Dental implantation risk assessment system based on artificial intelligence
Through conditional generation adversarial networks and multi-factor risk perception models, the problems of uneven data sets and dependence on doctors' experience in traditional dental implant risk assessment methods are solved, and personalized dental implant risk assessment is achieved, which improves the reliability and accuracy of the assessment.
Patent Information
- Application Number
- CN202510630547.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-16
- Publication Date
- 2025-08-08
AI Technical Summary
Traditional dental implant risk assessment methods rely on labeled data sets of scarce and high-risk cases, which leads to the model being susceptible to overfitting common cases, low reliability in the evaluation of scarce and complex anatomical features, and relying on physician experience and two-dimensional imaging measurements, which is highly subjective and difficult to capture the interaction between complex nonlinear relationships and multimodal data, and cannot adapt to individualized differences.
The conditional generation adversarial network is used to expand data, generate synthetic images that are strictly matched with the patient's individual characteristics, and build a multi-factor risk perception model, and accurately capture multi-dimensional risk factors through anatomical structure-guided feature extraction and dynamic risk factor decoupling.
It significantly improves the personalized ability of dental implant risk assessment, solves the generalization problem of model and the problem of strong subjectivity, can accurately capture multi-dimensional risk factors, and improves the reliability and personalized evaluation ability of the assessment.
Smart Images
Figure CN120452683A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of dental implant clinical data processing, and in particular to a dental implant risk assessment system based on artificial intelligence. Background Art
[0002] Dental implant risk assessment is to predict the success rate and potential risks of dental implant surgery by comprehensively analyzing the patient's clinical information, medical history, oral health status, physiological indicators and other multi-dimensional data; it can help doctors make more scientific decisions before surgery, reduce the incidence of postoperative complications, improve the patient's implant success rate and quality of life, while avoiding unnecessary waste of medical resources and treatment delays, and realize the formulation of personalized treatment plans.
[0003] However, traditional dental implant risk assessment methods have technical problems such as relying on limited annotated datasets of scarce high-risk cases, which makes the model prone to overfitting common cases and low reliability in assessing scarce and complex anatomical features; traditional dental implant risk assessment methods have technical problems such as relying on doctor experience and two-dimensional image measurement, which are highly subjective and lack quantification, making it difficult to capture complex nonlinear relationships and the interaction of multimodal data, resulting in one-sided risk assessment and inability to adapt to individual differences. Summary of the Invention
[0004] In response to the above situation, in order to overcome the shortcomings of the existing technology, the present invention provides an artificial intelligence-based dental implant risk assessment system. The traditional dental implant risk assessment method relies on a limited labeled data set of scarce high-risk cases, which makes the model prone to overfitting common cases and has low reliability in assessing scarce and complex anatomical features. This solution creatively uses a conditional generative adversarial network for data expansion. It can dynamically generate synthetic images that strictly match the individual characteristics of the patient while ensuring that the generated images conform to the real anatomical features, thereby solving the model generalization problem caused by the uneven distribution of cases in real data; the traditional dental implant risk assessment method relies on doctor experience and two-dimensional image measurement, which is highly subjective and insufficiently quantified, and makes it difficult to capture complex nonlinear relationships and the interaction of multimodal data, resulting in one-sided risk assessment and inability to adapt to individual differences. This solution creatively uses a multi-factor risk perception model as an implant risk model. Through anatomically guided feature extraction and dynamic risk factor decoupling, it can accurately capture multi-dimensional risk factors and significantly improve personalized assessment capabilities.
[0005] The technical solution adopted by the present invention is as follows: the artificial intelligence-based dental implant risk assessment system provided by the present invention includes an information acquisition module, a data optimization module, an image data enhancement module, an implant risk model construction module and a dental implant risk assessment module;
[0006] The information acquisition module obtains the original data set for risk assessment by performing data collection;
[0007] The data optimization module uses data optimization methods such as data trimming, feature engineering, data standardization and data set segmentation to obtain the data set to be evaluated, the preliminary training set and the preliminary test set;
[0008] The image data enhancement module performs data expansion by constructing a conditional generative adversarial network to obtain an enhanced training set and an enhanced test set;
[0009] The planting risk model construction module constructs a multi-factor risk perception model as a planting risk model to obtain a planting risk model;
[0010] The dental implant risk assessment module performs a dental implant risk assessment based on the dataset to be assessed by adopting the implant risk model to obtain implant risk reference data.
