Temporomandibular joint three-dimensional reconstruction system based on joint imaging diagnosis

Abstract

The invention relates to the technical field of medical image processing, and discloses a temporomandibular joint three-dimensional reconstruction system based on joint imaging diagnosis. The system utilizes CBCT equipment to collect temporal-mandibular joint tomographic image data, and extracts facial feature sites and maxillofacial joint standard model approval sites through a multi-scale residual network. And carrying out space registration by adopting an affine transformation and nonlinear optimization algorithm to generate a synchronous correction parameter, further carrying out deformation interpolation on the standard model of the maxillofacial joint to construct a maxillofacial model, and adjusting the position of a condylar process. And based on the deformed maxillofacial model, generating an occlusal plate structure model by using an isogeometric analysis method, and iteratively optimizing the degree of fit through an edge calculation frame. The system can accurately reconstruct the three-dimensional model of the temporomandibular joint, optimizes the adaptation of the biteplate, provides powerful support for the diagnosis and treatment of temporomandibular joint diseases, and facilitates the improvement of the diagnosis and treatment level.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and in particular to a temporomandibular joint three-dimensional reconstruction system based on joint imaging diagnosis. Background Art

[0002] The temporomandibular joint (TMJ), a vital component of the oral and maxillofacial system, directly impacts physiological functions such as chewing, swallowing, and speech. Accurately diagnosing TMJ disorders and developing effective treatment plans are crucial in clinical practice. However, traditional diagnostic methods have numerous limitations.

[0003] In terms of imaging, while early plain X-rays could provide some skeletal information, they were two-dimensional images that could not clearly demonstrate the complex three-dimensional structure of the temporomandibular joint. Soft tissues within the joint, such as the articular disc, were virtually impossible to observe. This made it difficult for doctors to fully understand joint pathology and could easily lead to missed or misdiagnosed cases. While CT scans have improved image resolution to a certain extent, revealing more detailed skeletal structures, their ability to distinguish soft tissues remains limited. Furthermore, the radiation dose is relatively high, and frequent use could potentially harm patients.

[0004] During disease diagnosis and treatment plan development, doctors rely solely on experience and limited imaging information due to a lack of accurate three-dimensional model support. For complex temporomandibular joint disorders, such as disc displacement and condylar bone changes, it's difficult to accurately assess the severity and progression of the condition. Treatment plans, whether conservative or surgical, lack reliable evidence, making personalized, precise treatment impossible. This results in treatment outcomes that fall short of expectations, prolonged recovery periods, and significant impacts on patients' quality of life.

[0005] In the field of occlusal splint treatment, the traditional occlusal splint fabrication process relies on manual manipulation and simple model measurements. This method lacks precise analysis of the patient's maxillofacial structure, resulting in a poor fit between the occlusal splint and the patient's actual maxillofacial model. Poorly fitting occlusal splints not only fail to effectively alleviate symptoms of temporomandibular joint (TMJ) disorders, but may also increase joint stress, causing new problems and further impacting the patient's treatment experience and recovery process.

[0006] With the continuous advancement of medical technology and the increasing demand for medical service quality, the development of a system that can accurately reconstruct a 3D model of the temporomandibular joint and achieve precise fitting of occlusal splints is urgent. This system must fully utilize advanced imaging technology and computer algorithms to address the pain points of traditional methods in the diagnosis and treatment process, providing clinicians with more accurate and effective diagnostic and treatment tools, and improving the overall diagnosis and treatment of temporomandibular joint disorders. Summary of the Invention

[0007] The purpose of the present invention is to provide a three-dimensional reconstruction system of the temporomandibular joint based on joint imaging diagnosis to solve the problems raised in the above background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a three-dimensional reconstruction system for the temporomandibular joint based on joint imaging diagnosis, the system comprising:

[0009] Image data acquisition module: used to obtain tomographic image data of the temporomandibular joint through CBCT equipment;

[0010] Feature extraction module: segmenting the key anatomical structures of the tomographic image data based on a multi-scale residual network, extracting facial feature sites and approved sites of the maxillofacial joint standard model;

[0011] Matching correction module: uses affine transformation and nonlinear optimization algorithm to spatially align the facial feature sites with the approved sites, and generates synchronization correction parameters based on dynamic weight constraints;

[0012] Model reconstruction module: Perform deformation interpolation on the standard model of the maxillofacial joint according to the synchronous correction parameters to construct the maxillofacial model, and adjust the position of the condyle to the middle zone of the articular disc through the gradient descent method;

[0013] Occlusal plate adaptation module: Based on the maxillofacial model after deformation interpolation, the occlusal plate structure model is generated using the isogeometric analysis method, and an edge computing framework is constructed to iteratively optimize the adaptability of the occlusal plate and the maxillofacial model.

