Fatty liver disease risk prediction method and device

By constructing a risk prediction model for fatty liver disease based on longitudinal health examination data, the problem of early screening of large-scale populations in the prior art is solved, and high accuracy and non-invasive risk prediction are achieved.

CN120108729AInactive Publication Date: 2025-06-06FIRST AFFILIATED HOSPITAL OF KUNMING MEDICAL UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510257275.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-03-05
Publication Date
2025-06-06
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art is difficult to achieve non-invasive and accurate risk prediction of fatty liver disease in large populations, especially in early screening.

Method used

By collecting and analyzing individual longitudinal health examination data, preprocessing the data, screening characteristic variables, constructing a risk prediction model for fatty liver disease, and using dynamic feature fusion algorithms and neural network functions to make predictions.

Benefits of technology

It improves the accuracy and reliability of the risk prediction of fatty liver disease, simplifies the complexity of the model, improves the diagnostic efficiency and accuracy, and is suitable for early screening of large-scale populations.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120108729A_ABST
    Figure CN120108729A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of fatty liver risk prediction, and discloses a fatty liver disease risk prediction method which comprises the following steps: step 1, collecting health physical examination data which comprises basic information of an individual, blood biochemical index data, body measurement index data and time information of multiple physical examinations, forming a non-independent longitudinal data set; 2, performing data preprocessing, including data cleaning and data standardization, on the collected non-independent longitudinal data set, and converting different types of data into formats suitable for analysis; by collecting and analyzing the longitudinal health examination data of the individual, preprocessing the data, screening the characteristic variables and constructing the fatty liver disease risk prediction model, the accuracy and reliability of fatty liver disease risk prediction are improved, the model complexity is simplified, and the diagnosis efficiency and accuracy are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of fatty liver risk prediction, and in particular to a method and device for predicting fatty liver disease risk. Background Art

[0002] With the change of modern lifestyle, the incidence of fatty liver has shown a significant upward trend worldwide. Fatty liver not only affects the normal function of the liver, but may also further develop into more serious liver diseases such as liver fibrosis, cirrhosis and even liver cancer, posing a serious threat to the patient's health.

[0003] At present, the clinical diagnosis of fatty liver mainly relies on imaging examinations such as liver ultrasound, CT, and liver biopsy. However, these methods have many limitations. Liver ultrasound and CT examinations have limited sensitivity for early fatty liver, and can often only be detected after liver fat accumulation reaches a certain level; although liver biopsy has high accuracy, it is invasive, may cause complications, and has low patient acceptance, making it difficult to carry out early screening in a large population. With the popularization of health examinations, a large amount of physical examination data has been accumulated, but these data cannot be effectively modeled and analyzed using conventional statistical analysis methods because they do not meet the independence assumption, resulting in low utilization efficiency of health examination data. Conventional prediction methods lack sufficient mining of dynamic change information in longitudinal data, making it difficult to accurately predict the risk of fatty liver.

[0004] In summary, how to make full use of physical examination data to establish an accurate, non-invasive and applicable method for predicting the risk of fatty liver disease in a large population has become a technical problem that needs to be urgently solved in the field of liver disease prevention and control. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a method and device for predicting the risk of fatty liver disease. By collecting and analyzing individual longitudinal health examination data, pre-processing the data and screening characteristic variables, a fatty liver disease risk prediction model is constructed, which improves the accuracy and reliability of fatty liver disease risk prediction, simplifies the complexity of the model, improves the diagnostic efficiency and accuracy, and solves the above-mentioned technical problems.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for predicting the risk of fatty liver disease, comprising the following steps:

[0007] Step 1: Collect health examination data, which includes basic information of individuals, blood biochemical index data, body measurement index data, and time information of multiple physical examinations to form a non-independent longitudinal data set;

[0008] Step 2: Data preprocessing of the collected non-independent longitudinal data sets, including data cleaning and data standardization, converting different types of data into a format suitable for analysis;