[0011] Furthermore, in the information acquisition module, the risk assessment original data set specifically includes a historical risk assessment original data set and a current risk assessment original data set. Both the historical risk assessment original data set and the current risk assessment original data set include imaging data and clinical data. The historical risk assessment original data set also includes risk level label data.
[0012] Furthermore, in the data optimization module, the data trimming is specifically to remove missing values and duplicate values in the original data, the feature engineering is specifically to perform feature selection, feature extraction and feature construction on the original data, the data standardization is specifically to perform data standardization on the original data using the minimum-maximum normalization method, and the data set segmentation is used to segment the data set;
[0013] The current risk assessment original data set is optimized through the data trimming, the feature engineering and the data standardization to obtain the data set to be evaluated, and the historical risk assessment original data set is optimized through the data trimming, the feature engineering, the data standardization and the data set segmentation to obtain a preliminary training set and a preliminary test set.
[0014] Furthermore, in the image data enhancement module, image data enhancement is specifically performed by constructing a conditional generative adversarial network model and performing image data expansion on the preliminary training set and the preliminary test set to obtain an enhanced training set and an enhanced test set. The conditional generative adversarial network model specifically includes a generator module and a discriminator module. The generator module is specifically a conditional U-Net, and the discriminator module is specifically a dense convolutional network.
[0015] The image data enhancement module specifically includes designing condition input, adjusting feature conditions, designing loss functions, constructing and training models, and expanding image data;
[0016] The design condition input is used to design the condition input required by the generator module, specifically integrating clinical data and risk level label data as the condition input of the generator module;
[0017] The feature condition adjustment is used to linearly adjust the intermediate layer features of the generator module based on the conditional input. Specifically, based on the conditional input of the generator module, the output features of each convolutional layer and upsampling layer in the generator module are processed using feature linear modulation. The formula used is as follows:
[0018] ;
[0019] Where, represents the characteristic linear modulation scaling coefficient, represents the characteristic linear modulation shift coefficient, Represents the multilayer perceptron operation function used to generate feature linear modulation parameters, represents the intermediate layer features of the modulated generator module, represents the intermediate layer features of the generator module, Represents element-by-element multiplication, and Cd represents the conditional input of the generator module;
[0020] The designed loss function specifically combines the adversarial loss and the bone density regression loss as the total model loss of the conditional generative adversarial network model. The bone density regression loss is used to constrain the bone density distribution of the generated image by the generator module to conform to the real anatomical characteristics. The loss function is designed using the following formula:
[0021] ;
[0022] Where, represents the adversarial loss value, represents the expected calculation function, represents the discriminator output function, represents the real image data, Indicates that the generator module generates image data, represents the regression loss of bone density, represents the fully connected layer function, represents the dense convolutional network feature extraction function of the discriminator module, Indicates the real bone density data, Represents the total loss value of the conditional generation adversarial network model, Indicates calculation of L2 norm;
[0023] The constructing and training the model specifically comprises constructing a conditional generative adversarial network model through the design condition input, the feature condition adjustment and the design loss function, training the model based on the preliminary training set, and verifying the model performance based on the preliminary test set to obtain the conditional generative adversarial network model;
[0024] The image data expansion specifically involves using the conditional generative adversarial network model to expand the image data of the preliminary training set and the preliminary test set to obtain an enhanced training set and an enhanced test set.
