Method and system for constructing periodontitis dynamic prognosis prediction model based on survival analysis

Abstract

The invention discloses a periodontitis dynamic prognosis prediction model construction method and system based on survival analysis, and the method comprises the steps: collecting and preprocessing periodontitis patient data, and constructing a structured multi-modal clinical data set; the method comprises the following steps: randomly dividing a multi-modal clinical data set into a training set and a test set according to patient IDs, only on the training set, carrying out single-factor Cox proportional risk regression analysis on all patient level data and tooth level data together, and screening indexes significantly related to periodontitis progress risks; taking periodontitis progress time as a dependent variable, and using the training set and the screened indexes to construct a time-dependent Cox periodontitis dynamic prognosis prediction model; and introducing an SHAP algorithm to analyze a prediction result of the time-dependent Cox periodontitis dynamic prognosis prediction model. The method can be used for realizing dynamic, accurate and explainable prediction of the periodontitis progress risk.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of medical informatization, and relates to a periodontitis dynamic prognosis prediction model construction method and system based on survival analysis. BACKGROUND

[0002] Periodontitis is a chronic infectious disease mediated by dental plaque biofilm. Its pathological nature is the inflammatory reaction caused by the disruption of the dynamic balance between plaque microorganisms and the host immune defense system. Its characteristics are the progressive destruction of periodontal support tissue (gingiva, periodontal membrane, alveolar bone and cementum), which can eventually lead to tooth loosening and even loss.

[0003] The complexity of periodontitis treatment prognosis is due to the pathological mechanism of multi-factor interaction and clinical heterogeneity. The individual difference of treatment effect is significant, and is jointly affected by patient level factors (such as age, smoking) and tooth level factors (such as root furcation lesion, mobility). At present, the prognosis evaluation of periodontitis mainly depends on the clinical experience of doctors and static evaluation standards, which has obvious limitations: Lack of dynamic prediction ability: the existing methods are mostly cross-sectional evaluation or only predict the end outcome (such as tooth loss), which cannot capture the dynamic risk evolution rule of the disease at different time points such as 1 year, 3 years and 5 years, and it is difficult to identify the critical point of the transition from stable period to active period, so as to miss the opportunity of early intervention.

[0004] Prediction model "black box" problem: even if machine learning or statistical model is applied, the prediction logic is often opaque, and the clinician cannot understand the quantitative contribution of each feature (such as root furcation lesion, clinical attachment loss) to the prognosis of specific patients, which makes it difficult for clinicians to develop individualized treatment plans and makes it difficult to land clinically.

[0005] Insufficient use of data: traditional prediction models fail to effectively integrate and analyze multi-modal data of patient level (global background) and tooth level (local features), ignoring the interaction between the two, which limits the improvement of prediction accuracy.

[0006] Survival analysis (such as Cox proportional hazards model) can effectively handle right-censored data and make time-dependent risk prediction. Explainable AI technology (such as SHAP) can crack the model "black box". However, simply combining Cox model with SHAP (Shapley Additive exPlanations) analysis for periodontitis prognosis prediction still has the following technical bottlenecks: due to the complex hierarchical structure and interaction effect between patient level factors and tooth level factors in periodontal clinical data, if not properly handled, it is easy to introduce noise, and the SHAP analysis result may be mixed and unclear, which cannot provide clear and targeted decision basis for the clinic.

[0007] Therefore, there is an urgent need in the art for a periodontitis prognosis prediction scheme that can effectively avoid data leakage, accurately screen key features, thereby constructing a high-precision dynamic prediction model, and providing clear clinical interpretation. SUMMARY

[0008] The present application aims to overcome the shortcomings of the prior art, and provides a periodontitis dynamic prognosis prediction model construction method and system based on survival analysis, to realize dynamic, accurate and interpretable prediction of periodontitis progression risk. It includes: 1) solving the effective integration and robust feature selection problem of patient-level and tooth-level multi-modal data in periodontitis data; 2) constructing a periodontitis dynamic prognosis prediction model with high discrimination (AUC) at multiple time points (such as 1, 3, and 5 years); 3) providing intuitive and individualized feature contribution analysis, so that clinicians can understand the driving factors behind each prediction and their changes over time, and support the development of personalized treatment plans.

