Gum health risk prediction method, electric toothbrush and program product

Abstract

The invention provides a gingival health risk prediction method, an electric toothbrush and a program product, and the method comprises the steps: inputting the obtained user historical case information into a pre-trained static risk assessment model for gingival health risk grading, and obtaining the initial probability of the gingival health risk; the method comprises the following steps: acquiring tooth brushing behavior data in real time through an electric toothbrush, preprocessing the tooth brushing behavior data, and inputting the preprocessed tooth brushing behavior data into a pre-trained dynamic risk correction model to obtain a correction coefficient of gingival health risk; and multiplying the correction coefficient by the initial probability to obtain a gingival health risk prediction result, dividing the gingival health risk prediction result into a plurality of risk levels according to a predetermined threshold, and generating an oral care suggestion according to the risk levels. By means of the technical scheme, more comprehensive and accurate oral health assessment and personalized nursing suggestions can be provided, and a user is effectively helped to find and improve gingival health problems in time.

Description

Technical Field

[0001] This invention relates to the field of gingival health prediction, and in particular to a method for predicting gingival health risks, an electric toothbrush, and a programmed product. Background Technology

[0002] As people place increasing importance on oral health, electric toothbrushes are becoming more and more popular due to their efficient cleaning capabilities. Current electric toothbrushes on the market primarily focus on improving the cleaning effect on the tooth surface and can record some basic brushing data, such as brushing time and frequency. However, these basic data have limitations in assessing gum health risks and cannot accurately predict the true health status of the gums. Furthermore, gum health problems are often difficult to detect in their early stages, and if targeted interventions are not implemented in a timely manner, they can lead to more serious oral health problems. Summary of the Invention

[0003] Embodiments of the present invention provide a method for predicting gum health risks, an electric toothbrush, and a program product to provide more comprehensive and accurate oral health assessments and personalized care recommendations, effectively helping users to promptly identify and improve gum health problems.

[0004] To achieve the above objectives, on the one hand, a method for predicting gingival health risks is provided, the method comprising: The acquired user historical medical records are input into a pre-trained static risk assessment model to classify gum health risk and obtain an initial probability of gum health risk. The historical medical records include: historical gum disease information to determine if the user has a history of gum disease; historical oral surgery information to determine if the user has a history of oral surgery; and family genetic information to determine if the user has a family history of oral diseases. The static risk assessment model calculates the initial probability using the following formula:

[0005] This represents the initial probability. Indicate whether there is a history of gum disease. Indicate whether there is a history of oral surgery. Indicate whether there is a family history of genetic disorders. and The values ​​are either 0 or 1, where 0 indicates a relevant medical history and 1 indicates no relevant medical history. , and This represents the parameters of the static risk assessment model determined through maximum likelihood estimation; Real-time brushing behavior data is acquired using an electric toothbrush. This data is preprocessed and then input into a pre-trained dynamic risk correction model to obtain a correction coefficient for gum health risk. The brushing behavior data includes: pressure data applied by the electric toothbrush to the teeth or gums; position coordinates of the electric toothbrush relative to its initial position when activated in the oral cavity; angle data of the electric toothbrush relative to its initial angle when activated in the oral cavity; and the movement speed of the electric toothbrush. The correction coefficient is:

[0006] K represents the correction coefficient, and SVM represents the Support Vector Machine algorithm. This represents the preprocessed brushing behavior data; The correction coefficient is multiplied by the initial probability to obtain the gum health risk prediction result. The gum health risk prediction result is divided into multiple risk levels according to a predetermined threshold, and oral care recommendations are generated based on the risk levels.

[0007] Preferably, in the method for predicting gingival health risks, the preprocessed brushing behavior data includes: average pressure, pressure standard deviation, brushing speed standard deviation, and the dwell time of the electric toothbrush on the gums; wherein, the preprocessing includes: data cleaning and data calibration of the brushing behavior data, followed by feature extraction; wherein, The data cleaning includes: marking and removing data in the brushing behavior data that exceeds a predetermined range; and smoothing the brushing behavior data using a sliding window or moving average filter. The data calibration includes: standardizing the brushing behavior data.

[0008] Preferably, in the method for predicting gum health risks, the average pressure is the average pressure exerted by the toothbrush on the teeth and gums during brushing, calculated using the following formula:

[0009] in, This represents the average pressure. Indicates time Real-time stress data acquisition That is the total time spent brushing your teeth; The pressure standard deviation is used to measure the degree of pressure fluctuation during brushing, and is calculated using the following formula.

[0010] in, This represents the standard deviation of the pressure. The standard deviation of brushing speed is used to measure the degree of fluctuation in brushing speed during the brushing process, and is calculated using the following formula.

[0011] in, Indicates the time of the toothbrush The magnitude of the velocity at any given moment This indicates the average speed during the brushing process.

[0012] Preferably, in the gum health risk prediction method, the electric toothbrush includes a nine-axis gyroscope. Accelerometer data is acquired using the accelerometer of the nine-axis gyroscope. The position coordinates and the movement velocity are calculated using the accelerometer data. The specific position of the electric toothbrush within the oral cavity is determined using the position coordinates, the movement velocity, and the angle data. The specific position determines the various tooth surfaces and gum areas covered by the electric toothbrush. The position coordinates are obtained by integrating the initial position of the electric toothbrush with the accelerometer data.

[0013] Preferably, in the gum health risk prediction method, the static risk assessment model is trained using a predetermined first dataset, which includes predetermined user historical case data and a first initial probability corresponding to the user historical case data. The training process includes: using the predetermined user historical cases as input to the static risk assessment model and the corresponding first initial probability as output to the static risk assessment model, and performing supervised learning training on the static risk assessment model.