[0025] Furthermore, in the implant risk model construction module, a model required for assessing dental implant risks is constructed, specifically a multi-factor risk perception model is constructed as an implant risk model;
[0026] The implant risk model construction module specifically includes dual-branch feature processing, gated feature fusion, risk factor extraction, obtaining model output, calculating model loss, and model construction and training;
[0027] The dual-branch feature processing is used to process imaging data features and clinical data features, and includes:
[0028] Image data branch construction is used to construct an image data branch for processing image data features, including:
[0029] Macro feature processing is used to capture the overall structure of the entire jaw, specifically by downsampling through large-stride convolution;
[0030] Mesoscopic feature processing is used to analyze the local morphology of the alveolar bone, specifically by expanding the receptive field through hollow convolution;
[0031] Micro-feature processing is used to extract gingival texture detail information, specifically by capturing detail information through 1×1×1 convolution;
[0032] Multi-scale feature fusion is used to integrate multi-scale information. The formula used is as follows:
[0033] ;
[0034] Where, Represents the multi-scale features of the image, Represents the splicing operation function, represents the three-dimensional convolution function, represents the 1×1×1 convolution kernel weight, Represents the macroscopic features of the image, Represents the mesoscopic features of the image, Represents the microscopic features of the image;
[0035] Clinical data branch construction, used to construct a clinical data branch for processing clinical data features, specifically dynamically adjusting the weights of clinical data features through Gini impurity;
[0036] The gated feature fusion is used to fuse the output features of the imaging data branch and the output features of the clinical data branch through a gating mechanism, and includes:
[0037] Calculate channel attention, specifically by calculating channel attention based on multi-scale features of the image;
[0038] Calculate spatial attention, specifically by calculating spatial attention based on multi-scale features of the image;
[0039] Gated fusion is specifically based on the gating mechanism to fuse multi-scale image features and clinical weighted features. The formula used is as follows:
[0040] ;
[0041] Where Gw represents the gate weight, represents the gate generation weight, represents the gate generation bias term, represents the fusion feature, represents the sigmoid function, represents clinically weighted features, represents channel attention, Indicates spatial attention;
[0042] The risk factor extraction is used to extract and generate multiple risk factor features, including:
[0043] Bone risk factor extraction, used to extract risk features of bone-related information, specifically by using an independent gated recurrent unit to process bone-related imaging data, the bone-related imaging data specifically including cone-beam CT imaging data, oral panoramic X-ray imaging data, and periapical imaging data;
[0044] Soft tissue risk factor extraction, used to extract risk features of soft tissue related information, specifically by using an independent gated recurrent unit to process soft tissue related image data, specifically intraoral scan image data;
[0045] Clinical risk factor extraction, used to extract risk features from clinical data, specifically by using a multi-layer perceptron to process clinical data;
[0046] Factor independence constraints are used to enforce independence between different risk factors. Specifically, bone risk factors, soft tissue risk factors, and clinical risk factors are concatenated into one matrix and the covariance regularization loss is minimized.
[0047] The model output is obtained to calculate the output of the multi-factor risk perception model, and the formula used is as follows:
[0048] ;
[0049] Where, represents the final risk characteristics, represents the fusion feature weight, represents the weight of the i-th risk factor, represents the i-th risk factor, represents the model prediction output, represents the softmax function;
[0050] The calculation model loss is specifically a combination of cross entropy loss and covariance regularization loss as the total model loss of the multi-factor risk perception model;
[0051] The model is constructed and trained, specifically by constructing a multi-factor risk perception model through the dual-branch feature processing, the gated feature fusion, the risk factor extraction, the acquisition of model output and the calculation of model loss, and training the model based on the enhanced training set, and verifying the model performance based on the enhanced test set to obtain a multi-factor risk perception model as a planting risk model.
[0052] Furthermore, in the dental implant risk assessment module, the implant risk model is specifically used to perform dental implant risk assessment based on the data set to be assessed, to obtain implant risk reference data, and the patient's dental implant risk is comprehensively assessed based on the implant risk reference data.
[0053] The beneficial effects achieved by the present invention using the above scheme are as follows:
[0054] (1) In response to the technical problems that traditional dental implant risk assessment methods rely on limited labeled data sets of scarce high-risk cases, which makes the model prone to overfitting common cases and has low reliability in evaluating scarce and complex anatomical features, this solution creatively uses a conditional generative adversarial network for data expansion. It can dynamically generate synthetic images that strictly match the individual characteristics of the patient while ensuring that the generated images conform to the real anatomical features, thereby solving the model generalization problem caused by the uneven distribution of cases in real data.