[0014] Preferably, the feature extraction module includes:

[0015] The tomographic image data is divided into coronal, sagittal and axial multi-plane sequences, which are input into the parallel convolution branch to extract the features of each plane;

[0016] A multi-scale residual network is constructed. The first branch uses a dilated convolution kernel to extract large-scale anatomical contour features, while the second branch uses a densely connected structure to capture local detail features.

[0017] The multi-plane feature maps are aligned through a cross-plane feature fusion mechanism, and a spatial pyramid pooling layer is used to generate a set of facial feature site coordinates;

[0018] The key points of the standard model of the maxillofacial joint are calibrated, and the approved sites are screened based on the curvature extreme value detection algorithm, and a set of approved sites containing spatial coordinates and topological relationships is output.

[0019] Preferably, the matching correction module adopts affine transformation and nonlinear optimization algorithm including:

[0020] Construct an affine transformation matrix to map facial feature points to the approved point coordinate system;

[0021] Define the matching error function, including Euclidean distance, normal vector angle and curvature similarity weight terms;

[0022] Design a dynamic weight constraint mechanism to adaptively adjust the contribution ratio of each weight term according to the local deformation gradient;

[0023] The transformation matrix parameters are iteratively optimized through the Levenberg-Marquardt algorithm, and the synchronization correction parameters and deviation values ​​are output.

[0024] Preferably, the deformation interpolation of the model reconstruction module includes:

[0025] The standard model of the maxillofacial joint is discretized into a tetrahedral mesh, and the node attributes include coordinates, curvature and biomechanical parameters;

[0026] Construct radial basis function interpolation field based on synchronous correction parameters to drive mesh node displacement;

[0027] The Laplace smoothness constraint term is introduced to suppress the grid distortion during the interpolation process;

[0028] The condyle position was adjusted by gradient descent method to minimize the spatial distance between the condyle center and the middle zone of the articular disc.

[0029] Preferably, the isogeometric analysis method of the bite plate adaptation module includes:

[0030] The maxillofacial model surface is parameterized as a non-uniform rational B-spline surface;

[0031] Generate the initial structural model of the occlusal splint based on the curvature matching criterion and extract the boundary of the occlusal contact area;

[0032] Build an edge computing framework and deploy a lightweight convolutional network at the edge of the bite plate to predict the fit;

[0033] The golden section method was used to adjust the thickness and curvature radius of the occlusal plate until the fit reached the preset threshold.

[0034] Preferably, the densely connected structure of the multi-scale residual network includes:

[0035] The local detail feature extraction process is decomposed into a multi-stage cascade sub-network, each stage contains a 3×3 convolution layer and a batch normalization layer;

[0036] Introducing skip connections between adjacent sub-networks to fuse shallow high-resolution features with deep semantic features;

[0037] The channel attention mechanism is used to adaptively weight the fused feature maps.

[0038] Preferably, the dynamic weight constraint mechanism includes:

[0039] The regional stiffness coefficient is calculated according to the local deformation gradient as the adjustment factor of the Euclidean distance weight term;

[0040] Dynamically adjust the weight ratio of the normal vector angle through the curvature change rate;

[0041] A fuzzy inference rule base is designed to map biomechanical parameters into dynamic weights of curvature similarity.

[0042] Preferably, the construction of the radial basis function interpolation field includes:

[0043] The Gaussian kernel function is selected as the basis function, the kernel radius is adaptively adjusted according to the grid density, and the synchronous correction parameters are used as the displacement constraints of the control points to construct the overdetermined equation system;

[0044] The basis function coefficients are solved by the singular value decomposition method to generate a continuous deformation field.

[0045] Preferably, the lightweight convolutional network includes:

[0046] A depthwise separable convolutional structure is used to reduce the number of parameters, and the input is the point cloud of the contact area between the bite plate and the maxillofacial model;

[0047] Design multi-tasking output heads while simultaneously predicting fit, stress distribution, and contact area.

