Blood ceramide and cardiovascular event correlation analysis system

Abstract

The invention relates to the technical field of cardiovascular disease detection, and discloses a blood ceramide and cardiovascular event correlation analysis system, a data acquisition module is used for acquiring a plurality of blood samples of different groups of patients, acquiring blood ceramide data by using a mass spectrometry detection method, and collecting clinical index data corresponding to the patients; the data processing module is used for cleaning the acquired blood ceramide data and clinical index data, removing abnormal values and missing values, and performing standardization processing on the data to obtain processed data; the feature extraction module is used for extracting target feature data associated with the cardiovascular event from the processed data; the correlation analysis module is used for inputting the target characteristic data into a blood ceramide and cardiovascular event correlation analysis model, predicting the risk of the cardiovascular event through the correlation analysis model, formulating a personalized prevention and treatment scheme and outputting a correlation analysis result; the present invention is helpful to reveal complex relationships between blood ceramides and cardiovascular events.

Description

Technical Field

[0001] The present invention relates to the technical field of cardiovascular disease detection, and particularly to a system for analyzing the association between blood ceramides and cardiovascular events. Background Art

[0002] Cardiovascular diseases are serious public health problems globally, characterized by high incidence and high mortality. Ceramides (CERA), as an important component of cell membranes, participate in various biological processes and play a role in the initiation and acceleration of cardiovascular endothelial injury. Their abnormal appearance in the blood is relatively early. Although there are currently methods such as mass spectrometry for detecting CERA, there are problems such as high requirements for equipment and operators, limitations in application models in aspects such as sample preservation and transportation, and there is still a lack of large-scale application research for East Asian populations. Existing models have deficiencies in feature extraction and modeling, which limit their promotion and practicality in clinical settings. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design a system for analyzing the association between blood ceramides and cardiovascular events.

[0004] The present invention provides a system for analyzing the association between blood ceramides and cardiovascular events, which includes: A data acquisition module, which is used to collect a number of blood samples of patients from different groups, obtain blood ceramide data by using mass spectrometry detection methods, and simultaneously collect the corresponding clinical index data of the patients; A data processing module, which is used to clean the obtained blood ceramide data and clinical index data, remove outliers and missing values, and perform standardization processing on the data by using a normalization method to obtain processed data; A feature extraction module, which is used to extract target feature data associated with cardiovascular events from the processed data by using the RFE algorithm; An association analysis module, which is used to input the target feature data into the association analysis model of blood ceramides and cardiovascular events, predict the risk of cardiovascular events through the association analysis model, formulate personalized prevention and treatment plans, and output the association analysis results.

[0005] Optionally, in the first implementation manner of the present invention, the data processing module includes: A missing value processing sub-module, which is used to traverse the obtained blood ceramide data and clinical index data, divide the variables into continuous variables, categorical variables, and ordinal variables, perform missing value processing on the continuous variables by using the multiple imputation method MICE algorithm, and perform missing value processing on the categorical variables by using the CHAID decision tree method; The first sorting sub-module is used to sort the data after missing value processing from small to large, and calculate the first quartile and the third quartile of the data; The outlier processing sub-module is used to calculate the interquartile range based on the first quartile and the third quartile, determine outliers based on the interquartile range, and mark them, and process the outliers by replacing them with upper and lower limit values; The normalization processing sub-module is used to perform normalization processing on the data after missing value processing and outlier processing by using the Z-score normalization method to obtain the processed data.

[0006] Optionally, in the second implementation manner of the present invention, the process of the multiple imputation method MICE algorithm includes: For continuous variables containing missing values, use the mean to fill in the missing values, generate an initial complete data set, set the number of iterations to 50 times, and enter the loop iteration; Use the remaining variables as predictive variables and the currently processed variable containing missing values as the response variable to construct a linear regression model; Use the constructed linear regression model to predict the missing values and obtain the prediction results of the missing values; Replace the missing values in the initial complete data set with the predicted missing values, update the data set, and after 50 iterations, obtain the final imputation result through the mean from the multiple complete data sets generated by multiple iterations, and fill it into the missing positions of the continuous variables.