[0009] To achieve the above-mentioned purpose, the present application adopts the following technical solutions: In a first aspect, the present application provides a periodontitis dynamic prognosis prediction model construction method based on survival analysis, comprising: Collecting periodontitis patient data including patient-level data and tooth-level data, performing Z-score standardization on continuous variables in the periodontitis patient data, and performing dummy variable encoding on categorical variables in the periodontitis patient data to construct a structured multi-modal clinical data set; Randomly dividing the multi-modal clinical data set into a training set and a test set according to patient ID, performing single-factor Cox proportional hazards regression analysis on all patient-level data and tooth-level data together only on the training set, screening variables that meet the significance level P<0.05, and obtaining a set of predictors significantly related to periodontitis progression risk; Taking periodontitis progression time as the dependent variable and the set of predictors as the covariate, a time-dependent Cox periodontitis dynamic prognosis prediction model is constructed; Introducing SHAP algorithm to perform multi-level and quantifiable analysis on the prediction results of the time-dependent Cox periodontitis dynamic prognosis prediction model, and realizing dynamic interpretation from global feature importance to individualized risk driving factors.

[0010] Specifically, the single-factor Cox proportional hazards regression analysis includes calculating the hazard ratio, 95% confidence interval and significance P value of each variable.

[0011] Specifically, the risk function of the time-dependent Cox periodontitis dynamic prognosis prediction model is represented as: h(t|X) = h0(t) x exp(β1X1 + β2X2 +... + βpXp), wherein h(t|X) is the risk function given the covariate X at time t, h0(t) is the baseline risk function at time t, X1, X2, …, Xp are the 1st, 2nd, …, pth covariates, β1, β2, …, βp are the regression coefficients of the 1st, 2nd, …, pth covariates; the time-dependent Cox periodontitis dynamic prognosis prediction model is used to calculate the conditional probability of periodontitis progression of an individual tooth at a future time point. p X p ), wherein h(t|X) is the risk function given the covariate X at time t, h0(t) is the baseline risk function at time t, X1, X2, …, X p are the 1st, 2nd, …, pth covariates, β1, β2, …, β p are the regression coefficients of the 1st, 2nd, …, pth covariates; the time-dependent Cox periodontitis dynamic prognosis prediction model is used to calculate the conditional probability of periodontitis progression of an individual tooth at a future time point.

[0012] Specifically, the time-dependent ROC curve is used to evaluate the discrimination ability of the time-dependent Cox periodontitis dynamic prognosis prediction model at different time points, and the performance of the time-dependent Cox periodontitis dynamic prognosis prediction model is quantified by calculating the area under the time-dependent curve.

[0013] Specifically, the absolute value mean of the SHAP value of each prediction factor is calculated, and a SHAP value importance ranking graph is generated; for the same tooth, a SHAP force diagram at different time points is drawn respectively to show the dynamic evolution process of the contribution degree of each prediction factor over time.

[0014] Specifically, according to the risk probability of an individual tooth at multiple time points output by the time-dependent Cox periodontitis dynamic prognosis prediction model and the determined optimal cutoff value, the tooth is divided into a dynamically changing high-risk group and a low-risk group; based on the analytical results of the SHAP algorithm, the risk driving factors of the high-risk group teeth at different time points are identified; according to the risk stratification and risk driving factor analysis results, personalized re-visit interval suggestions and treatment scheme optimization basis can be provided for clinicians (for example, for high-risk teeth driven by “root furcation lesions”, it is recommended to prioritize periodontal surgery treatment), so as to realize precise grading diagnosis and treatment of periodontitis.

[0015] Specifically, the patient-level data includes one or more of age, gender, education level, body mass index, smoking history, systemic disease history, periodontitis staging and grading data; the tooth-level data includes one or more of probing depth, clinical attachment loss, probing bleeding, root furcation lesion degree, mobility, tooth position, alveolar bone loss stage.

