Establishment method of peri-implantitis risk prediction model based on Meta analysis and machine learning algorithm, prediction method and device of peri-implantitis occurrence risk, and storage medium

Abstract

The invention relates to the field of computer data processing, and discloses a construction method of a periarthritis risk prediction model based on Meta analysis and a machine learning algorithm, a periarthritis occurrence risk prediction method and device, and a storage medium. The method comprises the following steps: systematically screening and evaluating periodontitis periodontitis risk factor literatures through a Meta analysis method, determining key risk factors and constructing a risk score table; acquiring clinical data of a patient to form a data set; performing assignment on the clinical data based on the risk score table, screening characteristic variables in the data set, and constructing a training set, a verification set and a test set based on the screened characteristic variables; constructing an initial prediction model based on a random forest algorithm; the initial prediction model is trained, multiple decision trees are constructed in parallel through a self-service sampling method, the initial prediction model is iteratively added into the trees, and finally the periarthritis risk prediction model with the root mean square error meeting the preset requirement is obtained.

Description

Technical Field

[0001] This application relates to the field of computer data processing technology, specifically to a method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms, a method and device for predicting the risk of peri-implantitis, and a storage medium. Background Technology

[0002] Peri-implantitis is one of the most common complications in implant restoration, clinically characterized by inflammation of the peri-implant soft tissue and progressive loss of hard tissue. This disease disrupts the health of the implant-bone interface, affecting the long-term stability of implant restoration. Therefore, exploring the influencing factors of peri-implantitis and implementing targeted prevention and control can improve the long-term success rate of implants. Studies have shown that peri-implantitis involves complex interactions of multiple risk factors, which promote inflammatory responses and bone resorption by influencing the peri-implant microenvironment; however, the specific mechanisms and weights of each factor are not fully understood. Currently, clinical risk assessment mainly relies on empirical judgment, lacking objective predictive tools. Existing predictive methods have significant limitations: single data sources cannot fully reflect the complexity of the disease; clinical data quality issues affect model accuracy; and there is a lack of effective integration of evidence-based medicine and clinical data. Traditional statistical methods cannot handle multi-factor interactions, resulting in insufficient generalization ability and interpretability of predictive models. Therefore, there is an urgent need to develop novel predictive systems based on meta-analysis and machine learning. By integrating multi-source data, constructing a scientific scoring system, and establishing high-precision predictive models, these systems can provide reliable decision support tools for clinicians, enabling early prevention and personalized intervention for peri-implantitis. Summary of the Invention

[0003] The purpose of this application is to provide a method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms, a method and device for predicting the risk of peri-implantitis, and a storage medium, in order to solve the problems in the prior art of needing to identify risk factors for peri-implantitis and predict the occurrence of peri-implantitis, as well as the problems of lack of evidence-based basis and insufficient prediction accuracy in the assessment of peri-implantitis risk factors.

[0004] To achieve the above objectives, this application provides a method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms, comprising the following steps:

[0005] A meta-analysis was used to systematically screen and evaluate literature on risk factors for peri-implantitis, identify key risk factors, and construct a risk scoring table.

[0006] Acquire patients' clinical data, form a dataset, and perform data preprocessing on the dataset;

[0007] The clinical data were assigned values ​​based on the risk scoring table. Univariate logistic regression and LASSO regression methods were used to screen the feature variables in the dataset. Based on the screened feature variables, training set, validation set and test set were constructed.

[0008] An initial prediction model was constructed based on the random forest algorithm;

[0009] The initial prediction model is trained based on the training set, validation set, and test set. Multiple decision trees are constructed in parallel using a bootstrap sampling method. The initial prediction model is iteratively added to fit the residual of the previous prediction by adding one tree at a time, and finally a peri-implantitis risk prediction model with a root mean square error that meets the preset requirements is obtained.

[0010] Optionally, the systematic screening and evaluation of literature on risk factors for peri-implantitis using meta-analysis to identify key risk factors specifically includes:

[0011] A systematic review was conducted to screen for risk factors associated with peri-implantitis from medical databases, including: gender, age, smoking status, plaque control level, history of periodontitis, history of diabetes, keratinized mucosal width, gingival biotype, implant diameter, implant length, implant site, jawbone position, bone augmentation, type of superstructure restoration, and superstructure retention method.

[0012] Meta-analysis of the above risk factors identified eight key risk factors with statistical significance, including: smoking, poor plaque control, history of periodontitis, history of diabetes, keratinized mucosal width ≤2mm, implant diameter ≤3.5mm, anterior implant site, and maxillary position.

[0013] Optionally, constructing the risk scoring table specifically includes:

[0014] Scoring is based on the regression coefficients of each key risk factor, and 10 times the regression coefficient is used as the score value for each key risk factor.

[0015] A risk scoring scale for peri-implantitis was constructed, with the following scores for key risk factors: smoking = 6.18 points, poor plaque control = 4.30 points, history of periodontitis = 17.69 points, history of diabetes = 3.84 points, width of keratinized mucosa ≤ 2 mm = 4.45 points, implant diameter ≤ 3.5 mm = 6.57 points, anterior implant site = 13.38 points, and maxillary position = 1.96 points.

[0016] The total score of the risk scoring table is the sum of the scores for each risk factor, used to quantitatively assess the risk level of peri-implantitis. The calculation formula is: Peri-implantitis risk score = 6.18 × smoking + 4.30 × poor plaque control + 17.69 × history of periodontitis + 3.84 × history of diabetes + 4.45 × width of keratinized mucosa ≤ 2mm + 6.57 × implant diameter ≤ 3.5mm + 13.38 × anterior implant site + 1.96 × maxillary position.

[0017] Optionally, the data preprocessing of the dataset specifically includes:

[0018] The dataset is cleaned to remove outliers and missing values;

[0019] Multiple interpolation methods are used to fill in unavoidable missing data;

[0020] Numerical variables are standardized, categorical variables are one-hot encoded, and a risk scoring table is used to assign values ​​to the variables.

[0021] Optionally, the step of using univariate logistic regression and LASSO regression methods to screen the feature variables in the dataset specifically includes:

[0022] Univariate logistic regression analysis was used to initially screen all candidate feature variables in the dataset, calculate the odds ratio (OR) and its 95% confidence interval for each variable, and screen out significant variables with P-values ​​less than 0.05.

