Patient matching type implant and construction method

Abstract

The invention provides a patient matching type implant and a construction method, and relates to the technical field of dental implant design. The method comprises the following steps: acquiring three-dimensional oral CT scanning data and physiological data of a patient, and generating implant design reference data matched with the patient through an implant structure generation model; constructing an implant recovery analysis model and a finite element analysis model, and designing an implant stability analysis function and an implant recovery analysis function to construct an initial target optimization function; updating the initial target optimization function according to the implant personalized demand data of the patient to generate a personalized target optimization function, and optimizing the implant design reference data by adopting a target optimization algorithm to generate implant design target data matched with the patient; and printing and generating a target implant matched with the patient according to the implant design target data matched with the patient. According to the invention, a high-matching-degree dental implant meeting individual requirements is provided for a patient.

Description

Technical Field

[0001] The present invention relates to the technical field of dental implant design, and particularly to a patient-matched implant and a construction method thereof. Background Art

[0002] The specifications of mass-produced dental implants are relatively unified. Although these implants can meet the needs of general patients to a certain extent, due to the differences in oral data and physiological characteristics of different patients, it is generally difficult for standardized implants to precisely match the individual needs of patients. Factors such as bone density and bone structure vary greatly among different patients. In the mass production mode, the design of implants does not fully consider these individual differences, resulting in a low matching degree between the implant and the patient's oral cavity, thus affecting the implantation effect and the patient's recovery process. To overcome this limitation, dental professionals usually carry out personalized customization design according to the specific situation of patients. However, the method of personalized customization based on professionals usually requires a long design and production cycle, and is relatively dependent on the experience of designers, with low efficiency and being easily limited by the personal experience of designers, making it difficult to ensure the accuracy and consistency of the design.

[0003] With the emergence of computer-aided design and manufacturing technologies, data-driven methods have been tried in the dental field to design implants by analyzing patients' imaging data. However, some technologies usually rely on limited training sample data. If the collected sample data cannot comprehensively cover the personalized needs of patients, the design solutions mined based on the existing sample data mostly provide local optimal solutions and cannot meet the actual needs of all patients. For example, the specific needs of different patients in terms of implant stability, recovery speed, etc. are different. If the design process cannot be flexibly adjusted according to individual differences, the designed implants will not fully meet the personalized needs of patients.

[0004] How to conduct all-round and multi-dimensional optimization analysis based on patients' oral structures and physiological data, and on the basis of considering patients' personalized needs, provide an efficient and accurate implant design solution and construct an implant with a high matching degree to patients' personalized needs has become an urgent problem to be solved. Summary of the Invention

[0005] To solve the above technical problems, the present invention proposes a patient-matched implant and a construction method thereof, which can achieve flexible adjustment for patients' personalized needs while ensuring design accuracy, provide a more accurate and personalized design solution, make the designed implant better adapt to the individual differences of patients, and improve the dental implantation experience of patients.

[0006] To achieve the above object, the present invention provides the following technical solutions: A patient-matched implant construction method, comprising: Obtain the three-dimensional oral CT scan data and physiological data of the patient, input the three-dimensional oral CT scan data and physiological data of the patient into the implant structure generation model, and generate implant design reference data matched with the patient through the implant structure generation model. The implant structure generation model includes an input layer, a feature extraction layer, a geometric shape constraint optimization layer, a material compatibility optimization layer, and an output layer; Construct a finite element analysis model based on the three-dimensional oral CT scan data of the patient, determine the implant stability analysis function based on the finite element analysis model, determine the implant restoration analysis function through the implant restoration analysis model, and construct an initial target optimization function according to the implant stability analysis function and the implant restoration analysis function; Obtain the implant personalized requirement data of the patient, update the initial target optimization function according to the implant personalized requirement data to generate a personalized target optimization function, and use the target optimization algorithm to optimize the implant design reference data based on the personalized target optimization function to generate implant design target data matched with the patient; Input the implant design target data matched with the patient into a 3D printing device, and process the implant design target data matched with the patient through the 3D printing device to print and generate a target implant matched with the patient.

[0007] Preferably, for the implant structure generation model, it further includes: The input layer is used to receive the three-dimensional oral CT scan data and physiological data of the patient. The feature extraction layer is used to perform feature extraction and feature fusion on the three-dimensional oral CT scan data and physiological data of the patient to generate implant design requirement data including oral structure features and physiological features. The geometric shape constraint optimization layer is used to perform structure adjustment and optimization based on geometric constraints on the initial structure of the implant based on the implant design requirement data to generate the target optimized structure of the implant. The material compatibility optimization layer is used to perform material matching optimization on the implant based on the implant design requirement data to generate the material property data of the implant. The output layer is used to fuse and output the optimization results of the geometric shape constraint optimization layer and the material compatibility optimization layer to generate implant design reference data matched with the three-dimensional oral CT scan data and physiological data of the patient. The output layer includes a structure design output unit and a material design output unit. The structure design output unit is used to output the target optimized structure of the implant, and the material design output unit is used to output the material property data of the implant; Construct a first training dataset by obtaining the historical dental implant data of multiple patients, and train the implant structure generation model using the first training dataset.

[0008] Preferably, for the geometric shape constraint optimization layer and the material compatibility optimization layer, it further includes: The geometric shape constraint optimization layer performs geometric constraint-based structural adjustment and optimization on the initial structure of the implant according to the implant design requirement data to generate the target optimized structure of the implant, including: The geometric shape constraint optimization layer includes a deformation unit and a constraint convolution unit. The constraint convolution unit adopts a multi-output design. The deformation unit processes the oral structure features in the implant design requirement data, and uses the adaptive grid technology to perform geometric structure optimization on the initial structure of the implant based on the oral structure features to generate a stage-optimized structure; Input the stage-optimized structure into the constraint convolution unit. The constraint convolution unit performs a convolution operation on the stage-optimized structure to extract the Gaussian distribution of the geometric convolution features, and calculates the geometric structure difference parameter between the geometric convolution features and the oral structure features. If the geometric structure difference parameter is greater than the geometric structure difference threshold, perform iterative optimization on the Gaussian distribution of the geometric convolution features in the local area based on the local correction unit to output the target optimized structure of the implant, otherwise directly output the Gaussian distribution of the geometric convolution features as the target optimized structure of the implant; The material compatibility optimization layer processes the physiological features in the implant design requirement data based on the MLP unit to generate the material property data of the implant that matches the physiological features.