[0016] In a second aspect, the present application provides a periodontitis dynamic prognosis prediction system based on survival analysis, which is used to realize the method of the first aspect, and includes the following modules: a clinical data collection and preprocessing module configured to receive and structurally store multi-modal clinical data at patient level and tooth level, and perform data cleaning, continuous variable standardization and categorical variable encoding operations; a dataset division and predictor screening module connected to the clinical data collection and preprocessing module, configured to randomly divide the preprocessed multi-modal clinical dataset into a training set and a test set according to patient ID, and perform a single-factor Cox proportional hazards regression analysis on all variables in the training set to screen a set of predictors; a model construction and verification module connected to the dataset division and predictor screening module, configured to receive the set of predictors, construct a time-dependent Cox periodontitis dynamic prognosis prediction model, and verify the performance of the model; a SHAP explainability analysis module connected to the model construction and verification module, configured to call the trained time-dependent Cox periodontitis dynamic prognosis prediction model, calculate a SHAP value importance ranking chart and a SHAP force chart of individual teeth at different time points; a clinical decision support module connected to the model construction and verification module and the SHAP explainability analysis module, configured to generate a risk stratification report and individualized diagnosis and treatment recommendations based on the risk probability predicted by the time-dependent Cox periodontitis dynamic prognosis prediction model and the risk driving factors analyzed by SHAP.

[0017] In a third aspect, the present application provides an electronic device comprising at least one processor and a memory connected in communication with the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the periodontitis dynamic prognosis prediction model construction method based on survival analysis.

[0018] In a fourth aspect, a computer readable storage medium having computer instructions stored thereon, the computer instructions being executed by a processor to implement the steps of the periodontitis dynamic prognosis prediction model construction method based on survival analysis.

[0019] Compared with the prior art, the present application has the following beneficial effects: Dynamic prognosis prediction capability: by constructing a time-dependent Cox periodontitis dynamic prognosis prediction model, the probabilistic prediction of the risk of tooth-level periodontitis progression at future consecutive time points (such as 1, 3, 5 years) is realized, breaking through the limitations of traditional static assessment or only providing relative risk ranking, and being able to capture the dynamic process of disease evolution, providing a time window for early intervention.

[0020] High precision and strong discrimination: The model performs well on the test set, with AUC values of 0.943, 0.903, and 0.858 for 1-year, 3-year, and 5-year predictions, respectively, demonstrating its superior prediction performance.

[0021] Deep interpretability and temporal dynamic insight: By combining SHAP algorithm with time dimension depth, not only the global feature importance ranking is provided, but also the individualized risk driving factor time series dynamic visualization is realized. Doctors can intuitively see the main contributing factors of the risk change of a specific tooth at different time points, understand the "why" and "when" of risk evolution, greatly enhancing the clinical credibility and usability of the model.

[0022] Rigorous data processing procedure and robust feature selection: Innovatively adopt the rigorous procedure of "first divide the data set according to patients, then perform prediction factor selection on the training set", through data isolation at the patient level, fundamentally avoid the "data leakage" caused by the same patient data being used for training and testing at the same time, ensure the unbiasedness and reliability of feature selection results and model performance evaluation, improve the generalization ability of the model.

[0023] Direct and quantitative clinical decision support: The dynamic risk prediction results are combined with the depth of explainability analysis to output not only the risk level, but also the key risk driving factors and their changes over time. This provides direct and quantitative scientific basis for clinical doctors to perform risk stratification, develop individualized follow-up plans, and precise intervention strategies (such as treatment targeting specific risk factors). BRIEF DESCRIPTION OF DRAWINGS

[0024] The drawings accompanying the specification of this application form a part thereof, serve to further provide a further understanding of the application, and together with the description of the exemplary embodiments of the application and the explanation thereof serve to explain the application without imposing undue limitation on the application.

[0025] Figure 1 is a flowchart of the periodontitis dynamic prognosis prediction model construction method based on survival analysis provided by embodiment 1 of the application.

[0026] Figure 2 is the time-dependent ROC curve of the time-dependent Cox model in embodiment 1 of the application at 1 year, 3 years, and 5 years.

[0027] Figure 3 is a global contribution ranking chart (feature importance chart) of each clinical feature to the periodontitis progression risk prediction based on SHAP value in embodiment 1 of the application.