[0009] Step 3: Screening out characteristic variables related to the risk of fatty liver disease from the pre-processed physical examination data, the characteristic variables including the change trend of body mass index, the dynamic change of alanine aminotransferase, the fluctuation of triglyceride index, the change of blood pressure and blood sugar index, the change of high-density lipoprotein cholesterol level and the change trend of low-density lipoprotein cholesterol level;

[0010] Step 4: Construct a fatty liver risk prediction model based on the screened characteristic variables and in combination with a dynamic characteristic fusion algorithm. The expression of the fatty liver risk prediction model is as follows:

[0011]

[0012] Among them, P represents the risk probability of fatty liver disease, B represents the body mass index at time t, and ΔL t Indicates the change in alanine aminotransferase at time t, Tg t represents the triglyceride index at time t, D t represents the blood pressure and blood sugar index at time t, Es t represents the change in high-density lipoprotein cholesterol level at time t, Us t represents the trend of low-density lipoprotein cholesterol level at time t, f is the comprehensive feature fusion and conversion function, and w 1 、w 2 、w 3 、w 4 、w 5 and w 6 They represent the weights of body mass index, change in alanine aminotransferase, triglyceride index, blood pressure, blood sugar index, high-density lipoprotein cholesterol level and low-density lipoprotein cholesterol level respectively;

[0013] Step 5: Using the constructed fatty liver risk prediction model, input the health examination data of the individual to be predicted on the day, and output the fatty liver disease risk prediction result of the individual, wherein the prediction result is the probability value of fatty liver disease.

[0014] Preferably, the collection cycle of the health examination data in step one is once a quarter, and the data preprocessing in step two also includes processing of missing values ​​and detection and correction of abnormal values.

[0015] Preferably, when screening characteristic variables in step three, the correlation coefficient between each variable and the fatty liver disease label is calculated, and the variable whose absolute value of correlation is greater than the set threshold is selected. The correlation coefficient calculation formula is:

[0016]

[0017] Among them, x i and i are the first i observations of variable x and disease label y, respectively, x′ and y′ are the means of variable x and disease label y, respectively, n is the number of observed samples, and the threshold is set to c(0 <c<1)。

[0018] Preferably, the comprehensive feature fusion and conversion function is a neural network function, and the parameters of the function are determined by training the neural network. The various weights in step 4 are obtained by training historical data using a random forest algorithm.

[0019] Preferably, the output formula of the neural network function is:

[0020]

[0021] Among them, sign is the sign function, δ i is the Lagrange multiplier, y i is the category label of the training sample, K(a i ,a) is the kernel function, d is the bias term, and i is the number of training samples.

[0022] The present invention also provides a device for predicting the risk of fatty liver disease, comprising:

[0023] The data collection module is used to collect basic information of individuals, blood biochemical index data, body measurement index data and time information of multiple physical examinations on a quarterly basis to form a non-independent longitudinal data set;

[0024] Data preprocessing module, which is used to clean the collected data, handle missing values, detect and correct outliers, and convert different types of data into a format suitable for analysis;

[0025] A feature screening module is used to screen out feature variables related to the risk of fatty liver disease from the preprocessed physical examination data, and calculate the correlation coefficient between each variable and the fatty liver disease label;

[0026] A model building module is used to construct a fatty liver risk prediction model by combining the screened feature variables with a dynamic feature fusion algorithm;

[0027] The prediction module uses the established fatty liver risk prediction model and inputs the health examination data of the individual to be predicted on the same day to calculate the fatty liver risk probability;

[0028] The output module is used to output the fatty liver risk probability predicted by the prediction module.

[0029] Preferably, the data collection module further includes a data storage unit for storing the collected health examination data, and the data storage unit adopts a distributed storage method.

[0030] Preferably, the neural network function in the model building module adopts a multi-layer perceptron structure, including an input layer, a hidden layer and an output layer.