[0009] Preferably, a first training dataset is constructed by obtaining the historical dental implant data of multiple patients, and an implant structure generation model is trained using the first training dataset, including: Extract multiple groups of first sample data from the historical dental implant data of multiple patients. Each group of first sample data includes sample oral scan data, sample physiological data, and corresponding sample implant design data, and construct a first training dataset according to the multiple groups of first sample data; According to the Chamfer distance between the stage-optimized structure generated by the deformation unit and the sample implant structure corresponding to the sample implant design data in the training dataset, construct the deformation optimization loss function of the geometric shape constraint optimization layer. According to the negative log-likelihood loss between the Gaussian distribution of the geometric convolution features generated by the constraint convolution unit and the sample implant structure corresponding to the sample implant design data in the training dataset, construct the constraint convolution optimization loss function of the geometric shape constraint optimization layer, and generate the geometric optimization loss function of the geometric shape constraint optimization layer that includes the deformation optimization loss and the constraint convolution optimization loss; According to the difference between the material property data generated by the MLP unit and the sample implant material corresponding to the sample implant design data in the training dataset, construct the material optimization loss function of the material compatibility optimization layer, and determine the total loss function of the implant structure generation model according to the geometric optimization loss function and the material optimization loss function; Calculate the total loss value of the model during the training of the implant structure generation model based on the first training dataset based on the total loss function. After the total loss value of the model falls within the preset loss range, complete the training of the implant structure generation model; otherwise, continue to iteratively train the implant structure generation model based on the first training dataset.

[0010] Preferably, use the objective optimization algorithm to optimize the implant design reference data based on the personalized objective optimization function to generate implant design target data matching the patient, including: Use the particle swarm optimization algorithm based on particle iteration limit to optimize the implant design reference data, including initializing particles based on the target optimization structure in the implant design reference data, adding the patient's three-dimensional oral CT scan data and physiological data to multiple particles, calculating the fitness of each particle using the personalized objective optimization function, updating the particle velocity and position according to the fitness of the particle and the implant design reference data, and generating implant design target data matching the patient through multiple iterations; Among them, calculating the fitness of each particle includes inputting each particle into the implant stability analysis function and the implant restoration analysis function respectively, generating the stability score of the particle through the finite element analysis model, generating the restoration score of the particle through the implant restoration analysis model, and substituting the stability score and restoration score of the particle into the personalized objective optimization function to calculate the fitness of the particle; For the personalized objective optimization function, the structure is as follows:

[0011] In the formula, is the personalized objective function, is the implant stability analysis function, used to output the stability score of the particle, is the implant restoration analysis function, used to output the restoration score of the particle, 、 are the stability weight and the restoration weight respectively.

[0012] Preferably, for the implant restoration analysis model, it further includes: Construct a second training dataset based on the historical dental implant data of multiple patients, and train the implant restoration analysis model through the second training dataset, including extracting multiple groups of second sample data from multiple groups of historical dental implant data, and each group of second sample data includes sample oral scan data, sample physiological data, sample implant structure data, and the corresponding sample implant restoration score; The implant restoration analysis model is a random forest model. Multiple groups of sample oral scan data, sample physiological data, and sample implant structure data in the second training dataset are used as input features for training the implant restoration analysis model. The sample implant restoration scores corresponding to the multiple groups of sample oral scan data, sample physiological data, and sample implant structure data are used as the training target for the implant restoration analysis model. The implant restoration analysis model is trained using the second training dataset.

[0013] Preferably, the iterative optimization of the local region of the geometric convolution feature based on the local correction unit to output the target optimized structure of the implant includes: After each local region optimization of the geometric convolution feature based on the local correction unit, calculate the geometric structure difference parameter between the optimized geometric convolution feature and the oral structure feature. If the geometric structure difference parameter is greater than the geometric structure difference threshold, perform local region optimization again based on the local correction unit. Iterate until the geometric structure difference parameter is not greater than the geometric structure difference threshold, and then output the geometric convolution feature generated after the last local region optimization as the target optimized structure of the implant.

[0014] A patient-matched implant is constructed based on the above-mentioned method for constructing a patient-matched implant.

[0015] The present invention has the following beneficial effects: The present invention processes the three-dimensional oral CT scan data and physiological data of the patient through the implant structure generation model, extracts the oral structure features of the patient, optimizes and generates implant structure data with a high degree of matching with the oral structure of the patient based on geometric shape constraints, extracts the physiological characteristics of the patient, and optimizes and generates material property data with a high degree of matching with the physiological characteristics of the patient based on material compatibility. Considering that the coverage of the model training samples is insufficient, directly generating an implant design plan may involve a locally optimal plan. Therefore, the output of the model is modified to generate implant design reference data with a structural reference range, and an optimization function for analyzing the stability and postoperative recovery speed of the implant is further constructed. Combining with the personalized requirement data of the patient, a personalized target optimization function is finally constructed. The implant design reference data is optimized and analyzed through an optimization algorithm, and finally implant design data highly matched with the patient is generated, so as to realize the construction of a target implant matched with the patient, improve the matching degree between the implant and the personalized requirements of the patient while meeting the clinical requirements of the implant, and enhance the dental implant experience of the patient. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic flow chart of a method for constructing a patient-matched implant provided by an embodiment of the present invention.

