A rapid prediction method for the bonding strength value of dental restoration materials based on machine learning

Abstract

The present invention relates to the technical field of testing the bonding strength of resin cement-bonded dental restorations, and particularly to a rapid prediction method for the bonding strength value of dental restoration materials based on machine learning. The steps are as follows: S1: Construct an optimal XGB prediction model; S2: Perform preliminary processing on the information to be predicted; S3: Input all relevant features; S4: Obtain the predicted bonding strength value range according to the binary classification result output by the operation model; S5: The prediction results provide a reference for the selection of restoration materials, the selection of bonding products, and the optimization and adjustment of bonding strategies; The present invention can be closely combined with the optimization of the bonding process of dental restoration materials. By real-time analyzing the physical and chemical performance parameters of the materials and combining its closed-loop control system with the prediction model, it provides a reference for the selection of restoration materials, the dynamic adjustment of bonding operation steps and operation parameters, realizes the optimal bonding effect of the restoration materials, improves the reliability and performance of the restorations in clinical applications, and extends the clinical service life of the restorations.

Description

Technical Field

[0001] The present invention relates to the technical field of testing the bonding strength of resin cement-dependent dental restorations, and in particular, to a rapid prediction method for the bonding strength value of dental restoration materials based on machine learning. Background Art

[0002] Materials such as porcelain, all-ceramic, and polymerized ceramic have excellent aesthetic, mechanical, and biocompatible properties, and are currently the most commonly used restorative materials for patients with clinical tooth defects and dentition defects.

[0003] Bonding is a key factor determining the success of restorative treatment and the clinical service life of restorations. Good bonding can not only provide stable retention for restorations, but also improve the aesthetic effect and marginal sealing of restorations.

[0004] Clinically, an ideal bonding effect can be achieved by increasing mechanical retention and chemical bonding. Resin cement is a conventional bonding material for dental restorations such as all-ceramic restorations and polymerized ceramic restorations. The usual bonding process involves pretreatment of the restoration surface, pretreatment of the tooth surface, and final cementation of the resin cement. For example, pretreatment of the surface of all-ceramic restorations and polymerized ceramic restorations usually includes hydrofluoric acid etching or sandblasting to roughen the surface and form a micro-mechanical interlock with the infiltrated resin cement; chemical conditioning of the surface, such as coating with a silane coupling agent or 10-methacryloyloxydecyl dihydrogen phosphate (MDP), can promote chemical bonding between the restorative material and the resin cement.

[0005] The most direct evaluation method for the bonding strength between restorative materials and resin cements is in vitro bonding mechanical experiments, such as (micro)tensile bonding strength and (micro)shear bonding strength, etc. However, in vitro bonding experiments are limited. Although there are relatively standardized test methods and instruments, this standardization, despite stipulating some parameters such as specimen size, loading speed, loading head shape, etc., and with the assistance of statistical analysis, is far from sufficient to avoid the influence of uncontrollable factors such as experimental instruments, operational technique sensitivity, and test environment between different experiments. The measured values or the comparative conclusions obtained between experiments may deviate greatly, and even completely opposite conclusions may occur. A very common situation is that even for investigations of the same adherend, there are hundreds or even more similar literatures, and each literature has a clear conclusion, but these literatures may present different conclusions, making it impossible to determine which conclusion is the closest to the actual situation. Even though each of these traditional bonding experiments has consumed a large amount of time and resources for intensive specimen production and repeated laboratory measurements, the experiment authors claim that their research conclusions are correct, but they have to admit that it is impossible to explain where the different conclusions in other studies come from. Thus, this deviation in conclusions between experiments can only be attributed to the inherent drawbacks of traditional in vitro experiments. The inefficiency, low sensitivity, and uncontrollability of bonding experiments make it quite difficult to compare the bonding results between experiments.