[0014] Preferably, in the gum health risk prediction method, the dynamic risk correction model is trained using a predetermined second dataset, which includes predetermined brushing behavior data and oral examination results corresponding to the predetermined brushing behavior data. The training process includes using the second dataset as input to the dynamic risk correction model, which establishes a mapping relationship between the predetermined brushing behavior data and the correction coefficient of gum health risk by minimizing classification error or regression error, cross-validation, kernel function, and optimizing hyperparameters.

[0015] Preferably, the method for predicting gingival health risks further includes: performing correlation analysis, sensitivity analysis, feature importance ranking, visualization analysis, and statistical analysis on the risk level to obtain key factors affecting gingival health risks, and generating corresponding oral care recommendations based on the key factors; The correlation analysis includes: calculating the correlation between the brushing behavior data and the gingival health risk prediction results using the Pearson correlation coefficient formula; The sensitivity analysis includes: changing the value of each brushing behavior data point and recording the changes in the gum health risk prediction results; The feature importance ranking includes: calculating the importance of each brushing behavior data point using a random forest and ranking them by importance; The visualization analysis includes: plotting a scatter plot of the brushing behavior data and the gum health risk prediction results; The statistical analysis includes analyzing each brushing behavior data point using t-tests or variance statistics. The key factors include at least one of the brushing behavior data; The oral care recommendations include: adjusting brushing pressure, changing brushing technique, increasing gum massage time and / or seeking timely medical attention.

[0016] Preferably, the method for predicting gingival health risks further includes: optimizing the static risk assessment model and the dynamic risk correction model based on the gingival health risk prediction results, wherein the optimization includes: adjusting model parameters, increasing the amount of training data, improving feature engineering, and / or model evaluation and validation; wherein, The adjustment of model parameters includes: learning rate adjustment, regularization parameter adjustment and / or tree model parameter adjustment; The improved feature engineering includes: feature screening, feature construction, and / or feature transformation; wherein, the feature screening includes: screening and analyzing the brushing behavior data based on the gingival health risk prediction results, and using the screened brushing behavior data as input to the static risk assessment model and the dynamic risk correction model; the feature construction includes: data augmentation of the brushing behavior data; the feature transformation includes: performing a logarithmic transformation on the pressure data in the brushing behavior data; and normalizing the position coordinates and movement speed in the brushing behavior data; The model evaluation and validation include cross-validation and multi-index evaluation.

[0017] On the other hand, embodiments of the present invention provide an electric toothbrush, wherein the electric toothbrush includes: a pressure sensor for acquiring pressure data of the electric toothbrush; a nine-axis gyroscope for acquiring position coordinates, angle data, and movement speed of the electric toothbrush; a memory and a processor, wherein the memory stores at least one program, the at least one program being executed by the processor to implement the gingival health risk prediction method as described above.

[0018] In another aspect, embodiments of the present invention provide a computer program product, including a computer program, characterized in that, when the computer program is executed by a processor, it implements the gingival health risk prediction method as described above.

[0019] The above technical solution has the following technical effects: In embodiments of the present invention, the initial probability of gum health risk is calculated by inputting the user's historical medical information into a static risk assessment model. Furthermore, the electric toothbrush collects real-time brushing behavior data from the user, such as pressure, position coordinates, angle, and movement speed, and obtains a correction coefficient through a dynamic risk correction model. The correction coefficient is multiplied by the initial probability to obtain a more accurate gum health risk prediction result. Based on a predetermined threshold, the gum health risk prediction result is divided into different risk levels, and targeted oral care recommendations are generated accordingly. Compared to traditional methods that rely solely on basic brushing data, embodiments of the present invention combine long-term health records with short-term behavioral changes, improving the accuracy and timeliness of gum health risk assessment and helping to detect and prevent potential gum problems early. Attached Figure Description

[0020] Figure 1 This is a schematic flowchart of a method for predicting gingival health risks according to an embodiment of the present invention. Detailed Implementation

[0021] To further illustrate the various embodiments, the present invention provides accompanying drawings. These drawings are part of the disclosure of the present invention, primarily used to illustrate the embodiments and to explain the operating principles of the embodiments in conjunction with the relevant descriptions in the specification. With reference to these drawings, those skilled in the art should be able to understand other possible implementations and the advantages of the present invention. Components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0022] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments.

[0023] Example 1: To provide a more comprehensive and accurate oral health assessment and personalized care recommendations, and to effectively help users identify and improve gum health problems in a timely manner, embodiments of the present invention provide a gum health risk prediction method. Figure 1 This is a schematic flowchart of a method for predicting gingival health risks according to an embodiment of the present invention. Figure 1 As shown, the method includes: The acquired user historical medical records are input into a pre-trained static risk assessment model to classify gum health risk and obtain an initial probability of gum health risk. The historical medical records include: historical gum disease information to determine if the user has a history of gum disease; historical oral surgery information to determine if the user has a history of oral surgery; and family genetic information to determine if the user has a family history of oral diseases. The static risk assessment model calculates the initial probability using the following formula:

[0024] Indicates the initial probability. Indicate whether there is a history of gum disease. Indicate whether there is a history of oral surgery. Indicate whether there is a family history of genetic disorders. and The values ​​are either 0 or 1, where 0 indicates a relevant medical history and 1 indicates no relevant medical history. , and This represents the parameters of the static risk assessment model determined through maximum likelihood estimation; Real-time brushing behavior data is acquired using an electric toothbrush. This data is preprocessed and then input into a pre-trained dynamic risk correction model to obtain correction coefficients for gum health risk. The brushing behavior data includes: pressure applied by the electric toothbrush to the teeth or gums; position coordinates of the electric toothbrush relative to its initial position in the mouth upon activation; angle data of the initial angle of the electric toothbrush relative to its initial position in the mouth upon activation; and the movement speed of the electric toothbrush. The correction coefficients are:

[0025] K represents the correction coefficient, and SVM represents the Support Vector Machine algorithm. This represents preprocessed brushing behavior data; among which, These are the input features for the dynamic risk correction model; The correction coefficient is multiplied by the initial probability to obtain the gum health risk prediction result. The gum health risk prediction result is divided into multiple risk levels according to a predetermined threshold, and oral care recommendations are generated based on the risk level.