[0017] Figure 2 Structural schematic diagram of the implant structure generation model provided by the embodiment of the present invention. Detailed implementation manners

[0018] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0019] Please refer to Figure 1 and Figure 2 , a patient-matched implant construction method provided by the embodiment of the present invention specifically includes the following steps: Step S1: Obtain the three-dimensional oral CT scan data and physiological data of the patient, input the three-dimensional oral CT scan data and physiological data of the patient into the implant structure generation model, and generate implant design reference data matching the patient through the implant structure generation model.

[0020] In this embodiment, the three-dimensional oral CT scan data of the patient can be obtained through medical imaging technology, which includes information such as the three-dimensional structure of the patient's oral cavity, tooth position, bone density, alveolar bone morphology, etc. And further collect the physiological data of the patient, such as age, gender, height, weight, physical health status, etc. The three-dimensional oral CT scan data and physiological data are the key data for designing personalized matching dental implants. Input the collected three-dimensional oral CT scan data and physiological data into the pre-trained implant structure generation model, and generate implant design reference data matching the patient through the model. The implant design reference data is the initially determined implant design data that is more suitable for the patient, which includes the three-dimensional structure and materials of the implant, etc. By improving the structure of the model, the output implant design data is not a specific fixed structure, but an analyzable reference range. For example, parameters such as the length and width of the implant are an interval range.

[0021] Step S2: Construct a finite element analysis model based on the three-dimensional oral CT scan data of the patient, determine the implant stability analysis function based on the finite element analysis model, determine the implant restoration analysis function through the implant restoration analysis model, and construct an initial target optimization function according to the implant stability analysis function and the implant restoration analysis function.

[0022] In this embodiment, the basic oral structure information of the patient is determined based on the three-dimensional oral CT scan data of the patient, and a finite element analysis model of the patient is constructed. For example, software such as ANSYS can be used to model the patient's oral structure in detail, and the stress distribution and deformation of different implants under different load conditions can be simulated. The stability analysis function of the implant is further determined according to the finite element analysis model.

[0023] Among them, the stability of the implant is related to factors such as the size and structure of the implant, the maximum stress it bears, the contact area with the bone, and the bone density. Specifically, the following method can be used to analyze the stability of the implant and construct the stability analysis function of the implant:

[0024] In the formula, is the stability analysis function of the implant, is the maximum stress value of the implant, is the yield strength of the implant material. When the stress exceeds the yield strength of the material, the implant may be damaged. The ratio of the maximum stress value to the yield strength can be well used to measure the stability of the implant. is the contact area between the implant and the bone. The larger the contact area between the implant and the bone, the more uniform the load distribution, which helps to reduce stress concentration and make the stability of the implant higher. is the bone density. The higher the bone density, the greater the strength of the bone, which can better support the implant and thus improve the stability. A lower bone density is likely to cause the implant to be unstable. , , are weight coefficients, which are used to balance the importance of the maximum stress value of the implant, the contact area between the implant and the bone, and the bone density in the process of evaluating the stability of the implant.

[0025] After determining the specific structural data of the implant, the stress data and contact area data of the implant can be obtained through the finite element analysis model, so as to analyze the stability of the implant under the current structure in combination with the bone density data of the patient, and obtain the corresponding stability score. The higher the stability score, the more stable the implant is under the current structure. Through this method, the quantitative analysis of the stability of the implant is realized. Selecting an implant with high stability can ensure that the implant can work stably in the patient's oral environment for a long time without loosening or displacement.

[0026] In this embodiment, for the implant restoration analysis function, it is mainly implemented based on the implant restoration analysis model. The implant restoration analysis model is a pre-trained model that can analyze the corresponding restoration speed according to the input oral structure data, physiological data, and the structural design data of the implant, and predict the restoration speed during the implant healing process based on the individual differences of the patient to generate a restoration score. The restoration score helps to evaluate whether the implant can effectively promote bone healing and reduce postoperative complications.

[0027] After determining the stability analysis function and the restoration analysis function, an initial objective optimization function for achieving the target optimization can be constructed based on the stability analysis function and the restoration analysis function. The initial objective optimization function comprehensively considers multiple factors such as the stability of the implant and the restoration speed. As the basic objective function in the optimization process, it can be used to guide the design of the implant to evaluate the design quality of the implant from the aspects of the stability of the implant and the estimated restoration effect it brings.

[0028] Step S3: Obtain the personalized demand data of the patient's implant. Update the initial objective optimization function according to the personalized demand data of the implant, and use the objective optimization algorithm to optimize the implant design reference data based on the personalized objective optimization function to generate the implant design target data that matches the patient.

[0029] In this embodiment, the personalized demand data of the patient's implant can specifically be the individual needs of the patient regarding the implant, such as regarding the restoration speed, long-term stability, etc. It can be collected through communication with the patient, preoperative examinations, health assessments, etc., reflecting the unique needs of the patient. For example, the stability and restoration speed of the implant are divided into multiple levels. The patient can choose the personalized design tendency regarding the implant according to the actual needs and in combination with the suggestions of professional physicians. For example, whether it is required to complete the implant implantation process quickly, or it is acceptable to have a longer surgical recovery or waiting time but the implant can have higher stability. Update the initial objective optimization function through the personalized demand data of the patient.

[0030] Exemplarily, the stability and restoration speed of the implant can be divided into five levels, and each level has a corresponding score. Determine the level scores corresponding to the stability and restoration speed respectively according to the personalized demand data of the implant, and then determine the relative importance of the implant stability analysis function and the implant restoration analysis function in the initial objective optimization function according to the ratio of the scores. For example, if the scores corresponding to the stability and restoration speed are 5 and 3, then the personalized weights of the implant stability analysis function and the implant restoration analysis function are obtained after normalizing the level scores, so as to realize the construction of the personalized objective optimization function based on the personalized demand data.

[0031] For the implant design reference data, since it contains the range data that can be selected under the user's current oral conditions and physiological conditions, after determining the patient's personalized needs, the implant design reference data is optimized based on the personalized target optimization function, analyzed using the target optimization algorithm, and finally the implant design target data matching the patient is generated. The target optimization algorithm will iteratively adjust the design parameters according to the personalized target optimization function until the optimal solution is reached, so that the finally determined implant design target data can meet the patient's personalized needs and clinical requirements.