[0048] Preferably, the channel attention mechanism includes:

[0049] Calculate the global average pooling value of the feature map in the channel dimension to generate the channel importance vector;

[0050] Generate channel weights through the fully connected layer and Sigmoid function;

[0051] Multiply the channel weights by the original feature map element-wise and output the weighted feature map.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] In terms of image data processing and feature extraction, CBCT equipment is used to acquire tomographic image data, and through a feature extraction module based on a multi-scale residual network, key anatomical structures can be efficiently and accurately segmented, and facial feature sites and approved sites of the maxillofacial joint standard model can be extracted. The parallel convolution branches and multi-scale feature extraction method of the multi-scale residual network can comprehensively capture image information from different planes and scales, and its accuracy is greatly improved compared to traditional image feature extraction methods. For example, when processing complex temporomandibular joint images, traditional methods may miss some small but critical pathological features, but this system can clearly identify and extract these features, providing a more reliable basis for subsequent diagnosis and treatment.

[0054] In the model construction process, the matching correction module uses affine transformation and nonlinear optimization algorithm, combined with dynamic weight constraints to generate synchronous correction parameters, so that the spatial registration of facial feature sites and approved sites is more accurate. This precise registration effectively reduces the error in the model construction process, and the constructed maxillofacial model is more in line with the patient's actual temporomandibular joint structure. At the same time, during the deformation interpolation process, the model reconstruction module discretizes the standard model of the maxillofacial joint into a tetrahedral mesh, introduces Laplace smoothing constraints to suppress mesh distortion, and uses the gradient descent method to adjust the condyle position to the middle zone of the articular disc, further optimizing the accuracy and rationality of the maxillofacial model. This is of great significance for simulating the real movement state of the temporomandibular joint, analyzing the force distribution of the joint, etc., and helps doctors to have a deeper understanding of the mechanism of joint pathology.

[0055] In terms of occlusal splint fitting, an isogeometric analysis method is used to generate an occlusal splint structural model based on the maxillofacial model after deformation interpolation, and an edge computing framework is constructed to iteratively optimize the fit. The isogeometric analysis method parameterizes the surface of the maxillofacial model into a non-uniform rational B-spline surface, which can more accurately describe the maxillofacial shape. The initial occlusal splint structural model generated based on this has a higher fit. The lightweight convolutional network in the edge computing framework can quickly and accurately predict the fit. Combined with the golden section method, the thickness and curvature radius of the occlusal splint are adjusted to ensure that the fit between the occlusal splint and the maxillofacial model reaches the preset threshold. A well-fitting occlusal splint plays a key role in the treatment of temporomandibular joint diseases. It can effectively adjust the occlusal relationship of the joint, reduce joint pressure, relieve pain and discomfort symptoms, and improve the patient's treatment effect and quality of life.

[0056] From the overall perspective of clinical application, the system provides doctors with a comprehensive and accurate auxiliary tool for the diagnosis and treatment of the temporomandibular joint. With the help of the reconstructed three-dimensional model, doctors can intuitively observe the morphology, structure and pathological conditions of the joint, and make more accurate diagnoses and disease assessments. When formulating a treatment plan, it is possible to formulate a personalized treatment plan based on the results of the model analysis to improve the pertinence and effectiveness of the treatment. For example, for patients who need surgical treatment, doctors can simulate the surgical process through the model before the operation, plan the surgical path, and improve the success rate and safety of the operation; for patients who adopt conservative treatment, a well-fitting occlusal splint can better play a therapeutic role, shorten the treatment cycle, and reduce the patient's pain and medical costs. In short, the system of the present invention plays an important role in promoting the diagnosis and treatment of temporomandibular joint diseases, and has broad clinical application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a diagram showing the working principle of the temporomandibular joint three-dimensional reconstruction system based on joint imaging diagnosis according to the present invention;

[0058] Figure 2 This is the working principle diagram of the feature extraction module;

[0059] Figure 3 To match the working principle diagram of the correction module;

[0060] Figure 4 A diagram showing how the model rebuild module works. DETAILED DESCRIPTION

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0062] See also Figures 1-4 The present invention provides a three-dimensional reconstruction system for the temporomandibular joint based on joint imaging diagnosis, and its overall implementation scheme is as follows:

[0063] Image Data Acquisition Module: CBCT equipment is used to scan the patient's temporomandibular joint (TMJ) and acquire cross-sectional imaging data. CBCT equipment provides high-resolution cross-sectional images, providing an accurate data foundation for subsequent analysis and processing. In practice, CBCT scanning parameters are adjusted based on the patient's specific condition to ensure clear and complete imaging data, encompassing all key areas of the TMJ.