[0026] The static risk assessment model and the dynamic risk correction model in the embodiments of this invention have the following inherent relationship: 1) The essence of the output probability of a static risk assessment model The initial probability output by the static risk assessment model ( This represents the user's baseline risk probability for gum health, calculated as follows: Input: Limited to long-term, unchanging medical history factors (history of gingival disease) oral surgery history Family history of genetic diseases ).

[0027] Output meaning: = The probability of a user having gum disease due to congenital / historical factors, ranging from 0 to 1, ignoring current brushing behavior.

[0028] Example: There is a family history of genetic disorders. And had previously suffered from periodontitis. users, It could reach 0.8, which is considered high risk.

[0029] 2) The core function of the dynamic correction model: behavioral risk correction. The dynamic risk correction model outputs correction coefficients. The amplification / reduction effect of quantifying brushing behavior on underlying risks: Input: Brushing behavior characteristics, such as pressure / speed fluctuations, etc. See details. .

[0030] Output meaning: Current brushing habits exacerbate underlying risks, such as brushing too vigorously; Current brushing habits reduce basic risks, such as proper brushing technique; Example: If the standard deviation of a user's brushing pressure Exceeding the limit, the dynamic risk correction model outputs... This indicates a 20% increase in risk.

[0031] 3. In one implementation, the mathematical and medical basis for the correlation between the two models is shown in Table 1 below: Table 1

[0032] As shown in the table above, it is reasonable to couple these two models to accurately predict users' gum health risks.

[0033] The static risk model outputs the baseline risk determined by medical history, while the dynamic risk correction model outputs the real-time adjustment coefficient of the baseline risk based on the current brushing behavior. The two are coupled through multiplication; the physical meaning of the coupled result is clear. This aligns with the medical logic of "long-term risk × short-term behavioral impact"; The two models are trained independently, achieving technical separation: the static risk assessment model uses medical history data, while the dynamic risk correction model uses behavior-clinical paired data; there is no input confusion between the two; and the coupling is reasonable.

[0034] Furthermore, in specific implementations, the clinical interpretability of the embodiments of the present invention is strong. For example, the positive or negative impact of brushing behavior on gingival health can be directly corresponded by the correction coefficient K>1 or K<1.

[0035] The impact of brushing on gum health is supported by medical evidence, for example: Stress data: Excessive stress is a direct cause of gingival recession (see "Oral Health and Overall Health: White Paper on Prevention and Management of Periodontal Disease", Chinese Stomatological Association, Clinical Guidelines).

[0036] Retention time: Insufficient cleaning of the gingival margin area is positively correlated with plaque accumulation at the gingival margin (J Clin Periodontol paper: Journal of Clinical Periodontology (IF6.8) Pub Date: 2000-09-13, DOI:10.1034 / j.1600-051x.2000.027009637.xL A Ximénez-Fyvie1, AD Haffajee , S Som, M Thompson , G Torresyap , SS Socransky).

[0037] Example 2: Embodiments of the present invention provide a method for predicting gingival health risks, comprising: The acquired user historical medical records are input into a pre-trained static risk assessment model to classify gum health risk and obtain an initial probability of gum health risk. The historical medical records include: historical gum disease information to determine if the user has a history of gum disease; historical oral surgery information to determine if the user has a history of oral surgery; and family genetic information to determine if the user has a family history of oral diseases. The static risk assessment model calculates the initial probability using the following formula:

[0038] Indicates the initial probability. Indicate whether there is a history of gum disease. Indicate whether there is a history of oral surgery. Indicate whether there is a family history of genetic disorders. and The values ​​are either 0 or 1, where 0 indicates a relevant medical history and 1 indicates no relevant medical history. , and This represents the parameters of the static risk assessment model determined through maximum likelihood estimation; Real-time brushing behavior data is acquired using an electric toothbrush. This data is preprocessed and then input into a pre-trained dynamic risk correction model to obtain correction coefficients for gum health risk. The brushing behavior data includes: pressure applied by the electric toothbrush to the teeth or gums; position coordinates of the electric toothbrush relative to its initial position in the mouth upon activation; angle data of the initial angle of the electric toothbrush relative to its initial position in the mouth upon activation; and the movement speed of the electric toothbrush. The correction coefficients are:

[0039] K represents the correction coefficient, and SVM represents the Support Vector Machine algorithm. This represents pre-processed brushing behavior data; The correction coefficient is multiplied by the initial probability to obtain the gum health risk prediction result. The gum health risk prediction result is divided into multiple risk levels according to a predetermined threshold, and oral care recommendations are generated based on the risk level.

