Cardiovascular data monitoring method for cardiovascular medicine department

Abstract

The invention discloses a cardiovascular data monitoring method for the cardiovascular medicine department, and relates to the technical field of cardiovascular monitoring, and the method comprises the steps: collecting and preprocessing multi-source cardiovascular data of a patient, and building a cardiovascular data set in combination with a conditional generative adversarial network; constructing a cardiovascular network framework based on the established cardiovascular data set, and optimizing cardiovascular network parameters by using a meta-learning algorithm to capture feature data in the cardiovascular data set; analyzing a causal relationship between the feature data and the cardiovascular event by using a causal discovery algorithm, and constructing a causal relationship graph based on an analysis result; and a fuzzy membership function and an entropy weight algorithm are fused, nodes in the causal relationship graph are adjusted, and a hierarchical monitoring strategy is generated in combination with a multi-objective evolutionary algorithm. According to the method, the multi-source cardiovascular data of the patient is collected and preprocessed, and the conditional generative adversarial network is introduced to generate high-quality and diversified synthetic data, so that the data scale is effectively expanded, and the balance and representativeness of data distribution are improved.

Description

Technical Field

[0001] The present invention relates to the field of cardiovascular monitoring technology, and in particular to a cardiovascular data monitoring method in cardiology. Background Art

[0002] Cardiovascular disease is one of the major diseases currently affecting the world. With the aging population and changes in lifestyle, the incidence of cardiovascular disease continues to rise. Therefore, early warning, dynamic monitoring, and risk assessment of cardiovascular disease have become important research areas in the field of clinical cardiology. Traditional cardiovascular health monitoring relies primarily on intermittent checkups of patients in hospitals. Data collection often has disadvantages such as large time delays, poor continuity, and an inability to dynamically reflect changes in the condition. Furthermore, monitoring indicators often rely on static indicators, such as single-point blood pressure, electrocardiogram results, and blood lipids, making it difficult to fully capture the dynamic evolution of cardiovascular physiological signals.

[0003] In recent years, with the continuous advancement of sensor technology, wireless communication technology, and artificial intelligence algorithms, real-time monitoring of cardiovascular data has gradually become a reality, providing a new technical means for early warning and continuous management of cardiovascular disease. However, most current monitoring systems still have certain limitations: they often rely on a single data source (such as the electrocardiogram (ECG)) and lack the integrated analysis of multimodal physiological parameters (such as blood pressure, blood oxygen saturation, respiratory rate, etc.), making it difficult to comprehensively and accurately reflect the patient's overall cardiovascular health status. In addition, individuals vary significantly in age, gender, genetic background, and other factors, all of which have a significant impact on the functional performance of the cardiovascular system. Traditional analysis methods often have difficulty effectively capturing the complex temporal characteristics of cardiovascular signals, such as long-term trends, short-term fluctuations, hysteresis effects, and cross-timescale dependencies, thus limiting the accuracy and reliability of disease prediction.

[0004] Currently, no effective solutions have been proposed for the problems in related technologies. Summary of the Invention

[0005] In response to the problems in the related art, the present invention proposes a cardiovascular data monitoring method for cardiovascular medicine to overcome the above-mentioned technical problems existing in the existing related art.

[0006] To this end, the specific technical solutions adopted in the present invention are as follows:

[0007] A cardiovascular data monitoring method for cardiology, the method comprising:

[0008] S1. Collect and preprocess multi-source cardiovascular data of patients and build a cardiovascular dataset using a conditional generative adversarial network.

[0009] S2. Based on the established cardiovascular dataset, a cardiovascular network framework is constructed, and the cardiovascular network parameters are optimized using a meta-learning algorithm to capture the characteristic data in the cardiovascular dataset;

[0010] S3. Use causal discovery algorithms to analyze the causal relationship between feature data and cardiovascular events, and construct a causal relationship diagram based on the analysis results;

[0011] S4. Fuzzy membership function and entropy weight algorithm are integrated to adjust the nodes in the causal relationship graph, and a hierarchical monitoring strategy is generated by combining the multi-objective evolutionary algorithm.