[0007] Optionally, in the third implementation manner of the present invention, the process of the CHAID decision tree method includes: Use the categorical variable containing missing values as the target variable and the remaining variables as explanatory variables to construct a training data set; Use the CHAID algorithm to construct a decision tree model. The CHAID algorithm determines the best split point by calculating the chi-square statistic between variables, continuously splits the training data set to form a tree structure, so that the distribution of the target variable within each node is uniform; Adopt the decision tree model, start from the root node, gradually traverse downward according to the values of the explanatory variables until reaching the leaf node, and use the category with the highest frequency in the leaf node as the prediction result of the missing value, and fill it into the missing position of the categorical variable.

[0008] Optionally, in the fourth implementation manner of the present invention, the feature extraction module includes: The second sorting sub-module is used to assign an importance score to each feature based on the processed data through a support vector machine, sort the features in descending order according to the feature importance score to generate a ranking list, remove the feature with the lowest ranking, and update the first feature set; A generation sub-module, configured to determine whether the first feature set reaches a preset retention quantity. If so, output the finally retained feature index set and its corresponding mapping table of original feature names, and generate a feature score table sorted in descending order of importance. A calculation sub-module, configured to screen out features associated with cardiovascular events from the feature score table, and remove redundant features by calculating the correlation matrix between features to obtain target feature data.

[0009] Optionally, in the fifth implementation manner of the present invention, the association analysis module includes: A capture sub-module, configured to input the target feature data into an association analysis model, analyze the data related to blood ceramides and cardiovascular events by combining multiple weak learners, capture potential linear and non-linear association features in the data, and transform the target feature data into a second feature set. A mining sub-module, configured to transform the second feature set into a graph structure, and deeply mine the non-linear relationship between factors related to blood ceramides and cardiovascular events, as well as the network effect generated by the interaction of various factors through the interaction calculation of nodes and edges. An output sub-module, configured to fuse the analyzed relationships through the Stacking integration algorithm, introduce time series analysis to dynamically adjust prediction parameters, use logistic regression to output the probability prediction value of cardiovascular events, and perform risk level classification. A formulation sub-module, configured to formulate personalized prevention and treatment plans after completing the risk prediction of cardiovascular events, and integrate and output the association analysis results.

[0010] Optionally, in the sixth implementation manner of the present invention, the formulation sub-module includes: Apply a multi-agent reinforcement learning algorithm to create multiple agents. The agents simulate the real clinical decision-making scenario and the effect of the simulation plan on reducing the risk of cardiovascular events, and the agents obtain corresponding rewards or punishments. Gradually obtain a better combination direction of plans through continuous iteration to determine the plan framework. Introduce a Bayesian optimization algorithm to screen out the optimal parameter combination in the parameter space through the Bayesian optimization algorithm. Sort out the personalized prevention and treatment plans refined by the Bayesian optimization algorithm, and integrate them with the risk prediction results of cardiovascular events to output the association analysis results.

[0011] Optionally, in the seventh implementation manner of the present invention, a method for implementing the above-mentioned association analysis system of blood ceramides and cardiovascular events includes the following steps: Collect a number of blood samples from patients in different groups, use a mass spectrometry detection method to obtain blood ceramide data, and at the same time collect the corresponding clinical index data of the patients. Clean the obtained blood ceramide data and clinical index data, remove outliers and missing values, and use the normalization method to standardize the data to obtain the processed data; Use the RFE algorithm to extract the target feature data associated with cardiovascular events from the processed data; Input the target feature data into the blood ceramide and cardiovascular event association analysis model, predict the risk of cardiovascular events through the association analysis model, and formulate personalized prevention and treatment plans, and output the association analysis results.

[0012] Optionally, in the eighth implementation manner of the present invention, a method for implementing the blood ceramide and cardiovascular event association analysis system includes the following steps: Traverse the obtained blood ceramide data and clinical index data, divide the variables into continuous variables, categorical variables, and ordinal variables, use the multiple imputation method MICE algorithm to process the missing values of continuous variables, and use the CHAID decision tree method to process the missing values of categorical variables; Sort the data after missing value processing from small to large, and calculate the first quartile and the third quartile of the data; Calculate the interquartile range based on the first quartile and the third quartile, determine outliers based on the interquartile range, and mark them, and process the outliers by replacing them with upper and lower limits; Use the Z-score normalization method to standardize the data after missing value processing and outlier processing to obtain the processed data.