[0023] The LASSO regression method was used to further screen the initially selected feature variables. Specifically, this included: constructing a LASSO penalized regression model, compressing the variable coefficients through L1 regularization, and compressing the regression coefficients of unrelated variables to 0; setting 10-fold cross-validation in LASSO regression to determine the optimal penalty parameter λ value, and executing the process using the R package.

[0024] The selected characteristic variables include at least one of the following: peri-implantitis risk score, gender, gingival biotype, superstructure retention method, and bone mineral density in the edentulous area.

[0025] Optionally, the parameters of the peri-implantitis risk prediction model include: number of trees: 100; minimum branch purity gain: 0.0; maximum tree depth: 6; metric: gini.

[0026] To achieve the above objectives, this application also provides a method for predicting peri-implantitis, comprising the following steps:

[0027] Acquire the patient's clinical characteristic data and input it into the peri-implantitis risk prediction model constructed by the construction method of the peri-implantitis risk prediction model based on meta-analysis and machine learning algorithm as described in any of the preceding claims. The characteristic data includes at least one of the following: peri-implantitis risk score, gender, gingival biotype, superstructure retention method, and bone mineral density in the edentulous area.

[0028] The peri-implantitis risk prediction model outputs a prediction result, giving the probability that the patient will develop peri-implantitis.

[0029] To achieve the above objectives, this application also provides an apparatus for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms, comprising: a memory; and a processor connected to the memory, the processor being configured to perform the steps of the method described above.

[0030] To achieve the above objectives, this application also provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a machine, implements the steps of the method described above.

[0031] The embodiments of this application have the following advantages:

[0032] Using the methods described above, based on a systematic review through meta-analysis and large-sample clinical data, univariate logistic regression analysis and LASSO regression were employed to screen for risk factors of peri-implantitis. A risk scoring table was constructed based on evidence-based medicine, and finally, a highly efficient peri-implantitis risk prediction model was built using machine learning algorithms. The random forest model demonstrated excellent predictive performance on the training, validation, and test sets, with mean area under the curve (AUC) values ​​of 1.000 (1.000-1.000), 0.986 (0.972-0.999), and 0.987 (0.979-0.995), respectively, significantly outperforming traditional prediction methods. The peri-implantitis risk prediction model constructed based on the meta-analysis random forest algorithm effectively addresses the problems in existing technologies regarding the need to identify peri-implantitis risk factors and predict its occurrence, as well as the lack of evidence-based support and insufficient prediction accuracy in assessing peri-implantitis risk factors. Attached Figure Description

[0033] To more clearly illustrate the embodiments of this application or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0034] Figure 1 A flowchart illustrating a method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms, provided for at least one embodiment of this application;

[0035] Figure 2 A schematic diagram of the LASSO coefficient path for a method of constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms, provided for at least one embodiment of this application;

[0036] Figure 3 A schematic diagram of the LASSO regression analysis cross-validation curve for a method of constructing a peri-implantitis risk prediction model based on a machine learning algorithm provided in at least one embodiment of this application.

[0037] Figure 4 A schematic diagram of the training set ROC curves comparing multiple machine learning models for constructing a machine learning-based peri-implantitis risk prediction model provided in at least one embodiment of this application.

[0038] Figure 5 A schematic diagram of the validation set ROC curves for a method of constructing a peri-implantitis risk prediction model based on a machine learning algorithm provided in at least one embodiment of this application.

[0039] Figure 6 A schematic diagram of the ROC curve of the training set of a random forest algorithm model for constructing a peri-implantitis risk prediction model based on machine learning algorithm provided in at least one embodiment of this application;

[0040] Figure 7 A schematic diagram of the ROC curve of the random forest algorithm model validation set for a method of constructing a peri-implantitis risk prediction model based on machine learning algorithm provided in at least one embodiment of this application;

[0041] Figure 8 A schematic diagram of a method for predicting peri-implantitis provided for at least one embodiment of this application;

[0042] Figure 9 SHAP feature importance diagram of a peri-implantitis prediction method provided for at least one embodiment of this application;

[0043] Figure 10 SHAP force map of a sample of a peri-implantitis prediction method provided for at least one embodiment of this application;

[0044] Figure 11 SHAP force map of another sample of a peri-implantitis prediction method provided for at least one embodiment of this application;

[0045] Figure 12 A block diagram of a device for constructing a peri-implantitis risk prediction model based on a machine learning algorithm, provided for at least one embodiment of this application. Detailed Implementation

[0046] The following specific embodiments illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0047] It should be noted that the steps in the claims and description of this application may be performed substantially in parallel or in reverse order under appropriate circumstances, depending on the function involved.

[0048] Furthermore, the technical features involved in the different embodiments of this application described below can be combined with each other as long as they do not conflict with each other.

[0049] One embodiment of this application provides a method for constructing a peri-implantitis risk prediction model based on machine learning algorithms, referencing... Figure 1 , Figure 1 The flowchart illustrates a method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms, provided in at least one embodiment of this application. It should be understood that the method may also include additional boxes not shown and / or the boxes shown may be omitted, and the scope of this application is not limited in this respect.

[0050] In step 101, key risk factors for peri-implantitis were identified through meta-analysis and a risk scoring table was constructed.

[0051] In some embodiments, a meta-analysis is used to systematically screen and evaluate literature on risk factors for peri-implantitis to identify key risk factors, including:

[0052] A systematic review was conducted to screen for risk factors associated with peri-implantitis from medical databases, including: gender, age, smoking status, plaque control level, history of periodontitis, history of diabetes, keratinized mucosal width, gingival biotype, implant diameter, implant length, implant site, jawbone position, bone augmentation, type of superstructure restoration, and superstructure retention method. A meta-analysis of these risk factors identified eight statistically significant key risk factors, including: smoking (RR=1.85, 95% CI: 1.32-2.59), poor plaque control (RR=2.41, 95% CI: 1.32-2.59), and... 1.67-3.48), history of periodontitis (RR=3.12, 95%CI: 2.25-4.33), history of diabetes (RR=1.97, 95%CI: 1.45-2.68), width of keratinized mucosa ≤2mm (RR=2.16, 95%CI: 1.58-2.95), implant diameter ≤3.5mm (RR=1.73, 95%CI: 1.26-2.38), anterior implant site (RR=1.89, 95%CI: 1.38-2.59), and maxillary position (RR=1.65, 95%CI: 1.21-2.25).