[0032] Step S4: Input the implant design target data matching the patient into the 3D printing device, and process the implant design target data matching the patient through the 3D printing device to print and generate the target implant matching the patient.

[0033] In this embodiment, the implant design target data matching the patient contains the precise geometric shape and material properties of the implant. Inputting it into the 3D printing device and realizing layer-by-layer printing through 3D printing technology can generate an implant that highly matches the patient's oral and physiological data. The 3D printing device can precisely construct the implant according to the digital design data to ensure its perfect fit with the patient's oral anatomical structure. The printing materials used are usually biocompatible materials such as titanium alloy, ceramics, etc. After determining the printing materials required according to the implant design target data matching the patient, the implant is printed and formed to ensure that the implant has good biocompatibility and mechanical properties. Through the 3D printing process, the target implant that highly matches the patient's oral structure and physiological needs is obtained.

[0034] It is worth noting that using a model based on patient data, such as training a deep learning model to generate personalized implant design data, can indeed generate implants that match the patient's bone geometry. However, since the model is trained based on the patient's oral data and physiological data such as age, bone density, etc., the generated personalized implant design often represents a relatively standardized suitable solution. Although this design conforms to the geometry of the patient's oral cavity, some personalized needs such as recovery speed and stability may not be fully considered. Although the collected sample data may contain the needs of different patients, for example, some patients may need sufficient stability and have no high requirements for the recovery and healing speed of the implant surgery, while some young patients may choose a lighter and more easily healed design, and some simply choose a generally applicable implant design. However, the data is still prone to incomplete coverage, that is, it is difficult to ensure that patients of all ages are covered for the same design needs. The personalized implant design data generated by the model more represents a relatively better design and it is difficult to fully include the patient's individual needs.

[0035] In actual applications, relatively better implant designs may vary significantly from the personalized needs of patients. For example, some patients may hope to shorten the implant cycle by designing a faster - recovering implant, choose a lighter design to reduce postoperative discomfort and accelerate recovery. Some patients may place more emphasis on the bite - force adaptability and long - term stability of the implant, avoiding complications that may be caused by overly rapid healing. Such patients may prefer to use more stable and larger - sized implants, and may even choose heavier materials to ensure that the implant can withstand greater chewing forces and pressures during long - term use.

[0036] In response to the above, in the process of realizing patient - matching implant construction, the present invention introduces uncertainty into the implant structure generation model. During the process of the model learning the implicit feature relationship between oral structure and physiological data and implant design data in the learning sample data, it learns the range of structural adaptability, that is, given the oral structure and physiological data, it learns the reference range of implant design data. Finally, after obtaining the patient's three - dimensional oral CT scan data and physiological data, the model is analyzed to generate an implant design scheme with a certain degree of flexibility and adaptability, that is, some parameters are given an optional range. While ensuring that the implant has basic adaptability, such as stability and other properties meeting the design requirements, a relatively flexible adjustment range is provided, so as to determine the best implant design scheme based on the target optimization algorithm, which can be further optimized according to the personalized needs of the patient. Specifically, in this process, an analysis function for quantitatively analyzing the stability and recovery speed of the implant is constructed, and combined with the personalized needs of the patient, a personalized target optimization function adapted to the patient's personalized needs is finally constructed. Then, the target optimization algorithm is used to perform optimization analysis in the reference design data generated by the model. Finally, according to the personalized needs of the patient, a dental implant design scheme that meets the patient's personalized needs is generated, so as to construct a target implant that highly matches the patient, improving the matching degree between the implant and the patient's personalized needs while meeting the clinical requirements of the implant, and ultimately achieving the effect of greatly enhancing the patient's dental implant experience.

[0037] In the above content, the implant structure generation model in step S1 is a deep - learning model through multi - layer processing and optimization, used to generate reference implant design data according to the patient's oral structure and physiological characteristics. The main components of this model include an input layer, a feature extraction layer, a geometric shape constraint optimization layer, a material compatibility optimization layer, and an output layer.

[0038] The input layer is used to receive the patient's three - dimensional oral CT scan data and physiological data, which contain reference data for designing and analyzing features such as the stability and recovery ability of the implant.

[0039] The feature extraction layer is used to extract and fuse features from the three-dimensional oral CT scan data and physiological data of the patient. For example, a convolutional neural network is used to process the input three-dimensional CT scan data, and multiple convolutional layers are used to extract features from the CT image data and the ReLU activation function is adopted to make the extracted features have high effectiveness and robustness. A feature encoder is used to perform embedding encoding on the physiological data, map the features to a high-dimensional space, and realize the fusion of the proposed physiological features and CT features, and finally generate a unified feature representation. For example, implant design requirement data containing oral structure features and physiological features is generated.

[0040] The geometric shape constraint optimization layer is used to perform geometric constraint-based structural adjustment and optimization on the initial structure of the implant based on the implant design requirement data to generate the target optimized structure of the implant.

[0041] Among them, the geometric shape constraint optimization layer includes two components: a deformation unit and a constraint convolution unit. The deformation unit is used to process the oral structure features of the patient and use the adaptive grid technology to optimize the geometric shape of the initial structure of the implant. This process ensures that the geometric shape of the implant matches the spatial structure of the patient's oral cavity and generates a stage-optimized structure. Among them, the shape of the deformation unit mainly adopts the deformation method, and the shape optimization of the implant is realized by learning operations such as translation, rotation, stretching, and compression. Specifically, the iterative optimization of the implant structure is realized based on the adaptive grid technology to initially generate an optimized structure that conforms to the patient's bone structure.