[0031] Compared with the prior art, the present invention provides a bronchoscopic lavage sampling anti-pollution device, which has the following beneficial effects:

[0032] The present invention collects and analyzes individual longitudinal health examination data, pre-processes the data and then screens characteristic variables to construct a fatty liver disease risk prediction model, thereby improving the accuracy and reliability of fatty liver disease risk prediction, simplifying the model complexity, and improving diagnostic efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0034] Figure 2 It is a logical schematic diagram of the device of the present invention; DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0036] See also Figure 1-Figure 2 The present invention provides a method for predicting the risk of fatty liver disease, comprising the following steps:

[0037] Step 101: Collect health examination data, which includes basic information of individuals, blood biochemical index data, body measurement index data, and time information of multiple physical examinations to form a non-independent longitudinal data set;

[0038] In this application, basic information of individuals (such as age, gender, etc.), blood biochemical index data (such as alanine aminotransferase, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, etc.), body measurement index data (such as weight, height, and then calculate body mass index) and time information of multiple physical examinations are collected once a quarter to form a non-independent longitudinal data set. For example, for a 40-year-old male, data are collected for four consecutive quarters, with weights of 70kg, 72kg, 73kg, and 75kg, corresponding to a height of 1.75m, and blood biochemical indicators of alanine aminotransferase of 30U / L, 32U / L, 35U / L, and 40U / L, respectively. Collecting these data to form a non-independent longitudinal data set has many advantages: first, it can capture the dynamic changes of individual indicators, such as body mass index, blood biochemical indicators, etc., which fluctuate over time and can accurately reflect the evolution of individual health status. For example, the quarterly changes in body mass index can reflect the trend of body fat accumulation, providing a dynamic perspective for analyzing the risk of fatty liver disease; second, compared with a single physical examination data, its rich information helps to accurately screen characteristic variables that are closely related to fatty liver disease. For example, the long-term dynamic changes of alanine aminotransferase can better reveal the relationship between liver function status and disease than a single value, thereby improving the accuracy and reliability of the risk prediction model.

[0039] Step 201: preprocessing the collected non-independent longitudinal data sets, including data cleaning and data standardization, to convert different types of data into a format suitable for analysis;

[0040] The collected non-independent longitudinal data sets are removed from obviously erroneous or unreasonable data, such as negative height, abnormal maximum values ​​of alanine aminotransferase far beyond the normal range, and other data that do not conform to the actual physiological conditions; when there are some indicators whose values ​​are not detected in a certain physical examination, the missing values ​​can be filled by the mean filling or the adjacent value filling; for example, if the body mass index is missing in a certain physical examination, it can be estimated and filled based on the mean of the individual's past physical examination data; if the blood pressure value is missing, it can be reasonably filled with reference to the changing trend and mean of the blood pressure values ​​of the recent physical examinations; statistical methods are used to identify and correct abnormal values. Taking blood pressure as an example, if the systolic pressure is too high above the normal range, it is determined whether it is a measurement error in combination with the individual's other health conditions and past data. If it is determined to be abnormal, it is corrected to a reasonable value or further checked and confirmed. For blood biochemical indicators, if the alanine aminotransferase value is too high and inconsistent with the individual's recent physical condition and other indicator changes, the data accuracy needs to be verified and re-tested and corrected if necessary. Finally, data of different magnitudes and properties are converted into a unified format suitable for analysis. In this way, by preprocessing the data, the interference of erroneous data can be eliminated, such as removing erroneously entered extreme biochemical index values, to ensure the basic quality and credibility of the data; processing missing values ​​can complete key information, avoid analysis bias caused by missing data, fill in missing physical index data in a reasonable manner, and maintain data integrity and analysis consistency; outlier detection and correction can accurately identify and correct data that deviates from the group to prevent it from distorting the analysis results and ensure that the data reflects the true health status; data standardization unifies data format and magnitude, eliminates the impact of indicator differences, improves the compatibility and comparability of different indicators in analysis, fatty liver risk prediction model construction and risk prediction, and enhances the stability and prediction accuracy of the fatty liver risk prediction model.

