Liver cancer risk prediction model and construction method

Abstract

The present invention discloses a liver cancer risk prediction model and construction method, which relates to the technical field of liver cancer risk prediction, in order to solve the problem of being unable to accurately obtain risk data of patients. The diversified signal forms of the present invention can attract the attention of medical personnel in an all-round way. Different signal forms can be targeted at different environments and situations, improve the flexibility and applicability of early warning, and automatically generate intervention suggestions based on the early warning intensity and early warning signal, which greatly reduces the workload of medical personnel. By using parameter tuning methods such as grid search and random search, the parameter space can be systematically explored to find the optimal model parameter configuration, thereby improving the predictive performance of the model. By reselecting and transforming features, redundant features can be removed to make the model more stable and efficient. Statistical tests and multiple regression analysis are used to evaluate the relationship between variables and liver cancer occurrence, which helps to accurately identify variables that have a significant impact on liver cancer occurrence.

Description

Technical Field

[0001] The present invention relates to the technical field of liver cancer risk prediction, and specifically to a liver cancer risk prediction model and a construction method. Background Art

[0002] Liver cancer risk prediction refers to estimating the possibility of an individual developing liver cancer in the future by analyzing various individual information, such as lifestyle habits, family medical history, biomarker levels, imaging examination results, etc.

[0003] Chinese patent application document with publication number CN116543898A discloses a method for constructing a risk prediction model for thyroid cancer for a target group. The method mainly obtains first sample data of subjects in the target group and performs variable cleaning on the first sample data to obtain the first sample data. The influencing factors of the target group's family history of hypertension are put into the process of constructing the risk prediction model, which is conducive to accurately and precisely predicting the thyroid cancer risk of the target group; constructs a sample information database based on the first sample data; based on statistical analysis, determines characteristic variables for predicting the risk of thyroid cancer in the target group based on the sample information database; wherein the characteristic variables include family history factors; constructs a risk prediction model for thyroid cancer for the target group based on the characteristic variables. This step can improve the effectiveness and reliability of the risk prediction model. Although the above patent document solves the problem of risk prediction, the following problems still exist in actual operation:

[0004] 1. After collecting patient data, no targeted risk analysis is conducted, and no further risk warnings and intervention decisions are made based on the risk analysis results, resulting in inaccurate predictions of patient disease risks.

[0005] 2. Failure to make more accurate risk assessments based on the risk profile of the data, and failure to conduct targeted model evaluation and optimization of the assessed risks, resulted in an inability to accurately assess the patient's risk level.

[0006] 3. Failure to formulate effective risk warning and intervention plans based on the patient’s risk causes and risk levels results in the patient being unable to undergo further risk diagnosis based on their condition. Summary of the Invention

[0007] The purpose of the present invention is to provide a liver cancer risk prediction model and construction method. The diverse signal forms can attract the attention of medical personnel in all directions. Different signal forms can be targeted at different environments and situations to improve the flexibility and applicability of early warning. Intervention suggestions are automatically generated according to the warning intensity and warning signal, which greatly reduces the workload of medical personnel. Parameter tuning methods such as grid search and random search can be used to systematically explore the parameter space and find the optimal model parameter configuration, thereby improving the predictive performance of the model. Reselecting and transforming features can remove redundant features, making the model more stable and efficient. Statistical tests and multivariate regression analysis are used to evaluate the relationship between variables and liver cancer occurrence, which helps to accurately identify variables that have a significant impact on liver cancer occurrence and can solve problems in the existing technology.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] Liver cancer risk prediction model, including risk assessment model and risk warning model;

[0010] Risk assessment models for:

[0011] Collecting the patient's liver cancer-related data and using the collected liver cancer-related data to quantitatively assess the risk of liver cancer. The quantitative assessment includes risk factor analysis, risk assessment calculation, and risk level classification based on the patient's liver cancer-related data. After the quantitative assessment is completed, a risk assessment model is obtained.

[0012] Risk early warning models for:

[0013] The patients are assessed for liver cancer risk warning according to the risk assessment model. The liver cancer risk warning assessment includes risk dynamic assessment, warning signal triggering and intervention suggestion generation. After the liver cancer risk warning assessment is completed, a risk warning model is obtained.

[0014] A method for constructing a liver cancer risk prediction model comprises the following steps:

[0015] S1: Collecting the patient's liver cancer-related data from the database, performing data preprocessing after the liver cancer-related data is collected, and obtaining the target collected data after the data preprocessing is completed;

[0016] S2: Perform risk factor analysis, risk assessment calculation, and risk level classification on the target collected data. After the risk factor analysis, risk assessment calculation, and risk level classification are completed, a risk assessment model is obtained.

[0017] S3: Evaluate and optimize the risk assessment model, and then deploy the risk assessment model to the risk prediction port after the model evaluation and optimization are completed.

[0018] Preferably, step S1 in the risk assessment model construction method includes:

[0019] Collect patients' liver cancer-related data from databases, including the hospital's electronic medical record system, health examination report system, and epidemiological survey system database;

[0020] Liver cancer-related data include clinical data, biomarker data, lifestyle information, and environmental exposure data;

[0021] Perform data cleaning and data standardization on liver cancer related data;

[0022] Data cleaning and data standardization include handling missing values, data denoising, and integrating data from different sources into a unified format;

[0023] After data preprocessing is completed, the target collected data is obtained.