[0042] The constraint convolution unit is used to perform convolution operations on the stage-optimized structure generated by the deformation unit to extract geometric convolution features. Geometric constraint conditions are introduced in this process to ensure that the finally generated implant shape meets the anatomical requirements of the patient's bone, realize the further optimization of the structure generated by the deformation unit, and make it geometrically more adaptable to the patient's oral structure. On the basis of the deformation operation, other constraint conditions such as local geometric consistency and contact surface constraints are combined to achieve more detailed local optimization. Specifically, convolution operations are used to process multiple local regions to optimize the local structure, and filters for geometric constraints are set in the convolution network. In traditional convolution operations, the filter usually only focuses on the local features of the image. In geometric constraint convolution, the filter will take into account the spatial position and neighborhood information of the point cloud, so that the convolution operation is not only based on the simple relationship between pixels or points, but also considers the geometric constraints. In this process, the geometric adaptability of the generated implant and the contact area of the patient's bone is optimized through a specific loss function, and the spatial features between the implant and the oral structure are captured through the constraint convolution operation, and then the detailed morphological adjustment of the implant is realized. Among them, the constraint convolution unit adopts a multi-output design, so that the finally output implant structure is range data.

[0043] The material compatibility optimization layer is used to optimize the material matching of the implant based on the implant design requirement data, and generate the material characteristic data of the implant. Among them, the role of the material compatibility optimization layer is to optimize the material selection of the implant based on the physiological characteristics of the patient, ensuring the biocompatibility and functionality of the material. The main component of this layer is the MLP unit (multi-layer perceptron), which processes the physiological characteristics of the patient through the MLP unit to generate the material characteristic data of the implant that matches the physiological characteristics. The MLP unit realizes the generation of the material characteristic data of the implant that matches the patient's physiological characteristics by learning the relationship between the implant materials suitable for patients with different physiological data in the sample data.

[0044] The output layer is used to fuse and output the optimization results of the geometric shape constraint optimization layer and the material compatibility optimization layer, and generate the implant design reference data that matches the patient's three-dimensional oral CT scan data and physiological data. Among them, the output layer includes a structural design output unit and a material design output unit. The structural design output unit is used to output the target optimized structure of the implant, that is, the implant structure data corrected by geometric shape optimization and local constraint optimization, so that the structure of the implant can perfectly match the patient's three-dimensional oral CT scan data to ensure the stability and comfort of the implant. The material design output unit is used to output the material characteristic data of the implant, that is, the material characteristic data that matches the patient's physiological characteristics output after material compatibility optimization. This data describes the material selection and processing method of the implant to ensure the compatibility and long-term stability of the implant in the patient's bone. The structural design output unit among them also adopts a multi-output design similar to the constraint convolution unit, so that the output result includes the corresponding structural parameter range. For example, a Gaussian distribution output layer is added to the last layer of the network, so that the output of the geometric feature is not just a value, but a mean and a standard deviation, so as to determine the corresponding parameter value range according to the mean and the standard deviation.

[0045] For the above implant structure generation model, it is obtained by constructing a first training dataset through the historical dental implant data of multiple patients and training with the first training dataset. Specifically, it includes the following contents: Extract multiple groups of first sample data from the historical dental implant data of multiple patients. Each group of first sample data includes sample oral scan data, sample physiological data, and corresponding sample implant design data. The first training dataset is constructed according to the multiple groups of first sample data.

[0046] In this embodiment, the historical dental implant data of the patient at least includes the patient's sample oral scan data, sample physiological data, sample implant design data, and dental implant surgery data, etc. The oral structure data and physiological data contain reference information for implant design. The implant design data is the design data such as the geometric shape, size, and material properties of the implant designed according to the patient's actual personal information. The data source can be formulated by professional doctors or dental experts based on the patient's oral condition and used as the true label to guide the model learning. The dental implant surgery data includes the patient's condition during the surgery using the implant, such as the stability of the implant during the implantation process and the patient's recovery time, etc. A first training dataset containing multiple groups of first sample data is constructed through these data. The structure of each group of sample data includes the correspondence between the oral scan data, physiological data, and implant design data.

[0047] During the process of training the implant structure generation model using the first training dataset, the goal of the model is to generate personalized implant design data by analyzing the input oral scan data and physiological data. To ensure that the model can accurately learn and optimize the correspondence between the oral scan data, physiological data, and implant design data during the training process, corresponding loss functions are designed for the local objectives of different network layers to reflect the difference between the implant design generated by the model and the actual design.

[0048] Specifically, for the geometric optimization loss of the geometric shape constraint optimization layer, the goal of the geometric shape constraint optimization layer is to generate an implant geometric shape that highly matches the patient's oral structure. It is necessary to learn the correspondence between the patient's oral scan data and the dental implant structure in the sample to learn how to optimize the geometric shape of the implant. In this process, the following two losses are introduced, namely the deformation optimization loss and the constrained convolution optimization loss.

[0049] The deformation optimization loss is used to measure the difference between the stage-optimized structure generated by the deformation unit and the sample implant structure corresponding to the implant design data of the samples in the training dataset. In this embodiment, the Chamfer distance between the two is calculated to measure the similarity between the point sets corresponding to the two structures. During the model training process, by minimizing the deformation optimization loss, the network parameters of the deformation unit can be gradually optimized. In this way, the deformation optimization loss function of the geometric shape constraint optimization layer is constructed. The constrained convolution optimization loss is used to evaluate the difference between the geometric convolution features generated by the constrained convolution unit and the implant design data in the actual samples. Since the constrained convolution unit adopts a multi-output design, for example, the output is specifically the Gaussian distribution of the geometric convolution features, the difference between the two is measured by calculating the negative log-likelihood loss. By minimizing the negative log-likelihood loss, the weight update of the constrained convolution unit is realized, so that the constrained convolution unit can generate implant structure data that highly matches the oral structure of the patient.

[0050] Finally, a geometric optimization loss function of the geometric shape constraint optimization layer, which includes the deformation optimization loss and the constrained convolution optimization loss, is generated. Through the combined action of the two loss functions, the geometric shape constraint optimization layer can gradually optimize the generated implant structure during the training process to ensure its close fit with the anatomical structure of the patient's oral cavity.