[0028] Figure 4 is a SHAP force chart for dynamically analyzing the risk driving factors of periodontitis progression in embodiment 1 of the application.

[0029] Figure 5 Figure 1 is a Kaplan-Meier survival curve of high-risk and low-risk groups of teeth based on model-predicted risk division in Embodiment 1 of the present application. DETAILED DESCRIPTION

[0030] The present application will be further described below in conjunction with the accompanying drawings and embodiments. The present embodiment takes the real clinical data of the Oral Medicine Diagnosis and Treatment Center of the Second Affiliated Hospital of Nanchang University as an example to explain the implementation process of the method in detail. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application. Embodiment 1

[0031] The present embodiment takes the real clinical data of the Oral Medicine Diagnosis and Treatment Center of the Second Affiliated Hospital of Nanchang University as an example to explain the implementation process of the periodontitis dynamic prognosis prediction model based on survival analysis in detail. The core of this method is to organically combine the time-dependent Cox proportional hazards regression model with SHAP explainability analysis to achieve dynamic, individualized, and explainable prediction of periodontitis progression risk.

[0032] S1: Collect periodontitis patient data including patient-level data and tooth-level data, perform Z-score standardization on continuous variables in the periodontitis patient data, perform dummy variable encoding on categorical variables in the periodontitis patient data, and construct a structured multi-modal clinical data set.

[0033] Retrospectively collect 162 cases of periodontitis patient data from the Oral Medicine Diagnosis and Treatment Center of the Second Affiliated Hospital of Nanchang University from January 2018 to December 2023, a total of 3017 teeth. The periodontitis patient data includes patient-level data and tooth-level data; the patient-level data includes age, gender, education level, body mass index (BMI), smoking history, history of systemic diseases such as diabetes and cardiovascular disease, periodontitis staging and grading data; the tooth-level data includes probing depth (PPD), clinical attachment loss (CAL), bleeding index (BI), degree of root furcation lesion, tooth mobility, tooth position, and radiographic bone loss (RBL) staging of each tooth; The structured processing includes: Data cleaning: exclude the third molar, teeth with no retention value at baseline, teeth with poor image quality that cannot be measured, and teeth lost due to non-periodontitis reasons.

[0034] Feature standardization: Z-score standardization is performed on continuous variables such as age, body mass index (BMI), maximum probing depth, and maximum clinical attachment loss.

[0035] Feature encoding: Categorical variables such as gender, education level, periodontitis stage and grade, and degree of furcation involvement were encoded as dummy variables. The following rules were followed to ensure the rigor of the encoding: For categorical variables with multiple dimensions, such as "tooth position", the different attribute dimensions (e.g. "anterior / posterior" and "maxillary / mandibular") should be treated as independent categorical variables and encoded as dummy variables separately. This approach can avoid information redundancy and logical confusion caused by dimension intersection. When performing dummy variable encoding, to prevent multicollinearity problems, for any categorical variable with n categories, only n-1 dummy variables are generated (i.e. using the "delete the first category" or similar strategy), with one category serving as the reference benchmark.

[0036] S2: Data set division and prediction factor set screening based on training set To fundamentally avoid data leakage, the 162 periodontitis patient data were first randomly divided into a training set and a test set according to patient ID, ensuring that the training set and test set patients were completely independent, fundamentally avoiding "data leakage" caused by the same patient's different teeth appearing in both the training set and the test set, thereby ensuring the unbiasedness and reliability of subsequent model performance evaluation. Subsequently, only in the training set, all patient-level data and tooth-level data were subjected to single-factor Cox proportional hazards regression analysis, calculating the hazard ratio (Hazard Ratio, HR), 95% confidence interval (95% CI), and significance P value of each variable, and screening out variables significantly related to periodontitis progression (P < 0.05) as the prediction factor set for subsequent construction of the time-dependent Cox periodontitis dynamic prognosis prediction model. The rigorous process of "first dividing the training set and test set, then screening" ensures that the feature screening step itself is also completed on the training set, and the generalization ability of the screening results (significant variables) will be verified on a completely independent test set, which is more robust than screening on the entire data set or stratified screening.