[0055] (2) In view of the technical problems that traditional dental implant risk assessment methods rely on doctors' experience and two-dimensional image measurement, are highly subjective and lack quantification, and are difficult to capture complex nonlinear relationships and the interaction of multimodal data, resulting in one-sided risk assessment and inability to adapt to individual differences, this scheme creatively adopts a multi-factor risk perception model as an implant risk model. Through anatomically guided feature extraction and dynamic risk factor decoupling, it can accurately capture multi-dimensional risk factors and significantly improve personalized assessment capabilities. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 A schematic diagram of the modules of the artificial intelligence-based dental implant risk assessment system provided by the present invention;
[0057] Figure 2 This is a flowchart of the data optimization module;
[0058] Figure 3 This is a flowchart of the image data enhancement module;
[0059] Figure 4 Schematic diagram of the process for building modules for the planting risk model;
[0060] Figure 5 Schematic diagram of the process of risk factor extraction for building modules of the implant risk model.
[0061] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION
[0062] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0063] In the description of the present invention, it should be understood that terms such as "upper", "lower", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.
[0064] Example 1, see Figure 1The artificial intelligence-based dental implant risk assessment system provided by the present invention includes an information acquisition module, a data optimization module, an image data enhancement module, an implant risk model construction module and a dental implant risk assessment module;
[0065] The information acquisition module obtains the original data set for risk assessment by performing data collection;
[0066] The data optimization module uses data optimization methods such as data trimming, feature engineering, data standardization and data set segmentation to obtain the data set to be evaluated, the preliminary training set and the preliminary test set;
[0067] The image data enhancement module performs data expansion by constructing a conditional generative adversarial network to obtain an enhanced training set and an enhanced test set;
[0068] The planting risk model construction module constructs a multi-factor risk perception model as a planting risk model to obtain a planting risk model;
[0069] The dental implant risk assessment module performs a dental implant risk assessment based on the dataset to be assessed by adopting the implant risk model to obtain implant risk reference data.
[0070] Example 2, see Figure 1 In the information acquisition module, the risk assessment original data set specifically includes a historical risk assessment original data set and a current risk assessment original data set. Both the historical risk assessment original data set and the current risk assessment original data set include imaging data and clinical data. The historical risk assessment original data set also includes risk level label data. The imaging data specifically includes cone beam CT imaging data, intraoral scanning imaging data, oral panoramic X-ray imaging data and periapical imaging data. The clinical data specifically includes patient basic information data, patient medical history data, patient health status data, patient bone density data, patient oral status data, surgery type data and implant type data. The risk level label data is specifically a risk level label, including no risk, low risk, medium risk and high risk.
[0071] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In the data optimization module, the data trimming is specifically to remove missing values and duplicate values in the original data. The feature engineering is specifically to perform feature selection, feature extraction and feature construction on the original data. The data standardization is specifically to use the minimum-maximum normalization method to perform data standardization on the original data. The data set segmentation is used to segment the data set.
[0072] The current risk assessment original data set is optimized through the data trimming, the feature engineering and the data standardization to obtain the data set to be evaluated, and the historical risk assessment original data set is optimized through the data trimming, the feature engineering, the data standardization and the data set segmentation to obtain a preliminary training set and a preliminary test set.
[0073] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In the image data enhancement module, the image data enhancement is performed by constructing a conditional generative adversarial network model and performing image data expansion on the preliminary training set and the preliminary test set to obtain an enhanced training set and an enhanced test set. The conditional generative adversarial network model specifically includes a generator module and a discriminator module. The generator module is specifically a conditional U-Net, and the discriminator module is specifically a dense convolutional network.