[0006] Meanwhile, research on changing the composition and structure of restorative materials to improve the mechanical strength of ceramics themselves has also been ongoing. In vitro bonding experiments evaluating a certain new bonding product or bonding strategy usually select one or at least a few types of restorative products as adherends, because the limitations of experimental cost and workload need to be considered. This brings a problem that the conclusions obtained from bonding experiments may not be applicable, or at least inaccurate, for other restorative materials with different chemical compositions from the restorative products selected in the experiments. Another problem is that when evaluating the effect of a certain active ingredient in a bonding product on the bonding of a restorative material alone, it is also impossible to evaluate whether this active ingredient is affected or interacted by other chemical components in the bonding product or system, thereby affecting the bonding performance. The compositions of various chemical components in different products are different. There are at least hundreds of bonding products available on the market, and there is definitely more than one claimed bonding strategy for each product. Coupled with the changing factors of the chemical composition of the target restorative material for bonding, although each clinician pursues the best combination plan to ensure clinical longevity, it is unrealistic to accurately select the best bonding product and strategy for a restorative material through traditional in vitro bonding experiments from this ever-changing combination.

[0007] The above situations urge for a more powerful bonding experiment method, which can evaluate the individual and interactive effects of factors such as chemical components in bonding products, components of restorative materials, and bonding strategies on bonding performance without relying on in vitro experiments, thereby facilitating the development of new restorative materials, new bonding products, and bonding strategies.

[0008] Therefore, aiming at the problems of insufficient accuracy in predicting the bonding strength of dental restorative materials, low data utilization efficiency, lack of comprehensive analysis tools, a rapid prediction method for the bonding strength value between dental restorative materials and resin cements based on machine learning can be designed. Summary of the Invention

[0009] In order to overcome the problems of insufficient accuracy in measuring the bonding strength value between current dental restorative materials and resin cements, low data utilization efficiency, lack of comprehensive analysis tools.

[0010] The technical solution of the present invention is: a rapid prediction method for the bonding strength value of dental restorative materials based on machine learning, and its steps are as follows:

[0011] The steps are as follows:

[0012] S1: Based on the in vitro bonding experiment data of dental restorative materials, construct an optimal XGB prediction model through the methods of stratified cross-validation, nested cross-validation, and grid search;

[0013] S2: Preliminarily process the information of the dental restorative material to be predicted, or the bonding strategy to be predicted, or the bonding operation method to be predicted, etc., to ensure data consistency, and the format and structure of the newly input data are the same as those of the training data, such as including the same features (variables), the same feature order, and the same data type, such as numerical type or categorical type;

[0014] S3: Input all relevant features of the adherend and related bonding treatment factors involved in the bonding performance of the restorative material into the constructed python algorithm, including the chemical composition and processing method of the restorative material, the roughening pretreatment method of the restorative material surface, the chemical composition of the treatment agent applied to the restorative material surface, the chemical composition of the adhesive applied to the restorative material surface, the chemical composition of the resin cement, and the bonding steps;

[0015] S4: Through XGB model analysis, obtain the predicted bonding strength value range according to the binary classification result output after the operation model;

[0016] S5: According to the predicted bonding strength value range combined with the clinically acceptable threshold, judge the bonding performance that the restorative material, or bonding strategy, or bonding method to be predicted can achieve, and provide a reference for the selection of restorative materials, the selection of bonding products, and the optimization and adjustment of bonding strategies.

[0017] Preferably, the preprocessed data is numerical, categorical, or text data that can be recognized by computer code.

[0018] Preferably, a rapid detection technique for predicting and evaluating the bonding strength between dental restorations relying on resin cements and resin cements is applied. This technique does not require in vitro experiments.

[0019] Advantages of the present invention:

[0020] 1. It can be closely combined with the optimization of dental restoration materials, corresponding bonding products, and bonding strategies. By real-time analyzing material physical and chemical property parameters such as chemical composition, surface roughening treatment methods, chemical composition of bonding products, and bonding steps, these key features are synchronously fed back to the selection of restoration materials and the optimization link of bonding strategies together with the prediction results of the model. Combining its closed-loop control system with the prediction model provides a reference for selecting restoration materials, dynamically adjusting bonding operation steps and operation parameters, so as to better optimize the bonding strength, achieve the best bonding effect of the restoration materials, improve the reliability and performance of the restorations in clinical applications, and extend the clinical service life of the restorations;