[0053] The construction of the risk scoring table specifically includes:

[0054] Scoring is based on the regression coefficients (β=ln(RR)) of each key risk factor, and 10 times the regression coefficient is used as the score value of each factor.

[0055] A peri-implantitis risk scoring scale was constructed, with the following scores for each risk factor: smoking = 6.18 points, poor plaque control = 4.30 points, history of periodontitis = 17.69 points, history of diabetes = 3.84 points, keratotic mucosal width ≤ 2 mm = 4.45 points, implant diameter ≤ 3.5 mm = 6.57 points, anterior implant site = 13.38 points, and maxillary position = 1.96 points. The results of the peri-implantitis risk scoring scale construction are shown in Table 1.

[0056] Table 1:

[0057]

[0058] The total score of the risk scoring table is the sum of the scores for each risk factor, used to quantitatively assess the risk level of peri-implantitis. The calculation formula is: Peri-implantitis risk score = 6.18 × smoking + 4.30 × poor plaque control + 17.69 × history of periodontitis + 3.84 × history of diabetes + 4.45 × width of keratinized mucosa ≤ 2 mm + 6.57 × implant diameter ≤ 3.5 mm + 13.38 × anterior implant site + 1.96 × maxillary position.

[0059] In step 102, the patient's clinical data is acquired to form a dataset, and the dataset is preprocessed.

[0060] In some embodiments, acquiring the patient's clinical data specifically includes:

[0061] Obtain data on factors influencing peri-implantitis, including: gender, age, history of hypertension, history of diabetes, periodontitis, smoking, frequent alcohol consumption, poor plaque control, mucosal width, gingival biotype, bone density in the edentulous area, implant diameter, implant length, implant shape, implant surface treatment, implant site, jawbone, bone augmentation, immediate implantation, type of superstructure restoration, superstructure retention method, abutment-implant connection method, and crown material, at least one of the following.

[0062] In some embodiments, the data preprocessing of the dataset specifically includes:

[0063] The dataset was cleaned to remove outliers and missing values, and multiple imputation was used to fill in unavoidable missing data. Numerical variables were standardized, categorical variables were one-hot encoded, and a risk scoring table was used to assign values ​​to the variables.

[0064] In step 103, univariate logistic regression and LASSO regression methods are used to screen the feature variables in the dataset, and training set, validation set and test set are constructed based on the screened feature variables.

[0065] In some embodiments, the filtering of feature variables in the dataset specifically includes:

[0066] The characteristic variables included in this application are: age, gender, history of hypertension, frequent alcohol consumption, bone mineral density in the edentulous area, implant shape, implant surface treatment method, bone augmentation, immediate implantation, type of superstructure restoration, superstructure retention method, crown material, abutment-implant connection method, and gingival biotype. Additionally, the risk score for peri-implantitis was calculated after assigning values ​​to variables such as smoking, poor plaque control, history of periodontitis, history of diabetes, keratinized mucosal width ≤2mm, implant diameter ≤3.5mm, anterior implant location, and maxillary position. Univariate logistic regression was performed to initially screen the above influencing factors, identifying categorical variables with a two-sided p-value less than 0.05, which were considered statistically significant. The results of the univariate logistic regression analysis are shown in Table 2.

[0067] Table 2:

[0068]

[0069] refer to Figure 2 and Figure 3 Lasso regression analysis was performed using the R package "glmnet4.1.2" to conduct secondary feature selection on the significant variables selected by univariate logistic regression. This method constructs an optimization function by introducing an L1 regularization penalty term, achieving the following functions within a 10-fold cross-validation framework: 1) compressing regression coefficients to effectively control model complexity and avoid overfitting; 2) automatically completing variable selection through a coefficient shrinkage mechanism, compressing the regression coefficients of redundant variables to zero, thereby eliminating the interference of multicollinearity on model stability. The analysis process strictly employs a 10-fold cross-validation strategy to determine the optimal penalty parameters, ensuring the reliability of the feature selection results.

[0070] Figure 2 The diagram shows the LASSO regression coefficient path, with the upper horizontal axis displaying the number of non-zero coefficients, the vertical axis reflecting the magnitude of the coefficients, and the lower horizontal axis representing the standardized coefficient vector. The coefficient trajectories of the six variables are presented as color curves, visually demonstrating the selection process of each variable under different constraints. Figure 3 In the cross-validation error plot, the vertical axis represents the prediction error (the lower the value, the better the model fit), the upper horizontal axis marks the number of variables corresponding to different λ values, and the lower horizontal axis represents the logarithmic value of λ. The minimum error point (lambda.min) marked by the dashed line on the left corresponds to the value 5 on the upper horizontal axis, indicating that 5 predictor variables were ultimately selected and retained. Both plots together verify the reliability of the variable selection process and the optimization effect of the model.

[0071] In some embodiments, the final selected characteristic variables include at least one of the following: gender, gingival biotype, superstructure retention method, bone mineral density in the edentulous area, and peri-implantitis risk score.

[0072] In step 104, an initial prediction model is constructed based on the random forest algorithm.

[0073] In step 105, the initial prediction model is trained based on the training set, validation set, and test set. Multiple decision trees are trained in parallel by randomly sampling multiple sample subsets with replacement from the training set using a bootstrap sampling method. The initial prediction model is iteratively added to fit the residual of the previous prediction by adding one tree at a time, and finally obtains a peri-implantitis risk prediction model with a root mean square error that meets the preset requirements.

[0074] Specifically, an initial prediction model was constructed based on various machine learning methods, including Logistic Regression (LR), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Random Forest (RF), Adaptive Boosting (AdaBoost), Gaussian Naive Bayes (GNB), Multilayer Perceptron (MLP), K-Nearest Neighbors (KNN), and Support Vector Machine (SVM). The Logistic Regression model parameters were: convergence metric: 1e-06, regularization type: l2, number of iterations: 100, regularization factor: 1.0. The XGBoost model parameters were: L2 regularization coefficient: 0.5, minimum bifurcation weight sum: 4, maximum tree depth: 8, learning rate: 0.3. The LightGBM model parameters are: maximum number of leaves: 5; maximum number of trees: 50; maximum tree depth: 20; learning rate: 0.001; algorithm type: GBDT. The RF model parameters are: number of trees: 100; minimum branch purity reward: 0.0; maximum tree depth: ; metric: Gini. The AdaBoost model parameters are: learning rate: 1.0; number of models: 50. The GNB model parameters are: prior probability: None; var_smoothing: 1e-07. The MLP model parameters are: number of iterations: 20; hidden layer width: (30, 30); nonlinear function: logistic. The KNN model parameters are: weight type: distance; number of nearest neighbors: 6. The SVM model parameters are: number of iterations: 20; hidden layer width: (30, 30); nonlinear function: logistic. The KNN model parameters are: convergence metric: 0.001; kernel type: RBF; regularization factor: 1.0. The comprehensive evaluation results based on AUC values ​​across the training, validation, and test sets show that the random forest model exhibits the best predictive performance. Figure 4 and Figure 5 As shown, by comparing the ROC curves of different algorithms (plotted with true positive rate on the ordinate and false positive rate on the abscissa), the performance of each model in terms of the balance between sensitivity and specificity can be intuitively compared. Among them, AUC, as a quantitative indicator of model discriminative efficacy, represents better classification accuracy with a higher value. Specifically, Figure 6 and Figure 7The study highlights the Receiver Operating Characteristic (ROC) curves of the Random Forest model on the training and validation sets, and its excellent AUC values ​​further confirm the algorithm's stable predictive ability.