[0037] The results showed that a total of 19 variables were significantly related to the risk of periodontitis progression (P < 0.05), including: 10 patient-level factors: age, gender, education level, diabetes, cardiovascular disease, rheumatoid arthritis, smoking, tooth brushing frequency, periodontitis stage, and periodontitis grade; 9 tooth-level factors: tooth position (anterior / posterior), tooth position (maxillary / mandibular), crown restoration, BI, maximum probing depth (maximum PPD), maximum clinical attachment loss (maximum CAL), furcation involvement, mobility, and alveolar bone loss stage (RBL stage).

[0038] The 19 variables are all included in the time-dependent Cox periodontitis dynamic prognosis prediction model. The key of this step is to divide the data at the patient level first, which effectively prevents the same patient's information from being used in the training and testing stages, ensuring the scientific rigor of all subsequent analyses.

[0039] S3: Taking the periodontitis progression time as the dependent variable and the set of prediction factors as the covariate, a time-dependent Cox periodontitis dynamic prognosis prediction model is constructed; Model construction: Taking the periodontitis progression time as the dependent variable and the 19 variables screened in step S2 as the covariates, a time-dependent Cox periodontitis dynamic prognosis prediction model is constructed. The risk function of the time-dependent Cox dynamic prognosis prediction model is represented as: h(t|X) = h0(t) x exp(β1X1+β2X2+...+βpXp), wherein h(t|X) is the risk function given the covariate X at time t, h0(t) is the baseline risk function at time t, X1, X2,..., Xp are the 1st, 2nd,..., pth covariates, β1, β2,..., βp are the regression coefficients of the 1st, 2nd,..., pth covariates; the core advantage of this model is that it can dynamically evaluate the change of risk over time and can predict the conditional probability P(T<t|X) of an individual tooth occurring an event at a specific time point t, rather than just giving a comprehensive hazard ratio. Unlike the ordinary Cox model which only outputs a comprehensive HR for risk ranking, the present application further calculates the conditional probability P(T<t|X) of an individual tooth occurring periodontitis progression at a specific future time point (such as 1 year, 3 years, 5 years) using the model. This makes the clinical monitoring jump from static relative risk assessment to dynamic, time-point-specific absolute risk prediction. p X p ), wherein h(t|X) is the risk function given the covariate X at time t, h0(t) is the baseline risk function at time t, X1, X2,..., X p are the 1st, 2nd,..., pth covariates, β1, β2,..., β p are the regression coefficients of the 1st, 2nd,..., pth covariates; the core advantage of this model is that it can dynamically evaluate the change of risk over time and can predict the conditional probability P(T<t|X) of an individual tooth occurring an event at a specific time point t, rather than just giving a comprehensive hazard ratio. Unlike the ordinary Cox model which only outputs a comprehensive HR for risk ranking, the present application further calculates the conditional probability P(T<t|X) of an individual tooth occurring periodontitis progression at a specific future time point (such as 1 year, 3 years, 5 years) using the model. This makes the clinical monitoring jump from static relative risk assessment to dynamic, time-point-specific absolute risk prediction.

[0040] Model validation: The time-dependent ROC curve is used to evaluate the discriminant ability of the time-dependent Cox periodontitis dynamic prognosis prediction model at different time points, and the area under the time-dependent curve (AUC) is calculated to quantify the performance of the model. This method is designed specifically to evaluate the prediction accuracy of survival analysis models at specific time points, which is superior to the overall evaluation using only the C-index. The results show that the AUC values of the time-dependent Cox periodontitis dynamic prognosis prediction model at 1 year, 3 years and 5 years reach 0.943, 0.903 and 0.858 respectively (see Figure 2 ), indicating that the constructed time-dependent Cox periodontitis dynamic prognosis prediction model has excellent discriminant ability in short-term and long-term prediction.

[0041] Dynamic risk stratification: Based on the above time-dependent risk probability, further according to the principle of maximizing the Youden index, the optimal cutoff value of the risk probability is determined for different time points (such as 1 year, 3 years, 5 years), so as to divide the teeth into high-risk group and low-risk group which changes dynamically over time.