[0064] Feature Extraction Module: This module processes the acquired sectional image data using a multi-scale residual network. It segments key anatomical structures and extracts facial landmarks and referenced locations for the standard maxillofacial joint model. Through in-depth analysis of the image data, it accurately identifies key features related to the temporomandibular joint, providing critical information for subsequent model construction and registration.

[0065] The Matching Correction Module uses affine transformations and nonlinear optimization algorithms to spatially align facial feature points with the approved points. During this process, synchronized correction parameters are generated based on dynamic weight constraints to ensure more accurate matching between points and minimize errors. By continuously optimizing these parameters, the reconstructed model is more consistent with the patient's TMJ.

[0066] Model Reconstruction Module: Based on the generated synchronous correction parameters, the standard maxillofacial joint model is deformed and interpolated to construct the maxillofacial model. Simultaneously, the condyle position is adjusted using the gradient descent method to locate it in the middle zone of the articular disc, thereby constructing a more accurate and physiological maxillofacial model.

[0067] Occlusal splint fitting module: Based on the deformed interpolated maxillofacial model, an isogeometric analysis method is used to generate an occlusal splint structural model. An edge computing framework is also built to iteratively optimize the fit between the occlusal splint and the maxillofacial model to achieve better treatment or diagnostic results.

[0068] Next, the present invention is further described in detail through the following five embodiments:

[0069] Example 1:

[0070] This example details the specific implementation of the feature extraction module. First, the tomographic image data is divided into multi-plane sequences based on the coronal, sagittal, and axial planes. These multi-plane sequences are then fed into parallel convolution branches. Parallel convolution branches can simultaneously extract features from image data in different planes, improving processing efficiency. During this process, each convolution branch extracts feature information specific to the characteristics of each plane.

[0071] Then, a multi-scale residual network is constructed. The first branch of the network uses a dilated convolution kernel. The dilated convolution kernel can expand the receptive field of the convolution kernel without increasing the number of parameters, thereby extracting anatomical contour features over a wide range. For example, in images of the temporomandibular joint, macroscopic features such as the overall shape and position of the joint can be captured. The second branch uses a densely connected structure to decompose the local detail feature extraction process into a multi-stage cascade sub-network, each stage containing a 3×3 convolution layer and a batch normalization layer. The 3×3 convolution layer can effectively extract local features, and the batch normalization layer normalizes the data to accelerate the convergence of the network.

[0072] Skip connections are introduced between adjacent sub-networks. This connection method can fuse shallow high-resolution features with deep semantic features. Shallow high-resolution features contain more detailed information, while deep semantic features have a higher level of abstraction. The fusion of the two can more comprehensively describe image characteristics. A channel attention mechanism is used to adaptively weight the fused feature map. The global average pooling value of the feature map is calculated in the channel dimension to generate a channel importance vector. Channel weights are generated through a fully connected layer and a sigmoid function. The channel weights are element-wise multiplied with the original feature map to output a weighted feature map. This can highlight channel information that is more important for classification or segmentation tasks, improving the accuracy of feature extraction.

[0073] Multi-plane feature maps are aligned through a cross-plane feature fusion mechanism. Feature maps from different planes contain different information, which can be integrated through fusion. A spatial pyramid pooling layer is used to generate a set of facial feature site coordinates. This layer can fuse features at different scales to generate more representative facial feature site coordinates.

[0074] For the standard maxillofacial joint model, key point calibration is first performed to mark key locations within the model. Then, the approval sites are selected based on a curvature extreme value detection algorithm. This algorithm calculates the curvature of the model surface and identifies the curvature extreme points as approval sites. The final output is a set of approval sites containing spatial coordinates and topological relationships.

[0075] Example 2:

[0076] This embodiment focuses on the specific operation of the affine transformation and nonlinear optimization algorithm in the matching correction module. When constructing the affine transformation matrix, the purpose is to map the facial feature points to the approved point coordinate system. The affine transformation matrix can be expressed as:

[0077]

[0078] in, is the coordinate vector of the facial feature points, is the linear part of the transformation matrix, which determines the comprehensive transformation of rotation, scaling and translation. By adjusting these parameters, the facial feature points are mapped in the approved point coordinate system.