[0075] Specifically, a Python function is provided to train the dataset using the Random Forest algorithm model. This function follows the objective function of the Random Forest and includes detailed comments explaining the implementation of each step.

[0076] Specific functions:

[0077] import numpy as np

[0078] from sklearn.ensemble import RandomForestRegressor

[0079] from sklearn.datasets import load_boston

[0080] from sklearn.model_selection import train_test_split

[0081] from sklearn.metrics import mean_squared_error

[0082] def train_RF_model(X, y, n_estimators=100):

[0083] """

[0084] The dataset was trained using a random forest model.

[0085] parameter:

[0086] X (numpy.ndarray): Feature dataset

[0087] y (numpy.ndarray): Labeled dataset

[0088] n_estimators (int): The number of trees in the forest, defaulting to 100.

[0089] return:

[0090] model (RandomForestRegressor): A trained random forest model

[0091] X_train (numpy.ndarray): Training set features

[0092] X_test (numpy.ndarray): Test set features

[0093] """

[0094] # 1. Data Partitioning - Random Forest has built-in cross-validation, so usually only one partition is needed.

[0095] X_train, X_val, y_train, y_val = train_test_split(

[0096] X, y, test_size=0.2, random_state=42)

[0097] # 2. Model Parameter Configuration

[0098] params = {'n_estimators': n_estimators, # Number of decision trees}

[0099] 'max_depth': 6, # Maximum depth of the tree

[0100] 'min_samples_split': 2, # Minimum number of samples required for node splitting

[0101] 'min_samples_leaf': 1, # Minimum number of samples for a leaf node

[0102] 'max_features': 'auto', # The proportion of features to consider (sqrt(n_features))

[0103] 'bootstrap': True, # Use bootstrap sampling

[0104] 'oob_score': True, # Calculate out-of-bag error

[0105] 'random_state': 42 # Random seed}

[0106] # 3. Model Training

[0107] model = RandomForestRegressor(**params)

[0108] model.fit(X_train, y_train)

[0109] # 4. Model Evaluation

[0110] train_pred = model.predict(X_train)

[0111] val_pred = model.predict(X_val)

[0112] print(f"Train RMSE: {np.sqrt(mean_squared_error(y_train, train_pred))}")

[0113] print(f"Validation RMSE: {np.sqrt(mean_squared_error(y_val, val_pred))}")

[0114] print(f"OOB Score: {model.oob_score_}")

[0115] return model, X_train, X_val

[0116] Code explanation:

[0117] Dataset splitting and transformation:

[0118] Use train_test_split to split the data into training and validation sets, and use random_state=42 to ensure reproducibility.

[0119] Parameter settings:

[0120] n_estimators: Sets the number of decision trees in the forest. The larger the value, the more stable the model, but the higher the computational cost. The default value is 100.

[0121] max_depth: Controls the maximum depth of a single decision tree, limiting the growth of the tree to prevent overfitting. Setting it to 6 means that the tree can have a maximum of 6 splits.

[0122] min_samples_split: Specifies the minimum number of samples required for a node to split. Setting it to 2 means that each node needs at least 2 samples before it can continue splitting.

[0123] min_samples_leaf: Defines the minimum number of samples required for a leaf node. Setting it to 1 means that each leaf node contains at least one sample.

[0124] `max_features`: Controls the number of features considered in each split. 'auto' defaults to using sqrt(n_features) features to enhance model diversity.

[0125] bootstrap: Enables Bootstrap sampling. Setting it to True means that random sampling with replacement is used during training of each tree.

[0126] oob_score: Enables out-of-bag error evaluation. Setting it to True allows for model validation using samples not used in training.

[0127] random_state: Fixed random seed. Setting it to 42 ensures the reproducibility of the experiment, and the same results can always be obtained under the same parameters.

[0128] Model training:

[0129] Model training is performed using `RandomForestRegressor(**params)`. The `**params` parameter dictionary is unpacked, and all preset parameters are passed in. The `fit()` method performs the following operations: it constructs a specified number (n_estimators) of decision trees in parallel, each tree using a subset of the bootstrap sampled data (approximately 63.2% of the original training data), and randomly selects a subset of features (controlled by `max_features`) each time a node splits.

[0130] Prediction and Assessment:

[0131] After model training, predictions are made on the training set. Each tree provides an independent prediction, and the final prediction is the average of all tree predictions (regression task). Calculated metrics include RMSE and OOB Score.

[0132] RMSE (Root Mean Square Error) is a common metric used in machine learning to evaluate the prediction error of a model. It measures the difference between the predicted value and the true value, and imposes a higher penalty on predictions with larger discrepancies. The steps for calculating RMSE are as follows:

[0133] Error calculation: For each predicted value, calculate the difference between it and the corresponding true value: (e_i =\hat{y}_i - y_i), where (\hat{y}_i) is the (i)th predicted value and (y_i) is the (i)th true value.

[0134] Squared Error: The square of each error is calculated to eliminate the cancellation of positive and negative errors: (e_i^2).

[0135] Calculate the mean squared error (MSE): Sum all the squared errors and then divide by the total number of data points (n) to obtain the mean squared error.