[0013] Optionally, in the ninth implementation manner of the present invention, a method for implementing the blood ceramide and cardiovascular event association analysis system includes the following steps: Based on the processed data, assign importance scores to each feature through a support vector machine, sort the features in descending order according to the feature importance scores to generate a ranking list, remove the feature with the lowest ranking, and update the first feature set; Determine whether the first feature set reaches the preset retention quantity. If so, output the finally retained feature index set and its corresponding original feature name mapping table, and generate a feature score table arranged in descending order of importance; Screen out the features associated with cardiovascular events from the feature score table, and remove redundant features by calculating the correlation matrix between features to obtain the target feature data.

[0014] In the technical solution provided by the present invention, a number of blood samples of patients from different groups are collected, and blood ceramide data is obtained by using a mass spectrometry detection method. At the same time, the corresponding clinical index data of the patients is collected; the obtained blood ceramide data and clinical index data are cleaned to remove outliers and missing values, and the data is standardized by using a normalization method to obtain processed data; the RFE algorithm is used to extract target feature data associated with cardiovascular events from the processed data; the target feature data is input into the correlation analysis model of blood ceramide and cardiovascular events, the risk of cardiovascular events is predicted through the correlation analysis model, and a personalized prevention and treatment plan is formulated, and the correlation analysis result is output; the present invention can efficiently integrate the blood ceramide data obtained by mass spectrometry detection with the patient clinical index information, extract target features highly related to cardiovascular events through data processing, and then construct a correlation analysis model with individual adaptability, improving the standardization of data processing and the stability of model prediction, helping to reveal the complex relationship between ceramide and cardiovascular events, enhancing the early warning ability of cardiovascular diseases, and at the same time providing a scientific basis for personalized health management, with significant social benefits and practical value. Brief Description of the Drawings

[0015] By reading the following detailed description of the preferred embodiments, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention.

[0016] Figure 1 It is a schematic structural diagram of a correlation analysis system for blood ceramide and cardiovascular events provided by an embodiment of the present invention; Figure 2 It is a schematic structural diagram of a data processing module provided by an embodiment of the present invention; Figure 3 It is a schematic structural diagram of a feature extraction module provided by an embodiment of the present invention. Detailed Embodiments

[0017] The terms "first", "second", "third", "fourth", etc. (if any) in the specification, claims and above drawings of the present invention are used to distinguish similar objects and do not have to be used to describe a specific order or sequence. It should be understood that such data may be interchanged under appropriate circumstances so that the embodiments described herein can be implemented in an order different from that shown or described herein. In addition, the terms "comprising" or "having" and any variation thereof are intended to cover non-exclusive inclusion. For example, a process, method, device, product or device comprising a series of steps or units does not have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0018] For ease of understanding, the specific process of the embodiment of the present invention will be described below. Please refer to Figure 1 The structural schematic diagram of the blood ceramide and cardiovascular event association analysis system provided by the embodiment of the present invention. The system includes: A data acquisition module, which is used to collect a number of blood samples of patients in different groups, obtain blood ceramide data by using a mass spectrometry detection method, and at the same time collect the corresponding clinical index data of the patients; A data processing module, which is used to clean the obtained blood ceramide data and clinical index data, remove outliers and missing values, and perform standardization processing on the data by using a normalization method to obtain processed data; A feature extraction module, which is used to extract target feature data associated with cardiovascular events from the processed data by using the RFE algorithm; An association analysis module, which is used to input the target feature data into the blood ceramide and cardiovascular event association analysis model, predict the risk of cardiovascular events through the association analysis model, formulate personalized prevention and treatment plans, and output the association analysis results.

[0019] In this embodiment, please refer to Figure 2 , the data processing module includes: A missing value processing sub-module, which is used to traverse the obtained blood ceramide data and clinical index data, divide the variables into continuous variables, categorical variables and ordinal variables, perform missing value processing on the continuous variables by using the multiple imputation method MICE algorithm, and perform missing value processing on the categorical variables by using the CHAID decision tree method; A first sorting sub-module, which is used to sort the data after missing value processing from small to large, and calculate the first quartile and the third quartile of the data; An outlier processing sub-module, which is used to calculate the interquartile range according to the first quartile and the third quartile, determine outliers based on the interquartile range, and mark them, and process the outliers by replacing them with upper and lower limits; A standardization processing sub-module, which is used to perform standardization processing on the data after missing value processing and outlier processing by using the Z-score normalization method to obtain processed data.