[0042] The method for determining the optimal decision cutoff value is based on the prediction results of the test set to objectively determine the threshold value for converting the probability value output by the model into the final classification label in a data-driven manner, rather than using 0.5 as the default value subjectively. The specific steps are as follows: using the trained time-dependent Cox periodontitis dynamic prognosis prediction model to predict the samples in the test set, obtaining the prediction probability P of each sample belonging to the positive example (such as disease); traverse all possible probability cutoff points T between 0 and 1 (set the step size to 0.01). For each candidate cutoff point T, predict all samples with P≥T as positive examples and P<T as negative examples, and compare them with the true labels to calculate the sensitivity and specificity at this cutoff point; adopt the maximization of the Youden index as the criterion for determining the optimal cutoff point. Calculate the Youden index J corresponding to each candidate cutoff point T, the calculation formula is: J = sensitivity + specificity - 1; finally compare the Youden indexes J calculated by all candidate cutoff points T, select the cutoff point T* that makes the J value maximum, which is the optimal decision cutoff value determined by the present application. Through the above method, the subjectivity and potential bias of using 0.5 as the default cutoff value are overcome.

[0043] Survival analysis further shows that the 5-year progression-free survival rate of the high-risk group teeth classified by the time-dependent Cox periodontitis dynamic prognosis prediction model is only 23.4%, while that of the low-risk group is as high as 95.1% (see Figure 5 ), and the risk difference is extremely significant (P<0.0001).

[0044] In summary, the time-dependent Cox periodontitis dynamic prognosis prediction model constructed in this step not only retains the statistical rigor of the Cox model, but also realizes the dynamic monitoring and early warning of periodontitis progression risk by outputting the risk probability at different time points and supporting the time-resolved evaluation and stratification strategy, breaking through the limitation of traditional models that only provide static or overall risk ranking.

[0045] S4: Introducing SHAP algorithm to analyze the prediction results of the time-dependent Cox periodontitis dynamic prognosis prediction model at multiple levels and quantitatively, realizing the dynamic interpretation from global feature importance to individual risk driving factors.

[0046] For any tooth sample at time point t, the prediction risk value f(x) is decomposed into the sum of feature contributions by SHAP algorithm: f(x)= E[f(x)]+Σφi ; where E[f(x)] is the average predicted risk value (benchmark value) of the time-dependent Cox periodontitis dynamic prognosis prediction model on all samples, φ i is the SHAP value of the i-th predictor, i.e., the contribution of the i-th predictor to the predicted risk value, a positive value indicating an increased risk and a negative value indicating a decreased risk; Global feature importance: Calculate the absolute value mean of the SHAP values of each predictor to generate a SHAP value importance ranking chart (see Figure 3 ), and identify the core risk factors affecting periodontal progression from a global perspective (such as root furcation lesions, clinical attachment loss, etc.). It is found that root furcation lesions, maximum CAL, age, periodontitis stage, and smoking are the top five risk factors with the highest contribution. This allows doctors to quickly grasp the key indicators affecting prognosis. This method is superior to traditional regression coefficients or P-value ranking, and can accurately reflect the actual contribution of each feature to the model prediction output.

[0047] Individualized dynamic analysis: For the same tooth, SHAP value importance ranking charts at different time points are drawn to show the dynamic evolution process of the contribution of each predictor over time. By deeply integrating SHAP analysis with the time dimension, SHAP force diagrams (see Figure 4 ) of a single tooth at 1, 3, and 5 years are drawn to visually display the dynamic evolution process of the contribution of each predictor over time. This allows clinicians not only to understand "which factors are important", but also to grasp "when and which factors play a leading role", thereby identifying the "critical features" that cause periodontitis to transition from the stable phase to the active phase.

[0048] For example, a tooth has a low risk (f(x) = -0.09) at 1 year, and protective factors (such as stage III periodontitis) dominate (see Figure 4 a in the figure); By 3 years, the risk has increased (f(x) = 1.72), and I-degree root furcation lesions and smoking have become the main risk drivers (see Figure 4 b in the figure); By 5 years, the risk has significantly increased (f(x) = 4.62), and the contribution of I-degree root furcation lesions has increased dramatically (see Figure 4 c in the figure).