[0136] [ MSE = \frac{1}{n} \sum_{i=1}^{n} e_i^2 ]

[0137] Square Root Error (RMSE): Finally, perform the square root operation on the MSE:

[0138] [ RMSE = \sqrt{MSE} ]

[0139] The formula can be summarized as: [ RMSE = \sqrt{ \frac{1}{n} \sum_{i=1}^{n} (\hat{y}_i - y_i)^2} ]

[0140] The smaller the RMSE, the closer the model's predictions are to the actual values, and the better the model's performance. It's important to note that the units of RMSE are the same as those of the target variable, thus providing a degree of intuitiveness.

[0141] The Out-of-Bag Score (OOB Score) is a model performance evaluation metric specific to the Random Forest algorithm. It utilizes the naturally generated, untrained samples (i.e., "out-of-bag" samples) during the Bootstrap sampling process for internal validation. This metric provides unbiased estimates similar to cross-validation without requiring additional validation set partitioning, combining computational efficiency and high data utilization. The calculation steps for the OOB Score are as follows:

[0142] Bootstrap sampling: For each decision tree in the forest: sample with replacement from the original training set to obtain approximately 63.2% of the samples as a training subset, and the remaining approximately 36.8% of the samples are the OOB samples of that tree: OOB sample set = Total set - Bootstrap sample set.

[0143] OOB prediction: For each sample xi, collect the prediction results of all decision trees that treat xi as an OOB sample, and denote Tii as the number of trees that treat xi as an OOB sample.

[0144] Result aggregation: Classification task (voting mechanism):

[0145] ,

[0146] Where n: the total number of samples in the original training set.

[0147] i: Sample index, i = 1, 2, ..., n

[0148] The summation sign over all samples.

[0149] y i : The true label (true category) of the i-th sample.

[0150] II(·): Indicator function; value is 1 when the condition inside the parentheses is true, otherwise it is 0.

[0151] mode: Take the mode (the value that appears most frequently).

[0152] The predicted label of the t-th decision tree for the i-th sample.

[0153] t: Decision tree index

[0154] T i : Use the i-th sample as the index set of the decision tree for OOB samples.

[0155] { }t∈T i : All t∈T i The set of prediction results for sample i from the tree.

[0156] Regression task (average mechanism):

[0157] .

[0158] Final output: The classification task returns the accuracy (between 0 and 1), and the regression task returns the R² score (which may be negative, and the closer to 1 the better).

[0159] Where n: the total number of samples in the original training set.

[0160] i: Sample index, i = 1, 2, ..., n

[0161] The summation sign over all samples.

[0162] y i : The true value (continuous value) of the i-th sample.

[0163] | T i |:Set T iSize (the number of trees that treat sample i as an OOB sample).

[0164] For set T i The summation symbol for all trees in the array.

[0165] : The predicted value (continuous value) of the t-th decision tree for the i-th sample.

[0166] The average of all true values ​​in the sample. .

[0167] Summary of mathematical expressions: Standard calculation formula for regression tasks:

[0168] ,

[0169] Where n: the total number of samples in the original training set.

[0170] i: Sample index, i = 1, 2, ..., n

[0171] The summation sign over all samples.

[0172] y i : The true value (continuous value) of the i-th sample.

[0173] : The out-of-bag predicted value of the i-th sample,

[0174] : The average of the true values ​​of all samples.

[0175] In the above formula:

[0176] ,

[0177] in, : The out-of-bag predicted value of the i-th sample,

[0178] | T i |:Set T i Size (the number of trees that treat sample i as an OOB sample).

[0179] For set T i The summation symbol for all trees in the array.

[0180] f t : The t-th decision tree model,

[0181] x i : The feature vector of the i-th sample

[0182] f t (x i ): The t-th decision tree for sample x i The predicted output.

[0183] Specifically, the initial prediction model, constructed using the random forest algorithm, is trained on the training set, validation set, and test set. The initial prediction model iteratively adds trees, fitting the residuals of the previous prediction with each addition, ultimately yielding a peri-implantitis risk prediction model.

[0184] The specific process follows the objective function Obj as: [\text{Obj}(\theta) = \sum_{i=1}^nl(y_i, \hat{y}i^{(t)}) + \sum{k=1}^T \Omega(f_k)]. Here, (l) is the loss function, used to measure the deviation between the predicted value (\hat{y}_i^{(t)}) and the actual value (y_i); while (\Omega(f_k)) is the regularization term, used to control the complexity of the model, in the form: [\Omega(f_k) = \gamma T + \frac{1}{2} \lambda \sum_{j=1}^T w_j^2]. Here, (\gamma) and (\lambda) are regularization parameters, (T) is the number of trees, and (w_j) is the node weight of the tree.

[0185] In some embodiments, the peri-implantitis risk prediction model is expressed as the following formula: [ \hat{y} = \sum_{k=1}^{K} f_k(\mathbf{x}) ].

[0186] Where (\hat{y} ) is the predicted value, (K ) is the number of trees, and ( f_k(\mathbf{x}) ) is the output of the (k)th tree on the input feature (\mathbf{x} ).

[0187] In some embodiments, the parameters of the peri-implantitis risk prediction model include: number of trees: 100; minimum forking purity gain: 0.0; maximum tree depth: 6; metric: gini.

[0188] An embodiment of this application also provides a method for predicting peri-implantitis, including the steps of: obtaining relevant feature values ​​of the patient and inputting them into the peri-implantitis risk prediction model constructed by the aforementioned method embodiment of the peri-implantitis risk prediction model based on machine learning algorithm. The feature values ​​include at least one of gender, gingival biotype, superstructure retention mode, bone density in the edentulous area, and peri-implantitis risk score.

[0189] The peri-implantitis risk prediction model outputs a prediction result, giving the patient the probability of developing peri-implantitis.

[0190] The risk probability value is compared with a preset threshold, and a high-risk warning is generated when the threshold is exceeded.