[0024] Preferably, step S1 in the risk assessment model construction method further includes:

[0025] Extract data sets corresponding to different data sources;

[0026] Retrieve the cleaning process data corresponding to the data cleaning of each data set, wherein the cleaning process data includes the number of missing values ​​and their corresponding weight values ​​and format conversion time;

[0027] The data quality coefficients corresponding to different data sources are obtained by using the number of missing values ​​in the data sets corresponding to different data sources, their corresponding weight values, and the format conversion time;

[0028] The data quality coefficient is obtained by the following formula:

[0029]

[0030] Among them, J represents the data quality coefficient; n represents the number of missing values ​​corresponding to the data source; w ni represents the data weight value corresponding to the i-th missing value; m represents the number of valid data corresponding to the data source; w mi Indicates the data weight value corresponding to the i-th valid data; T i Indicates the format conversion time corresponding to the i-th valid data; T c Indicates the preset conversion time reference value; z indicates the total number of collected data corresponding to the data source;

[0031] The data quality coefficients corresponding to the different data sources are used to limit the collection ratio of liver cancer related data collected by different data sources.

[0032] Preferably, the data quality coefficients corresponding to the different data sources are used to limit the proportion of liver cancer-related data collected by different data sources, including:

[0033] Compare the data quality coefficients corresponding to different data sources with the preset data quality coefficient thresholds;

[0034] A data source whose data quality coefficient is not lower than a preset data quality coefficient threshold is used as a target data source;

[0035] Extract the proportion of the current collected data of the target data source in the total collected data;

[0036] Obtaining an upper limit value of the collected data ratio corresponding to the target data source by using the data quality coefficient corresponding to the target data source in combination with the data ratio;

[0037] The upper limit of the collected data ratio is obtained by the following formula:

[0038]

[0039] Among them, B represents the upper limit of the proportion of collected data corresponding to the target data source; B c Indicates the proportion of the current collected data of the target data source to the total collected data; x indicates the number of data sources other than the current target data source; B xi represents the proportion of the collected data corresponding to the i-th data source other than the current target data source in the total collected data; J m Indicates the data quality coefficient corresponding to the target data source; J i Indicates the data quality coefficient corresponding to the i-th data source other than the current target data source;

[0040] The data ratio of the target data source corresponding to the collected data is controlled according to the upper limit of the collected data ratio as a standard.

[0041] Preferably, step S2 in the risk assessment model construction method includes:

[0042] First, the target data is collected for risk factor analysis, which includes exploratory data analysis, variable screening, and feature selection;

[0043] Exploratory analysis involves calculating the mean, median, standard deviation, minimum, and maximum values ​​of the target data, and visualizing the calculation results in the form of charts. The visualization then reveals the distribution and anomalies of the target data.

[0044] Variable screening involves performing univariate and multivariate analyses based on the abnormalities of the target collected data. Univariate analysis involves using statistical tests to evaluate the relationship between each variable and liver cancer occurrence. Multivariate analysis involves using multiple regression analysis to evaluate the combined effects of multiple variables on liver cancer occurrence.

[0045] Feature selection involves using the stepwise regression method to select feature data that have a significant impact on liver cancer occurrence from the results of univariate and multivariate analyses of the target collected data obtained in variable screening.

[0046] The data after feature selection in the target collected data are marked as risk factor analysis results.

[0047] Preferably, step S2 in the risk assessment model construction method further includes:

[0048] Perform risk assessment calculations based on the risk factor analysis results. Before performing risk assessment calculations, an assessment model is constructed based on the risk factor analysis results. The assessment model is a Cox proportional hazards model, a random forest model, or a gradient boosting tree model.

[0049] The obtained evaluation model is trained, and the trained evaluation model is used to evaluate and calculate the risk factor analysis results;

[0050] After the evaluation calculation is completed, a risk assessment model of the target collected data is obtained;

[0051] Confirm the assessment threshold in the risk assessment model and classify the risk level according to the confirmed threshold;

[0052] The risk level is divided into a threshold range based on the assessment threshold, and a risk level is assigned to each assessment threshold. The risk levels include low risk, medium risk and high risk;

[0053] After each assessment threshold is assigned to the corresponding risk level, a risk assessment model for the target collection data is obtained.

[0054] Preferably, step S3 in the risk assessment model construction method includes:

[0055] The risk assessment model was evaluated using cross-validation method for performance indicators, including ROC curve, AUC value, precision, recall rate, F1 score, Brier score and calibration curve;

[0056] Based on the performance indicator evaluation results, the risk assessment model is optimized using parameter tuning methods, including grid search and random search.

[0057] Re-select features for the risk assessment model after parameter tuning, and standardize, normalize, or logarithmically transform the selected features;

[0058] The risk assessment model after feature reselection is integrated using the ensemble learning method for model fusion;

[0059] After the model fusion is completed, the risk assessment model with model evaluation and optimization is obtained;

[0060] Save the risk assessment model after model evaluation and optimization in a persistent format, such as PMML, ONNX, pickle file, or custom binary format;

[0061] Migrate the model evaluation and optimized risk assessment model in a persistent format to the server, which is connected to the risk prediction port signal;

[0062] The server transmits the model evaluation and optimized risk assessment model in a persistent format to the risk prediction port.