[0020] In this embodiment, the process of the MICE algorithm of multiple imputation method includes: for continuous variables with missing values, filling the missing values with the mean value to generate an initial complete data set, setting the number of iterations to 50 times, and entering the loop iteration; using the remaining variables as predictor variables and the variable with missing values being processed currently as the response variable to construct a linear regression model; using the constructed linear regression model to predict the missing values to obtain the prediction results of the missing values; replacing the missing values in the initial complete data set with the predicted missing values to update the data set. After 50 iterations are completed, the final imputation result is obtained through the mean value from multiple complete data sets generated in multiple iterations and filled into the missing positions of the continuous variables.

[0021] In this embodiment, the process of the CHAID decision tree method includes: taking the categorical variable with missing values as the target variable and the remaining variables as explanatory variables to construct a training data set; using the CHAID algorithm to construct a decision tree model. The CHAID algorithm determines the best split point by calculating the chi-square statistic between variables and continuously splits the training data set to form a tree structure, making the distribution of the target variable within each node uniform; adopting the decision tree model, starting from the root node, traversing step by step according to the values of the explanatory variables until reaching the leaf node, and taking the category with the highest frequency in the leaf node as the prediction result of the missing value and filling it into the missing position of the categorical variable. In this embodiment, please refer to Figure 3 , the feature extraction module includes: The second sorting sub-module is used to assign an importance score to each feature based on the processed data through a support vector machine, sort the features in descending order according to the feature importance score to generate a ranking list, remove the feature with the lowest ranking, and update the first feature set; The generation sub-module is used to determine whether the first feature set reaches the preset retention quantity. If so, output the set of finally retained feature indices and its corresponding mapping table of original feature names, and generate a feature score table arranged in descending order of importance; The calculation sub-module is used to screen out the features associated with cardiovascular events from the feature score table and remove redundant features by calculating the correlation matrix between features to obtain the target feature data.

[0022] In this embodiment, the association analysis module includes: The capture sub-module is used to input the target feature data into the association analysis model, analyze the data related to blood ceramides and cardiovascular events by combining multiple weak learners, capture the potential linear and non-linear association features in the data, and convert the target feature data into a second feature set; A mining sub-module for converting the second feature set into a graph structure, and through the interaction calculation of nodes and edges, deeply mining the non-linear relationship between blood ceramides and cardiovascular event-related factors, as well as the network effects generated by the interaction of various factors; An output sub-module for fusing the relationships obtained through analysis by the Stacking integration algorithm, introducing time series analysis to dynamically adjust prediction parameters, using logistic regression to output the probability prediction value of cardiovascular events, and performing risk level classification; A formulation sub-module for formulating personalized prevention and treatment plans after completing the risk prediction of cardiovascular events, and integrating and outputting the results of association analysis.

[0023] In this embodiment, the formulation sub-module includes: Using a multi-agent reinforcement learning algorithm to create multiple agents. The agents simulate the real clinical decision-making scenario and the effect of the simulation plan on reducing the risk of cardiovascular events, and the agents obtain corresponding rewards or punishments; Gradually obtain a better combination direction of the plan through continuous iteration, and determine the plan framework; Introduce the Bayesian optimization algorithm, and screen out the optimal parameter combination in the parameter space through the Bayesian optimization algorithm; Sort out the personalized prevention and treatment plans refined by the Bayesian optimization algorithm, and integrate them with the risk prediction results of cardiovascular events to output the results of association analysis.

[0024] In this embodiment, a number of blood samples (such as 400) of different groups including natural populations, patients with stable coronary heart disease, patients with angina pectoris, patients with acute myocardial infarction (AMI), etc. are collected, and mass spectrometry detection methods are used to obtain blood ceramide data. At the same time, clinical index data such as coronary artery stenosis degree, angina pectoris degree, electrocardiogram, and myocardial enzyme of the patients are collected.