[0041] Step 301: Screening out characteristic variables related to the risk of fatty liver disease from the pre-processed physical examination data, the characteristic variables including the change trend of body mass index, the dynamic change of alanine aminotransferase, the fluctuation of triglyceride index, the change of blood pressure and blood sugar index, the change of high-density lipoprotein cholesterol level and the change trend of low-density lipoprotein cholesterol level;

[0042] Step 302: Calculate the correlation coefficient between each variable and the fatty liver disease label, and select the variable whose absolute value of correlation is greater than the set threshold. The correlation coefficient calculation formula is:

[0043]

[0044] Among them, x i and i are the first i observations of variable x and disease label y, respectively, x′ and y′ are the means of variable x and disease label y, respectively, n is the number of observation samples, and the threshold is set to c(0 <c<1)。

[0045] In this application, the above formula is used to calculate the correlation coefficients between various variables and the fatty liver disease label. For the body mass index variable, the body mass index values x of a large number of individuals at different time points are obtained. i and the corresponding disease label values y. i , and the mean values x' and y' are calculated and substituted into the formula to obtain the correlation coefficient. If the absolute value of this coefficient exceeds the set threshold (0 < c < 1), for example, the coefficient of the body mass index change trend is 0.6 (the threshold is set at 0.5), then it is selected. Similarly, for variables such as the dynamic change of alanine aminotransferase and the fluctuation of triglyceride index, they are calculated and screened one by one. This method is scientific and rigorous, based on the statistical analysis of a large number of sample data, accurately identifying variables strongly associated with fatty liver disease, providing key support for the subsequent accurate construction of a prediction model, and improving the prediction accuracy of the model.

[0046] Step 401: Train a neural network to determine the comprehensive feature fusion and transformation function. Use the random forest algorithm to train the historical data to obtain the weights of the body weight coefficient, the change amount of alanine aminotransferase, the triglyceride index, blood pressure, blood sugar index, the change of high-density lipoprotein cholesterol level, and the change trend of density lipoprotein cholesterol level. The output formula of the neural network function is:

[0047]

[0048] where sign is the sign function, δ i is the Lagrange multiplier, y i is the class label of the training sample, K(a i , a) is the kernel function, d is the bias term, and i is the number of training samples;

[0049] By collecting various feature variable data of a large number of individuals as training samples (such as the body mass index change trend, the dynamic change of alanine aminotransferase, etc.) and the corresponding fatty liver disease conditions (as class labels), during the training process, these parameter values are continuously adjusted to make the function output accurately reflect the relationship between the input feature variables and the disease label as much as possible. The backpropagation algorithm is used to adjust the weights of each layer of neurons from the output layer to the input layer according to the difference between the prediction result and the actual disease label, and then optimize the Lagrange multiplier, kernel function parameters, bias term, etc. until the model achieves good performance on the training set, such as the accuracy, recall rate and other indicators meet the requirements, so as to determine the comprehensive feature fusion and transformation function.

[0050] Step 402: Based on the selected feature variables and combined with the dynamic feature fusion algorithm, construct a fatty liver risk prediction model. The expression of the fatty liver risk prediction model is as follows:

[0051]

[0052] Among them, P represents the risk probability of fatty liver disease, B represents the body mass index at time t, and ΔL t Indicates the change in alanine aminotransferase at time t, Tg t represents the triglyceride index at time t, D t represents the blood pressure and blood sugar index at time t, Es t represents the change in high-density lipoprotein cholesterol level at time t, Us t represents the trend of low-density lipoprotein cholesterol level at time t, f is the comprehensive feature fusion and conversion function, and w 1 、w 2 、w 3 、w 4 、w 5 and w 6 They represent the weights of body mass index, change in alanine aminotransferase, triglyceride index, blood pressure, blood sugar index, high-density lipoprotein cholesterol level and low-density lipoprotein cholesterol level respectively;

[0053] After determining the comprehensive feature fusion and conversion function, the summation values ​​(such as the change in alanine aminotransferase ), multiplied by the corresponding weight sum, and converted by function to obtain the risk probability value P of fatty liver disease between 0 and 1, such as 0.3, which means 30% disease probability, providing key quantitative basis for individual health assessment and disease prevention and control; by calculating the risk probability based on individual health examination data, such as early warning for those with continuous increase in body mass index and abnormal fluctuation of alanine aminotransferase, it helps medical staff intervene in the asymptomatic period of the disease, implement diet adjustment, exercise guidance and other interventions, and reduce the risk of disease; in terms of clinical diagnosis, it assists doctors in quickly assessing the condition, provides quantitative reference for diagnostic decisions, avoids subjective bias based on experience, and improves diagnostic efficiency and accuracy.