[0040] Preferably, a high-risk level is defined as a gum health risk prediction result greater than or equal to 0.7; a medium-risk level is defined as a gum health risk prediction result less than 0.7 but greater than or equal to 0.3; and a low-risk level is defined as a gum health risk prediction result less than 0.3. Preferably, a high-precision pressure sensor is provided in the brush head of the electric toothbrush to detect the pressure applied to the teeth and gums by the electric toothbrush in real time. The pressure data is converted into an electrical signal for transmission through principles such as piezoresistive or capacitive effects. This pressure sensor has high sensitivity and a wide measurement range, accurately capturing subtle changes in pressure during brushing. Furthermore, the electric toothbrush is equipped with a nine-axis gyroscope, which integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The three-axis accelerometer measures the linear acceleration of the electric toothbrush along three orthogonal axes, the three-axis gyroscope measures the angular velocity of the toothbrush around the three axes, and the three-axis magnetometer measures the components of the Earth's magnetic field along the three axes. By fusing data from these three sensors, the spatial position, posture, and motion state of the electric toothbrush can be accurately calculated, further enabling precise measurement of its position, angle, and speed during brushing. Analysis of the nine-axis gyroscope data determines the toothbrush's specific position within the mouth, assessing whether it covers all tooth surfaces and the gum line, while also acquiring its trajectory and speed changes. Simultaneously, the pressure sensor is calibrated using standard weights, and the measurement accuracy of the nine-axis gyroscope is verified and adjusted using high-precision positioning equipment.

[0041] Specifically, during each brushing session, the electric toothbrush simultaneously activates a pressure sensor and a nine-axis gyroscope to continuously collect brushing behavior data, including but not limited to brushing pressure, toothbrush position coordinates, angle changes, and movement speed. This data is transmitted in real-time via wireless communication methods such as Bluetooth and Wi-Fi to a compatible smart device, such as a smartphone, for storage and processing through a corresponding app on the device. The smart device can also be set with regular reminders to remind users to brush their teeth on time, check gum health risk assessment results, and undergo oral examinations, enhancing users' awareness and management of oral health.

[0042] Preferably, the preprocessed brushing behavior data includes: average pressure, standard deviation of pressure, standard deviation of brushing speed, and dwell time of the electric toothbrush on the gums; wherein, the preprocessing includes: data cleaning and data calibration of the brushing behavior data, followed by feature extraction; wherein, Data cleaning is used to remove outliers and errors from brushing behavior data. This includes: monitoring the raw data output from the pressure sensor and the nine-axis gyroscope; if the data shows abnormal fluctuations or exceeds a reasonable range (i.e., a predetermined range), such as pressure values ​​exceeding the maximum pressure threshold that the gums can withstand, position coordinates exceeding the oral cavity space range, or abnormal acceleration values, these data are marked as abnormal and removed; and smoothing the brushing behavior data using sliding window or moving average filtering to reduce noise interference and improve data quality. Data calibration is used to ensure the accuracy and consistency of data from different sensors, ensuring that the data from different sensors have the same dimensions and scale range, facilitating subsequent analysis. This includes standardizing brushing behavior data; for example, using the Z-score standardization method to convert brushing behavior data into standard normal distribution data with a mean of 0 and a standard deviation of 1, the formula is:

[0043] in, For brushing behavior data, This represents the mean of the brushing behavior data. The standard deviation of brushing behavior data. Standardized brushing behavior data; Feature extraction is used to extract characteristic parameters related to gingival health risks, including: extracting stable parameters of brushing pressure, duration of electric toothbrush stay in the gingival region, and standard deviation of brushing speed.

[0044] Specifically, average pressure is the average pressure applied by the toothbrush to the teeth and gums during brushing. It reflects whether the brushing force is appropriate and is calculated using the following formula.

[0045] in, Indicates average pressure. Indicates time Real-time stress data acquisition That is the total time spent brushing your teeth; Specifically, the standard deviation of pressure is used to measure the degree of fluctuation in pressure during brushing, i.e., to assess the stability of force. Large fluctuations can easily damage the gums. The standard deviation of pressure is calculated using the following formula.

[0046] in, Indicates the standard deviation of pressure; Mean pressure and pressure standard deviation are extracted from the raw pressure time-series data; Specifically, the standard deviation of brushing speed is used to measure the degree of fluctuation in brushing speed during the brushing process, and is calculated using the following formula:

[0047] in, Indicates the time of the toothbrush The magnitude of the velocity at a given moment is the instantaneous velocity at that moment, which is also the magnitude of the vector velocity. In one specific implementation, this value is obtained through a nine-axis gyroscope. The average speed during brushing is a scalar quantity calculated by integrating the instantaneous speed. T is the total brushing time per session, which is recorded by a timer in one implementation. The standard deviation of brushing speed is a quantified value of the degree of speed fluctuation, directly reflecting the stability of the brushing motion. The larger this value, the more uneven the brushing appears, i.e., brushing behavior that is sometimes fast and sometimes slow, which can easily lead to incomplete cleaning or gum damage.

[0048] In practice, the original data is acquired by discretely sampling the brushing speed in the actual electric toothbrush processor.

[0049] These are the input features for the dynamic risk correction model; in one specific implementation, .

[0050] Preferably, the electric toothbrush includes a nine-axis gyroscope. Accelerometer data is acquired through the accelerometer of the nine-axis gyroscope. Position coordinates and motion velocity are calculated using the accelerometer data. The specific position of the electric toothbrush in the oral cavity is determined using the position coordinates, the motion velocity, and the angle data. The specific position determines the various tooth surfaces and gum areas covered by the electric toothbrush. The position coordinates are obtained by integrating the initial position of the electric toothbrush with the accelerometer data.

[0051] Specifically, angle data refers to the angles around the x, y, and z axes of an electric toothbrush relative to its initial posture when activated in the mouth. In other words, angle data describes how the electric toothbrush changes direction during use. The x-axis typically refers to left-right rotation (roll); the y-axis to front-back rotation (pitch); and the z-axis to rotation perpendicular to the toothbrush head surface (yaw). For example, when a user adjusts the toothbrush angle to better clean different areas of the teeth or gums while brushing, these angle changes can be precisely recorded.