[0063] Preferably, the method for constructing a risk warning model includes the following steps:

[0064] S4: Integrate the risk assessment results in the risk assessment model, and set up a dynamic assessment mechanism for the integrated risk assessment results. After the dynamic assessment mechanism is set up, the patient risk prediction data is obtained;

[0065] S5: Determine the warning intensity of the patient risk prediction data, generate a warning signal based on the warning intensity, generate automated intervention suggestions based on the generated warning signal, and construct a risk warning model based on the generated automated intervention suggestions;

[0066] In step S4, the risk assessment results in the risk assessment model are integrated, and a dynamic assessment mechanism is set up for the integrated risk assessment results, including:

[0067] Confirm the risk assessment results of the patients in the risk assessment model, integrate the data sets after confirmation, and obtain the risk data sets after the data sets are integrated;

[0068] A dynamic assessment mechanism is set up for the risk data set. The dynamic assessment mechanism is set up to determine the frequency of risk assessment for patients. After the frequency is confirmed, trigger conditions are set up. The trigger conditions are set up so that when the assessment threshold is used to assign a risk level, an early warning is automatically triggered according to the assigned risk level.

[0069] After the dynamic assessment mechanism is set up, patient risk prediction data can be obtained.

[0070] Preferably, step S5 of the risk warning model construction method includes:

[0071] The risk level assigned to the patient in the patient risk prediction data is used to determine the warning intensity. The higher the risk level, the higher the warning intensity.

[0072] Different warning signals are distinguished according to the warning intensity, among which the warning signals include visual signals, text descriptions and sound signals;

[0073] Generate automated intervention recommendations for patients based on warning intensity and warning signals;

[0074] Automated intervention recommendations include guidance on treatment options, healthcare facility contact information, and emergency instructions;

[0075] The generated automated intervention suggestions are converted into visual form, and the automated intervention suggestions after the visual conversion are transmitted to the mobile terminals of medical staff and patients for display.

[0076] Compared with the prior art, the present invention has the following beneficial effects:

[0077] 1. The liver cancer risk prediction model and construction method provided by this invention uses statistical tests and multiple regression analysis to evaluate the relationship between variables and liver cancer development. This method helps accurately identify variables that have a significant impact on liver cancer development, improving the accuracy of the analysis. The feature selection process uses a stepwise regression screening method to select feature data that have a significant impact on liver cancer development from the variable screening results. This helps reduce noise and redundant information, improving the model's predictive performance.

[0078] 2. The liver cancer risk prediction model and construction method provided by the present invention adopt parameter tuning methods such as grid search and random search, which can systematically explore the parameter space and find the optimal model parameter configuration, thereby improving the prediction performance of the model. By reselecting and transforming features, redundant features can be removed, feature quality can be improved, and the model can be made more stable and efficient.

[0079] 3. The liver cancer risk prediction model and construction method provided by the present invention set trigger conditions so that when the assessment threshold reaches a specific risk level, the early warning mechanism can be automatically triggered without human intervention, thereby improving the efficiency and accuracy of the early warning. Automated early warning helps to timely discover and deal with potential risks, providing strong support for clinical decision-making. The diverse signal forms can attract the attention of medical staff in all directions. Different signal forms can be targeted at different environments and situations, improving the flexibility and applicability of early warnings. Intervention recommendations are automatically generated according to the warning intensity and warning signals, which greatly reduces the workload of medical staff. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1Schematic diagram of the steps for constructing a risk assessment model in constructing a liver cancer risk prediction model of the present invention;

[0081] Figure 2 This is a schematic diagram of the steps for constructing a risk warning model in constructing a liver cancer risk prediction model of the present invention;

[0082] Figure 3 This is a schematic diagram of the process of constructing the liver cancer risk prediction model of the present invention. DETAILED DESCRIPTION

[0083] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0084] In order to solve the problem in existing technologies that patient data is collected without targeted risk analysis, and further risk warning and intervention decision-making are not made based on the risk analysis results, which leads to inaccurate prediction of patient disease risk, please refer to Figure 1-Figure 3 , this embodiment provides the following technical solutions:

[0085] Liver cancer risk prediction model, including risk assessment model and risk warning model;

[0086] Risk assessment models for:

[0087] Collecting the patient's liver cancer-related data and using the collected liver cancer-related data to quantitatively assess the risk of liver cancer. The quantitative assessment includes risk factor analysis, risk assessment calculation, and risk level classification based on the patient's liver cancer-related data. After the quantitative assessment is completed, a risk assessment model is obtained.

[0088] Risk early warning models for:

[0089] The patients are assessed for liver cancer risk warning according to the risk assessment model. The liver cancer risk warning assessment includes risk dynamic assessment, warning signal triggering and intervention suggestion generation. After the liver cancer risk warning assessment is completed, a risk warning model is obtained.

[0090] Specifically, the risk assessment model collects liver cancer-related data from patients and conducts personalized quantitative assessments. This means that each patient will receive a dedicated risk assessment based on their unique situation, improving the accuracy and pertinence of the assessment. Through risk assessment calculations and risk level classification, patients can receive a clear risk assessment result. The risk warning model can conduct dynamic risk assessments and monitor changes in a patient's liver cancer risk in real time. This helps to promptly identify rising risk trends and take timely intervention measures. When a patient's liver cancer risk reaches or exceeds a preset threshold, the risk warning model will promptly trigger a warning signal. The risk warning model not only provides risk warnings but also generates personalized intervention recommendations based on the patient's specific situation.