[0012] Optionally, collecting and preprocessing multi-source cardiovascular data of patients and combining them with a conditional generative adversarial network to establish a cardiovascular dataset include:

[0013] S11. Collect multi-source cardiovascular data of the patient and use time synchronization technology and feature alignment technology to align the multi-source cardiovascular data in time and feature dimensions;

[0014] S12, using a bandpass filter to denoise the aligned multi-source cardiovascular data;

[0015] S13. Based on the denoised multi-source cardiovascular data, construct a conditional vector and convert the format of the conditional vector using embedded coding technology;

[0016] S14. Use the format-converted conditional vector as input and use the conditional generative adversarial network to generate a cardiovascular dataset.

[0017] Optionally, based on the established cardiovascular dataset, a cardiovascular network framework is constructed, and the cardiovascular network parameters are optimized using a meta-learning algorithm to capture the characteristic data in the cardiovascular dataset, including:

[0018] S21. Based on the generated cardiovascular dataset, build a cardiovascular feature modeling network that integrates long short-term memory network and attention mechanism, and initialize the cardiovascular network parameters;

[0019] S22. Combine meta-learning algorithms to build a multi-task meta-learning framework, and use the multi-task meta-learning framework to perform gradient updates and parameter aggregation for each task to optimize cardiovascular network parameters;

[0020] S23. Based on the optimized cardiovascular network parameters, a deep learning algorithm is used to capture the characteristic data in the cardiovascular dataset.

[0021] Optionally, using a causal discovery algorithm to analyze the causal relationship between the characteristic data and cardiovascular events, and constructing a causal relationship diagram based on the analysis results includes:

[0022] S31. Based on the extracted feature data, an adaptive dynamic time window mechanism is constructed. Combined with a multi-scale causal convolutional network, the feature data is modeled at different time scales to capture the lag effect and cross-temporal dependency of the feature data and extract temporal dynamic features.

[0023] S32. Based on the extracted temporal dynamic features, a constraint-based causal discovery algorithm is used to identify the potential causal structure between the feature data and cardiovascular events, and to generate an initial set of causal hypotheses;

[0024] S33. Based on the generated initial causal hypothesis set, a hybrid verification framework including statistical testing and logical reasoning is constructed. In combination with the conditional independence test algorithm, the initial causal hypotheses are gradually verified and screened to generate a candidate causal relationship set.

[0025] S34. Based on the generated set of candidate causal relationships, a causal relationship graph representing the evolution process of cardiovascular risk is constructed with feature data as graph nodes and causal relationships as directed edges.

[0026] Optionally, based on the extracted feature data, an adaptive dynamic time window mechanism is constructed, and combined with a multi-scale causal convolutional network, the feature data is modeled at different time scales to capture the lag effect and cross-time dependency of the feature data. The extracted time dynamic features include:

[0027] S311. Based on the volatility of the feature data, an adaptive time window mechanism based on a sliding window and dynamic threshold adjustment is constructed to adjust the length and step size of the time window in real time;

[0028] S312. Using the constructed adaptive time window mechanism, segment the feature data, extract the local dynamic features within each time window, and identify the changing trend in the feature data;

[0029] S313. Combining the changing trends in feature data with a multi-scale causal convolutional network, by adjusting the convolution kernel scale and dilation rate, we can capture the lag effect and cross-temporal dependency between feature data to simulate the dynamic evolution of feature data.

[0030] S314. According to the simulation results of the dynamic evolution process, the evolution rules of the feature data at different time scales are identified, and time series feature data are extracted based on the identification results.