[0052] Specifically, position coordinates refer to the spatial position of the electric toothbrush relative to its initial position when it is activated within the user's mouth during brushing. This describes the dynamic movement trajectory of the electric toothbrush during brushing, including changes in the forward / backward, left / right, and up / down directions. Combining position coordinates with data from other sensors, such as pressure sensors, can determine the extent of contact between the electric toothbrush and the teeth or gums. For example, if the pressure sensor detects a pressure value exceeding a predetermined pressure threshold, it means the electric toothbrush is applying pressure to the teeth or gums, indicating contact.

[0053] In one specific embodiment, the position coordinates of the nine-axis gyroscope in the electric toothbrush are obtained by integrating accelerometer data. The accelerometer measures the linear acceleration of the electric toothbrush along three orthogonal axes (x, y, z). By integrating the acceleration data twice, the position change of the electric toothbrush can be obtained, and then combined with the initial position to obtain the real-time position coordinates. The specific formula is as follows: In time interval The formula for calculating the position coordinates of an electric toothbrush is as follows:

[0054] in, It is the accelerometer in time The acceleration vector measured at time t. It is the velocity vector obtained by integrating acceleration. These are the initial coordinates of the toothbrush.

[0055] The velocity is obtained by integrating the accelerometer data once. The specific formula is as follows: In time interval The formula for calculating the speed of an electric toothbrush is:

[0056] in, It is the initial velocity vector of the toothbrush. It is the accelerometer in time The acceleration vector measured at any given time.

[0057] In one specific implementation, the training of the dynamic risk correction model has the following characteristics: Training data: The second dataset is a mapping from "brushing behavior characteristics" to "clinical gingival status at the same time," such as the gingival bleeding index. Example data is shown in Table 2 below: Table 2

[0058] Model learning objective: SVM directly establishes the association between brushing behavior and current gum status through this type of data, forming a non-indirect association static risk assessment model.

[0059] Among them, the final risk value The medical explanation, where K is the correction factor: Basic scenario: If there is no behavioral influence (this is the ideal situation of brushing teeth). The risk is entirely determined by the medical history, that is... .

[0060] Behavioral impact: When aggressive brushing is detected, the model obtains... The actual risk for users becomes For example, if brushing behavior is ideal, the initial probability of having gum health risks, i.e., the basic probability of gum health risks, is 0.4. If K=1.5, the corrected probability of having gum health risks rises to 0.6.

[0061] Preferably, a static risk assessment model is constructed by combining the user's historical medical record information, including but not limited to past history of gingival disease, oral surgery history, and family history of gingival disease. Based on this medical record data, the user's gingival health risk is initially classified to obtain an initial probability of the user's gingival health risk. The static risk assessment model is trained using a predetermined first dataset, which includes predetermined user historical case data and corresponding initial probabilities. The training process includes using predetermined user historical cases as input to the static risk assessment model and the corresponding initial probabilities as output, performing supervised learning training on the static risk assessment model. Preferably, machine learning algorithms, such as neural networks and support vector machines, are used to train a dynamic risk correction model on a large amount of user brushing behavior data and their corresponding gingival health status (obtained through oral examination results). This dynamic risk correction model can dynamically correct the static risk assessment result, i.e., the initial probability, based on real-time brushing behavior data to more accurately predict the current gingival health risk status. The dynamic risk correction model is trained using a predetermined second dataset, which includes predetermined brushing behavior data and corresponding oral examination results. The training process includes using the second dataset as input to the dynamic risk correction model, which establishes a mapping relationship between the predetermined brushing behavior data and the correction coefficient of gingival health risk by minimizing classification error or regression error, cross-validation, kernel function, and optimizing hyperparameters.

[0062] Preferably, the static risk assessment model is a logistic regression model; the dynamic risk correction model is a classification or regression model constructed using machine learning algorithms. Selecting a suitable machine learning algorithm is a crucial step in training the static risk assessment model and the dynamic risk correction model. The selection strategy needs to comprehensively consider the characteristics of the data and the specific requirements of the prediction objective. Specifically, the machine learning algorithms for the static risk assessment model and the dynamic risk correction model are selected based on a predetermined strategy, taking into account the data characteristics and prediction objectives. The data characteristics include: linearity versus nonlinearity, data size, data distribution characteristics, and data dimensionality; the prediction objectives include: classification problems, regression problems, and clustering problems. The predetermined strategy includes: selecting machine learning algorithms based on data characteristics; prioritizing linear models such as linear regression, LDA, and linear SVM when the data is linear, as they are computationally efficient and easy to interpret; selecting models capable of handling nonlinear relationships, such as decision trees and their ensemble algorithms, neural networks, and nonlinear SVMs, to better fit the complex patterns of the data; considering computational efficiency when dealing with large datasets, choosing linear models, tree models (such as GBDT), or neural network frameworks with distributed computing capabilities; and avoiding complex datasets when dealing with small datasets. For model selection, choose a simple model and combine it with regularization techniques, such as ridge regression and Lasso regression, to prevent overfitting. When the data is normally distributed, you can use models based on parametric assumptions, such as linear regression and t-tests. When the data is non-normally distributed, choose a model suitable for the specific distribution or a non-parametric model, such as Poisson regression, negative binomial regression, and decision trees. When the data is low-dimensional, a simple model is usually sufficient, such as linear regression and Naive Bayes. When the data is high-dimensional, use models capable of handling high-dimensional data, such as SVM and random forests, combined with feature selection or dimensionality reduction techniques. The choice of machine learning algorithm depends on the prediction objective. When the prediction objective is binary classification, logistic regression, decision trees, random forests, SVM, neural networks, etc., are selected based on the data characteristics. When the prediction objective is multi-class classification, multi-class versions of the above algorithms are selected, such as CNN for image multi-class classification. When the prediction objective is regression, linear regression, multinomial regression, decision trees, random forests, GBDT, neural networks, etc., are selected based on the linear or non-linear characteristics of the data. When the prediction objective is clustering, k-means clustering, hierarchical clustering, DBSCAN, etc., are selected based on the data distribution and cluster characteristics.