[0091] In order to solve the problem in the existing technology that after receiving the patient's data, a more accurate risk judgment is not made based on the risk situation of the data, and a targeted model evaluation and optimization is not performed on the judged risk, which leads to the inability to accurately obtain the patient's risk level, please refer to Figure 1-Figure 3 , this embodiment provides the following technical solutions:

[0092] A method for constructing a liver cancer risk prediction model, including a method for constructing a risk assessment model, comprises the following steps:

[0093] S1: Collecting the patient's liver cancer-related data from the database, performing data preprocessing after the liver cancer-related data is collected, and obtaining the target collected data after the data preprocessing is completed;

[0094] Among them, by collecting multi-dimensional information such as clinical data, biomarker data, lifestyle information, and environmental exposure data, we can have a more comprehensive understanding of the patient's liver cancer status and its related factors;

[0095] S2: Perform risk factor analysis, risk assessment calculation, and risk level classification on the target collected data. After the risk factor analysis, risk assessment calculation, and risk level classification are completed, a risk assessment model is obtained.

[0096] Among them, dividing the risk levels according to the threshold range of the assessment threshold helps to intuitively understand the risk levels of different assessment results;

[0097] S3: Evaluate and optimize the risk assessment model, and then deploy the risk assessment model to the risk prediction port after the evaluation and optimization are completed.

[0098] The entire process from model development to deployment is systematically planned to ensure the repeatability and verifiability of each step, which is conducive to subsequent model maintenance and updating.

[0099] The S1 step in the risk assessment model construction method includes:

[0100] Collect patients' liver cancer-related data from databases, including the hospital's electronic medical record system, health examination report system, and epidemiological survey system database;

[0101] Liver cancer-related data include clinical data, biomarker data, lifestyle information, and environmental exposure data;

[0102] Perform data cleaning and data standardization on liver cancer related data;

[0103] Data cleaning and data standardization include handling missing values, data denoising, and integrating data from different sources into a unified format;

[0104] After data preprocessing is completed, the target collected data is obtained.

[0105] Specifically, by collecting multi-dimensional information such as clinical data, biomarker data, lifestyle information, and environmental exposure data, we can gain a more comprehensive understanding of the patient's liver cancer status and related factors. The data cleaning process can identify and handle problems such as missing values ​​and outliers, thereby improving the accuracy and reliability of the data. Data standardization helps eliminate format differences between data from different sources and ensures data consistency and comparability. Cleaned and standardized data can be more conveniently used for subsequent data analysis, data mining, and machine learning tasks. Unified data formats and representations help simplify the analysis process and improve the accuracy and efficiency of the analysis. Through the analysis and mining of these data, new biomarkers, therapeutic targets, etc. can be discovered, providing a scientific basis for the precise and personalized treatment of liver cancer.

[0106] Specifically, the S1 step in the risk assessment model construction method also includes:

[0107] Extract data sets corresponding to different data sources;

[0108] Retrieve the cleaning process data corresponding to the data cleaning of each data set, wherein the cleaning process data includes the number of missing values ​​and their corresponding weight values ​​and format conversion time;

[0109] The data quality coefficients corresponding to different data sources are obtained by using the number of missing values ​​in the data sets corresponding to different data sources, their corresponding weight values, and the format conversion time;

[0110] The data quality coefficient is obtained by the following formula:

[0111]

[0112] Among them, J represents the data quality coefficient; n represents the number of missing values ​​corresponding to the data source; wni represents the data weight value corresponding to the i-th missing value; m represents the number of valid data corresponding to the data source; w mi Indicates the data weight value corresponding to the i-th valid data; T i Indicates the format conversion time corresponding to the i-th valid data; T c Indicates the preset conversion time reference value; z indicates the total number of collected data corresponding to the data source;

[0113] The data quality coefficients corresponding to the different data sources are used to limit the collection ratio of liver cancer related data collected by different data sources.

[0114] The technical effect of the above-mentioned technical solution is that, by comprehensively considering the number of missing values ​​and their weights, the number of valid data and their weights, and the format conversion time, it can quantitatively assess the data quality of different data sources. This quantitative assessment provides an important reference for subsequent data processing and analysis. The calculated data quality coefficient can be used to limit the proportion of liver cancer-related data collected from different data sources. This means that during the data collection process, higher-quality data sources can be prioritized, reducing or avoiding the collection of data from lower-quality sources, thereby improving overall data quality and reliability. By quantitatively assessing data quality, high-quality data can be screened before data processing and analysis, reducing the processing and analysis time for low-quality data and improving data processing efficiency. This technical solution not only considers traditional data quality indicators such as missing values ​​and format conversion time, but also introduces data weights, making the data quality assessment more comprehensive and accurate. This helps enhance data usability and provides more reliable support for subsequent data analysis and decision-making. By assessing format conversion time, this technical solution encourages data source format standardization, making data from different data sources easier to integrate and compare. This helps promote data standardization and improve data interoperability and comparability. The formula and data quality coefficient calculation method in this technical solution are flexible and scalable. As data sources increase and data processing technology continues to develop, the parameters and indicators in the formula can be further adjusted and optimized to adapt to new data processing needs.

[0115] In summary, this technical solution, by quantitatively assessing the data quality of different data sources and limiting the proportion of data collected accordingly, helps improve data processing efficiency, enhance data availability and standardization, and also offers a degree of flexibility and scalability. These technical effects are of great significance for improving the quality and reliability of liver cancer-related data.