[0054] Step 501: using the constructed fatty liver risk prediction model, inputting the health examination data of the individual to be predicted on the day, and outputting the fatty liver disease risk prediction result of the individual, where the prediction result is the probability value of fatty liver disease.

[0055] As a specific implementation of the above-mentioned method for predicting the risk of fatty liver disease, the present application also provides a device for predicting the risk of fatty liver disease, such as Figure 2 As shown, the device comprises:

[0056] The data collection module is used to collect basic information of individuals, blood biochemical index data, body measurement index data and time information of multiple physical examinations on a quarterly basis to form a non-independent longitudinal data set;

[0057] The data collection module is the cornerstone of the entire fatty liver disease risk prediction device. The data collection module collects data from multiple channels, connects with the hospital physical examination system, and automatically extracts individual basic information (age, gender, etc.), blood biochemical indicators (alanine aminotransferase, triglycerides, etc.) and body measurement indicators (height, weight, etc.) data according to the set period, and simultaneously collects the physical examination timestamp to accurately form a longitudinal data sequence; it can also interact with the database of the health management agency to expand the breadth of data and ensure the continuity and integrity of the data. Collect multi-dimensional data on a quarterly basis to construct a non-independent longitudinal data set to provide comprehensive and dynamic information for subsequent accurate predictions. Longitudinal data can capture the changing trends of individual indicators, the quarterly fluctuations of body mass index, and the dynamic evolution of blood biochemical indicators. This information is crucial to analyzing the risk factors of fatty liver disease, can accurately reflect the individual health trajectory, enhance the sensitivity and accuracy of the prediction model for disease risk assessment, improve the reliability of prediction, and lay a solid foundation for early disease prevention and control.

[0058] Data preprocessing module, which is used to clean the collected data, handle missing values, detect and correct outliers, and convert different types of data into a format suitable for analysis;

[0059] The data preprocessing module cleans the data and accurately removes erroneous or unreasonable data, such as eliminating extreme biochemical index values ​​that are entered incorrectly, to ensure a solid foundation for data quality and avoid interference from erroneous data on analysis; when dealing with missing values, key information is completed based on the mean, adjacent values, or regression model to ensure data integrity, prevent missing values ​​from causing analysis bias, and maintain data logical coherence and analysis credibility; in the outlier detection and correction stage, statistical methods are used to accurately identify and correct deviated group data, such as correcting abnormal blood pressure values ​​caused by measurement errors, to ensure that the data truthfully reflects individual health status and improve analysis accuracy; through data standardization, various indicators are unified in format and magnitude to eliminate analysis barriers caused by indicator differences, improve the compatibility and comparability of different indicators in model construction and analysis, and lay a solid data foundation for accurately predicting the risk of fatty liver disease.

[0060] A feature screening module is used to screen out feature variables related to the risk of fatty liver disease from the preprocessed physical examination data, and calculate the correlation coefficient between each variable and the fatty liver disease label;

[0061] By accurately screening out key characteristic variables, we lay the foundation for building an efficient prediction model, avoid interference from irrelevant variables, simplify the complexity of the model, and improve computational efficiency and interpretability; such as excluding accidental fluctuation indicators, focusing on core variables such as body mass index trends and dynamic changes in transaminase, accurately characterizing factors associated with disease risk, enhancing the accuracy and stability of model predictions, and strongly supporting the accurate estimation of the risk of fatty liver disease, providing a reliable basis for the formulation of disease prevention and control strategies.

