definition and assembly reference point; determine the contact area, posture direction and placement position of the prosthesis, and complete the three-dimensional positioning and posture matching of the prosthesis in the patient model as a surgical planning plan.

[0109] It should be noted that after completing the bone surface matching and contact surface fitting of the maxillary and mandibular prostheses, the relative position and posture of the two parts of the prosthesis in three-dimensional space are calculated through spatial geometric registration methods to achieve fully automatic output of the preoperative plan. Prosthesis relative posture registration and automated surgical planning include:

[0110] Set up a standardized coordinate system for the prosthesis. Set up a local coordinate system for the maxillary fossa prosthesis and the mandibular ramus prosthesis separately: the contact surface is the XY plane, the Z axis points into the bone, and the center point of the key hole or anatomical feature point is the coordinate origin; ensure that the coordinate axis is aligned with the prosthesis geometry;

[0111] Determine the spatial corresponding points between the prostheses. The posterior and lateral 1 / 3 of the concave surface of the maxillary fossa prosthesis should coincide with the highest point of the convex surface of the condylar head of the mandibular ramus prosthesis. Convert them to the same world coordinate system and set simultaneous common point constraints.

[0112] The spatial posture constraint is introduced, and three plane posture constraints are constructed based on the fact that the upper edge plane of the maxillary prosthesis should be parallel to the orbitoauricular plane and the prosthesis contact surface should be parallel to the lateral surface of the bone.

[0113] Based on the simultaneous common point constraints and three plane posture constraints, a six-variable linear equation system is formed, and the rotation matrix and translation vector of the upper and lower prostheses are obtained based on the six-variable linear equation system;

[0114] Apply rotation and translation parameters to the STL data of the standard prosthesis model to automatically and accurately place the prosthesis at the corresponding position of the patient's bone model; output the STL data as a preoperative digital planning file, which can be used for surgical navigation, guide design or 3D printing.

[0115] Specifically, if Figure 4 As shown, the design logic of the maxillary fossa prosthesis is:

[0116] The prosthesis contact area was extracted from the point cloud of the right zygomatic arch area in the 3D model. The range of the screening area was determined as the initial analysis object. The directed bounding box of the point cloud in this area was recalculated to extract the main axis and boundary information.

[0117] The boundary definition information of the contact area between the prosthesis and the bone surface is framed to obtain the lateral section; the boundary definition information includes the front boundary definition, the upper boundary definition, and the posterior boundary definition, specifically including:

[0118] Anterior limit: take the most prominent point in the point cloud as the anterior edge of the zygomatic arch, and remove all point cloud data in front of this point;

[0119] Upper bound limitation: Search upward from the front bound to find the point closest to the upper bound of the bounding box and remove points above the upper bound;

[0120] Posterior boundary definition: Slice upward from the bottom of the bounding box, calculate the point density mutation point, and identify it as the bony edge of the external auditory canal;

[0121] The lateral section was divided into several small squares, and the anatomical key points closest to the bone surface were found in each square. All these points were collected as representative points of the zygomatic arch bone surface, and the section equation was fitted using the least squares method as the lateral fitting surface.

[0122] In the same way, generate the upper edge fitting surface in the upper boundary direction;

[0123] The intersection of the lateral fitting surface and the upper edge fitting surface is used as the prosthesis placement direction, and the spatial posture of the prosthesis is determined based on the prosthesis placement direction. The final output is the rotation matrix of the prosthesis relative to the patient's three-dimensional model, which is used for precise prosthesis placement.

[0124] For example, in the design of a right maxillary fossa prosthesis, we select the point cloud of the front 50%, bottom 50%, and right 20% of the bounding box, which contains the anatomical structure of the zygomatic arch. The resulting point set is shown in Equation 2.

[0125]

[0126] Recalculate the directed bounding box of the filtered point cloud, whose three main axes and upper and lower boundaries are recorded as and

[0127] The bounding box plane corresponding to the sagittal plane of the model in the point cloud is used as the initial plane, and the point closest to the initial plane is taken as the reference point. This point is the most prominent part of the zygomatic arch and is marked as p. zarch .