[0191] In some embodiments, the method further includes: interpreting the peri-implantitis risk prediction model using SHAP values, demonstrating important features and their impact on peri-implantitis, and calculating the SHAP value (φ_j) for each feature value (x_j) in the peri-implantitis risk prediction model to reflect the average contribution of different feature values ​​to the prediction results. (See reference) Figure 8 The SHAP summary plot can show the non-linear relationship between each influencing factor and peri-implantitis. A positive correlation exists between the peri-implantitis risk score and male sex, meaning a higher risk score corresponds to a greater risk of peri-implantitis; furthermore, men have a higher risk of peri-implantitis than women. The SHAP summary plot combines feature importance with feature effect. Each point on the SHAP summary plot represents a feature and an instance's Shapley value. The position on the vertical axis is determined by the feature, and the position on the horizontal axis is determined by the Shapley value. Blue to red represents feature values ​​from low to high. Overlapping points jitter along the vertical axis, thus revealing the distribution of Shapley values ​​for each feature. These features are ordered according to their importance. Reference Figure 9 The feature importance ranking shows that gender, gingival biotype, superstructure retention method, bone mineral density in the edentulous area, and peri-implantitis risk score are the features of the peri-implantitis risk prediction model. Among them, the peri-implantitis risk score is the most important feature, changing the predicted absolute probability of peri-implantitis by an average of approximately 27 percentage points. (Reference) Figure 10 and Figure 11 The model was further explained using two samples, in which... Figure 10 The predicted probability for the first sample was 0.78. A high peri-implantitis risk score, adhesive fixation, and Class III bone mineral density at the implantation site had a positive effect on the results, while female sex had an inhibitory effect on the results. Figure 11The predicted probability of the second sample was 0.09, the gender was female, and the low risk score of peri-implantitis had an inhibitory effect on the results.

[0192] Specifically, for a particular feature (j), its Shapley value is calculated using the following formula:

[0193] [ \phi_j = \sum_{S \subseteq N \setminus {j}} \frac{|S|! (|N|-|S|-1)!}{|N|!} \left[ f_S(\mathbf{x}_S \cup {x_j}) - f_S(\mathbf{x}_S) \right] ]

[0194] in:

[0195] (N) is the set of features.

[0196] (S \subseteq N \setminus { j} ) represents all feature subsets that do not contain feature ( j ).

[0197] (f_S(\mathbf{x}_S)) is the prediction made by the model using only the feature subset (S).

[0198] ( f_S(\mathbf{x}_S \cup { x_j}) ) is the model prediction resulting from the combined effect of feature subset ( S ) and feature ( j ).

[0199] In the peri-implantitis risk prediction model, the SHAP value for each feature is calculated to regress to feature importance. The SHAP library automatically handles these calculations; below is the Python implementation code:

[0200] from sklearn.ensemble import RandomForestRegressor

[0201] import shap

[0202] import matplotlib.pyplot as plt

[0203] from sklearn.datasets import load_boston

[0204] from sklearn.model_selection import train_test_split

[0205] # Load the dataset (dataset for peri-implantitis)

[0206] boston = load_boston()

[0207] X, y = boston.data, boston.target

[0208] # Splitting the dataset

[0209] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

[0210] # Training a random forest model

[0211] model = RandomForestRegressor( n_estimators=100, max_depth=6,min_samples_split=2, max_features=0.8, bootstrap=True,oob_score=True, random_state=42)

[0212] model.fit(X_train, y_train)

[0213] # Interpreting using SHAP values

[0214] explainer = shap.TreeExplainer(model)

[0215] shap_values ​​= explainer.shap_values(X_train)

[0216] # Draw a feature importance graph

[0217] shap.summary_plot(shap_values,X_train,feature_names=boston.feature_names)

[0218] # Plot the SHAP value of a single sample

[0219] shap.initjs()

[0220] shap.force_plot(explainer.expected_value, shap_values[0,:], X_train[0,:], feature_names=boston.feature_names)

[0221] in:

[0222] Loading and preparing data:

[0223] Dataset on peri-implantitis.

[0224] Use train_test_split to split the data into training and test sets.

[0225] Training the random forest model:

[0226] The model was trained using RandomForestRegressor, with the objective function chosen as squared error (reg:squarederror).

[0227] Calculate the SHAP value:

[0228] Create a TreeExplainer object to interpret the random forest model.

[0229] Use shap_values ​​to obtain the SHAP values ​​of the training set data.

[0230] Plotting SHAP values:

[0231] `shap.summary_plot` displays the overall feature importance.

[0232] `shap.force_plot` displays the specific impact of a single sample's features on the prediction results.

[0233] Using the methods described above, based on a systematic review through meta-analysis and large-sample clinical data, univariate logistic regression analysis and LASSO regression were employed to screen for risk factors of peri-implantitis. A risk scoring table was constructed based on evidence-based medicine, and finally, a highly efficient peri-implantitis risk prediction model was built using machine learning algorithms. The random forest model demonstrated excellent predictive performance on the training, validation, and test sets, with average AUC values ​​of 1.000 (1.000-1.000), 0.986 (0.972-0.999), and 0.987 (0.979-0.995), respectively, significantly outperforming traditional prediction methods. The peri-implantitis risk prediction model constructed based on the meta-analysis random forest algorithm effectively addresses the problems in existing technologies regarding the need to identify peri-implantitis risk factors and predict its occurrence, as well as the lack of evidence-based support and insufficient prediction accuracy in assessing peri-implantitis risk factors.

[0234] Figure 12 This application provides a block diagram of an apparatus for constructing a risk prediction model for peri-implantitis based on a machine learning algorithm, according to at least one embodiment. The apparatus includes: a memory 201; and a processor 202 connected to the memory 201, the processor 202 being configured to: systematically screen and evaluate literature on peri-implantitis risk factors using meta-analysis methods, identify key risk factors, and construct a risk scoring table;

[0235] The clinical data were assigned values ​​based on the risk scoring table. Univariate logistic regression and LASSO regression methods were used to screen the feature variables in the dataset. Based on the screened feature variables, training set, validation set and test set were constructed.

[0236] An initial prediction model was constructed based on the random forest algorithm;

[0237] The initial prediction model is trained based on the training set, validation set, and test set. Multiple decision trees are constructed in parallel using a bootstrap sampling method. When splitting a node, a subset of features is randomly selected and the optimal splitting feature is determined based on the Gini coefficient. Finally, the prediction results of each decision tree are integrated through a majority voting mechanism to obtain a peri-implantitis risk prediction model with an AUC value that meets the preset requirements.