[0079] Define the matching error function, which includes Euclidean distance, normal vector angle and curvature similarity weight. Euclidean distance is used to measure the spatial distance between two sites. The formula is Where (x1, y1, z1) and (x2, y2, z2) are the coordinates of the two points. The normal vector angle is used to reflect the direction difference of the surface at the point, and the curvature similarity is used to measure the similarity of the surface curvature at the point.

[0080] A dynamic weight constraint mechanism is designed to adaptively adjust the contribution ratio of each weight term based on the local deformation gradient. A regional stiffness coefficient is calculated based on the local deformation gradient and used as a modulating factor for the Euclidean distance weight term. When the local deformation gradient is large, indicating that the region is undergoing drastic changes, the influence of the Euclidean distance weight term is appropriately reduced, and more consideration is given to the normal vector angle and curvature similarity. The normal vector angle weight ratio is dynamically adjusted using the curvature change rate, with the normal vector angle weight correspondingly increased in regions with large curvature change rates. A fuzzy inference rule library is designed to map biomechanical parameters to dynamic weights for curvature similarity. Biomechanical parameters can reflect information such as the force applied to the joint. This information is used to adjust the curvature similarity weights for a more reasonable match.

[0081] The transformation matrix parameters are iteratively optimized using the Levenberg-Marquardt algorithm. This algorithm combines the advantages of gradient descent and Gauss-Newton methods, using Gauss-Newton to accelerate convergence when approaching the optimal solution and gradient descent to ensure algorithm stability when far from the optimal solution. During the iterative process, the transformation matrix parameters are continuously adjusted until certain convergence conditions are met, ultimately outputting the synchronized correction parameters and deviation values.

[0082] Example 3:

[0083] This embodiment details the deformation interpolation process of the model reconstruction module. First, the standard maxillofacial joint model is discretized into a tetrahedral mesh, with each node having coordinates, curvature, and biomechanical parameters. The coordinates are used to determine the position of the node in space, the curvature reflects the degree of curvature of the model surface at the node, and the biomechanical parameters are related to the mechanical properties of the joint, such as the elastic modulus.

[0084] The radial basis function interpolation field is constructed based on the synchronous correction parameters. The Gaussian kernel function is selected as the basis function, and the formula is:

[0085]

[0086] Where r is the distance between nodes, and σ is the core radius, which is adaptively adjusted based on the mesh density. Using the synchronization correction parameters as control point displacement constraints, an overdetermined system of equations is constructed. The basis function coefficients are solved using the singular value decomposition method to generate a continuous deformation field. The singular value decomposition method decomposes a matrix into the product of three matrices, making it easy to solve the overdetermined system of equations and obtain the basis function coefficients, which drive the mesh node displacements.

[0087] During the interpolation process, the Laplace smooth constraint term is introduced to suppress mesh distortion. The Laplace smooth constraint term can be expressed as where v i and v j are the coordinates of adjacent nodes, N j is the set of neighboring nodes of node j. Through this constraint, the mesh remains relatively smooth during deformation to avoid excessive distortion.

[0088] The condyle position is adjusted by gradient descent method to minimize the spatial distance between the condyle center and the middle zone of the articular disc. The objective function is defined as:

[0089]

[0090] where d k is the distance between the condyle center and the kth sampling point in the mid-zone of the articular disc, and n is the number of sampling points. By iteratively updating the condyle position coordinate v, calculating the gradient of the objective function, and adjusting the coordinates along the negative direction of the gradient, the condyle position is optimized until the objective function converges to a smaller value.

[0091] Example 4:

[0092] This example details the isogeometric analysis method for the occlusal splint adaptation module. First, the maxillofacial model surface is parameterized as a non-uniform rational B-spline surface. Non-uniform rational B-spline surfaces can flexibly represent complex geometric shapes and accurately describe the maxillofacial model surface through control points and weights. During the parameterization process, the positions and weights of the control points, as well as parameters such as the node vectors, are determined to accurately construct the surface model.

[0093] The initial structure of the bite plate is generated based on the curvature matching principle, and the boundaries of the occlusal contact area are extracted. The curvature matching principle generates the initial structure of the bite plate in areas with similar curvature based on the curvature distribution of the maxillofacial model surface. This ensures a good fit between the bite plate and the maxillofacial model from the initial stage. The boundaries of the occlusal contact area are extracted to facilitate subsequent analysis and optimization of the occlusal contact.