[0128]

[0129] This part is used as the front boundary of the maxillary fossa prosthesis bounding box, and the points before the front boundary are discarded, that is, Then find the point closest to the upper bound of the bounding box near the front bound As shown in the following formula:

[0130]

[0131] Will The vertical axis coordinate is used as the new upper bound, and the points above the upper bound are discarded, that is,

[0132] From the bottom of the bounding box Initially, slice the slices with a step length of 0.625 mm and calculate S for each slice. i The number of points N(S i ), and compared with the number of points in the previous slice, when the ratio of the number of points increases exceeds a certain threshold ε, that is, When the cut surface intersects with the bony edge of the external auditory canal, the last cut surface before the intersection is marked as S. k , and its plane equation is:

[0133]

[0134] Seeking S k The intersection G with the STL model M is S k ∩M, whose edge is calculate The most prominent point in the sagittal axis is marked as p ear This point satisfies:

[0135]

[0136] The point 2 mm forward from this point is taken as the posterior limit of the maxillary fossa prosthesis, and the point beyond the posterior limit is discarded.

[0137] Generation of the prosthesis-bone contact plane

[0138] The filtered target area point cloud is recorded as Recalculate The three main axes and upper and lower boundaries of the directed bounding box are recorded as and The lateral sagittal plane of the bounding box Divide it into several 1mm×1mm squares, and use S ij If S ij Point p in ij=(x ij ,y ij ,z ij ) T Satisfy the constraints:

[0139]

[0140] Project all points in the point cloud to the surface, and select the points projected to each square whose distance is closest and less than a certain threshold. These points represent the characteristics of the patient's zygomatic arch surface, denoted as As shown in Formula 5:

[0141]

[0142] Point set P up The number of points in is m, and the coordinates of the kth point are Based on these m points, the closest distance plane S is generated up =n up ·p+d up , this plane is the lateral section closest to the zygomatic arch surface. up The solution can be obtained by using the least squares method, and the matrix A and vector b are constructed as shown in Formula 6.

[0143]

[0144] Then S up The normal vector is: n up =(A T A) -1 A T b=(a up ,b up ,c up ) T .

[0145] The constant term is:

[0146] Get S up Then, take the upper cross section of the bounding box As the reference plane for projection, repeat the above operation to obtain the upper edge of the orbital ear plane Serves as the upper limit of the maxillary fossa prosthesis.

[0147] Zygomatic arch lateral section S up and the upper edge of the orbitoauricular plane The intersection line is a straight line L up , that is, in plane S up Upper straight line L upThe maxillary fossa prosthesis is designed in the direction of the prosthesis. Since the direction of the contact plane between the prosthesis and the bone and the direction of the fixing screw are already determined, the posture of the maxillary fossa prosthesis can be uniquely determined, that is, the rotation matrix R of the maxillary fossa prosthesis coordinate system relative to the world coordinate system of the patient model up can be uniquely determined.

[0148] like Figure 5 As shown, the placement logic of the mandibular ramus prosthesis is:

[0149] The lateral surface of the mandible was screened from the point cloud of the region of interest, the directed bounding box was calculated, the condyle, coracoid process and mandibular angle were determined, the mandibular lateral section was generated and the bone-prosthesis fitting surface was optimized.

[0150] The placement angle is determined by the plane and line direction and the rotation matrix, the standardized prosthesis coordinate system is determined based on the prosthesis position of geometric constraints, and the contact surface and hole center are set.

[0151] By using the common point constraint (the lowest point of the concave surface of the glenoid fossa coincides with the highest point of the convex surface of the mandibular ramus) and the plane constraint, a linear equation system is established to solve the translation matrix.