[0063] In practical applications, data characteristics and prediction objectives are often intertwined. Therefore, it is necessary to consider both factors when selecting a model. For example, in a large-scale data classification problem with nonlinear characteristics, a trade-off may need to be struck between decision tree ensemble algorithms (such as random forests and GBDT) and neural networks. If computational resources are limited, or model interpretability is required, decision tree ensemble algorithms may be more suitable; if higher prediction accuracy is pursued and sufficient computational resources are available, neural networks may be a better choice. Furthermore, the performance of different models under specific data and prediction objectives can be compared through cross-validation and model evaluation metrics such as accuracy, recall, F1 score, and mean squared error, thereby selecting the optimal model. For example, the formula for calculating accuracy is:

[0064] In this context, TP represents a true positive, TN represents a true negative, FP represents a false positive, and FN represents a false negative.

[0065] Preferably, combining the static risk assessment model and the dynamic risk correction model can comprehensively consider the user's historical medical records and real-time brushing behavior data, output a gum health risk assessment result, and present it to the user in the form of an intuitive risk level. The risk level is then analyzed by correlation analysis, sensitivity analysis, feature importance ranking, visualization analysis and statistical analysis to generate oral care recommendations. Specifically, correlation analysis is used to measure the strength of the linear relationship between two variables. In gum health risk prediction, the correlation between brushing behavior data and the initial probability or gum health risk prediction result can be calculated, including: calculating the correlation between brushing behavior data and gum health risk prediction result using the Pearson correlation coefficient formula; the Pearson correlation coefficient formula is:

[0066] in, and These are the observed values ​​of two variables. and These are their means.

[0067] In one specific embodiment, by calculating the correlation between brushing pressure and the predicted gum health risk, it can be determined whether brushing pressure is a key factor affecting gum health risk. If the absolute value of the correlation coefficient is large, such as greater than 0.5, it indicates a strong correlation between the two.

[0068] Specifically, sensitivity analysis is used to assess how sensitive the output of a dynamic risk correction model is to changes in input variables. In gum health risk prediction, it can evaluate the impact of different brushing behavior data on the prediction results, identifying the factors with the greatest influence. This includes changing the value of each brushing behavior data point and recording the changes in the prediction results; for example, gradually increasing brushing pressure and recording the changes in the prediction results. If the gum health risk prediction results increase significantly with increasing pressure, it indicates that brushing pressure is a sensitive factor.

[0069] Specifically, feature importance ranking is used to determine which brushing behavior data contributes most to the prediction of gum health risk. For example, when the dynamic risk correction model is a random forest or gradient boosting tree model, the ranking can be achieved by calculating the average importance of each brushing behavior data point when constructing the decision tree. Taking random forest as an example, feature importance can be determined by calculating the average number of times each brushing behavior data point is used to split nodes across all decision trees and the average purity gain resulting from the splits. By ranking feature importance, it is intuitive to see which brushing behavior data points are most critical for predicting gum health risk.

[0070] Specifically, visualization analysis helps to intuitively discover patterns and key factors in the data, including: plotting scatter plots of brushing behavior data and gum health risk prediction results, with the horizontal axis representing brushing pressure and the vertical axis representing gum health risk prediction results, and observing the distribution trend of the data points. If the data points show a clear upward trend, it indicates a positive correlation between brushing pressure and gum health risk. Similarly, graphs showing the relationship between other brushing behavior data and gum health risk prediction results can be plotted.

[0071] Specifically, the statistical analysis includes analyzing each brushing behavior data point using t-tests or statistical methods of variance, and comparing the mean differences in the predicted gum health risk under different brushing behavior data. If the mean predicted gum health risk of a certain factor differs significantly at different levels, it indicates that the factor has a significant impact on gum health risk.

[0072] Specifically, oral care recommendations include: adjusting brushing pressure, changing brushing technique, increasing gum massage time and / or seeking timely medical attention.

[0073] In one specific embodiment, a smart device paired with the electric toothbrush provides users with gum health risk information based on gum health risk prediction results and corresponding risk levels. When a high gum health risk is predicted, the smart device alerts the user through various means such as graphics, text, and sound, and provides corresponding oral care suggestions, such as adjusting brushing pressure, changing brushing techniques, increasing gum massage time, and seeking timely medical attention. The smart device can also upload the user's gum health risk assessment results and related brushing behavior data to a cloud server to establish the user's oral health record. Dentists or professional oral care personnel can access the cloud server to view the user's oral health data and provide remote oral health consultation and guidance services.