[0049] The SHAP force diagram visually presents the key factors leading to changes in risk status and their evolution rules by comparing the predicted feature contributions of a single tooth at 1 year, 3 years, and 5 years. This time-series contribution evolution visualization clearly reveals the "critical features" that cause the tooth to transition from the stable phase to the active phase, which cannot be achieved by ordinary Cox models or static SHAP analysis.

[0050] By quantifying the risk driving factors of individual teeth at different time points and their contribution changes, the present application realizes the leap from "population risk prediction" to "individualized mechanism interpretation". The analysis results can be directly used to explain why a certain tooth has a higher risk at a specific time point and provide a targeted basis for intervention measures (for example: if the risk is mainly driven by "periodontal furcation lesions" at 3 years, it suggests that local scaling or periodontal surgery intervention should be prioritized).

[0051] The deep combination of SHAP explainability analysis and time-dependent prediction model in this step realizes dynamic, individualized, and mechanistic analysis of periodontitis prognosis, breaks through the limitations of traditional models that only provide static or overall explanations, and provides key insights for clinical precision intervention.

[0052] According to the risk probabilities of individual teeth at multiple time points output by the time-dependent Cox periodontitis dynamic prognosis prediction model and the determined optimal cutoff value, the teeth are divided into high-risk and low-risk groups that change dynamically; based on the analysis results of the SHAP algorithm, the risk driving factors of high-risk teeth at different time points are identified. For example, for the tooth that turned into a high-risk tooth at 5 years in step S4, the system not only warns, but also clearly points out that "I degree root furcation lesion" is the current main intervenable risk factor. Doctors can take targeted measures accordingly, such as performing deeper root furcation debridement or considering surgical treatment on the tooth, rather than generally strengthening full-mouth scaling. At the same time, the system suggests shortening the patient's follow-up interval (such as from 1 year to 6 months), realizing true precision medicine. Embodiment 2

[0053] The present embodiment provides a periodontitis dynamic prognosis prediction system based on survival analysis, comprising: A clinical data acquisition and preprocessing module for receiving and structuring the storage of multi-modal clinical data at the patient level and tooth level, and performing data cleaning, continuous variable standardization and classification variable encoding operations; A data set division and prediction factor screening module connected to the clinical data acquisition and preprocessing module for randomly dividing the preprocessed multi-modal clinical data set into a training set and a test set according to the patient ID, and performing a single factor Cox proportional risk regression analysis on all variables in the training set to screen out a set of prediction factors; A model construction and verification module connected to the data set division and prediction factor screening module for receiving the set of prediction factors, constructing a time-dependent Cox periodontitis dynamic prognosis prediction model, and verifying the model performance; a SHAP explainability analysis module connected to the model construction and verification module, configured to call the trained time-dependent Cox periodontal disease dynamic prognosis prediction model, calculate a SHAP value importance ranking chart and a SHAP force diagram of an individual tooth at different time points; a clinical decision support module connected to the model construction and verification module and the SHAP explainability analysis module, configured to generate a risk stratification report and individualized diagnosis and treatment recommendations based on the risk probability predicted by the time-dependent Cox periodontal disease dynamic prognosis prediction model and the risk driving factors analyzed by the SHAP. Embodiment 3

[0054] The embodiment provides an electronic device, comprising at least one processor and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the periodontal disease dynamic prognosis prediction model construction method based on survival analysis in Embodiment 1. Embodiment 4

[0055] The embodiment provides a computer readable storage medium having computer instructions stored thereon, and the computer instructions are executed by a processor to implement the steps of the periodontal disease dynamic prognosis prediction model construction method based on survival analysis in Embodiment 1.

[0056] The above merely describes the preferred embodiments of the present application and is not used to limit the present application. For those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for constructing a dynamic prognostic prediction model for periodontitis based on survival analysis, characterized in that, include: We collected periodontitis patient data, including patient-level data and tooth-level data, performed Z-score standardization on continuous variables in the periodontitis patient data, and performed dummy variable encoding on categorical variables in the periodontitis patient data to construct a structured multimodal clinical dataset. The multimodal clinical dataset was randomly divided into training and test sets according to patient ID. Only on the training set, all patient-level data and tooth-level data were subjected to univariate Cox proportional hazards regression analysis. Variables that met the significance level P<0.05 were screened to obtain a set of predictors that were significantly associated with the risk of periodontitis progression. Using the time of periodontitis progression as the dependent variable and the set of predictors as covariates, a time-dependent Cox dynamic prognostic prediction model for periodontitis was constructed. The SHAP algorithm is introduced to perform multi-level and quantifiable analysis of the prediction results of the time-dependent Cox periodontitis dynamic prognosis prediction model, realizing a dynamic interpretation from the importance of global features to individualized risk drivers.