[0152] Because the mandibular anatomy is much simpler than the maxillary anatomy and the mandibular ramus prosthesis only needs to fit the lateral surface of the mandibular ramus, the design rules for the mandibular ramus prosthesis are relatively simple and are summarized as follows:

[0153] The STL model of the patient's mandibular area was filled with point clouds, and the principal component analysis (PCA) algorithm was used to calculate the directed bounding box of the point cloud and generate 6 bounding box planes.

[0154] Taking the right mandibular prosthesis design as an example, the point cloud of the right 20% area of ​​the bounding box is selected. This part of the point cloud contains the anatomical structure of the lateral surface of the mandible. The filtered point cloud set is recorded as Recalculate the directed bounding box of the filtered point cloud, whose three main axes and upper and lower boundaries are recorded as and

[0155] Take the point p in the point cloud that is closest to the top surface of the bounding box con (condyle), the point p closest to the front of the bounding box cor (coracoid process) and the point p closest to the back of the bounding box ang (mandibular angle), as shown in the following formula.

[0156]

[0157] The plane S formed by these three points jaw =n jaw ·p+d jaw Represents the lateral section of the mandibular surface, the line L from the mandibular angle to the condylejaw =p con -p ang It represents the direction of the mandibular ramus prosthesis. jaw The normal vector is n jaw =(p con -p cor )×(p con -p ang ), the constant term is d jaw =-n jaw ·p con .

[0158] Because the condyle is generally more convex, the generated plane S jaw It may not fit perfectly on the outer side of the mandible. The first 20% of the point cloud close to the top surface of the bounding box is removed, and the remaining point cloud is S jaw As the reference plane for projection, perform step 5 in maxillary fossa prosthesis design to obtain a plane that fits better with the outer surface of the mandible.

[0159] and L jaw exist Projection on Next, you can Upper straight line The mandibular ramus prosthesis is designed in the direction of the prosthesis. Because the direction of the contact plane between the prosthesis and the bone and the direction of the fixing screw are already determined, the posture of the mandibular ramus prosthesis can be uniquely determined, that is, the rotation matrix R of the maxillary fossa prosthesis coordinate system relative to the world coordinate system of the patient model jaw can be uniquely determined.

[0160] Prosthesis Position Determination Based on Geometric Constraints

[0161] Standardize the models of the maxillary fossa and mandibular ramus prostheses. The contact surface between the prosthesis and the zygomatic arch or mandible is defined as the yOz plane, with the direction pointing into the bone as the positive x-axis. For the maxillary fossa prosthesis, the center of the leftmost hole is defined as the origin, and the line connecting the centers of holes at the same horizontal position is defined as the positive y-axis. For the mandibular ramus prosthesis, the center of the lowest hole is defined as the origin, and the line connecting the centers of holes at the same vertical position is defined as the positive z-axis.

[0162] In the standardized coordinate system, the coordinates of the center points of each hole are as follows: Figure 6 The coordinates of the posterior outer 1 / 3 point of the concave surface of the maxillary fossa prosthesis in the maxillary prosthesis coordinate system are p matchup =(7.8,5.2,8.6) TThe distance between the upper edge of the prosthesis and the center point is 3 mm; the coordinate of the highest point of the convex surface of the mandibular ramus prosthesis in the mandibular prosthesis coordinate system is p matchjaw =(3.0,6.0,44.8) T The goal of design is to make these two points coincide so that the joint head and socket are assembled correctly.

[0163] Because the above algorithm has determined the posture of the two parts of the prosthesis, that is, the rotation matrix R of the prosthesis coordinate system relative to the world coordinate system up and R jaw can be uniquely determined, so we only need to determine the translation matrix t of the two parts of the prosthesis relative to the world coordinate system up =(x up ,y up ,z up ) T and t jaw =(x jaw ,y jaw ,z jaw ) T , a unique surgical plan can be generated. The constraints on these six parameters are the common point constraint mentioned above, the plane constraint of the upper edge of the maxillary prosthesis, and the constraints on the planes where the two parts of the prosthesis are located. The normal vectors of all planes are assumed to be normalized. The constraint equations can be listed as follows:

[0164]

[0165] Substitute R up =(u1,u2,u3) T , R jaw =(w1,w2,w3) T , where u1,u2,u3,w1,w2,w3 are all 1×3 vectors. Converting the above equations into a linear system, we get:

[0166]

[0167] Because the three constraint planes S up 、 They are not parallel to each other, so the normal vectors of the three planes are linearly independent. Therefore, the coefficient matrix and augmented matrix rank of the above linear equations are both 6. It can be seen that the linear equations have a unique solution, thus uniquely determining the placement position of the two parts of the prosthesis. up and t jaw .