[0074] Preferably, the static risk assessment model and the dynamic risk correction model are optimized based on the gingival health risk prediction results. The optimization process includes: adjusting model parameters, increasing the amount of training data, improving feature engineering, and / or model evaluation and validation. Specifically, adjusting model parameters includes: learning rate adjustment, regularization parameter adjustment, and / or tree model parameter adjustment. Among these, finding a suitable learning rate through cross-validation ensures that the static risk assessment model and the dynamic risk correction model converge quickly during training without becoming unstable due to an excessively large learning rate. In logistic regression or neural networks, increasing the weight of the L2 regularization term can smooth the parameters of the static risk assessment model and the dynamic risk correction model, improving their generalization ability. For tree models such as decision trees, random forests, and GBDT, the complexity of the static risk assessment model and the dynamic risk correction model can be controlled by adjusting parameters such as tree depth and minimum number of split samples. For example, limiting the maximum tree depth can prevent the static risk assessment model and the dynamic risk correction model from becoming too complex and causing overfitting.

[0075] Specifically, increasing the amount of training data includes: (1) Data augmentation: For existing brushing behavior data, more training samples can be generated through data augmentation techniques. For example, random noise can be added to brushing pressure data, and small random perturbations can be made to toothbrush position and speed data to simulate more brushing scenarios; (2) Collect more user data: Expand the user group and collect more user data with different ages, genders, oral health conditions, etc., so that the static risk assessment model and dynamic risk correction model can learn a wider range of brushing behavior patterns and improve the robustness and generalization ability of the static risk assessment model and dynamic risk correction model; (3) Balance the dataset: If the distribution of samples of certain categories in the dataset, such as high risk, medium risk, and low risk, is unbalanced, the number of samples of each category can be made more balanced by sampling the minority class or undersampling the majority class, so as to avoid the bias caused by data imbalance in the static risk assessment model and dynamic risk correction model.

[0076] Specifically, improved feature engineering includes: feature selection, feature construction, and / or feature transformation; among which, Feature selection includes: filtering and analyzing brushing behavior data based on gingival health risk prediction results, and using the filtered brushing behavior data as input for static risk assessment models and dynamic risk correction models; for example, based on key factors identified in the gingival health risk prediction analysis, such as brushing pressure and toothbrush dwell time, brushing behavior data that is highly correlated with gingival health risk is selected, while redundant or irrelevant brushing behavior data is removed.

[0077] Feature construction includes: data augmentation of brushing behavior data; that is, constructing new brushing behavior data based on existing brushing behavior data. For example, calculating the rate of pressure change during brushing, the proportion of time the toothbrush stays in different areas, etc., these new brushing behavior data may better reflect the impact of brushing behavior on gum health.

[0078] Feature transformations include: performing a logarithmic transformation on the pressure data in the brushing behavior data to make it closer to a normal distribution; and normalizing the position coordinates and movement speed in the brushing behavior data to ensure that the dimensions of different brushing behavior data are consistent, which is beneficial to the training and convergence of the model. Specifically, model fusion includes: (1) Ensemble learning: integrating multiple different models, such as weighted averaging or voting on the gingival health risk prediction results of models such as logistic regression, decision tree, and neural network to obtain the final gingival health risk prediction result; ensemble learning can make full use of the advantages of each model and improve the prediction accuracy and stability of the model. (2) Stacked model: constructing a stacked model, that is, first using multiple base models to predict the data, then using the prediction results of the base models as input, and then training a meta-model to fuse and optimize the prediction results of the base models; stacked model can further improve the performance of the model.

[0079] Specifically, model evaluation and validation include cross-validation, multi-metric evaluation, and continuous monitoring and updating. Cross-validation involves using k-fold cross-validation, dividing the dataset into k subsets. Each time, k-1 subsets are used as the training set, and the remaining subset is used as the validation set. This process is repeated k times, and the average result is used as the evaluation metric for the static risk assessment model or dynamic risk correction model. Cross-validation fully utilizes the data and avoids inaccurate model evaluation results due to the randomness of data partitioning. Multi-metric evaluation includes combining commonly used classification metrics such as precision, recall, and F1 score with metrics such as ROC curves and AUC scores to comprehensively evaluate the performance of the static risk assessment model or dynamic risk correction model. For regression problems, metrics such as mean squared error (MSE) and mean absolute error (MAE) can be used. Continuous monitoring and updating involves continuously monitoring the predictive performance of the static risk assessment model or dynamic risk correction model during practical application, regularly collecting new data to retrain and update the model to adapt to changes in brushing behavior data and dynamic changes in users' oral health status.

[0080] Example 3: The present invention also provides an electric toothbrush, comprising: a pressure sensor for acquiring pressure data of the electric toothbrush; a nine-axis gyroscope for acquiring position coordinates, angle data and movement speed of the electric toothbrush; a memory and a processor, wherein the memory stores at least one program, and the at least one program is executed by the processor to implement any of the above-mentioned gingival health risk prediction methods.

[0081] Example 4: The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps described above.

[0082] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A method for predicting gingival health risks, characterized in that, include: The acquired user historical medical records are input into a pre-trained static risk assessment model to classify gum health risk and obtain an initial probability of gum health risk. The historical medical records include: historical gum disease information to determine if the user has a history of gum disease; historical oral surgery information to determine if the user has a history of oral surgery; and family genetic information to determine if the user has a family history of oral diseases. The static risk assessment model calculates the initial probability using the following formula: This represents the initial probability. Indicate whether there is a history of gum disease. Indicate whether there is a history of oral surgery. Indicate whether there is a family history of genetic disorders. and The values ​​are either 0 or 1, where 0 indicates a relevant medical history and 1 indicates no relevant medical history. , and This represents the parameters of the static risk assessment model determined through maximum likelihood estimation; Real-time brushing behavior data is acquired using an electric toothbrush. This data is preprocessed and then input into a pre-trained dynamic risk correction model to obtain a correction coefficient for gum health risk. The brushing behavior data includes: pressure data applied by the electric toothbrush to the teeth or gums; position coordinates of the electric toothbrush relative to its initial position when activated in the oral cavity; angle data of the electric toothbrush relative to its initial angle when activated in the oral cavity; and the movement speed of the electric toothbrush. The correction coefficient is: K represents the correction coefficient, and SVM represents the Support Vector Machine algorithm. This represents the preprocessed brushing behavior data; The correction coefficient is multiplied by the initial probability to obtain the gum health risk prediction result. The gum health risk prediction result is divided into multiple risk levels according to a predetermined threshold, and oral care recommendations are generated based on the risk levels.