[0062] A model building module is used to construct a fatty liver risk prediction model by combining the screened feature variables with a dynamic feature fusion algorithm;

[0063] The model building module combines the selected characteristic variables, such as the trend of changes in body mass index and the dynamic changes of alanine aminotransferase, with the dynamic feature fusion algorithm to build a fatty liver risk prediction model; a neural network function with a multi-layer perceptron structure (including an input layer to receive characteristic variables, a hidden layer to process conversions, and an output layer to output results) is used to train historical data with a random forest algorithm to determine weights. For example, a large number of individual historical physical examination characteristic variable data that have been marked as sick or not are grouped, a group of weights are used for random forest training, a group of verification optimization, and the neural network function parameters are optimized by the back propagation algorithm. The neuron connection weights are adjusted according to the difference between the predicted and actual disease labels, and the function parameters are accurately determined to complete the construction of the fatty liver risk prediction model.

[0064] The prediction module uses the established fatty liver risk prediction model and inputs the health examination data of the individual to be predicted on the same day to calculate the fatty liver risk probability;

[0065] The prediction module is the core output link of the fatty liver disease risk prediction device. It uses a complete fatty liver risk prediction model to quickly process the health examination data of the individual to be predicted on the same day. For example, it receives key indicators such as the individual's body mass index, alanine aminotransferase value, triglyceride index, etc. on the same day, and inputs them into the model to calculate the accurate fatty liver risk probability; this probability value intuitively presents the possibility of an individual's disease, and provides a key quantitative basis for medical staff to formulate health intervention plans and clinical diagnosis decisions.

[0066] The output module is used to output the fatty liver risk probability predicted by the prediction module.

[0067] The output module receives the fatty liver risk probability calculated by the prediction module and converts the abstract data into an intuitive presentation form to the user's client. It can be displayed as a numerical value (such as 30% probability of disease) in a medical app or physical examination report, helping individuals to clearly understand the risk of disease and enhance awareness and participation in health management.

[0068] Furthermore, the data collection module also includes a data storage unit for storing the collected health examination data, and the data storage unit adopts a distributed storage method.

[0069] The massive amount of physical examination data increases dramatically with time and the number of individuals. Distributed storage can be elastically expanded, storage resources can be allocated on demand, and it can accommodate data torrents. It can ensure high data availability and reliability, multi-node redundant backup, even if some nodes fail, the data can be accessed instantly without loss, and the device system can be maintained in stable operation. At the same time, when multiple users read or write data concurrently, the distributed architecture uses a parallel processing mechanism to improve the reading and writing speed, optimize performance, and ensure fast data access. In addition, the powerful fault-tolerant capability can automatically detect and repair node failures to ensure data integrity and security.

[0070] Furthermore, the neural network function in the model building module adopts a multi-layer perceptron structure, including an input layer, a hidden layer, and an output layer.

[0071] The input layer can accept rich and diverse feature variable data such as changes in body mass index and dynamic values ​​of transaminase, providing a multi-information cornerstone for the model; the hidden layer plays a key conversion function, through the complex combination of multiple neurons and nonlinear activation function processing, deeply explores the intrinsic correlation and potential patterns of feature variables, and accurately abstracts high-level feature representations that are closely related to the risk of fatty liver disease, such as capturing the coordinated changes of different indicators; the output layer accurately outputs the risk probability value based on this, realizing efficient mapping from input data to quantitative prediction of disease risk; this structure improves the model's expressive ability, can accurately fit complex nonlinear relationships, model complex and changeable fatty liver disease risk factors, and enhance the model's prediction accuracy.