[0168] At this point, the translation and rotation matrices of the maxillary fossa prosthesis and the mandibular ramus prosthesis relative to the world coordinate system have all been obtained. The standardized prosthesis STL model is imported and the translation and rotation transformation is performed. as well as Place the prosthesis in the corresponding position in the patient model to complete the entire surgical planning process.

[0169] After the design is completed, it is necessary to ensure that the maxillary fossa prosthesis is not too close to the bony external auditory canal, and that the lower edge of the mandibular ramus prosthesis does not exceed the mandibular angle. If these conditions are not met, the design must be redesigned or a different prosthesis model must be used.

[0170] Example 2

[0171] This example, based on Example 1, verifies the quality and efficiency of the planning algorithm by comparing planning time and the area of ​​contact between the prosthesis's outer surface and the zygomatic arch or mandibular outer surface. The experimental data, obtained from the Ninth People's Hospital affiliated with Shanghai Jiao Tong University School of Medicine, includes STL models of the maxilla and mandible of 12 patients, as well as a Biomet standard temporomandibular joint prosthesis model, consisting of a small maxillary fossa prosthesis and a medium-narrow mandibular ramus prosthesis.

[0172] Standardized artificial joint replacement preoperative design was performed based on the maxillofacial CT data of 12 patients in the validation set. The contact areas of the maxillary fossa prosthesis, mandibular ramus prosthesis and bone tissue between the automated design group and the doctor-designed group were compared.

[0173] Grade evaluation: Senior physicians with extensive clinical experience will evaluate the automatically generated surgical plans. The specific evaluation criteria are as follows:

[0174] Advantages: No adjustment is required, fully compliant with design principles;

[0175] Good: Conforms to the design principles but requires fine-tuning of the prosthesis position;

[0176] Unqualified: The output result is completely unusable;

[0177] All plans evaluated as excellent or good are qualified plans.

[0178] The results showed that the design time for all cases was less than 30 seconds, with an average of approximately 21 seconds, demonstrating the algorithm's sufficient time efficiency. There was no significant difference in the prosthesis-bone contact area between the automatically planned and manually planned prosthesis-bone contact areas. All plans were qualified, with four rated "excellent" and 20 rated "good" requiring fine-tuning.

[0179] The artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method provided in an embodiment of the present invention is used to execute a lightweight design method suitable for plastic plate structures provided in the above-mentioned embodiments of the present invention. The specific methods and processes for each structure to realize the corresponding functions included in the artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method are detailed in the above-mentioned embodiment of a lightweight design method suitable for plastic plate structures, and will not be repeated here.

[0180] The above description is merely illustrative of certain exemplary embodiments of the present invention. It goes without saying that those skilled in the art will be able to modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and description are illustrative in nature and should not be construed as limiting the scope of protection of the claims.

Claims

1. An artificial intelligence-assisted method for automatically placing a standard artificial temporomandibular joint, characterized in that: The following steps are involved: Training is used to extract the 3D reconstruction model of the mandibular and zygomatic arch joint prosthesis placement area; Using an improved 3D U-Net neural network, the regions of interest (ROIs) of the mandibular fossa and mandibular ramus are segmented and extracted to obtain the spatial geometry of the prosthesis installation. The ROIs are the three-dimensional markers of the mandibular bone and zygomatic arch where the prosthesis is in direct contact with the force. Based on the spatial geometric structure of the prosthesis installation, the point cloud data of the lateral bone surface of the zygomatic arch and the mandibular ramus is extracted, the anatomical key points are identified, and the standard prosthesis fitting surface is generated through geometric fitting and constraint optimization; Determine the relative position relationship between the maxillary fossa prosthesis and the mandibular ramus prosthesis, clarify the installation direction, hole definition and assembly reference point; determine the contact area, posture direction and placement position of the prosthesis, and complete the three-dimensional positioning and posture matching of the prosthesis in the patient model as a surgical planning plan.