[0025] In this embodiment, each variable in the obtained blood ceramide data and clinical index data is checked in sequence; according to the characteristics of variable values, they are divided into continuous variables (such as age, blood pressure value), categorical variables (such as gender, disease type), and ordinal variables (such as cardiac function classification); for continuous variables, the multiple imputation method (MICE algorithm) is used to handle missing values; the MICE algorithm first fills the missing values initially with simple methods such as the mean or median, and then enters the iterative process. In each iteration, for each continuous variable with missing values, a linear regression model is constructed using the remaining variables as predictors, and the missing values are predicted and replaced using the model. The iteration is repeated continuously until the convergence condition is met; for categorical variables, the CHAID decision tree method is used, with other variables as inputs, and node splitting is performed based on the chi-square test to construct a decision tree model. According to the tree structure and the majority class principle of leaf nodes, the missing values are predicted and filled; the data after missing value processing is sorted separately for each variable from smallest to largest; for the sorted data, according to the calculation rule of quartiles, the first quartile (Q1, that is, the value at the first 25% position after data sorting) and the third quartile (Q3, that is, the value at the first 75% position after data sorting) of the data are determined; if the number of data is n, when n×0.25 or n×0.75 is an integer, Q1 or Q3 takes the average value of the data at the corresponding position and the next position; if it is a non-integer, it is rounded up, and the data at the corresponding position is Q1 or Q3; according to the calculated first quartile Q1 and third quartile Q3, the interquartile range (IQR = Q3 - Q1) is calculated; an outlier determination range is set, and the data less than Q1 - 1.5×IQR or greater than Q3 + 1.5×IQR is determined as an outlier, and these outliers are marked; for the marked outliers, the method of replacing with upper and lower limits is used for processing, that is, the outlier less than Q1 - 1.5×IQR is replaced with Q1 - 1.5×IQR, and the outlier greater than Q3 + 1.5×IQR is replaced with Q3 + 1.5×IQR; the Z-score normalization method is used to standardize the data after missing value processing and outlier processing; for each data point, it is calculated according to the formula Z = (X - μ) / σ, where X is the value of the original data point, μ is the mean of the variable data, and σ is the standard deviation of the variable data; through this formula, the data is converted into standard normal distribution data with a mean of 0 and a standard deviation of 1, so as to eliminate the differences in dimension and value range between different variables and obtain the finally processed data for subsequent analysis and use.

[0026] In this embodiment, the processed data is used as input, and a support vector machine (SVM) model is selected to assign importance scores to each feature. For a linear SVM, it finds the hyperplane that can separate the data to the greatest extent by optimizing the objective function. The importance of a feature is determined by the absolute value of the weight vector. The larger the absolute value of the weight, the greater the influence of the corresponding feature on the classification decision, and the higher the importance score of this feature. For a non-linear SVM, the kernel function is used to map the data into a high-dimensional space, and by calculating the kernel coefficients and support vectors, the contribution degree of each feature to the classification boundary is determined, which is used as the basis for the feature importance score. Based on the obtained importance scores, all features are sorted in descending order to generate a feature ranking list. The feature with the lowest ranking is removed from the ranking list, and the first feature set is updated to complete one iteration of feature screening. Set a preset number of retained features, which can be determined according to experience or the results of preliminary data analysis. After each feature is removed and the first feature set is updated, check whether the number of features in the first feature set reaches the preset number of retained features. If the preset number is reached, stop the feature screening process, output the set of indices of the finally retained features, and establish a mapping table between this index set and the feature names in the original data to facilitate subsequent identification and analysis of features. At the same time, generate a feature score table sorted in descending order of importance, which clearly shows the importance score of each retained feature. From the generated feature score table, in combination with medical knowledge and research purposes, select the features that may be associated with cardiovascular events. To further optimize the feature set, calculate the correlation matrix between these selected features. The elements in the matrix represent the correlation coefficients between pairwise features, reflecting the degree of linear correlation between features. Set a correlation threshold (such as 0.8). If the absolute value of the correlation coefficient between two features is greater than this threshold, it is considered that these two features have strong redundancy. For redundant feature pairs, retain the feature with the higher importance score and remove the other feature, and finally obtain the target feature data, so that the feature set can both reflect the association with cardiovascular events and have a low redundancy.