[0051] Regarding the material optimization loss of the material compatibility optimization layer, the task of the material compatibility optimization layer is to optimize the material selection of the implant according to the physiological characteristics of the patient to ensure the best biocompatibility and long-term stability of the implant. Based on this characteristic, the material optimization loss of the material compatibility optimization layer is defined. This loss evaluates the suitability of the material selection by comparing the difference between the generated implant material characteristic data and the material data of the sample implants in the training data. The material compatibility optimization layer mainly performs data processing and material characteristic generation based on MLP units. The MLP unit is essentially a classification model that can predict suitable implant materials such as titanium alloy, ceramic, etc. according to the input physiological data. The MLP unit is very efficient in processing such cases with multiple input features and outputting discrete material categories. For each material, a probability value is given according to the physiological data, indicating which material is more suitable for the patient. In this embodiment, the mean squared error loss is specifically used to measure the error between the prediction result of the MLP unit and the true result corresponding to the sample data, so as to construct the material optimization loss function of the material compatibility optimization layer.

[0052] Construct the total loss function of the implant structure generation model through the geometric optimization loss function and the material optimization loss function. The model can optimize the geometric shape and material properties of the implant. By minimizing the difference between the geometric shape and material properties, the best implant design can be achieved. During the model training process, the total loss of the model is iteratively analyzed, and the weights of different network layers are updated based on forward propagation. After the total loss value of the model falls within the preset loss range, it indicates that the model has learned how to generate personalized implant designs according to the patient's oral scan data and physiological data, thus completing the training of the implant structure generation model. Otherwise, continue to iteratively train the implant structure generation model based on the first training dataset.

[0053] It should be noted that after training the implant structure generation model with the first training dataset, during the inference process of the model, the geometric shape constraint optimization layer can automatically generate data that is relatively well-matched to the patient's oral structure based on the deformation unit and the constraint convolution unit. To further improve the matching degree between the implant structure generated by the geometric shape constraint optimization layer and the patient's oral structure, after generating the corresponding geometric structure, the geometric adaptability between the implant shape and the bone is further locally inspected. The role is to slightly adjust the existing implant shape, specifically through geometric error calculation and fine-tuning, so that the geometric shape of the implant is more in line with the bone structure at the local level.

[0054] Specifically, for the Gaussian distribution data output by the geometric shape constraint optimization layer, calculate the structural difference parameter between the output implant structure and the oral structure features. For example, calculate the negative log-likelihood loss between the output Gaussian distribution data and the oral structure features to obtain the geometric structure difference parameter. If the geometric structure difference parameter is not greater than the preset geometric structure difference threshold, it indicates that the optimized structure is relatively ideal and can be directly output. If the geometric structure difference parameter is greater than the preset geometric structure difference threshold, it indicates that the optimization result is not yet ideal and further optimization is required.

[0055] The process of further optimization is to iteratively optimize the Gaussian distribution of the geometric convolution features in the local area based on the local correction unit to output the target optimized structure of the implant.

[0056] Specifically, if the geometric structure difference parameter is greater than the relevant threshold, the local correction unit will iteratively optimize the local area of the implant to improve the local fit between the implant and the oral structure.

[0057] In this process, the geometric structure difference parameters obtained through calculation are used to determine which local areas need to be stretched or shrunk. For example, if the gap between the contact surface of a certain area and the bone is large, the surface of the implant is stretched to make it contact with the bone. If the contact in a certain area is too tight, a shrinking operation is performed. Fine-tuning can use geometric transformation methods such as affine transformation. And after each local area optimization of the geometric convolution features based on the local correction unit, the geometric structure difference parameters between the optimized overall structure and the oral structure features are calculated. If the geometric structure difference parameters are greater than the geometric structure difference threshold, local area optimization is performed again based on the local correction unit. After iterating until the geometric structure difference parameters are not greater than the geometric structure difference threshold, the geometric convolution features generated after the last local area optimization are output as the target optimized structure of the implant. In the case where the geometric structure difference parameters are not greater than the geometric structure difference threshold, the corresponding structure is directly output. Through the above process, local fine-tuning of the structure generated by the model can be achieved, so that the finally output target optimized structure of the implant can better fit the oral structure of the patient.

[0058] In the above content, for the optimization of the implant design reference data using the target optimization algorithm based on the personalized target optimization function in step S3 to generate the implant design target data that matches the patient, the optimization process is specifically implemented based on the particle swarm optimization algorithm (PSO). Through multiple iterations of optimization and updating of the positions and velocities of the particles, the optimal design is finally achieved. The specific implementation process is as follows: The particle swarm optimization algorithm based on particle iteration limit is used to optimize the implant design reference data, including initializing the particles based on the target optimized structure in the implant design reference data. In this process, for the fixed data in the implant design reference data, it is directly used as the initial position data of the particles. For the data containing the range of structure parameters, such as the specific size data of the implant, etc., it is initialized according to the range defined by the implant design reference data, thus completing the initialization of the particle positions. Each particle represents a possible implant design solution. The position of the particle represents a combination of implant design parameters such as geometric shape, size, material properties, etc. The velocity of the particle represents the possible change trend of the design parameters during the optimization process. At the same time, the patient's three-dimensional oral CT scan data and physiological data are also added to multiple particles, specifically added to the position data of the particles. These data are only used as basic data and are not adjusted during the optimization process. They are only used as fixed parameters to calculate and analyze the fitness of each particle to guide the particles to find the implant design that best meets the patient's needs in the solution space. The iteration process only optimizes and adjusts according to the implant structure data involving range changes in the implant design reference data to determine the optimal solution.

[0059] During the iterative optimization process, a personalized objective optimization function is used to calculate the fitness of each particle. Based on the fitness of the particle and the reference data for implant design, the velocity and position of the particle are updated. Through multiple iterations, the implant design target data matching the patient is generated.

[0060] In this embodiment, the fitness of a particle is an index for evaluating the quality of the particle as a solution for implant design. The process of calculating the fitness of each particle includes performing stability analysis and restoration analysis on the implant corresponding to the particle, and calculating the stability score and restoration score.