2. The artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method according to claim 1 is characterized in that: The construction logic of the 3D reconstruction model: Extracting maxillofacial CT image data of the patient, wherein the maxillofacial CT image data is DICOM format image data; the maxillofacial CT image data includes 2D sequence views of cross-section, sagittal and coronal planes; Maxillofacial CT image data is 3D reconstructed using 3D reconstruction software to generate a 3D reconstruction model; the 3D reconstruction software is ProPlan CMF; The three-dimensional reconstruction model is trained to extract the target areas of the mandible and zygomatic arch for placing the joint prosthesis, and shape modeling is performed in the target areas to obtain shape analysis results.

3. The artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method according to claim 2, characterized in that: The logic for obtaining the shape analysis results: Acquire maxillofacial CT image data, perform threshold segmentation, binarization, and 3D reconstruction on the maxillofacial CT image data, extract the target areas of the mandible and zygomatic arch in the oral and maxillofacial region for placement of joint prostheses, and annotate and mesh the target areas to form a polygonal vertex point set; Generalized Prototype Analysis is performed on the grid point sets of multiple samples to eliminate differences in position, scale, and rotation, ensuring that the shape data are aligned in a unified reference frame. Based on the aligned vertex data, the principal component analysis method is used to extract eigenvectors. Shape analysis results are extracted based on feature vector combinations and used for automatic prosthesis design and placement tasks.

4. The artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method according to claim 3 is characterized in that: The construction logic based on the improved 3D U-Net is: Based on the 3D reconstruction model and shape analysis results, the regions of interest of the mandible and zygomatic arch were extracted; the image data of the regions of interest were converted into NIFTI format image data and input into the improved 3D U-Net network; We designed an improved 3D U-Net neural network structure, which adopts a symmetrical encoder-decoder architecture, combines multi-layer convolution, max pooling, and deconvolution operations, adds an additional feature fusion layer between the encoder and decoder, and adds a batch normalization (BN) layer after each convolution layer. Design an optimized loss function, combine the LR region loss and the LB boundary loss to set the optimized loss function, and form a weighted combination of the two to form the final loss function, which optimizes segmentation accuracy, accelerates training and prevents overfitting; Output the key structure labels of the regions of interest of the mandible and zygomatic arch respectively.

5. The artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method according to claim 4 is characterized in that: The logic of prosthesis preoperative planning and design is: A standard temporomandibular joint prosthesis model and its corresponding geometric parameters were selected. Based on the results of rigid region of interest and shape analysis, the contact areas between the mandibular and maxillary surfaces and the prosthesis were extracted. The implantation path, direction, and spatial posture were determined by combining the geometric constraint relationships between the prosthesis components. The positioning and alignment of the prosthesis were automatically planned to obtain a prosthesis matching solution.

6. The artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method according to claim 5, characterized in that: The process of bone surface point cloud modeling and prosthesis contact surface fitting is as follows: Construct a directed bounding box based on the bone surface point cloud data to clarify the directions of the sagittal, coronal and horizontal planes; The sagittal symmetry plane of the patient's maxillofacial region is calculated based on the directed bounding box of the bone surface, and the left-right side conversion is realized to identify the anatomical key points of the region of interest, including the most prominent point of the zygomatic arch, the edge of the bony external auditory canal, and the condyle, coracoid process, and mandibular angle of the mandibular region. Based on anatomical key points, a point cloud subset containing the prosthesis contact area is selected. The selected point cloud subset is projected onto the bounding box plane, and the point closest to the bone surface in each sub-area is found through meshing. The least squares method is used to fit the point set and solve the contact surface plane equation. Geometric constraints for bone-prosthesis contact were set, including the distance constraint between the posterior edge of the glenoid fossa prosthesis and the bony external auditory canal. The top surface of the glenoid fossa prosthesis was parallel to the orbitoauricular plane and aligned as closely as possible with the lateral surface of the zygomatic arch. Based on the geometric constraints, the lateral section of the zygomatic arch and the upper edge of the orbitoauricular plane were generated. The prosthesis rotation matrix was determined by the intersection of the two planes, corresponding to the placement angle of the glenoid fossa prosthesis in three-dimensional space.