[0238] In some embodiments, the processor 202 is further configured to: systematically screen and evaluate literature on risk factors for peri-implantitis using meta-analysis methods to identify key risk factors, specifically including:

[0239] A systematic review was conducted to screen for risk factors associated with peri-implantitis from medical databases, including: gender, age, smoking status, plaque control level, history of periodontitis, history of diabetes, keratinized mucosal width, gingival biotype, implant diameter, implant length, implant site, jawbone position, bone augmentation, type of superstructure restoration, and superstructure retention method. A meta-analysis of these risk factors identified eight statistically significant key risk factors, including: smoking (RR=1.85, 95% CI: 1.32-2.59), poor plaque control (RR=2.41, 95% CI: 1.32-2.59), and... 1.67-3.48), history of periodontitis (RR=3.12, 95%CI: 2.25-4.33), history of diabetes (RR=1.97, 95%CI: 1.45-2.68), width of keratinized mucosa ≤2mm (RR=2.16, 95%CI: 1.58-2.95), implant diameter ≤3.5mm (RR=1.73, 95%CI: 1.26-2.38), anterior implant site (RR=1.89, 95%CI: 1.38-2.59), and maxillary position (RR=1.65, 95%CI: 1.21-2.25).

[0240] In some embodiments, the processor 202 is further configured to: construct a risk scoring table, specifically including: assigning scores based on the regression coefficients (β=ln(RR)) of each key risk factor, and using 10 times the regression coefficient as the score value of each factor; constructing a peri-implantitis risk scoring table, wherein the score values ​​of each risk factor are: smoking 6.18 points, poor plaque control 4.3 points, history of periodontitis 17.69 points, history of diabetes 3.84 points, keratotic mucosal width ≤2mm 4.45 points, implant diameter ≤3.5mm 6.57 points, anterior implant site 13.38 points, maxillary position 1.96 points;

[0241] The total score of the risk scoring table is the sum of the scores for each risk factor, used to quantitatively assess the risk level of peri-implantitis.

[0242] In some embodiments, the processor 202 is further configured to: acquire the patient's clinical data, specifically including:

[0243] Obtain data on factors influencing peri-implantitis, including: gender, age, history of hypertension, history of diabetes, periodontitis, smoking, frequent alcohol consumption, poor plaque control, mucosal width, gingival biotype, bone density in the edentulous area, implant diameter, implant length, implant shape, implant surface treatment, implant site, jawbone, bone augmentation, immediate implantation, type of superstructure restoration, superstructure retention method, abutment-implant connection method, and crown material.

[0244] In some embodiments, the processor 202 is further configured to: perform data preprocessing on the dataset, specifically including: cleaning the dataset to remove outliers and missing values; using multiple imputation to fill in unavoidable missing data; standardizing numerical variables; performing one-hot encoding on categorical variables; and assigning values ​​to variables using a risk scoring table.

[0245] In some embodiments, the processor 202 is further configured to: filter the feature variables in the dataset using univariate logistic regression and LASSO regression methods, specifically including:

[0246] Univariate logistic regression analysis was used to initially screen all candidate feature variables in the dataset, calculate the odds ratio (OR) and its 95% confidence interval for each variable, and screen out significant variables with P-values ​​less than 0.05.

[0247] The LASSO regression method was used to further screen the initially selected feature variables. Specifically, this included: constructing a LASSO penalized regression model, compressing the variable coefficients through L1 regularization, and compressing the coefficients of unrelated variables to 0; setting 10-fold cross-validation in LASSO regression to determine the optimal penalty parameter λ value, and executing the process using the R package.

[0248] In some embodiments, the processor 202 is further configured to: the selected feature variables include: peri-implantitis risk score, gender, gingival biotype, superstructure retention method, and bone mineral density in the edentulous area.

[0249] In some embodiments, the processor 202 is further configured such that the parameters of the peri-implantitis risk prediction model include: n_estimators=100, max_depth=6, min_samples_split=2, max_features=0.8, bootstrap=True, oob_score=True, random_state=42.

[0250] For specific implementation methods, please refer to the aforementioned method embodiments, which will not be repeated here.

[0251] The embodiments of this application cover various forms, including methods, apparatus, systems, and computer program products. Specifically, the computer program product is a storage medium containing computer-executable instructions used to implement the technical solutions proposed in this application.

[0252] Computer-readable storage media are physical media capable of storing and retaining instructions for use by instruction execution devices. These media include, but are not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, and combinations thereof. Specific examples include portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact discs (CD-ROM), digital multifunction discs (DVD), memory sticks, floppy disks, mechanical encoding devices, and devices that store instructions, such as punch cards or recessed structures. It is important to note that computer-readable storage media do not include transient signals, such as radio waves, electromagnetic waves, or electrical signals transmitted through wires.

[0253] Computer-readable program instructions can be downloaded to computing / processing devices or external storage devices via a network (such as the Internet, local area network, wide area network, or wireless network), or obtained directly from computer-readable storage media. The network may include components such as copper wires, fiber optics, wireless transmission, routers, firewalls, switches, gateway computers, or edge servers. Each device's network adapter or interface is responsible for receiving and forwarding these instructions, which are stored in the device's computer-readable storage media.

[0254] The computer program instructions that execute the operations described in this application may include assembly instructions, ISA instructions, machine instructions, microcode, firmware instructions, status data, or source code or object code written in one or more programming languages ​​(such as Smalltalk, C++, C, etc.). The program instructions may execute entirely on the user's computer, partially on the user's computer or a remote computer, or run as a standalone software package. The remote computer is connected to the user's computer via a local area network, wide area network, or the Internet. In some cases, customized electronic circuits (such as FPGAs, PLAs, etc.) execute these instructions using the status information of the computer-readable program instructions, thereby achieving the functions described in the application.

[0255] Various aspects of this application are described by flowcharts or block diagrams of methods, apparatus (systems), and computer program products. It should be understood that each block in the flowchart or block diagram, and combinations thereof, can be implemented by computer-readable program instructions.

[0256] These computer-readable program instructions can be transmitted to the processing unit of a general-purpose computer, special-purpose computer, or other programmable data processing device, thereby causing these devices to perform the functions specified in the flowchart or block diagram when executing the instructions. These program instructions can also be stored in a computer-readable storage medium, causing the device to operate in a specific manner to perform the functions or operations described in the flowchart or block diagram. The computer-readable program instructions can also be loaded onto a computer or other programmable data processing device to perform a series of operational steps, thereby achieving the functions or actions specified in the flowchart or block diagram.

[0257] The flowcharts and block diagrams in the accompanying drawings illustrate the implementation structure and operation of the systems, methods, and computer program products of this application. Each block in a flowchart or block diagram represents a module, program segment, or part of an instruction, containing instructions that perform a specific logical function. In some implementations, the functions in the blocks may be executed in a different order than shown in the diagram. For example, two blocks may be executed in parallel or in reverse order, depending on the functions involved. Furthermore, the blocks and combinations thereof in the flowcharts or block diagrams can be implemented using a dedicated hardware system or through a combination of dedicated hardware and computer instructions.