[0031] Optionally, based on the extracted temporal dynamic features, a constraint-based causal discovery algorithm is used to identify the potential causal structure between the feature data and cardiovascular events, and to generate an initial set of causal hypotheses including:

[0032] S321. Establish a set of conditional variables based on the extracted temporal dynamic features, and gradually expand the set of conditional variables through a causal discovery algorithm to construct an undirected skeleton graph between the conditional variables;

[0033] S322. Based on the constructed undirected skeleton graph, a separation criterion is used to identify collision structures, and based on the identified collision structures, the edge direction propagation rule is used to infer the direction of directed edges between conditional variables to determine the causal path between the temporal dynamic features;

[0034] S323. Embed the preset constraints into the causal reasoning process to eliminate unreasonable causal paths that do not meet the constraints of the loyalty assumption;

[0035] S324. Based on the elimination results, the causal paths after elimination are scored using the Bayesian Information Criterion, and causal paths with high scores are prioritized to generate an initial set of causal hypotheses that characterize the evolution of cardiovascular risk.

[0036] Optionally, the constraints include: time prior constraints, medical domain knowledge constraints, data type and variable attribute constraints, fidelity assumption constraints, and structural topology constraints.

[0037] Optionally, the calculation formula for scoring the causal path after elimination using the Bayesian Information Criterion is:

[0038]

[0039] Where BIC represents the Bayesian Information Criterion; G represents the causal path to be scored; m represents the number of temporal dynamic feature variables; i represents the index value; represents the maximum likelihood function value of the i-th node under the given parent node condition; θ i represents the parameters of the i-th node; D represents the time dynamic characteristics; d i represents the parameter dimension of the i-th node; n represents the total number of time windows of the time series.

[0040] Optionally, the fuzzy membership function and entropy weight method are integrated to adjust the nodes in the causal network, and a hierarchical monitoring strategy is generated by combining a multi-objective evolutionary algorithm, including:

[0041] S41. Based on the constructed causal relationship graph, extract the indicators of each node and construct a corresponding fuzzy membership function for each indicator;

[0042] S42. Using the constructed fuzzy membership function, calculate the membership value of each node at different levels, and obtain the fuzzy representation matrix of the node with respect to various indicators;

[0043] S43, normalizing the obtained fuzzy representation matrix, and using the entropy weight method to calculate the information entropy and corresponding weight of each indicator, to generate a weighted decision matrix that integrates the fuzzy membership and entropy weight information;

[0044] S44. Based on the information reflected in the weighted decision matrix, evaluate and determine the monitoring priority weight of each node in the causal relationship graph, optimize the causal relationship graph by adjusting the importance weight of the node, and establish a multi-objective optimization function for node monitoring;

[0045] S45. Use a multi-objective evolutionary algorithm to solve the multi-objective optimization function, obtain the Pareto optimal solution set, and obtain the hierarchical monitoring strategy corresponding to each solution.

[0046] Optionally, the formula of the fuzzy membership function is:

[0047]

[0048] Where μ A (x) represents the membership value of indicator x in level A; x represents the indicator value to be evaluated; a represents the threshold parameter of the membership function; b represents the threshold parameter of the membership function; c represents the threshold parameter of the membership function; d represents the threshold parameter of the membership function.

[0049] The beneficial effects of the present invention are:

[0050] 1. This invention collects and preprocesses multi-source cardiovascular data of patients and introduces a conditional generative adversarial network to generate high-quality and diverse synthetic data, effectively expanding the data scale and improving the balance and representativeness of data distribution, thereby improving the accuracy and robustness of disease prediction.

[0051] 2. The present invention analyzes the extracted feature data through a causal discovery algorithm and constructs a graph structure that characterizes the causal relationship between feature variables and cardiovascular events. This not only helps to reveal the potential mechanism of disease evolution, but also provides scientific support and an explainable theoretical basis for formulating personalized diagnosis and treatment plans.