[0116] Specifically, the data quality coefficients corresponding to the different data sources are used to limit the proportion of liver cancer-related data collected by different data sources, including:

[0117] Compare the data quality coefficients corresponding to different data sources with the preset data quality coefficient thresholds;

[0118] A data source whose data quality coefficient is not lower than a preset data quality coefficient threshold is used as a target data source;

[0119] Extract the proportion of the current collected data of the target data source in the total collected data;

[0120] Obtaining an upper limit value of the collected data ratio corresponding to the target data source by using the data quality coefficient corresponding to the target data source in combination with the data ratio;

[0121] The upper limit of the collected data ratio is obtained by the following formula:

[0122]

[0123] Among them, B represents the upper limit of the proportion of collected data corresponding to the target data source; B c Indicates the proportion of the current collected data of the target data source to the total collected data; x indicates the number of data sources other than the current target data source; B xi represents the proportion of the collected data corresponding to the i-th data source other than the current target data source in the total collected data; J m Indicates the data quality coefficient corresponding to the target data source; J i Indicates the data quality coefficient corresponding to the i-th data source other than the current target data source;

[0124] The data ratio of the target data source corresponding to the collected data is controlled according to the upper limit of the collected data ratio as a standard.

[0125] The technical effect of the above-mentioned technical solution is that by comparing the data quality coefficients of different data sources against a preset threshold, the solution can select higher-quality data sources as target data sources. This data quality-based data collection strategy helps ensure the reliability and accuracy of collected data. This technical solution not only considers the current data collection ratio of the target data source but also combines the data collection ratios and data quality coefficients of other data sources to calculate the upper limit of the data collection ratio corresponding to the target data source through a formula. This dynamic adjustment mechanism can adjust the data collection ratio in real time based on the data quality of different data sources, thereby optimizing the data collection strategy. By limiting the collection ratio of low-quality data sources, the solution can reduce the collection of invalid or low-quality data, thereby improving data collection efficiency. Furthermore, due to the high data quality of the target data source, subsequent data processing and analysis will be more efficient. This technical solution, by limiting the data collection ratio through the data quality coefficient, effectively incentivizes data sources. To obtain more data collection opportunities, data sources will strive to improve data quality to meet the preset data quality coefficient threshold. This technical solution provides a scientific data management method that quantitatively assesses data quality and adjusts data collection ratios accordingly, making data management more standardized and orderly. This helps improve the overall level of data management and provides more reliable support for subsequent data analysis and decision-making. The formulas and data quality coefficient calculation methods in this technical solution are adaptable and can be adjusted and optimized according to actual conditions. As data sources increase and data processing technology continues to develop, this technical solution can continue to adapt to new data processing needs.

[0126] In summary, this technical solution provides an effective method for collecting and managing liver cancer-related data by implementing a data collection strategy based on data quality, dynamically adjusting the data collection ratio, improving data collection efficiency, promoting data quality improvement, and enhancing the scientific nature and adaptability of data management. These technical benefits are of great significance for improving data quality, optimizing data collection strategies, and enhancing data management capabilities.

[0127] Preferably, step S2 in the risk assessment model construction method includes:

[0128] First, the target data is collected for risk factor analysis, which includes exploratory data analysis, variable screening, and feature selection;

[0129] Exploratory analysis involves calculating the mean, median, standard deviation, minimum, and maximum values ​​of the target data, and visualizing the calculation results in the form of charts. The visualization then reveals the distribution and anomalies of the target data.

[0130] Variable screening involves performing univariate and multivariate analyses based on the abnormalities of the target collected data. Univariate analysis involves using statistical tests to evaluate the relationship between each variable and liver cancer occurrence. Multivariate analysis involves using multiple regression analysis to evaluate the combined effects of multiple variables on liver cancer occurrence.

[0131] Feature selection involves using the stepwise regression method to select feature data that have a significant impact on liver cancer occurrence from the results of univariate and multivariate analyses of the target collected data obtained in variable screening.

[0132] The data after feature selection in the target collected data are marked as risk factor analysis results.

[0133] Perform risk assessment calculations based on the risk factor analysis results. Before performing risk assessment calculations, an assessment model is constructed based on the risk factor analysis results. The assessment model is a Cox proportional hazards model, a random forest model, or a gradient boosting tree model.

[0134] The obtained evaluation model is trained, and the trained evaluation model is used to evaluate and calculate the risk factor analysis results;

[0135] After the evaluation calculation is completed, a risk assessment model of the target collected data is obtained;

[0136] Confirm the assessment threshold in the risk assessment model and classify the risk level according to the confirmed threshold;

[0137] The risk level is divided into a threshold range based on the assessment threshold, and a risk level is assigned to each assessment threshold. The risk levels include low risk, medium risk and high risk;

[0138] After each assessment threshold is assigned to the corresponding risk level, a risk assessment model for the target collection data is obtained.