7. The artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method according to claim 6, characterized in that: Prosthesis relative pose registration and automated surgical planning, including: Set up a standardized coordinate system for the prosthesis. Set up a local coordinate system for the maxillary fossa prosthesis and the mandibular ramus prosthesis separately: the contact surface is the XY plane, the Z axis points into the bone, and the center point of the key hole or anatomical feature point is the coordinate origin; ensure that the coordinate axis is aligned with the prosthesis geometry; Determine the spatial corresponding points between the prostheses. The lowest point of the concave surface of the maxillary fossa prosthesis should coincide with the highest point of the convex surface of the mandibular ramus prosthesis. Convert them to the same world coordinate system and set the simultaneous common point constraint. The spatial posture constraint is introduced, and three plane posture constraints are constructed based on the fact that the upper edge plane of the maxillary prosthesis should be parallel to the orbitoauricular plane and the prosthesis contact surface should be parallel to the lateral surface of the bone. Based on the simultaneous common point constraints and three plane posture constraints, a six-variable linear equation system is formed, and the rotation matrix and translation vector of the upper and lower prostheses are obtained based on the six-variable linear equation system; Apply rotation and translation parameters to the STL data of the standard prosthesis model to automatically and accurately place the prosthesis at the corresponding position of the patient's bone model; output the STL data as a preoperative digital planning file, which can be used for surgical navigation, guide design or 3D printing.

8. The artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method according to claim 7, characterized in that: The design logic of the maxillary fossa prosthesis is: The prosthesis contact area was extracted from the point cloud of the right zygomatic arch area in the 3D model. The range of the screening area was determined as the initial analysis object. The directed bounding box of the point cloud in this area was recalculated to extract the main axis and boundary information. The boundary information of the contact area between the prosthesis and the bone surface is framed to obtain the lateral section; The lateral section was divided into several small squares, and the anatomical key points closest to the bone surface were found in each square. All these points were collected as representative points of the zygomatic arch bone surface, and the section equation was fitted using the least squares method as the lateral fitting surface. In the same way, generate the upper edge fitting surface in the upper boundary direction; The intersection of the lateral fitting surface and the upper edge fitting surface is used as the prosthesis placement direction, and the spatial posture of the prosthesis is determined based on the prosthesis placement direction. The final output is the rotation matrix of the prosthesis relative to the patient's three-dimensional model, which is used for precise prosthesis placement.

9. The artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method according to claim 8, characterized in that: The mandibular ramus prosthesis placement logic is: The lateral surface of the mandible is screened from the point cloud of the region of interest, the directed bounding box is calculated, the condyle, coracoid process and mandibular angle are determined, the mandibular lateral section is generated and the bone-prosthesis fitting surface is optimized; Determine the placement angle through the plane and line direction and the rotation matrix, determine the standardized prosthesis coordinate system based on the prosthesis position under geometric constraints, and set the contact surface and hole center; By using the common point constraint (the lowest point of the concave surface of the glenoid fossa coincides with the highest point of the convex surface of the mandibular ramus) and the plane constraint, a linear equation system is established to solve the translation matrix.

10. A computer program product stored on a computer-readable medium, characterized in that: It includes a computer-readable program that, when executed on an electronic device, provides a user input interface to implement the artificial intelligence-assisted standard artificial temporomandibular joint automatic placement method as described in any one of claims 1 to 9.

Images