[0052] 3. The present invention combines the fuzzy membership function with the entropy weight algorithm to perform dynamic quantitative evaluation of the nodes in the causal graph, and adopts a multi-objective evolutionary algorithm to generate a personalized hierarchical monitoring strategy, thereby fully considering the differences between individual patients and improving the flexibility and scientific nature of the monitoring plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0054] Figure 1 The present invention is a flowchart of a cardiovascular data monitoring method in cardiology according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] To further illustrate each embodiment, the present invention provides drawings, which are part of the disclosure of the present invention. They are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. By referring to these contents, ordinary technicians in this field should be able to understand other possible implementation methods and the advantages of the present invention.

[0056] According to an embodiment of the present invention, a cardiovascular data monitoring method for cardiology department is provided.

[0057] The present invention will now be further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to the cardiovascular data monitoring method of the cardiology department of an embodiment of the present invention, the method includes:

[0058] S1. Collect and preprocess multi-source cardiovascular data of patients, and establish a cardiovascular dataset by combining it with a conditional generative adversarial network.

[0059] In this optional embodiment, collecting and preprocessing multi-source cardiovascular data of a patient and establishing a cardiovascular dataset in combination with a conditional generative adversarial network include:

[0060] S11. Collect multi-source cardiovascular data of the patient and use time synchronization technology and feature alignment technology to align the multi-source cardiovascular data in time and feature dimensions;

[0061] S12, using a bandpass filter to denoise the aligned multi-source cardiovascular data;

[0062] S13. Based on the denoised multi-source cardiovascular data, construct a conditional vector and convert the format of the conditional vector using embedded coding technology;

[0063] S14. Use the format-converted conditional vector as input and use the conditional generative adversarial network to generate a cardiovascular dataset.

[0064] It should be noted that the specific examples of collecting and preprocessing multi-source cardiovascular data of patients and establishing a cardiovascular dataset by combining a conditional generative adversarial network are as follows:

[0065] First, multi-source cardiovascular data (such as electrocardiogram (ECG) signals and blood pressure monitoring data) are time-aligned through a timestamp synchronization strategy and a sliding window method to eliminate time deviations between devices (for example, the -9.8-second deviation correction between smart bracelets and PSG devices), and feature dimension alignment is achieved by combining interpolation. A 0.05-40 Hz bandpass filter is then used to remove baseline drift and electromyographic interference, improving the signal-to-noise ratio by approximately 20%. Embedded coding technology is then used to convert the denoised multimodal data (such as heart rate and voltage timing features) into conditional vectors, incorporating individual patient information. Finally, a conditional generative adversarial network (such as a structure that integrates an autocorrelation residual module and a spatiotemporal convolutional network) is input to generate a balanced data set containing abnormal heart rates, which increases the accuracy of the cardiovascular disease diagnosis model to 69.3%, while reducing data acquisition costs and enhancing model generalization capabilities.

[0066] S2. Based on the established cardiovascular dataset, a cardiovascular network framework is constructed, and the cardiovascular network parameters are optimized using a meta-learning algorithm to capture the characteristic data in the cardiovascular dataset.

[0067] In this optional embodiment, a cardiovascular network framework is constructed based on the established cardiovascular dataset, and a meta-learning algorithm is used to optimize cardiovascular network parameters to capture characteristic data in the cardiovascular dataset, including:

[0068] S21. Based on the generated cardiovascular dataset, build a cardiovascular feature modeling network that integrates long short-term memory network and attention mechanism, and initialize the cardiovascular network parameters;

[0069] S22. Combine meta-learning algorithms to build a multi-task meta-learning framework, and use the multi-task meta-learning framework to perform gradient updates and parameter aggregation for each task to optimize cardiovascular network parameters;

[0070] S23. Based on the optimized cardiovascular network parameters, a deep learning algorithm is used to capture the characteristic data in the cardiovascular dataset.