2. The method for predicting gingival health risks according to claim 1, characterized in that, The preprocessed brushing behavior data includes: average pressure, pressure standard deviation, brushing speed standard deviation, and the dwell time of the electric toothbrush on the gums; wherein, the preprocessing includes: data cleaning and calibration of the brushing behavior data, followed by feature extraction; wherein, The data cleaning includes: marking and removing data in the brushing behavior data that exceeds a predetermined range; and smoothing the brushing behavior data using a sliding window or moving average filter. The data calibration includes: standardizing the brushing behavior data.

3. The method for predicting gingival health risks according to claim 2, characterized in that, The average pressure is the average pressure exerted by the toothbrush on the teeth and gums during brushing, calculated using the following formula: in, This represents the average pressure. Indicates time Real-time stress data acquisition That is the total time spent brushing your teeth; The pressure standard deviation is used to measure the degree of pressure fluctuation during brushing, and is calculated using the following formula. in, This represents the standard deviation of the pressure. The standard deviation of brushing speed is used to measure the degree of fluctuation in brushing speed during the brushing process, and is calculated using the following formula. in, Indicates the time of the toothbrush The magnitude of the velocity at any given moment This indicates the average speed during the brushing process.

4. The method for predicting gingival health risks according to claim 1, characterized in that, The electric toothbrush includes a nine-axis gyroscope. Accelerometer data is acquired through the accelerometer of the nine-axis gyroscope. The position coordinates and the movement velocity are calculated using the accelerometer data. The specific position of the electric toothbrush in the oral cavity is determined using the position coordinates, the movement velocity, and the angle data. The specific position determines the various tooth surfaces and gum areas covered by the electric toothbrush. The position coordinates are obtained by integrating the initial position of the electric toothbrush with the accelerometer data.

5. The method for predicting gingival health risks according to claim 1, characterized in that, The static risk assessment model is trained using a predetermined first dataset, which includes predetermined user historical case data and corresponding first initial probabilities. The training process includes: using the predetermined user historical cases as input to the static risk assessment model and the corresponding first initial probabilities as output to the static risk assessment model, and performing supervised learning training on the static risk assessment model.

6. The method for predicting gingival health risks according to claim 1, characterized in that, The dynamic risk correction model is trained using a predetermined second dataset, which includes predetermined brushing behavior data and oral examination results corresponding to the predetermined brushing behavior data. The training process includes using the second dataset as input to the dynamic risk correction model, which establishes a mapping relationship between the predetermined brushing behavior data and the correction coefficient of gingival health risk by minimizing classification error or regression error, cross-validation, kernel function, and optimizing hyperparameters.

7. The method for predicting gingival health risks according to claim 1, characterized in that, Also includes: Correlation analysis, sensitivity analysis, feature importance ranking, visualization analysis, and statistical analysis are performed on the risk levels to obtain key factors affecting gingival health risks, and corresponding oral care recommendations are generated based on the key factors. The correlation analysis includes: calculating the correlation between the brushing behavior data and the gingival health risk prediction results using the Pearson correlation coefficient formula; The sensitivity analysis includes: changing the value of each brushing behavior data point and recording the changes in the gum health risk prediction results; The feature importance ranking includes: calculating the importance of each brushing behavior data point using a random forest and ranking them by importance; The visualization analysis includes: plotting a scatter plot of the brushing behavior data and the gum health risk prediction results; The statistical analysis includes analyzing each brushing behavior data point using t-tests or variance statistics. The key factors include at least one of the brushing behavior data; The oral care recommendations include: adjusting brushing pressure, changing brushing technique, increasing gum massage time and / or seeking timely medical attention.

8. The method for predicting gingival health risks according to claim 1, characterized in that, Also includes: The static risk assessment model and the dynamic risk correction model are optimized based on the gingival health risk prediction results. The optimization includes: adjusting model parameters, increasing the amount of training data, improving feature engineering, and / or model evaluation and validation. The adjustment of model parameters includes: learning rate adjustment, regularization parameter adjustment and / or tree model parameter adjustment; The improved feature engineering includes: feature screening, feature construction, and / or feature transformation; wherein, the feature screening includes: screening and analyzing the brushing behavior data based on the gingival health risk prediction results, and using the screened brushing behavior data as input to the static risk assessment model and the dynamic risk correction model; the feature construction includes: data augmentation of the brushing behavior data; the feature transformation includes: performing a logarithmic transformation on the pressure data in the brushing behavior data; and normalizing the position coordinates and movement speed in the brushing behavior data; The model evaluation and validation include cross-validation and multi-index evaluation.

9. An electric toothbrush, characterized in that, The electric toothbrush includes: A pressure sensor is used to acquire pressure data from the electric toothbrush; A nine-axis gyroscope is used to acquire the position coordinates of the electric toothbrush, the angle data of the electric toothbrush, and the movement speed of the electric toothbrush; A memory and a processor, the memory storing at least one program, the at least one program being executed by the processor to implement the gingival health risk prediction method as described in any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the gingival health risk prediction method as described in any one of claims 1 to 8.

Images