[0074] The image data enhancement module specifically includes designing condition input, adjusting feature conditions, designing loss functions, constructing and training models, and expanding image data;
[0075] The design condition input is used to design the condition input required by the generator module, specifically integrating clinical data and risk level label data as the condition input of the generator module. The formula used is as follows:
[0076] ;
[0077] Where Cd represents the conditional input of the generator module, represents the multilayer perceptron function for processing clinical data, c represents the conditional information sampled from the clinical data distribution, represents the embedding processing function, and lab represents the risk level label data;
[0078] The feature condition adjustment is used to linearly adjust the intermediate layer features of the generator module based on the conditional input. Specifically, based on the conditional input of the generator module, the output features of each convolutional layer and upsampling layer in the generator module are processed using feature linear modulation. The formula used is as follows:
[0079] ;
[0080] Where, represents the characteristic linear modulation scaling coefficient, represents the characteristic linear modulation shift coefficient, Represents the multilayer perceptron operation function used to generate feature linear modulation parameters, represents the intermediate layer features of the modulated generator module, represents the intermediate layer features of the generator module, represents element-wise multiplication;
[0081] The designed loss function specifically combines the adversarial loss and the bone density regression loss as the total model loss of the conditional generative adversarial network model. The bone density regression loss is used to constrain the bone density distribution of the generated image by the generator module to conform to the real anatomical characteristics. The loss function is designed using the following formula:
[0082] ;
[0083] Where, represents the adversarial loss value, represents the expected calculation function, represents the discriminator output function, represents the real image data, Indicates that the generator module generates image data, represents the regression loss of bone density, represents the fully connected layer function, represents the dense convolutional network feature extraction function of the discriminator module, Indicates the real bone density data, Represents the total loss value of the conditional generation adversarial network model, Indicates calculation of L2 norm;
[0084] The constructing and training the model specifically comprises constructing a conditional generative adversarial network model through the design condition input, the feature condition adjustment and the design loss function, training the model based on the preliminary training set, and verifying the model performance based on the preliminary test set to obtain the conditional generative adversarial network model;
[0085] The image data expansion specifically involves using the conditional generative adversarial network model to expand the image data of the preliminary training set and the preliminary test set to obtain an enhanced training set and an enhanced test set.
[0086] By performing the above operations, this solution creatively uses a conditional generative adversarial network for data expansion to address the technical problems of traditional dental implant risk assessment methods, which rely on limited labeled datasets of scarce high-risk cases, resulting in the model easily overfitting common cases and low reliability in evaluating scarce and complex anatomical features. This solution can dynamically generate synthetic images that strictly match the individual characteristics of the patient while ensuring that the generated images conform to the real anatomical features, thereby solving the model generalization problem caused by the uneven distribution of cases in real data.
[0087] Example 5, see Figure 1 、 Figure 4 and Figure 5, based on the above embodiment, this embodiment is used to construct a model required for assessing dental implant risks in the implant risk model construction module, specifically to construct a multi-factor risk perception model as the implant risk model;
[0088] The implant risk model construction module specifically includes dual-branch feature processing, gated feature fusion, risk factor extraction, obtaining model output, calculating model loss, and model construction and training;
[0089] The dual-branch feature processing is used to process imaging data features and clinical data features, and includes:
[0090] Image data branch construction is used to construct an image data branch for processing image data features, including:
[0091] Macro feature processing is used to capture the overall structure of the entire jaw. Specifically, it is performed by downsampling through large-stride convolution. The formula used is as follows:
[0092] ;
[0093] Where, Represents the macroscopic features of the image, represents the ReLU activation function, represents the batch normalization function, represents the three-dimensional convolution function, represents the image input features, Represents the 3×3×3 convolution kernel weight;
[0094] Mesoscopic feature processing is used to analyze the local morphology of the alveolar bone. Specifically, the receptive field is expanded through hollow convolution. The formula used is as follows:
[0095] ;
[0096] Where, Represents the mesoscopic features of the image, represents the three-dimensional dilated convolution function, Indicates that the expansion rate is 2, Represents the weight of the 5×5×5 hole convolution kernel;