[0071] It should be noted that the specific examples of constructing a cardiovascular network framework based on the established cardiovascular dataset and optimizing cardiovascular network parameters using a meta-learning algorithm to capture the characteristic data in the cardiovascular dataset are as follows:

[0072] First, a feature modeling network integrating LSTM and self-attention mechanisms was constructed based on a generated cardiovascular dataset (e.g., an LSTM layer with a time step of 256 was used to extract temporal dependencies of blood pressure, combined with a multi-head self-attention mechanism to focus on key physiological features). The network parameters were set using the Xavier initialization strategy. Next, a meta-gradient-based multi-task learning framework (referring to the Task Relation Network (TRN) architecture) was employed. A dynamic task scheduler was used to assign weights to six subtasks, including coronary artery disease, hypertension, and heart failure (e.g., a weight of 0.35 for myocardial ischemia and 0.28 for arrhythmia). Second-order gradient optimization was used to achieve cross-task parameter aggregation, accelerating model convergence by 40% after 100 iterations. Finally, the optimized network captured multidimensional features (e.g., the accuracy of identifying ST-segment deviations on electrocardiograms increased to 89%, and the Dice coefficient for plaque segmentation on ultrasound images reached 85.7%), enabling cross-modal data association modeling. This model reduced the error in the Framingham risk score by 19.3% compared to traditional single-task models, and improved generalization for sample-scarce conditions (e.g., Brugada syndrome) by 32%.

[0073] S3. Use causal discovery algorithms to analyze the causal relationship between characteristic data and cardiovascular events, and construct a causal relationship diagram based on the analysis results.

[0074] In this optional embodiment, using a causal discovery algorithm to analyze the causal relationship between the characteristic data and cardiovascular events, and constructing a causal relationship graph based on the analysis results includes:

[0075] S31. Based on the extracted feature data, an adaptive dynamic time window mechanism is constructed, and combined with a multi-scale causal convolutional network, the feature data is modeled at different time scales to capture the lag effect and cross-time dependency of the feature data and extract temporal dynamic features.

[0076] In this optional embodiment, an adaptive dynamic time window mechanism is constructed based on the extracted feature data, and a multi-scale causal convolutional network is combined to model the feature data at different time scales to capture the lag effect and cross-time dependency of the feature data. Extracting temporal dynamic features includes:

[0077] S311. Based on the volatility of the feature data, an adaptive time window mechanism based on a sliding window and dynamic threshold adjustment is constructed to adjust the length and step size of the time window in real time;

[0078] S312. Using the constructed adaptive time window mechanism, segment the feature data, extract the local dynamic features within each time window, and identify the changing trend in the feature data;

[0079] S313. Combining the changing trends in feature data with a multi-scale causal convolutional network, by adjusting the convolution kernel scale and dilation rate, we can capture the lag effect and cross-temporal dependency between feature data to simulate the dynamic evolution of feature data.

[0080] S314. According to the simulation results of the dynamic evolution process, the evolution rules of the feature data at different time scales are identified, and time series feature data are extracted based on the identification results.

[0081] S32. Based on the extracted temporal dynamic features, a constraint-based causal discovery algorithm is used to identify the potential causal structure between the feature data and cardiovascular events, and generate an initial set of causal hypotheses.

[0082] In this optional embodiment, based on the extracted temporal dynamic features, a constraint-based causal discovery algorithm is used to identify the potential causal structure between the feature data and cardiovascular events, and to generate an initial set of causal hypotheses including:

[0083] S321. Based on the extracted temporal dynamic features, a set of conditional variables is established, and the set of conditional variables is gradually expanded through a causal discovery algorithm to construct an undirected skeleton graph between the conditional variables.

[0084] S322. According to the constructed undirected skeleton graph, the collision structure is identified using the separation criterion, and based on the identified collision structure, the directed edge direction between the conditional variables is inferred using the edge direction propagation rule to determine the causal path between the temporal dynamic features.

[0085] S323. Embed the preset constraints into the causal reasoning process to eliminate unreasonable causal paths that do not meet the constraints of the loyalty assumption.