[0021] 2. In order to construct the optimal XGB model, a number of improvements have been made in the construction and optimization of the model. First, aiming at the data complexity of predicting the bonding strength of restoration materials, combining feature engineering methods and feature importance analysis, by deeply analyzing multiple factors affecting the bonding strength (such as chemical composition of restoration materials, type of adhesives, surface treatment methods, etc.), key features are screened, the data dimension is reduced, and the prediction ability of the model is improved. Secondly, the dataset is divided into a training set and a test set. Stratified cross-validation, nested cross-validation, and grid search are carried out on the test set to improve data utilization; further evaluation of the prediction accuracy and generalization ability of the model is carried out on the test set to improve the credibility of the model. Through these improvements, the XGB model not only shows stronger stability when dealing with high-dimensional data, but also enables the model to have better generalization ability in practical applications and can adapt to changes in different materials and processes;

[0022] 3. The model can predict the bonding strength range, improve the accuracy and reliability of the prediction; efficiently process and utilize experimental data, improve data utilization; comprehensively analyze multiple influencing factors, provide a comprehensive and systematic analysis tool; select restoration materials and appropriate bonding processes and optimize them to achieve the best bonding strength and stability of the restorations. Description of the Drawings

[0023] Figure 1 The flowchart of the rapid prediction method for the bonding strength value of dental restoration materials based on machine learning according to the present invention is shown;

[0024] Figure 2Shown is the screening of the prediction model for the bonding strength of lithium metasilicate glass-ceramics implemented in the present invention;

[0025] Figure 3 Shown are the results of analyzing the feature importance of lithium metasilicate glass-ceramics based on six machine learning models implemented in the present invention;

[0026] Figure 4 Shown is the average importance score of lithium metasilicate glass-ceramics based on machine learning implemented in the present invention;

[0027] Figure 5 Shown is the visualization of the ROC curve on the training set and test set of the XGB model implemented in the present invention. Detailed implementation manners

[0028] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0029] Embodiment 1

[0030] Please refer to Figure 1 , the present invention provides an embodiment: a rapid prediction method for the bonding strength value of dental restoration materials based on machine learning, and the steps are as follows:

[0031] S1: Based on the in vitro bonding experiment data of dental restoration materials, construct an optimal XGB prediction model through the methods of stratified cross-validation, nested cross-validation, and grid search;

[0032] S2: Preliminarily process the dental restoration materials to be predicted, or the bonding strategies to be predicted, or the information of the bonding methods to be predicted: the preprocessed data is numerical, categorical, or text data that can be recognized by computer code, etc.; ensure data consistency, and the format and structure of the newly input data are the same as those of the training data, such as including the same features (variables), the same feature order, and the same data type, such as numerical or categorical;

[0033] S3: Input all relevant features of the adherends and related bonding treatment factors involved in the bonding performance of dental restoration materials into the constructed python algorithm, including the chemical composition and processing method of the restoration materials, the roughening pretreatment method of the restoration material surface, the chemical composition of the treatment agent applied to the restoration material surface, the chemical composition of the adhesive applied to the restoration material surface, the chemical composition of the resin cement, and the bonding steps;

[0034] S4: Through XGB model analysis, obtain the high or low predicted bonding strength value range according to the binary classification result "1" or "0" output after the operation model;

[0035] S5: Based on the predicted bonding strength value range and combined with the clinically acceptable threshold, judge the bonding performance that can be achieved by the dental restoration material bonding strategy or bonding method to be predicted, providing a reference for the selection of restoration materials, the selection of bonding products, and the optimization and adjustment of bonding strategies.

[0036] Example 2

[0037] Please refer to Figure 2 , the present invention provides an example: the establishment and screening of a prediction model for the bonding strength of lithium metasilicate glass-ceramics based on machine learning, including the following steps:

[0038] 1. Data collection

[0039] On the basis of Example 1, taking lithium metasilicate reinforced glass-ceramics as an example, considering that traditional bonding experiments require a large amount of time and cost, and the tested bonding strategies and bonding methods are limited, this experiment collected data related to the in vitro bonding experiment of lithium disilicate reinforced glass-ceramics from past experimental data or articles to construct a data set;

[0040] 2. Feature selection

[0041] Use feature engineering to extract useful information from the original data, including feature construction, feature encoding, and feature scaling;