[0097] Micro-feature processing is used to extract gingival texture detail information. Specifically, 1×1×1 convolution is used to capture detail information. The formula used is as follows:
[0098] ;
[0099] Where, Represents the microscopic features of the image, Represents the 1×1×1 convolution kernel weight;
[0100] Multi-scale feature fusion is used to integrate multi-scale information. The formula used is as follows:
[0101] ;
[0102] Where, Represents the multi-scale features of the image, Represents the splicing operation function;
[0103] Clinical data branch construction is used to construct a clinical data branch for processing clinical data features. Specifically, the weights of clinical data features are dynamically adjusted through Gini impurity. The formula used is as follows:
[0104] ;
[0105] Where, represents the Gini impurity calculation function, represents the nth clinical data feature, N represents the total number of clinical data features, represents the proportion of the nth clinical data feature in all clinical data features, Fs represents the feature importance score, represents the mth clinical data feature, represents the total clinical data characteristics, represents clinically weighted characteristics;
[0106] The gated feature fusion is used to fuse the output features of the imaging data branch and the output features of the clinical data branch through a gating mechanism, and includes:
[0107] Calculate channel attention, specifically based on the multi-scale features of the image, and the formula used is as follows:
[0108] ;
[0109] Where, represents channel attention, represents the sigmoid function, represents the channel attention generation weight, represents the average pooling function, Represents the channel attention generation bias term;
[0110] Calculate spatial attention, specifically based on the multi-scale features of the image. The formula used is as follows:
[0111] ;
[0112] Where, represents spatial attention, represents the maximum pooling function;
[0113] Gated fusion is specifically based on the gating mechanism to fuse multi-scale image features and clinical weighted features. The formula used is as follows:
[0114] ;
[0115] Where Gw represents the gate weight, represents the gate generation weight, represents the gate generation bias term, represents fusion features;
[0116] The risk factor extraction is used to extract and generate multiple risk factor features, including:
[0117] Bone risk factor extraction is used to extract risk features of bone-related information. Specifically, an independent gated recurrent unit is used to process bone-related imaging data. The bone-related imaging data specifically includes cone-beam CT imaging data, oral panoramic X-ray imaging data, and periapical imaging data. The bone risk factor extraction formula is as follows:
[0118] ;
[0119] Where, represents bone risk factors, represents the gated recurrent unit function, represents the flattening function, Represents bone-related imaging data, represents the parameters of the gated recurrent unit for processing bone-related imaging data;
[0120] Soft tissue risk factor extraction is used to extract risk characteristics of soft tissue related information. Specifically, an independent gated recurrent unit is used to process soft tissue related image data. The soft tissue related image data is specifically intraoral scan image data. The soft tissue risk factor extraction formula is as follows:
[0121] ;
[0122] Where, represents the soft tissue risk factor, Indicates soft tissue related imaging data, represents the parameters of the gated recurrent unit for processing soft tissue related imaging data;
[0123] Clinical risk factor extraction is used to extract risk characteristics of clinical data. Specifically, a multi-layer perceptron is used to process clinical data. The formula used is as follows:
[0124] ;
[0125] Where, represents clinical risk factors, represents the multilayer perceptron operation function used to generate clinical risk factors;
[0126] The factor independence constraint is used to enforce independence between different risk factors. Specifically, the bone risk factor, soft tissue risk factor, and clinical risk factor are concatenated into one matrix and the covariance regularization loss is minimized. The covariance regularization loss is calculated using the following formula:
[0127] ;
[0128] In the formula, Cm represents the risk factor covariance matrix, Sa represents the number of samples, Rm represents the risk factor concatenation matrix, and T represents the transposition operation. represents the covariance regularization loss value, represents the regularization hyperparameter, Indicates the calculation of Frobenius norm;
[0129] The model output is obtained to calculate the output of the multi-factor risk perception model, and the formula used is as follows:
[0130] ;
[0131] Where, represents the final risk characteristics, represents the fusion feature weight, represents the weight of the i-th risk factor, represents the i-th risk factor, represents the model prediction output, represents the softmax function;
[0132] The calculation model loss is specifically a combination of cross entropy loss and covariance regularization loss as the total model loss of the multi-factor risk perception model. The formula used is as follows:
[0133] ;
[0134] Where, represents the total loss value of the multi-factor risk perception model, Represents the cross entropy loss value;
[0135] The model is constructed and trained, specifically by constructing a multi-factor risk perception model through the dual-branch feature processing, the gated feature fusion, the risk factor extraction, the acquisition of model output and the calculation of model loss, and training the model based on the enhanced training set, and verifying the model performance based on the enhanced test set to obtain a multi-factor risk perception model as a planting risk model.