[0258] Unless otherwise stated, all features mentioned in this specification can be replaced by other alternative features that achieve the same or similar functionality. Therefore, each disclosed feature is merely an example corresponding to a set of equivalent or similar features. The use of phrases such as "furthermore," "preferably," or "even further" typically indicates an extension of another embodiment based on existing embodiments, combined with the foregoing embodiments to form a new implementation. Different further configurations can be arbitrarily combined within the same embodiment to form new embodiments.

[0259] Although this application has been described in detail above through general description and specific embodiments, some modifications or improvements can be made based on it, and these changes will be obvious to those skilled in the art. Therefore, as long as they do not depart from the core spirit of this application, all modifications or improvements made shall fall within the protection scope of this application.

Claims

1. A method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms, characterized in that, Includes the following steps: A meta-analysis was used to systematically screen and evaluate literature on risk factors for peri-implantitis, identify key risk factors, and construct a risk scoring table. Acquire patients' clinical data, form a dataset, and perform data preprocessing on the dataset; The clinical data were assigned values ​​based on the risk scoring table. Univariate logistic regression and LASSO regression methods were used to screen the feature variables in the dataset. Based on the screened feature variables, training set, validation set and test set were constructed. An initial prediction model was constructed based on the random forest algorithm; The initial prediction model is trained based on the training set, validation set, and test set. Multiple decision trees are constructed in parallel using a bootstrap sampling method. The initial prediction model is iteratively added to fit the residual of the previous prediction by adding one tree at a time, and finally a peri-implantitis risk prediction model with a root mean square error that meets the preset requirements is obtained.

2. The method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms according to claim 1, characterized in that, The meta-analysis method was used to systematically screen and evaluate literature on risk factors for peri-implantitis, identifying key risk factors, specifically including: A systematic review was conducted to screen for risk factors associated with peri-implantitis from medical databases, including: gender, age, smoking status, plaque control level, history of periodontitis, history of diabetes, keratinized mucosal width, gingival biotype, implant diameter, implant length, implant site, jawbone position, bone augmentation, type of superstructure restoration, and superstructure retention method. Meta-analysis of the above risk factors identified eight key risk factors with statistical significance, including: smoking, poor plaque control, history of periodontitis, history of diabetes, keratinized mucosal width ≤2mm, implant diameter ≤3.5mm, anterior implant site, and maxillary position.

3. The method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms according to claim 2, characterized in that, The construction of the risk scoring table specifically includes: Scoring is based on the regression coefficients of each key risk factor, and 10 times the regression coefficient is used as the score value for each key risk factor. A risk scoring scale for peri-implantitis was constructed, with the following scores for key risk factors: smoking = 6.18 points, poor plaque control = 4.30 points, history of periodontitis = 17.69 points, history of diabetes = 3.84 points, width of keratinized mucosa ≤ 2 mm = 4.45 points, implant diameter ≤ 3.5 mm = 6.57 points, anterior implant site = 13.38 points, and maxillary position = 1.96 points. The total score of the risk scoring table is the sum of the scores for each risk factor, used to quantitatively assess the risk level of peri-implantitis. The calculation formula is: Peri-implantitis risk score = 6.18 × smoking + 4.30 × poor plaque control + 17.69 × history of periodontitis + 3.84 × history of diabetes + 4.45 × width of keratinized mucosa ≤ 2mm + 6.57 × implant diameter ≤ 3.5mm + 13.38 × anterior implant site + 1.96 × maxillary position.

4. The method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms according to claim 1, characterized in that, The data preprocessing of the dataset specifically includes: The dataset is cleaned to remove outliers and missing values; Multiple interpolation methods are used to fill in unavoidable missing data; Numerical variables are standardized, categorical variables are one-hot encoded, and a risk scoring table is used to assign values ​​to the variables.

5. The method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms according to claim 1, characterized in that, The method of using univariate logistic regression and LASSO regression to screen the feature variables in the dataset specifically includes: Univariate logistic regression analysis was used to initially screen all candidate feature variables in the dataset, calculate the odds ratio and 95% confidence interval of each variable, and screen out significant variables with P-values ​​less than 0.

05. The LASSO regression method was used to further screen the initially selected feature variables. Specifically, this included: constructing a LASSO penalized regression model, compressing the variable coefficients through L1 regularization, and compressing the regression coefficients of unrelated variables to 0; setting 10-fold cross-validation in LASSO regression to determine the optimal penalty parameter λ value, and executing the process using the R package. The selected characteristic variables include at least one of the following: peri-implantitis risk score, gender, gingival biotype, superstructure retention method, and bone mineral density in the edentulous area.

6. The method for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms according to claim 1, characterized in that, The parameters of the peri-implantitis risk prediction model include: number of trees: 100; minimum branch purity gain: 0.0; maximum tree depth: 6; metric: gini.

7. A method for predicting the risk of peri-implantitis, characterized in that, Including the following steps: Acquire the patient’s clinical characteristic data and input it into the peri-implantitis risk prediction model based on meta-analysis and machine learning algorithm as described in any one of claims 1 to 6. The characteristic data includes at least one of the following: peri-implantitis risk score, gender, gingival biotype, superstructure retention method, and bone mineral density in the edentulous area. The model for predicting the risk of peri-implantitis processes the input clinical feature data and outputs prediction results, providing a probability value of the patient's risk of developing peri-implantitis. The risk probability value is compared with a preset threshold, and a high-risk warning is generated when the threshold is exceeded.

8. The method for predicting the risk of peri-implantitis according to claim 7, characterized in that, Also includes: The SHAP value is used to interpret the peri-implantitis risk prediction model, specifically including: calculating the SHAP value of each feature variable in the prediction model, quantifying the average contribution of each feature to the prediction result; and generating a ranking of feature importance. Visualization tools are used to show the dependency relationship between each feature value and the SHAP value, and the trend of the impact of feature value changes on the probability of peri-implantitis risk is presented intuitively. The interpretation results based on SHAP values ​​provide a quantitative basis for the formulation of clinical preventive intervention measures.

9. A device for constructing a peri-implantitis risk prediction model based on meta-analysis and machine learning algorithms, characterized in that, include: Memory; A processor connected to the memory, the processor being configured to perform the steps of the method for constructing a peri-implantitis risk prediction model as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, It stores computer program instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 8.

Images