[0042] Perform feature engineering and feature importance analysis on the relevant features in the data set to ensure that the input data has high correlation and reduce the data dimension to improve the prediction performance of the model;

[0043] Six machine learning models (LR, DT, RF, ET, GB, and XGB) were selected for feature importance analysis ( Figure 3 ); each model provides a feature importance score for each input feature, denoted as I i,j , where i = 1, 2,..., 6 represents the model index, and j = 1, 2,..., n represents the feature index;

[0044] Since the magnitudes of the feature importance scores of each model are different, the maximum-minimum normalization (min-max normalization) was used to normalize the feature importance scores of each model to ensure comparability. The normalization formula is as follows:

[0045]

[0046] Among them, represents the normalized importance score of feature X i in model M i , I min,i and I max,i are the minimum and maximum values of the feature importance scores of model M iThe minimum and maximum importance scores;

[0047] After normalization, the average importance score of each feature is obtained by calculating the average of the six model normalization scores, and the formula is as follows:

[0048]

[0049] Where, represents the average importance score of feature X i ;

[0050] Finally, sort according to the average importance scores of the features to determine the relative importance of each feature in predicting the classification of bond strength ( Figure 4 ), features with higher average scores are considered to be more influential in model prediction and also have greater improvement significance in the optimization of the bonding process. Reconstruct the dataset with the features having the top average feature importance scores to reduce data complexity and improve prediction accuracy;

[0051] 3. Model training and evaluation

[0052] Divide the dataset into a training set (80%) and a test set (20%). The training set is used to construct a machine learning model, and the test set is used to verify the performance of the model to ensure that the model still performs well on unseen data;

[0053] (1) Model training

[0054] Use a binary classification model to predict the bond strength of lithium metasilicate glass-ceramics, construct an XGB model, and use grid search for parameter optimization to find the best combination of hyperparameters on the training set; specifically, use stratified k-fold cross-validation (random parameter is 42) on the training set to evaluate the generalization performance of the model through different data partitions, ensuring that the distribution of the bond strength categories is balanced in each fold; grid search cross-validation is used to adjust the model hyperparameters to improve the prediction ability of the model;

[0055] (2) Model evaluation

[0056] Use the AUC-ROC curve (area under the receiver operating characteristic curve) to evaluate the classification effect of the model. This metric can comprehensively reflect the performance of the model. Further verify the classification performance of the model through accuracy to ensure the precision of the prediction results. To reduce the influence of random parameters on the AUC score, draw the AUC scores of stratified cross-validation with random parameters from 0 to 29 as a box plot. At the same time, to better evaluate the model performance, nested cross-validation is also adopted;

[0057] ROC (Receiver Operating Characteristic) Curve Generation Method: For a binary classification model, the model classifies the input data set where x i is the sample feature, y i ∈ {0, 1} is the label, and the model outputs a predicted value for the probability or score that the sample x i is classified as the positive class (i.e., );

[0058] ROC Curve Construction Steps: By setting a classification threshold t, the predicted value is converted into a discrete label i.e.: when , otherwise where t ∈ [0, 1] is an adjustable threshold. In most cases, the classification model defaults to setting the threshold to 0.5, which means that if the model predicts the probability of a sample to be greater than or equal to 0.5, it classifies the sample as the positive class;

[0059] True Positive Rate (TPR) and False Positive Rate (FPR) Calculation: For each threshold t, the True Positive Rate (TPR) and False Positive Rate (FPR) are calculated by the following formulas:

[0060]

[0061]

[0062] where I(·) is the indicator function, which takes the value of 1 when the condition inside the parentheses holds, and 0 otherwise. According to different thresholds t, an ROC curve is constructed, where each point is (FPR(t), TPR(t));

[0063] Area Under the ROC Curve (AUC) Calculation: Let t 1 , t 2 , …, t k be a sequence of thresholds arranged in descending order, and the corresponding TPR and FPR values are (TPR 1 , FPR 1 ), (TPR 2 , FPR 2 ), …, (TPR k , FPR k ), then the area under the ROC curve (AUC) can be calculated by the trapezoidal method, and the formula is:

[0064]

[0065] The above method constructs a framework that can generate an ROC curve and calculate the AUC value, where the ROC curve reflects the performance changes of the classification model at different thresholds, and the AUC value represents the comprehensive measure of the model's ability to distinguish positive and negative class samples;

[0066] 4. Model Validation

[0067] Obtain the performance metrics of the AUC-ROC curve and accuracy on the test set;

[0068] The optimal parameters and evaluation results of the model are shown in Table 1. Through comprehensive analysis, it can be concluded that XGB is the optimal model, which has good generalization ability and accuracy on both the training set and the test set. Figure 5 It is the visualization of the ROC curve of the XGB model on the training set and the test set.