[0027] In this embodiment, the target feature data is input into the association analysis model, and multiple different types of weak learners are selected, such as decision trees, logistic regression, support vector machines, etc.; each weak learner trains and analyzes the data related to blood ceramide and cardiovascular events respectively. The decision tree constructs a tree structure by recursively partitioning the data space and finding the best split point of the data, so as to capture the non-linear relationship in the data; logistic regression constructs a linear model to fit the linear relationship between the data features and the probability of cardiovascular events; support vector machines classify the data in the feature space by finding the optimal hyperplane and mine the non-linear boundary of the data; after each weak learner completes training, it outputs the analysis results of the data respectively; these results are combined to form a new feature representation, that is, the second feature set; the combination method can be to splice information such as the output probability, predicted class or feature importance of each weak learner, so that the new feature set not only contains the feature information of the original data, but also integrates the potential linear and non-linear association features captured by multiple weak learners from different perspectives; the second feature set is transformed into a graph structure, each feature is regarded as a node in the graph, and the association relationship between features is regarded as an edge; determine the construction rule of the edge, which can be based on measurement indicators such as the correlation coefficient and mutual information between features. When the association measurement between two features exceeds a certain threshold, an edge connection is established between the corresponding nodes; on the constructed graph structure, the non-linear relationship and network effect are deeply mined through the interaction calculation of nodes and edges; adopt the message passing mechanism in the graph neural network, each node receives the information passed through the edge from the neighbor nodes in each round of iteration, and this information includes the features of the neighbor nodes and the weight information of the edge; the node updates its own feature representation according to the received information, and the update process can use a neural network for non-linear transformation; after multiple rounds of message passing and node feature update, the feature representation of each node not only contains its own original information, but also integrates the information of the surrounding nodes, so as to capture the complex network effect generated by the interaction of various factors and mine the deeper non-linear relationship between the factors related to blood ceramide and cardiovascular events; use the Stacking ensemble algorithm to fuse the relationships mined through the graph structure; first, divide the data into a training set and a test set, and train multiple base learners on the training set. The base learners can select different models from the first step, such as random forests, gradient boosting trees, etc.; after each base learner is trained on the training set, it makes predictions on the training set and the test set respectively to obtain the prediction results; use the prediction results of the base learners on the training set as new features to train a meta-learner, and the meta-learner usually selects a logistic regression model; during the training process, the meta-learner learns how to combine the prediction results of the base learners to obtain a more accurate prediction; introduce the time series analysis method, considering the characteristics that the blood ceramide level and the risk of cardiovascular events may change over time, analyze the newly input data in chronological order.Extract time series features such as trend features and periodic features through methods such as sliding windows; dynamically adjust the prediction parameters in the Stacking ensemble model according to the features obtained from time series analysis, such as the weight coefficients in the logistic regression model; finally, use the adjusted logistic regression model to output the probability prediction value of cardiovascular events, and perform risk level classification according to the preset probability threshold, such as low risk, medium risk, and high risk.

[0028] In this embodiment, multiple agents are created according to different roles and factors involved in clinical decision-making, such as a drug therapy agent, a lifestyle intervention agent, a monitoring plan agent, etc.; the action space of each agent is defined. The actions of the drug therapy agent include the selection of different types of drugs and dose adjustment; the lifestyle intervention agent covers exercise types, dietary structure adjustment, etc.; the monitoring plan agent involves detection items and detection frequency setting; the state space is set, and the patient's blood ceramide data, clinical indicators (age, blood pressure, medical history, etc.) and the currently adopted plan combination are used as the state; a real clinical decision-making scenario is simulated, allowing the agent to take actions in this scenario, and according to the effect of the simulated plan on reducing the risk of cardiovascular events, the agent is given rewards or punishments according to a pre-set reward function; the reward function comprehensively considers factors such as the degree of risk reduction, treatment cost, and probability of side effects. A significant risk reduction gives a high reward, while severe side effects or excessive costs impose penalties; multiple agents continuously execute actions in the simulated scenario and update their own strategies according to the rewards or punishments feedback by the environment; each agent records historical states, actions, and reward data, and uses the experience replay mechanism to randomly sample from the historical data for learning; methods such as deep neural networks or policy gradients are used to approximately represent the agent's policy function and value function, and the network parameters are updated by optimizing the loss function (such as maximizing the expectation of the cumulative reward in the policy gradient algorithm) to adjust the policy; agents cooperate and compete with each other, share some experience data, and jointly explore different plan combinations; in multiple iteration processes, gradually learn a more beneficial action selection method for reducing the risk of cardiovascular events, converge to a better plan combination direction, and determine a preliminary plan framework, including the main types of intervention measures and the general implementation direction; based on the determined preliminary plan framework, define the plan parameter space, such as the value range of drug doses, the interval of exercise intensity, the optional values of monitoring frequency, etc.; initialize the Bayesian optimization algorithm, first randomly sample a small number of points in the parameter space for evaluation, and obtain the effects of the plans corresponding to these parameter combinations in the simulated scenario (which can be obtained by calling the simulation evaluation mechanism constructed by the previous multi-agent reinforcement learning); according to these initial evaluation results, construct a probability model of the objective function, and commonly use a Gaussian process model to describe the relationship between parameters and plan effects. This model can give the mean and variance estimates of plan effects at different parameter values; use an acquisition function (such as expected improvement, upper confidence bound, etc.) to select the next parameter point to be evaluated in the parameter space. The acquisition function comprehensively considers the balance between exploring new regions and exploiting known better regions; evaluate the newly selected parameter point, update the probability model, repeat the process of selecting parameter points and updating the model, continuously search in the parameter space, and gradually screen out the optimal parameter combination to refine the personalized prevention and treatment plan; integrate the optimal parameter combination obtained by the Bayesian optimization algorithm into the personalized prevention and treatment plan, and clarify specific drug types, accurate doses, detailed lifestyle intervention contents, accurate monitoring arrangements, etc.;Meanwhile, obtain the cardiovascular event risk prediction results based on models such as deep forest, including risk probabilities and risk levels for different time spans; integrate the refined personalized plan with the risk prediction results, and present the correlation analysis results in a clear and intuitive manner, such as preparing a report containing risk prediction, plan content, and expected effects, or generating a visual chart to show the mechanism of action and expected trends of the plan in reducing risks, and finally output the complete correlation analysis results to provide a comprehensive basis for clinical decision-making.