[0072] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting the risk of fatty liver disease, characterized in that: include, Step 1: Collect health examination data, which includes basic information of individuals, blood biochemical index data, body measurement index data, and time information of multiple physical examinations to form a non-independent longitudinal data set; Step 2: Data preprocessing of the collected non-independent longitudinal data sets, including data cleaning and data standardization, converting different types of data into a format suitable for analysis; Step 3: Screening out characteristic variables related to the risk of fatty liver disease from the pre-processed physical examination data, the characteristic variables including the change trend of body mass index, the dynamic change of alanine aminotransferase, the fluctuation of triglyceride index, the change of blood pressure and blood sugar index, the change of high-density lipoprotein cholesterol level and the change trend of low-density lipoprotein cholesterol level; Step 4: Construct a fatty liver risk prediction model based on the screened characteristic variables and in combination with a dynamic characteristic fusion algorithm. The expression of the fatty liver risk prediction model is as follows: Among them, P represents the risk probability of fatty liver disease, B represents the body mass index at time t, and ΔL t Indicates the change in alanine aminotransferase at time t, Tg t represents the triglyceride index at time t, D t represents the blood pressure and blood sugar index at time t, Es t represents the change in high-density lipoprotein cholesterol level at time t, Us t represents the change trend of low-density lipoprotein cholesterol level at time t, f is the comprehensive feature fusion and conversion function, w1, w2, w3, w4, w5 and w6 represent the weights of body mass index, change of alanine aminotransferase, triglyceride index, blood pressure, blood sugar index, high-density lipoprotein cholesterol level and low-density lipoprotein cholesterol level respectively; Step 5: Using the constructed fatty liver risk prediction model, input the health examination data of the individual to be predicted on the day, and output the fatty liver disease risk prediction result of the individual, wherein the prediction result is the probability value of fatty liver disease.

2. A method for predicting fatty liver disease risk according to claim 1, characterized in that: The collection cycle of the health examination data in step 1 is once every quarter, and the data preprocessing in step 2 also includes the processing of missing values ​​and the detection and correction of abnormal values.

3. A method for predicting fatty liver disease risk according to claim 1, characterized in that: When screening characteristic variables in step 3, the correlation coefficient between each variable and the fatty liver disease label is calculated, and the variable whose absolute value of correlation is greater than the set threshold is selected. The correlation coefficient calculation formula is: Among them, x i and i are the first i observations of variable x and disease label y, respectively, x′ and y′ are the means of variable x and disease label y, respectively, n is the number of observed samples, and the threshold is set to c(0 <c<1)。 4. A method for predicting fatty liver disease risk according to claim 1, characterized in that: The comprehensive feature fusion and conversion function is a neural network function, and the parameters of the function are determined by training the neural network. The various weights in step 4 are obtained by training historical data using a random forest algorithm.

5. A method for predicting the risk of fatty liver disease according to claim 4, characterized in that: The output formula of the neural network function is: Among them, sign is the sign function, δ i is the Lagrange multiplier, y i is the category label of the training sample, K(a i ,a) is the kernel function, d is the bias term, and i is the number of training samples.

6. A device for predicting the risk of fatty liver disease, applicable to the method for predicting the risk of fatty liver disease according to any one of claims 1 to 5, characterized in that: include, The data collection module is used to collect basic information of individuals, blood biochemical index data, body measurement index data and time information of multiple physical examinations on a quarterly basis to form a non-independent longitudinal data set; Data preprocessing module, which is used to clean the collected data, handle missing values, detect and correct outliers, and convert different types of data into a format suitable for analysis; A feature screening module is used to screen out feature variables related to the risk of fatty liver disease from the preprocessed physical examination data, and calculate the correlation coefficient between each variable and the fatty liver disease label; A model building module is used to construct a fatty liver risk prediction model by combining the screened feature variables with a dynamic feature fusion algorithm; The prediction module uses the established fatty liver risk prediction model and inputs the health examination data of the individual to be predicted on the same day to calculate the fatty liver risk probability; The output module is used to output the fatty liver risk probability predicted by the prediction module.

7. The device for predicting the risk of fatty liver disease according to claim 7, characterized in that: The data collection module also includes a data storage unit for storing the collected health examination data, and the data storage unit adopts a distributed storage method.

8. The device for predicting the risk of fatty liver disease according to claim 7, characterized in that: The neural network function in the model building module adopts a multi-layer perceptron structure, including an input layer, a hidden layer and an output layer.