[0094] An edge computing framework was constructed, and a lightweight convolutional network was deployed at the edge of the bite splint to predict fit. This lightweight convolutional network uses a depthwise separable convolutional structure to reduce the number of parameters. The input is a point cloud of the contact area between the bite splint and the maxillofacial model. Depthwise separable convolution decomposes traditional convolution into depthwise and pointwise convolutions, significantly reducing the number of parameters and computational complexity without sacrificing accuracy. A multi-task output head was designed to simultaneously predict fit, stress distribution, and contact area. This network enables rapid and accurate acquisition of fit information between the bite splint and the maxillofacial model.

[0095] The golden section method is used to adjust the thickness and radius of curvature of the occlusal splint until the fit reaches a preset threshold. The golden section method is an optimization algorithm that finds the optimal solution of a function by iterating within a certain interval. When adjusting the occlusal splint thickness t and radius of curvature r, the next adjustment value is selected according to the golden section ratio based on the change in fit. For example, let the initial interval be [a, b], and the two points selected for the first adjustment be a+0.382(ba) and a+0.618(ba). Based on the fit at these two points, the search interval is narrowed, and iterations continue until the fit meets the preset threshold.

[0096] Example 5:

[0097] This embodiment combines elements not fully described in the previous embodiments to provide a comprehensive description of the entire system and provide additional details. During image data acquisition, in addition to adjusting the scanning parameters of the CBCT device, the acquired data must also be preprocessed. Preprocessing includes operations such as noise removal and image grayscale correction to improve image data quality and provide a better data foundation for subsequent feature extraction and analysis.

[0098] In the feature extraction module, the training process of the multi-scale residual network is also crucial. Using a large amount of temporomandibular joint image data for training, the network parameters are continuously adjusted through a backpropagation algorithm, enabling the network to accurately extract facial feature points and verification points. During training, appropriate parameters such as the learning rate and loss function are set to ensure the network's convergence speed and accuracy.

[0099] In the matching correction module, the construction of the fuzzy inference rule base within the dynamic weight constraint mechanism requires extensive experimental data and clinical experience. By analyzing temporomandibular joint models under different conditions and determining the relationship between biomechanical parameters and curvature similarity weights, reasonable fuzzy inference rules can be constructed.

[0100] In the model reconstruction module, when constructing the radial basis function interpolation field, in addition to selecting the Gaussian kernel function, other basis functions, such as thin plate spline functions, can be tried, and the optimal basis function can be selected based on the actual results. Furthermore, when adjusting the condylar position, more physiological parameters, such as muscle force distribution, can be incorporated to make the reconstructed maxillofacial model more consistent with the actual physiological structure.

[0101] In the bite splint fitting module, when constructing a lightweight convolutional network, the network structure can be further optimized, for example by introducing an attention mechanism, to improve the network's accuracy in predicting the fit between the bite splint and the maxillofacial model. When adjusting bite splint parameters using the golden section method, other optimization algorithms, such as simulated annealing, can be combined to improve optimization efficiency and more quickly find the optimal bite splint parameters.

[0102] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0103] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional reconstruction system for the temporomandibular joint based on joint imaging diagnosis, characterized in that: include: Image data acquisition module: used to obtain tomographic image data of the temporomandibular joint through CBCT equipment; Feature extraction module: Based on the multi-scale residual network, the key anatomical structure segmentation of the tomographic image data is performed to extract the facial feature sites and the approved sites of the maxillofacial joint standard model; Matching correction module: uses affine transformation and nonlinear optimization algorithm to spatially align the facial feature sites with the approved sites, and generates synchronization correction parameters based on dynamic weight constraints; Model reconstruction module: Perform deformation interpolation on the standard model of the maxillofacial joint according to the synchronous correction parameters to construct the maxillofacial model, and adjust the position of the condyle to the middle zone of the articular disc through the gradient descent method; Occlusal plate adaptation module: Based on the maxillofacial model after deformation interpolation, the occlusal plate structure model is generated using the isogeometric analysis method, and an edge computing framework is constructed to iteratively optimize the adaptability of the occlusal plate and the maxillofacial model.

2. A 3D reconstruction system for the temporomandibular joint based on joint imaging diagnosis according to claim 1, characterized in that: The feature extraction module includes: The tomographic image data is divided into coronal, sagittal and axial multi-plane sequences, which are input into the parallel convolution branch to extract the features of each plane; A multi-scale residual network is constructed. The first branch uses a dilated convolution kernel to extract large-scale anatomical contour features, while the second branch uses a densely connected structure to capture local detail features. The multi-plane feature maps are aligned through a cross-plane feature fusion mechanism, and a spatial pyramid pooling layer is used to generate a set of facial feature site coordinates; The key points of the standard model of the maxillofacial joint are calibrated, and the approved sites are screened based on the curvature extreme value detection algorithm, and a set of approved sites containing spatial coordinates and topological relationships is output.