[0029] The above has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A system for analyzing the association between blood ceramide and cardiovascular events, characterized in that, The system includes: A data acquisition module, which is used to collect a number of blood samples from patients in different groups, obtain blood ceramide data by using mass spectrometry detection methods, and collect the corresponding clinical index data of the patients at the same time; A data processing module, which is used to clean the obtained blood ceramide data and clinical index data, remove outliers and missing values, and perform standardization processing on the data by using a normalization method to obtain processed data; A feature extraction module, which is used to extract target feature data associated with cardiovascular events from the processed data by using the RFE algorithm; An association analysis module, which is used to input the target feature data into the association analysis model of blood ceramide and cardiovascular events, predict the risk of cardiovascular events through the association analysis model, and formulate personalized prevention and treatment plans, and output the association analysis results.

2. The system for analyzing the association between blood ceramides and cardiovascular events according to claim 1, wherein The data processing module includes: A missing value processing sub-module, which is used to traverse the obtained blood ceramide data and clinical index data, divide the variables into continuous variables, categorical variables and ordinal variables, perform missing value processing on the continuous variables by using the multiple imputation method MICE algorithm, and perform missing value processing on the categorical variables by using the CHAID decision tree method; A first sorting sub-module, which is used to sort the data after missing value processing from small to large, and calculate the first quartile and the third quartile of the data; An outlier processing sub-module, which is used to calculate the interquartile range according to the first quartile and the third quartile, determine outliers based on the interquartile range, and mark them, and process the outliers by replacing them with upper and lower limit values; A standardization processing sub-module, which is used to perform standardization processing on the data after missing value processing and outlier processing by using the Z-score normalization method to obtain processed data.

3. The blood ceramide and cardiovascular event correlation analysis system according to claim 2, wherein The process of the multiple imputation method MICE algorithm includes: For continuous variables containing missing values, fill the missing values with the mean value to generate an initial complete data set, set the number of iterations to 50 times, and enter the loop iteration; Use the remaining variables as predictor variables and the currently processed variable containing missing values as the response variable to construct a linear regression model; Use the constructed linear regression model to predict the missing values and obtain the prediction results of the missing values; Replace the missing values in the initial complete data set with the predicted missing values, update the data set, and after 50 iterations, obtain the final imputation result through the mean value from multiple complete data sets generated by multiple iterations, and fill it into the missing positions of the continuous variables.

4. The blood ceramide and cardiovascular event correlation analysis system according to claim 2, wherein The process of the CHAID decision tree method includes: Use the categorical variable containing missing values as the target variable and the remaining variables as the explanatory variables to construct a training data set; Use the CHAID algorithm to construct a decision tree model. The CHAID algorithm determines the best split point by calculating the chi-square statistic between variables, continuously splits the training data set to form a tree structure, so that the distribution of the target variable within each node is uniform; Adopt the decision tree model, start from the root node, gradually traverse downward according to the values of the explanatory variables until reaching the leaf node, and use the category with the highest frequency in the leaf node as the prediction result of the missing value, and fill it into the missing position of the categorical variable.