2. The method according to claim 1, characterized in that, The univariate Cox proportional hazards regression analysis includes: calculating the hazard ratio, 95% confidence interval, and significance p-value for each variable.

3. The method according to claim 1, characterized in that, The risk function of the time-dependent Cox periodontitis dynamic prognostic prediction model is expressed as: h(t|X)=h0(t)×exp(β1X1+β2X2+...+β p X p ), where h(t|X) is the hazard function at time t given the covariate X, h0(t) is the baseline hazard function at time t, and X1, X2, ..., X p These are the 1st, 2nd, ..., pth covariates, β1, β2, ..., β p These are the regression coefficients of the 1st, 2nd, ..., pth covariates, respectively; the time-dependent Cox dynamic prognostic prediction model for periodontitis is used to calculate the conditional probability of periodontitis progression in an individual's teeth at future time points.

4. The method according to claim 3, characterized in that, Also includes: The time-dependent ROC curve was used to evaluate the discriminative ability of the time-dependent Cox periodontitis dynamic prognostic prediction model at different time points, and the performance of the time-dependent Cox periodontitis dynamic prognostic prediction model was quantified by calculating the area under the time-dependent curve.

5. The method according to claim 1, characterized in that, The analysis includes: calculating the absolute mean of the SHAP values ​​of each predictor and generating a ranking chart of SHAP value importance; for the same tooth, drawing SHAP force maps at different time points to show the dynamic evolution of the contribution of each predictor over time.

6. The method according to claim 1, characterized in that, Based on the risk probabilities of individual teeth at multiple time points and the determined optimal cutoff value output by the time-dependent Cox periodontitis dynamic prognosis prediction model, teeth are divided into dynamically changing high-risk and low-risk groups. Based on the analysis results of the SHAP algorithm, the risk drivers of high-risk group teeth at different time points were identified.

7. The method according to claim 1, characterized in that, The patient-level data includes one or more of the following: age, gender, education level, body mass index, smoking history, history of systemic diseases, and periodontitis staging and grading data; the tooth-level data includes one or more of the following: probing depth, clinical attachment loss, probing bleeding, degree of root furcation lesions, mobility, tooth position, and alveolar bone loss staging.

8. A dynamic prognostic prediction system for periodontitis based on survival analysis for implementing the method as described in any one of claims 1 to 7, characterized in that, Includes the following modules: The clinical data acquisition and preprocessing module is used to receive and structure multimodal clinical data at the patient and dental levels, and to perform data cleaning, continuous variable standardization, and categorical variable coding operations. The dataset partitioning and predictor selection module is connected to the clinical data acquisition and preprocessing module. It is used to randomly partition the preprocessed multimodal clinical dataset into a training set and a test set according to the patient ID, and perform univariate Cox proportional hazards regression analysis on all variables on the training set to select a set of predictors. The model building and validation module is connected to the dataset partitioning and predictor selection module. It is used to receive the predictor set, build a time-dependent Cox periodontitis dynamic prognosis prediction model, and validate the model performance. The SHAP interpretability analysis module, connected to the model building and validation module, is used to call the trained time-dependent Cox periodontitis dynamic prognosis prediction model to calculate the SHAP value importance ranking chart and the SHAP force chart of individual teeth at different time points. The clinical decision support module connects the model building and validation module and the SHAP interpretability analysis module. It is used to generate risk stratification reports and individualized treatment recommendations based on the risk probabilities predicted by the time-dependent Cox periodontitis dynamic prognosis prediction model and the risk driving factors resolved by SHAP.

9. An electronic device, comprising: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, characterized in that the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instructions are executed by the processor, they implement the steps of the method as described in any one of claims 1 to 7.

Images