[0136] By performing the above operations, this solution creatively adopts a multi-factor risk perception model as an implant risk model, which can accurately capture multi-dimensional risk factors through anatomically guided feature extraction and dynamic risk factor decoupling, and significantly improve the ability of personalized assessment.
[0137] Example 6, see Figure 1 This embodiment is based on the above embodiment. In the dental implant risk assessment module, the implant risk model is used to perform a dental implant risk assessment based on the data set to be evaluated, to obtain implant risk reference data, and to comprehensively assess the patient's dental implant risk based on the implant risk reference data.
[0138] 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," "comprising," 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.
[0139] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.
[0140] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.
Claims
1. The artificial intelligence-based dental implant risk assessment system is characterized by: The system includes an information acquisition module, a data optimization module, an image data enhancement module, an implant risk model construction module and a dental implant risk assessment module; The information acquisition module obtains a risk assessment original data set through data collection, wherein the risk assessment original data set specifically includes a historical risk assessment original data set and a current risk assessment original data set; The data optimization module uses data optimization methods such as data trimming, feature engineering, data standardization and data set segmentation to obtain the data set to be evaluated, the preliminary training set and the preliminary test set; The image data enhancement module performs data expansion by constructing a conditional generative adversarial network to obtain an enhanced training set and an enhanced test set. The specific contents include designing condition input, adjusting feature conditions, designing loss functions, constructing and training models, and expanding image data. The feature condition adjustment is used to linearly adjust the intermediate layer features of the generator module based on the conditional input, specifically, based on the conditional input of the generator module, the output features of each convolutional layer and upsampling layer in the generator module are processed using feature linear modulation; The designed loss function specifically combines the adversarial loss and the bone density regression loss as the total model loss of the conditional generative adversarial network model. The bone density regression loss is used to constrain the bone density distribution of the generated image by the generator module to conform to the real anatomical characteristics; The planting risk model construction module constructs a multi-factor risk perception model as a planting risk model to obtain a planting risk model. The specific contents include dual-branch feature processing, gated feature fusion, risk factor extraction, obtaining model output, calculating model loss, and model construction and training; The dental implant risk assessment module performs a dental implant risk assessment by using the implant risk model to obtain implant risk reference data.
2. The artificial intelligence-based dental implant risk assessment system according to claim 1, characterized in that: The dual-branch feature processing is used to process imaging data features and clinical data features, and includes: Image data branch construction is used to construct an image data branch for processing image data features, including: Macro feature processing is used to capture the overall structure of the entire jaw, specifically by downsampling through large-stride convolution; Mesoscopic feature processing is used to analyze the local morphology of the alveolar bone, specifically by expanding the receptive field through hollow convolution; Micro-feature processing is used to extract gingival texture detail information, specifically by capturing detail information through 1×1×1 convolution; Multi-scale feature fusion is used to integrate multi-scale information. The formula used is as follows: ； Where, Represents the multi-scale features of the image, Represents the splicing operation function, represents the three-dimensional convolution function, represents the 1×1×1 convolution kernel weight, Represents the macroscopic features of the image, Represents the mesoscopic features of the image, Represents the microscopic features of the image; Clinical data branch construction, used to construct a clinical data branch for processing clinical data features, specifically dynamically adjusting the weights of clinical data features through Gini impurity; The gated feature fusion is used to fuse the output features of the imaging data branch and the output features of the clinical data branch through a gating mechanism, and includes: Calculate channel attention, specifically by calculating channel attention based on multi-scale features of the image; Calculate spatial attention, specifically by calculating spatial attention based on multi-scale features of the image; Gated fusion is the fusion of multi-scale image features and clinical weighted features based on the gating mechanism. The formula used is as follows: ； Where Gw represents the gate weight, represents the gate generation weight, represents the gate generation bias term, represents the fusion feature, represents the sigmoid function, represents clinically weighted features, represents channel attention, Indicates spatial attention; The risk factor extraction is used to extract and generate multiple risk factor features, including: Bone risk factor extraction, used to extract risk features of bone-related information, specifically by using an