5. The system for analyzing the association between blood ceramides and cardiovascular events according to claim 1, wherein, The feature extraction module includes: A second sorting sub-module, which is used to assign an importance score to each feature by using a support vector machine based on the processed data, sort the features in descending order according to the feature importance score to generate a ranking list, remove the feature with the lowest ranking, and update the first feature set; A generation sub-module, which is used to determine whether the first feature set reaches a preset retention quantity. If so, it outputs the finally retained feature index set and its corresponding original feature name mapping table, and generates a feature score table arranged in descending order of importance; A calculation sub-module, which is used to screen out the features associated with cardiovascular events from the feature score table, and remove redundant features by calculating the correlation matrix between features to obtain target feature data.

6. The blood ceramide and cardiovascular event correlation analysis system according to claim 1, characterized in that, The association analysis module includes: A capture sub-module, which is used to input the target feature data into an association analysis model, analyze the data related to blood ceramides and cardiovascular events by combining multiple weak learners, capture potential linear and non-linear association features in the data, and convert the target feature data into a second feature set; A mining sub-module, which is used to convert the second feature set into a graph structure, and deeply mine the non-linear relationship between blood ceramides and cardiovascular event-related factors, as well as the network effect generated by the interaction of various factors through the interaction calculation of nodes and edges; An output sub-module, which is used to fuse the analyzed relationships through a Stacking integration algorithm, introduce time series analysis to dynamically adjust prediction parameters, use logistic regression to output the probability prediction value of cardiovascular events, and perform risk level classification; A formulation sub-module, which is used to formulate personalized prevention and treatment plans after completing the risk prediction of cardiovascular events, and integrate and output the association analysis results.

7. The system for analyzing the association between blood ceramides and cardiovascular events according to claim 6, characterized in that The formulation sub-module includes: Using a multi-agent reinforcement learning algorithm to create multiple agents. The agents simulate the real clinical decision-making scenario and the effect of the simulation plan on reducing the risk of cardiovascular events, and the agents obtain corresponding rewards or punishments; Gradually obtain a better combination direction of the plan through continuous iteration, and determine the plan framework; Introduce the Bayesian optimization algorithm, and screen out the optimal parameter combination in the parameter space through the Bayesian optimization algorithm; Sort out the personalized prevention and treatment plans refined by the Bayesian optimization algorithm, and integrate them with the cardiovascular event risk prediction results to output the association analysis results.

8. A method for implementing a system for analyzing the association between blood ceramides and cardiovascular events as described in claim 1, characterized in that, The method includes the following steps: Collect a number of blood samples from patients in different groups, use a mass spectrometry detection method to obtain blood ceramide data, and collect the corresponding clinical index data of the patients at the same time; Clean the obtained blood ceramide data and clinical index data, remove outliers and missing values, and perform standardization processing on the data using a normalization method to obtain processed data; Use the RFE algorithm to extract target feature data associated with cardiovascular events from the processed data; Input the target feature data into the blood ceramide and cardiovascular event association analysis model, predict the risk of cardiovascular events through the association analysis model, formulate personalized prevention and treatment plans, and output the association analysis results.

9. A method for implementing a system for analyzing the association between blood ceramides and cardiovascular events as described in claim 1, characterized in that, The method includes the following steps: Traverse the obtained blood ceramide data and clinical index data, divide the variables into continuous variables, categorical variables, and ordinal variables, use the MICE algorithm of multiple imputation method to handle missing values for continuous variables, and use the CHAID decision tree method to handle missing values for categorical variables; Sort the data after missing value handling from smallest to largest, and calculate the first quartile and the third quartile of the data; Calculate the interquartile range based on the first quartile and the third quartile, determine outliers based on the interquartile range, and mark them, and handle the outliers by replacing them with upper and lower limits; Use the Z-score normalization method to standardize the data after missing value handling and outlier handling to obtain the processed data.

10. A method for implementing a system for analyzing the association between blood ceramide and cardiovascular events as described in claim 1, characterized in that, This method includes the following steps: Based on the processed data, assign importance scores to each feature through a support vector machine, sort the features in descending order according to the feature importance scores to generate a ranking list, remove the feature with the lowest ranking, and update the first feature set; Determine whether the first feature set reaches the preset retention quantity. If so, output the finally retained feature index set and its corresponding original feature name mapping table, and generate a feature score table sorted in descending order of importance; Screen out the features associated with cardiovascular events from the feature score table, and remove redundant features by calculating the correlation matrix between features to obtain the target feature data.

Images