3. The three-dimensional reconstruction system of the temporomandibular joint based on joint imaging diagnosis according to claim 1, characterized in that: The matching correction module adopts affine transformation and nonlinear optimization algorithm including: Construct an affine transformation matrix to map facial feature points to the approved point coordinate system; Define the matching error function, including Euclidean distance, normal vector angle and curvature similarity weight terms; Design a dynamic weight constraint mechanism to adaptively adjust the contribution ratio of each weight term according to the local deformation gradient; The transformation matrix parameters are iteratively optimized through the Levenberg-Marquardt algorithm, and the synchronization correction parameters and deviation values ​​are output.

4. The system for three-dimensional reconstruction of the temporomandibular joint based on joint imaging diagnosis according to claim 1, characterized in that: The deformation interpolation of the model reconstruction module includes: The standard model of the maxillofacial joint is discretized into a tetrahedral mesh, and the node attributes include coordinates, curvature and biomechanical parameters; Construct radial basis function interpolation field based on synchronous correction parameters to drive mesh node displacement; The Laplace smoothness constraint term is introduced to suppress the grid distortion during the interpolation process; The condyle position was adjusted by gradient descent method to minimize the spatial distance between the condyle center and the middle zone of the articular disc.

5. The system for three-dimensional reconstruction of the temporomandibular joint based on joint imaging diagnosis according to claim 1, characterized in that: The isogeometric analysis method of the bite plate adaptation module includes: The maxillofacial model surface is parameterized as a non-uniform rational B-spline surface; Generate the initial structural model of the occlusal splint based on the curvature matching criterion and extract the boundary of the occlusal contact area; Build an edge computing framework and deploy a lightweight convolutional network at the edge of the bite plate to predict the fit; The golden section method was used to adjust the thickness and curvature radius of the occlusal plate until the fit reached the preset threshold.

6. The system for three-dimensional reconstruction of the temporomandibular joint based on joint imaging diagnosis according to claim 2, characterized in that: The dense connection structure of the multi-scale residual network includes: The local detail feature extraction process is decomposed into a multi-stage cascade sub-network, each stage contains a 3×3 convolution layer and a batch normalization layer; Introducing skip connections between adjacent sub-networks to fuse shallow high-resolution features with deep semantic features; The channel attention mechanism is used to adaptively weight the fused feature maps.

7. The system for three-dimensional reconstruction of the temporomandibular joint based on joint imaging diagnosis according to claim 3, characterized in that: The dynamic weight constraint mechanism includes: The regional stiffness coefficient is calculated according to the local deformation gradient as the adjustment factor of the Euclidean distance weight term; Dynamically adjust the weight ratio of the normal vector angle through the curvature change rate; A fuzzy inference rule base is designed to map biomechanical parameters into dynamic weights of curvature similarity.

8. The system for three-dimensional reconstruction of the temporomandibular joint based on joint imaging diagnosis according to claim 4, characterized in that: The construction of the radial basis function interpolation field includes: The Gaussian kernel function is selected as the basis function, the kernel radius is adaptively adjusted according to the grid density, and the synchronous correction parameters are used as the displacement constraints of the control points to construct the overdetermined equation system; The basis function coefficients are solved by the singular value decomposition method to generate a continuous deformation field.

9. The system for 3D reconstruction of the temporomandibular joint based on joint imaging diagnosis according to claim 5, characterized in that: The lightweight convolutional network includes: A depthwise separable convolutional structure is used to reduce the number of parameters, and the input is the point cloud of the contact area between the bite plate and the maxillofacial model; Design multi-tasking output heads while simultaneously predicting fit, stress distribution, and contact area.

10. The system for three-dimensional reconstruction of the temporomandibular joint based on joint imaging diagnosis according to claim 6, characterized in that: The channel attention mechanism includes: Calculate the global average pooling value of the feature map in the channel dimension to generate the channel importance vector; generate the channel weight through the fully connected layer and the Sigmoid function; Multiply the channel weights by the original feature map element-wise and output the weighted feature map.

Images