[0139] Specifically, starting with exploratory data analysis, the study proceeded with variable screening and feature selection, ensuring a comprehensive and systematic analysis. This progressively more in-depth analysis approach helped identify key risk factors and provided a solid foundation for subsequent risk assessment. By calculating the mean, median, standard deviation, minimum, and maximum values ​​and visualizing them in graphical form, the data distribution and anomalies can be intuitively understood. This helps identify potential issues and trends in the data, providing clues for subsequent analysis. The variable screening process included univariate and multivariate analyses, using statistical tests and multiple regression analysis to assess the relationship between variables and liver cancer development. This approach helped accurately identify variables with significant impacts on liver cancer development, improving the accuracy of the analysis. The feature selection process utilized a stepwise regression method to select features with significant impacts on liver cancer development from the variable screening results. This helped reduce noise and redundant information, improving the model's predictive performance. The trained assessment model was then used to perform risk assessment calculations, resulting in an accurate risk assessment model. This facilitated precise risk assessment of the target data, providing a basis for subsequent decision-making. Risk levels were categorized into low, medium, and high risk levels based on the assessment threshold range. This clear division helps to intuitively understand the risk levels of different assessment results and facilitates the adoption of appropriate risk management measures.

[0140] The S3 step in the risk assessment model construction method includes:

[0141] The risk assessment model was evaluated using cross-validation method for performance indicators, including ROC curve, AUC value, precision, recall rate, F1 score, Brier score and calibration curve;

[0142] Based on the performance indicator evaluation results, the risk assessment model is optimized using parameter tuning methods, including grid search and random search.

[0143] Re-select features for the risk assessment model after parameter tuning, and standardize, normalize, or logarithmically transform the selected features;

[0144] The risk assessment model after feature reselection is integrated using the ensemble learning method for model fusion;

[0145] After the model fusion is completed, the risk assessment model with model evaluation and optimization is obtained;

[0146] Save the risk assessment model after model evaluation and optimization in a persistent format, such as PMML, ONNX, pickle file, or custom binary format;

[0147] Migrate the model evaluation and optimized risk assessment model in a persistent format to the server, which is connected to the risk prediction port signal;

[0148] The server transmits the model evaluation and optimized risk assessment model in a persistent format to the risk prediction port.

[0149] Specifically, multiple performance indicators such as ROC curve, AUC value, precision, recall rate, F1 score, Brier score and calibration curve are used to comprehensively evaluate the performance of the model to ensure that the model performs well in different dimensions. Parameter tuning methods such as grid search and random search can be used to systematically explore the parameter space and find the optimal model parameter configuration, thereby improving the predictive performance of the model. Reselecting and transforming features (such as standardization, normalization, logarithmic transformation) can remove redundant features, improve feature quality, and make the model more stable and efficient. Using ensemble learning methods for model fusion can further improve the predictive accuracy and robustness of the model, because ensemble learning can combine the advantages of multiple models and reduce the overfitting of a single model. Risk assessment models can be saved as PMML, ONNX, pickle files or custom binary formats to facilitate cross-platform and cross-language migration and use of the model, thereby improving the flexibility and scalability of the model. The model in a persistent format can be migrated to the server and connected to the risk prediction port signal, so that the model can be easily integrated into the existing business system to achieve real-time or batch risk prediction. The entire process from model development to deployment has been systematically planned to ensure the repeatability and verifiability of each step, which is conducive to subsequent model maintenance and updates. By continuously optimizing and deploying risk assessment models, enterprises can more accurately identify and manage risks, improve the scientific nature and efficiency of business decisions, and thus bring greater business value.

[0150] In order to solve the problem in existing technologies that there is no effective risk warning and intervention plan based on the patient's risk cause and risk level, which results in the patient being unable to conduct further risk diagnosis based on their condition, please refer to Figure 1-Figure 3 , this embodiment provides the following technical solutions:

[0151] The method for constructing a risk warning model includes the following steps:

[0152] S4: Integrate the risk assessment results in the risk assessment model, and set up a dynamic assessment mechanism for the integrated risk assessment results. After the dynamic assessment mechanism is set up, the patient risk prediction data is obtained;

[0153] Among them, automated early warning helps to timely detect and deal with potential risks, providing strong support for clinical decision-making;

[0154] S5: Determine the warning intensity of the patient risk prediction data, generate a warning signal based on the warning intensity, generate automated intervention suggestions based on the generated warning signal, and construct a risk warning model based on the generated automated intervention suggestions;

[0155] Among them, by transmitting the intervention suggestions after visual conversion to the mobile terminals of medical staff and patients for display, the instant transmission and sharing of information is achieved;

[0156] In step S4, the risk assessment results in the risk assessment model are integrated, and a dynamic assessment mechanism is set up for the integrated risk assessment results, including:

[0157] Confirm the risk assessment results of the patients in the risk assessment model, integrate the data sets after confirmation, and obtain the risk data sets after the data sets are integrated;

[0158] A dynamic assessment mechanism is set up for the risk data set. The dynamic assessment mechanism is set up to determine the frequency of risk assessment for patients. After the frequency is confirmed, trigger conditions are set up. The trigger conditions are set up so that when the assessment threshold is used to assign a risk level, an early warning is automatically triggered according to the assigned risk level.

[0159] After the dynamic assessment mechanism is set up, patient risk prediction data can be obtained.