independent gated recurrent unit to process bone-related imaging data, the bone-related imaging data specifically including cone-beam CT imaging data, oral panoramic X-ray imaging data, and periapical imaging data; Soft tissue risk factor extraction, used to extract risk features of soft tissue related information, specifically by using an independent gated recurrent unit to process soft tissue related image data, specifically intraoral scan image data; Clinical risk factor extraction, used to extract risk features from clinical data, specifically by using a multi-layer perceptron to process clinical data; Factor independence constraints are used to enforce independence between different risk factors. Specifically, bone risk factors, soft tissue risk factors, and clinical risk factors are concatenated into one matrix and the covariance regularization loss is minimized. The model output is obtained to calculate the output of the multi-factor risk perception model, and the formula used is as follows: ； Where, represents the final risk characteristics, represents the fusion feature weight, represents the weight of the i-th risk factor, represents the i-th risk factor, represents the model prediction output, represents the softmax function; The calculation model loss is specifically a combination of cross entropy loss and covariance regularization loss as the total model loss of the multi-factor risk perception model; The model is constructed and trained, specifically by constructing a multi-factor risk perception model through the dual-branch feature processing, the gated feature fusion, the risk factor extraction, the acquisition of model output and the calculation of model loss, and training the model based on the enhanced training set, and verifying the model performance based on the enhanced test set to obtain a multi-factor risk perception model as a planting risk model.
3. The artificial intelligence-based dental implant risk assessment system according to claim 1, characterized in that: The design condition input is used to design the condition input required by the generator module, specifically integrating clinical data and risk level label data as the condition input of the generator module; The characteristic condition adjustment is performed using the following formula: ； Where, represents the characteristic linear modulation scaling coefficient, represents the characteristic linear modulation shift coefficient, Represents the multilayer perceptron operation function used to generate feature linear modulation parameters, represents the intermediate layer features of the modulated generator module, represents the intermediate layer features of the generator module, Represents element-by-element multiplication, and Cd represents the conditional input of the generator module; The design loss function uses the following formula: ； Where, represents the adversarial loss value, represents the expected calculation function, represents the discriminator output function, represents the real image data, Indicates that the generator module generates image data, represents the regression loss of bone density, represents the fully connected layer function, represents the dense convolutional network feature extraction function of the discriminator module, Indicates the real bone density data, Represents the total loss value of the conditional generation adversarial network model, Indicates calculation of L2 norm; The constructing and training the model specifically comprises constructing a conditional generative adversarial network model through the design condition input, the feature condition adjustment and the design loss function, training the model based on the preliminary training set, and verifying the model performance based on the preliminary test set to obtain the conditional generative adversarial network model; The image data expansion specifically involves using the conditional generative adversarial network model to expand the image data of the preliminary training set and the preliminary test set to obtain an enhanced training set and an enhanced test set.
4. The artificial intelligence-based dental implant risk assessment system according to claim 1, characterized in that: In the information acquisition module, both the historical risk assessment original data set and the current risk assessment original data set include imaging data and clinical data. The historical risk assessment original data set also includes risk level label data.
5. The artificial intelligence-based dental implant risk assessment system according to claim 1, characterized in that: In the data optimization module, the data trimming is specifically to remove missing values and duplicate values in the original data, the feature engineering is specifically to perform feature selection, feature extraction and feature construction on the original data, the data standardization is specifically to use the minimum-maximum normalization method to standardize the original data, and the data set segmentation is used to segment the data set; The current risk assessment original data set is optimized through the data trimming, the feature engineering and the data standardization to obtain the data set to be evaluated, and the historical risk assessment original data set is optimized through the data trimming, the feature engineering, the data standardization and the data set segmentation to obtain a preliminary training set and a preliminary test set.
6. The artificial intelligence-based dental implant risk assessment system according to claim 1, characterized in that: In the dental implant risk assessment module, specifically, the implant risk model is used to perform dental implant risk assessment based on the data set to be assessed, to obtain implant risk reference data, and the patient's dental implant risk is comprehensively assessed based on the implant risk reference data.