[0061] Specifically, each particle is input into the implant stability analysis function and the implant restoration analysis function respectively, and the stability score of the particle is generated based on the finite element analysis model. Specifically, data such as the maximum stress and contact area corresponding to the particle are generated according to the finite element analysis model, and the stability score of the particle is calculated through the previously constructed implant stability analysis function. And the implant-related data corresponding to the particle is input into the implant restoration analysis model, and the restoration score of the particle is generated through the implant restoration analysis model. Finally, the stability score and restoration score of the particle are substituted into the personalized objective optimization function to calculate the fitness of the particle.

[0062] For the personalized objective optimization function, the structure is as follows:

[0063] In the formula, is the personalized objective function, is the implant stability analysis function, which is used to output the stability score of the particle, is the implant restoration analysis function, which is used to output the restoration score of the particle, 、 are the stability weight and the restoration weight respectively, which are specifically quantified and set based on the personalized demand data of the patient's implant. The fitness of the particle calculated through the personalized objective optimization function. Specifically considering the personalized needs of the patient, such as paying more attention to stability or restoration effect, it can well analyze the matching situation between each particle and the patient. The higher the fitness, the more the design scheme corresponding to the particle meets the personalized needs of the patient.

[0064] After multiple iterations, the particle swarm optimization algorithm will gradually approach the optimal solution, that is, generate an implant design target data that most meets the personalized needs of the patient. This data not only considers the oral structure characteristics of the patient, but also comprehensively considers personalized physiological needs such as stability and restoration speed, ensuring that the implant can provide the best stability and restoration effect. The finally generated implant design target data will be input into the 3D printing device to complete the printing and manufacturing of the patient-matched implant.

[0065] In the above content, for the implant restoration analysis model, through multi-dimensional data input, the restoration effect corresponding to the patient after dental implant surgery using the implant can be comprehensively evaluated. Among them, the model construction and training include: A second training dataset is constructed based on the historical dental implant data of multiple patients, and the implant restoration analysis model is trained through the second training dataset. Specifically, this process includes extracting multiple groups of second sample data from multiple groups of historical dental implant data. Each group of second sample data includes sample oral scan data, sample physiological data, sample implant structure data, and the corresponding sample implant restoration score. The sample implant restoration score can be quantitatively set in combination with the actual restoration duration data of the patients in the historical dental implant data. The correlation relationships among the sample oral scan data, sample physiological data, and sample implant structure data included therein and the sample implant restoration score are learned and mined by training the implant restoration analysis model, and finally, the corresponding restoration score is predicted based on the input oral structure data, physiological data, and implant structure design data.

[0066] In this embodiment, the implant restoration analysis model is specifically a random forest model. This model makes predictions based on the voting mechanism of multiple decision trees and gradually learns how to predict the restoration effect score of the implant from the oral scan data, physiological data, and implant structure design data through the relationship between the features and labels in the training dataset. During the model training process, multiple groups of sample oral scan data, sample physiological data, and sample implant structure data in the second training dataset are used as the input features for training the implant restoration analysis model, and the sample implant restoration scores corresponding to multiple groups of sample oral scan data, sample physiological data, and sample implant structure data are used as the targets for training the implant restoration analysis model. The implant restoration analysis model is trained using the second training dataset. Finally, through the implant restoration analysis model, the construction of a personalized objective optimization function is realized. During the process of calculating the fitness of the particles through the personalized objective optimization function, the restoration score corresponding to the particle is output through the implant restoration analysis model, so as to calculate the fitness data of each particle and complete the optimization process of the particles.

[0067] The embodiment of the present invention also provides a patient-matched implant, which is specifically constructed by using the patient-matched implant construction method as described above. This implant can improve the matching degree between the implant and the personalized needs of the patient while meeting the clinical requirements of the implant, and finally achieve the effect of greatly improving the patient's dental implant experience.

[0068] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The parts not described in detail in this specification belong to the prior art well known to those skilled in the art.

Claims

1. A method for constructing a patient-matched implant, characterized in that: include: Acquire the patient's three-dimensional oral CT scan data and physiological data, input the patient's three-dimensional oral CT scan data and physiological data into the implant structure generation model, and generate implant design reference data matching the patient through the implant structure generation model, wherein the implant structure generation model includes an input layer, a feature extraction layer, a geometric shape constraint optimization layer, a material compatibility optimization layer, and an output layer; A finite element analysis model is constructed according to the patient's three-dimensional oral CT scan data, an implant stability analysis function is determined based on the finite element analysis model, an implant recovery analysis function is determined through the implant recovery analysis model, and an initial target optimization function is constructed according to the implant stability analysis function and the implant recovery analysis function; Obtain the patient's personalized implant demand data, update the initial target optimization function according to the personalized implant demand data to generate a personalized target optimization function, optimize the implant design reference data using a target optimization algorithm based on the personalized target optimization function, and generate implant design target data that matches the patient; The target implant design data matching the patient is input into a 3D printing device, the target implant design data matching the patient is processed by the 3D printing device, and a target implant matching the patient is printed and generated.

2. A method for constructing a patient-matched implant according to claim 1, characterized in that: For implant structure generation models, also includes: The input layer is used to receive the patient's three-dimensional oral CT scan data and physiological data, the feature extraction layer is used to extract and fuse the patient's three-dimensional oral CT scan data and physiological data to generate implant design requirement data containing oral structure features and physiological features, the geometric shape constraint optimization layer is used to perform structural adjustment optimization based on geometric constraints on the initial structure of the implant based on the implant design requirement data to generate the target optimized structure of the implant, the material compatibility optimization layer is used to perform material matching optimization on the implant based on the implant design requirement data to generate the material property data of the implant, the output layer is used to fuse and output the optimization results of the geometric shape constraint optimization layer and the material compatibility optimization layer to generate implant design reference data that matches the patient's three-dimensional oral CT scan data and physiological data, the output layer includes a structure design output unit and a material design output unit, the structure design output unit is used to output the target optimized structure of the implant, and the material design output unit is used to output the material property data of the implant; A first training data set is constructed by acquiring historical dental implant data of multiple patients, and an implant structure generation model is trained using the first training data set.