[0160] Specifically, by integrating the results of the risk assessment model, a more comprehensive risk dataset can be obtained, thereby more accurately reflecting the patient's risk status. The dynamic assessment mechanism enables timely adjustments to the risk assessment as the patient's condition changes, improving the timeliness of the assessment. The trigger conditions automatically trigger an early warning mechanism when the assessment threshold reaches a specific risk level, eliminating the need for manual intervention and improving the efficiency and accuracy of the early warning. Automated early warnings help promptly identify and address potential risks, providing strong support for clinical decision-making. Based on risk assessment results and risk levels, medical institutions can more effectively allocate resources, prioritize management and intervention for high-risk patients, and reduce potential losses. The dynamic assessment mechanism enables medical institutions to continuously monitor risk changes and promptly adjust risk management strategies to ensure patient safety. Based on the integrated risk dataset and dynamic assessment results, medical staff can make more scientific and reasonable medical decisions, improving treatment outcomes and patient satisfaction. Risk prediction data provides important reference for clinical pathway planning and treatment plan selection. By implementing this solution, medical institutions can establish a comprehensive risk assessment and management system, improving overall management capabilities.

[0161] The S5 step of the risk warning model construction method includes:

[0162] The risk level assigned to the patient in the patient risk prediction data is used to determine the warning intensity. The higher the risk level, the higher the warning intensity.

[0163] Different warning signals are distinguished according to the warning intensity, among which the warning signals include visual signals, text descriptions and sound signals;

[0164] Generate automated intervention recommendations for patients based on warning intensity and warning signals;

[0165] Automated intervention recommendations include guidance on treatment options, healthcare facility contact information, and emergency instructions;

[0166] The generated automated intervention suggestions are converted into visual form, and the automated intervention suggestions after the visual conversion are transmitted to the mobile terminals of medical staff and patients for display.

[0167] Specifically, by directly linking a patient's assigned risk level with the severity of the alert, the system accurately reflects the patient's risk profile. Higher risk levels correspond to higher alert severity, helping medical staff quickly identify high-risk patients and implement timely and effective interventions. Alert signals include visual, textual, and audible signals, enabling comprehensive attention from medical staff. Different signal formats can be tailored to specific environments and situations, enhancing the flexibility and adaptability of the alert. Automatically generating intervention recommendations based on the severity and severity of the alert significantly reduces the workload for medical staff. These recommendations include treatment plan guidelines, medical institution contact information, and emergency instructions, providing comprehensive reference and support for medical staff. The generated automated intervention recommendations are visualized for easier understanding. These visualized recommendations are then transmitted to medical staff and patients' mobile devices for display, enabling instant information sharing. This approach improves information accessibility and usability, helping medical staff and patients quickly access and respond to critical information. Accurate risk alerts and timely intervention recommendations help reduce healthcare risks and improve patient safety and satisfaction.

[0168] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0169] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. A liver cancer risk prediction model, characterized in that: Including risk assessment model and risk early warning model; Risk assessment models for: Collecting the patient's liver cancer-related data and using the collected liver cancer-related data to quantitatively assess the risk of liver cancer. The quantitative assessment includes risk factor analysis, risk assessment calculation, and risk level classification based on the patient's liver cancer-related data. After the quantitative assessment is completed, a risk assessment model is obtained. Risk early warning models for: Conduct a liver cancer risk warning assessment on patients based on the risk assessment model. This includes dynamic risk assessment, warning signal triggering, and generation of intervention recommendations. Upon completion of the liver cancer risk warning assessment, a risk warning model is generated. The liver cancer risk prediction model also includes a method for constructing the liver cancer risk prediction model: In the method for constructing a liver cancer risk prediction model, the method for constructing a risk assessment model includes the following steps: S1: Collecting the patient's liver cancer-related data from the database, performing data preprocessing after the liver cancer-related data is collected, and obtaining the target collected data after the data preprocessing is completed; S2: Perform risk factor analysis, risk assessment calculation, and risk level classification on the target collected data. After the risk factor analysis, risk assessment calculation, and risk level classification are completed, a risk assessment model is obtained. S3: Evaluate and optimize the risk assessment model, and then deploy the risk assessment model to the risk prediction port after the evaluation and optimization are completed. The S1 step in the risk assessment model construction method includes: Collect patients' liver cancer-related data from databases, including the hospital's electronic medical record system, health examination report system, and epidemiological survey system database; Liver cancer-related data include clinical data, biomarker data, lifestyle information, and environmental exposure data; Perform data cleaning and data standardization on liver cancer related data; Data cleaning and data standardization include handling missing values, data denoising, and integrating data from different sources into a unified format; After data preprocessing is completed, the target collected data is obtained; Extract data sets corresponding to different data sources; Retrieve the cleaning process data corresponding to the data cleaning of each data set, wherein the cleaning process data includes the number of missing values ​​and their corresponding weight values ​​and format conversion time; The data quality coefficients corresponding to different data sources are obtained by using the number of missing values ​​in the data sets corresponding to different data sources, their corresponding weight values, and the format conversion time; The data quality coefficient is obtained by the following formula: Where J represents the data quality coefficient; n represents the number of missing values ​​corresponding to the data source; represents the data weight value corresponding to the i-th missing value; m represents the number of valid data corresponding to the data source; Indicates the data weight value corresponding to the i-th valid data; Indicates the format conversion duration corresponding to the i-th valid data; Indicates the preset conversion time reference value; z indicates the total number of collected data corresponding to the data source; Using the data quality coefficients corresponding to the different data sources to limit the proportion of liver cancer-related data collected by different data sources; The data quality coefficients corresponding to the different data sources are used to limit the proportion of liver cancer-related data collected by different data sources, including: Compare the data quality coefficients corresponding to different data sources with the preset data quality coefficient thresholds; A data source whose data quality coefficient is not lower than a preset data quality coefficient threshold is used as a target data source; Extract the proportion of the current collected data of the target data source in the total collected data; Obtaining an upper limit value of the collected data ratio corresponding to the target data source by using the data quality coefficient corresponding to the target data source in combination with the data ratio; The upper limit of the collected data ratio is obtained by the following formula: Among them, B represents the upper limit of the proportion of collected data corresponding to the target data source; Indicates the proportion of the target data source's current collected data to the total collected data; x indicates the number of data sources other than the current target data source; Indicates the proportion of collected data corresponding to the i-th data source other than the current target data source in the total collected data; Indicates the data quality coefficient corresponding to the target data source; Indicates the data quality coefficient corresponding to the i-th data source other than the current target data source; According to the upper limit of the collected data ratio, the data ratio of the target data source corresponding to the collected data is controlled; The method for constructing a risk warning model includes the following steps: S4: Integrate the risk assessment results in the risk assessment model, and set up a dynamic assessment mechanism for the integrated risk assessment results. After the dynamic assessment mechanism is set up, the patient risk prediction data is obtained; S5: Determine the warning intensity of the patient risk prediction data, generate a warning signal based on the warning intensity, generate automated intervention suggestions based on the generated warning signal, and construct a risk warning model based on the generated automated intervention suggestions; In step S4, the risk assessment results in the risk assessment model are integrated, and a dynamic assessment mechanism is set up for the integrated risk assessment results, including: Confirm the risk assessment results of the patients in the risk assessment model, integrate the data sets after confirmation, and obtain the risk data sets after the data sets are integrated; A dynamic assessment mechanism is set up for the risk data set. The dynamic assessment mechanism is set up to determine the frequency of risk assessment for patients. After the frequency is confirmed, trigger conditions are set up. The trigger conditions are set up so that when the assessment threshold is used to assign a risk level, an early warning is automatically triggered according to the assigned risk level. After the dynamic assessment mechanism is set up, patient risk prediction data can be obtained.