[0069] Example 3

[0070] Reliable prediction of the bonding strength value of lithium metasilicate glass-ceramics using the XGB model

[0071] Take the known bonding data of lithium metasilicate glass-ceramics that are not included in the training or test set as separate validation data, determine information such as its bonding products, glass-ceramic composition, bonding strategy, etc. By inputting key feature values such as hydrofluoric acid concentration and treatment time, silane coupling agent composition, bonding agent composition (such as Bis-GMA, UDMA, TEGDMA, filler), and cement composition (such as solvent type, Bis-GMA, UDMA) into the already constructed optimal XGB model (parameters: colsample_bytree: 1.0, learning_rate: 0.2, max_depth: 5, n_estimators: 200, subsample: 0.9), the predicted range of the bonding strength value can be obtained (Table 1). The specific output result is 0 (low strength value) or 1 (high strength value). The obtained result is highly accurate compared with the actual situation, which corroborates the prediction accuracy verified by the model.

[0072] Table 1 Performance evaluation of multiple models based on lithium metasilicate glass-ceramic data.

[0073]

[0074]

[0075] Through the above steps, the following technical effects are achieved:

[0076] (1) Bonding strength prediction: The main technical effect of the present invention is to accurately predict the bonding strength of dental restoration materials using a machine learning model, so as to guide the selection and use of materials.

[0077] (2) Optimize decision-making: By using a machine learning-based classification model, it is possible to accurately distinguish the bonding strength categories of dental restoration materials, improving the selection efficiency and accuracy of materials in practical applications.

[0078] (3) Enhance the accuracy and robustness of the model: Adopt hierarchical cross-validation (CV) and nested cross-validation along with performance evaluation metrics (AUC-ROC curve, accuracy) to ensure the robustness of model evaluation and improve the generalization ability of the model.

[0079] The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings. However, the present invention is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art.

Claims

1. A method for rapidly predicting the bonding strength value of dental restoration materials based on machine learning, characterized in that: The steps are as follows: S1: The optimal XGB prediction model was constructed based on the in vitro bonding experimental data of dental restorative materials by using stratified cross-validation, nested cross-validation and grid search methods; S2: Preliminary processing of information such as the dental restoration material to be predicted, or the bonding strategy to be predicted, or the bonding operation method to be predicted to ensure data consistency. The format and structure of the new input data are consistent with the training data, such as containing the same features (variables), the same feature order, and the same data type, such as numerical or categorical; S3: Input all relevant features of the adherends and related bonding treatment factors involved in the bonding performance of the repair material into the constructed python algorithm, including the chemical composition and processing method of the repair material, the roughening pretreatment method of the repair material surface, the chemical composition of the treatment agent applied to the repair material surface, the chemical composition of the adhesive applied to the repair material surface, the chemical composition of the resin cement, and the bonding steps; S4: Through XGB model analysis, the predicted bonding strength value range is obtained according to the binary classification results output after the operation model; S5: Based on the predicted bonding strength value range combined with the clinically acceptable threshold, determine the bonding results that can be achieved by the predicted restorative material, bonding strategy, or bonding method, and provide a reference for the selection of restorative materials, bonding products, and optimization and adjustment of bonding strategies.

2. The method for rapid prediction of bonding strength value of dental restorative materials based on machine learning according to claim 1, characterized in that: The preprocessed data is numerical, categorical or textual data that can be recognized by computer code.

3. The method for rapid prediction of bonding strength value of dental restoration materials based on machine learning according to claim 1, characterized in that: A rapid testing technology used to predict and evaluate the bonding strength between dental restorations that rely on resin cement and resin cement. This technology does not require in vitro experiments.

Images