3. A method for constructing a patient-matched implant according to claim 2, characterized in that: For geometry-constrained and material-compatible optimization layers, this also includes: The geometric shape constraint optimization layer adjusts and optimizes the initial structure of the implant based on the geometric constraints based on the implant design requirement data, and generates the target optimized structure of the implant, including: The geometric shape constraint optimization layer includes a deformation unit and a constraint convolution unit. The constraint convolution unit adopts a multi-output design. The deformation unit is used to process the oral structure characteristics in the implant design requirement data. The adaptive grid technology is used to optimize the geometric structure of the initial structure of the implant based on the oral structure characteristics, and the structure is optimized in the generation stage. The stage optimization structure is input into the constrained convolution unit, and the stage optimization structure is convolved by the constrained convolution unit to extract the Gaussian distribution of the geometric convolution feature, and the geometric structure difference parameter between the geometric convolution feature and the oral structure feature is calculated. If the geometric structure difference parameter is greater than the geometric structure difference threshold, the Gaussian distribution of the geometric convolution feature is iteratively optimized in the local area based on the local correction unit to output the target optimization structure of the implant, otherwise the Gaussian distribution of the geometric convolution feature is directly output as the target optimization structure of the implant; The material compatibility optimization layer processes the physiological characteristics in the implant design requirement data based on the MLP unit and generates material property data of the implant that matches the physiological characteristics.

4. A method for constructing a patient-matched implant according to claim 3, characterized in that: A first training data set is constructed by acquiring historical dental implant data of multiple patients, and an implant structure generation model is obtained by training the first training data set, including: Extracting multiple groups of first sample data from historical dental implant data of multiple patients, each group of first sample data includes sample oral scan data, sample physiological data and corresponding sample implant design data, and constructing a first training data set based on the multiple groups of first sample data; According to the Chamfer distance between the stage optimization structure generated by the deformation unit and the sample implant structure corresponding to the sample implant design data in the training data set, a deformation optimization loss function of the geometry constraint optimization layer is constructed; according to the negative log-likelihood loss between the Gaussian distribution of the geometric convolution feature generated by the constrained convolution unit and the sample implant structure corresponding to the sample implant design data in the training data set, a constrained convolution optimization loss function of the geometry constraint optimization layer is constructed, and a geometry optimization loss function of the geometry constraint optimization layer including deformation optimization loss and constrained convolution optimization loss is generated; According to the difference between the material property data generated by the MLP unit and the sample implant material corresponding to the sample implant design data in the training data set, the material optimization loss function of the material compatibility optimization layer is constructed, and the total loss function of the implant structure generation model is determined according to the geometric optimization loss function and the material optimization loss function; The total loss value of the model in the process of training the implant structure generation model based on the first training data set is calculated based on the total loss function. The training of the implant structure generation model is completed after the total loss value of the model is within a preset loss range. Otherwise, the implant structure generation model continues to be iteratively trained based on the first training data set.

5. A method for constructing a patient-matched implant according to claim 2, characterized in that: Based on the personalized target optimization function, the target optimization algorithm is used to optimize the implant design reference data to generate implant design target data that matches the patient, including: A particle swarm optimization algorithm based on particle iteration restriction is used to optimize the implant design reference data, including initializing particles based on the target optimization structure in the implant design reference data, adding the patient's three-dimensional oral CT scan data and physiological data to multiple particles, and using a personalized target optimization function to calculate the fitness of each particle. The particle speed and position are updated according to the particle fitness and the implant design reference data, and the implant design target data matching the patient is generated through multiple iterations. Wherein, calculating the fitness of each particle includes inputting each particle into an implant stability analysis function and an implant recovery analysis function respectively, generating a stability score of the particle through a finite element analysis model, generating a recovery score of the particle through an implant recovery analysis model, and substituting the stability score and the recovery score of the particle into a personalized target optimization function to calculate the fitness of the particle; For the personalized target optimization function, the structure is as follows: In the formula, is the personalized objective function, It is an implant stability analysis function, used to output the stability score of particles. It is an implant restoration analysis function, used to output the restoration score of particles. , are stability weight and recovery weight respectively.

6. A method for constructing a patient-matched implant according to claim 5, characterized in that: For implant restoration analysis models, also included: A second training data set is constructed based on historical dental implant data of multiple patients, and an implant recovery analysis model is trained using the second training data set, including extracting multiple sets of second sample data from the multiple sets of historical dental implant data, each set of second sample data comprising sample oral scan data, sample physiological data, sample implant structure data, and a corresponding sample implant recovery score; The implant recovery analysis model is a random forest model. Multiple groups of sample oral scan data, sample physiological data and sample implant structure data in the second training data set are used as input features for training the implant recovery analysis model. The sample implant recovery scores corresponding to the multiple groups of sample oral scan data, sample physiological data and sample implant structure data are used as the targets for training the implant recovery analysis model. The implant recovery analysis model is obtained by training using the second training data set.

7. A method for constructing a patient-matched implant according to claim 3, characterized in that: The local area iterative optimization of the geometric convolution features based on the local correction unit to output the target optimized structure of the implant includes: Each time the geometric convolution feature is locally optimized based on the local correction unit, the geometric structure difference parameter between the optimized geometric convolution feature and the oral structure feature is calculated. If the geometric structure difference parameter is greater than the geometric structure difference threshold, the local area optimization is performed again based on the local correction unit. After iterating until the geometric structure difference parameter is no greater than the geometric structure difference threshold, the geometric convolution feature generated after the last local area optimization is output as the target optimized structure of the implant.

8. A patient-matched implant, characterized in that: The patient-matched implant is constructed based on a patient-matched implant construction method as described in any one of claims 1-7.

Images