2. The method for constructing a liver cancer risk prediction model according to claim 1, wherein: The S2 step in the risk assessment model construction method includes: First, the target data is collected for risk factor analysis, which includes exploratory data analysis, variable screening, and feature selection; Exploratory analysis involves calculating the mean, median, standard deviation, minimum, and maximum values ​​of the target data, and visualizing the calculation results in the form of charts. The visualization then reveals the distribution and anomalies of the target data. Variable screening involves performing univariate and multivariate analyses based on the abnormalities of the target collected data. Univariate analysis involves using statistical tests to evaluate the relationship between each variable and liver cancer occurrence. Multivariate analysis involves using multiple regression analysis to evaluate the combined effects of multiple variables on liver cancer occurrence. Feature selection involves using the stepwise regression method to select feature data that have a significant impact on liver cancer occurrence from the results of univariate and multivariate analyses of the target collected data obtained in variable screening. The data after feature selection in the target collected data are marked as risk factor analysis results.

3. The method for constructing a liver cancer risk prediction model according to claim 2, wherein: The S2 step in the risk assessment model construction method also includes: Perform risk assessment calculations based on the risk factor analysis results. Before performing risk assessment calculations, an assessment model is constructed based on the risk factor analysis results. The assessment model is a Cox proportional hazards model, a random forest model, or a gradient boosting tree model. The obtained evaluation model is trained, and the trained evaluation model is used to evaluate and calculate the risk factor analysis results; After the evaluation calculation is completed, a risk assessment model of the target collected data is obtained; Confirm the assessment threshold in the risk assessment model and classify the risk level according to the confirmed threshold; The risk level is divided into a threshold range based on the assessment threshold, and a risk level is assigned to each assessment threshold. The risk levels include low risk, medium risk and high risk; After each assessment threshold is assigned to the corresponding risk level, a risk assessment model for the target collection data is obtained.

4. The method for constructing a liver cancer risk prediction model according to claim 3, wherein: The S3 step in the risk assessment model construction method includes: The risk assessment model was evaluated using cross-validation method for performance indicators, including ROC curve, AUC value, precision, recall rate, F1 score, Brier score and calibration curve; Based on the performance indicator evaluation results, the risk assessment model is optimized using parameter tuning methods, including grid search and random search. Re-select features for the risk assessment model after parameter tuning, and standardize, normalize, or logarithmically transform the selected features; The risk assessment model after feature reselection is integrated using the ensemble learning method for model fusion; After the model fusion is completed, the risk assessment model with model evaluation and optimization is obtained; Save the risk assessment model after model evaluation and optimization in a persistent format, such as PMML, ONNX, pickle file, or custom binary format; Migrate the model evaluation and optimized risk assessment model in a persistent format to the server, which is connected to the risk prediction port signal; The server transmits the model evaluation and optimized risk assessment model in a persistent format to the risk prediction port.

5. The method for constructing a liver cancer risk prediction model according to claim 4, wherein: The S5 step of the risk warning model construction method includes: The risk level assigned to the patient in the patient risk prediction data is used to determine the warning intensity. The higher the risk level, the higher the warning intensity. Different warning signals are distinguished according to the warning intensity, among which the warning signals include visual signals, text descriptions and sound signals; Generate automated intervention recommendations for patients based on warning intensity and warning signals; Automated intervention recommendations include guidance on treatment options, healthcare facility contact information, and emergency instructions; The generated automated intervention suggestions are converted into visual form, and the automated intervention suggestions after the visual conversion are transmitted to the mobile terminals of medical staff and patients for display.

Images