[0086] In this optional embodiment, the constraints include: time prior constraints, medical domain knowledge constraints, data type and variable attribute constraints, fidelity assumption constraints, and structural topology constraints.

[0087] S324. Based on the elimination results, the causal paths after elimination are scored using the Bayesian Information Criterion, and causal paths with high scores are prioritized to generate an initial set of causal hypotheses that characterize the evolution of cardiovascular risk.

[0088] In this optional embodiment, the calculation formula for scoring the causal paths after elimination using the Bayesian Information Criterion is:

[0089]

[0090] Where BIC represents the Bayesian Information Criterion; G represents the causal path to be scored; m represents the number of temporal dynamic feature variables; i represents the index value; represents the maximum likelihood function value of the i-th node under the given parent node condition; θ i represents the parameters of the i-th node; D represents the time dynamic characteristics; d i represents the parameter dimension of the i-th node; n represents the total number of time windows of the time series.

[0091] S33. Based on the generated initial causal hypothesis set, a hybrid verification framework including statistical testing and logical reasoning is constructed, and combined with the conditional independence test algorithm, the initial causal hypotheses are gradually verified and screened to generate a set of candidate causal relationships.

[0092] S34. Based on the generated set of candidate causal relationships, a causal relationship graph representing the evolution process of cardiovascular risk is constructed with feature data as graph nodes and causal relationships as directed edges.

[0093] It should be noted that the specific embodiment of using the causal discovery algorithm to analyze the causal relationship between characteristic data and cardiovascular events and constructing a causal relationship graph based on the analysis results is as follows:

[0094] First, a window adjustment strategy is constructed based on the dynamic time window algorithm (such as the dynamic sequence generation principle of combining Flink's adaptive window mechanism with Hive's dynamic time window). The window span (30 seconds to 5 minutes) is dynamically adjusted according to the characteristics of the cardiovascular data stream (such as the 0.05-40Hz frequency band fluctuation of heart rate variability). The multi-scale causal convolutional network is integrated (referring to the 1×U one-dimensional convolution and V×W two-dimensional convolution parallel structure of the multi-scale CNN). Through three convolutional layers (kernel sizes are 5×5, 3×3, and 2×2, respectively), minute-level, hour-level, and day-level time scale features are extracted, which increases the accuracy of capturing the lag effect of ST segment deviation features to 91%. Then, the PC algorithm (based on conditional independence test and V-structure directional rule) is used to dynamically adjust the window span (30 seconds to 5 minutes) according to the characteristics of the cardiovascular data stream (such as the 0.05-40Hz frequency band fluctuation of heart rate variability). The method identified key causal paths (such as low-density lipoprotein → arterial plaque thickness → myocardial ischemia) from the causal discovery of 32-dimensional features such as lipids, generating a causal network containing 78 initial hypotheses. A hybrid verification framework was then constructed (integrating the p-value threshold of 0.01 for hypothesis testing and logical constraints). Through Markov blanket feature screening and post-nonlinear model verification, 32 spurious correlation paths were eliminated, bringing the F1 value of the causal relationship to 85.4%. The final cardiovascular risk evolution graph contained 45 nodes and 112 directed edges, successfully revealing the key evolutionary chain of morning blood pressure surge → left ventricular hypertrophy → heart failure. Compared with traditional methods, the AUC in Framingham risk prediction was improved by 17.2%, and the sensitivity for early warning of asymptomatic lesions was increased to 89.3%.

[0095] S4. Fuzzy membership function and entropy weight algorithm are integrated to adjust the nodes in the causal relationship graph, and a hierarchical monitoring strategy is generated by combining the multi-objective evolutionary algorithm.

[0096] In this optional embodiment, the fuzzy membership function and the entropy weight method are integrated to adjust the nodes in the causal network, and a hierarchical monitoring strategy is generated in combination with a multi-objective evolutionary algorithm, including:

[0097] S41. Based on the constructed causal relationship graph, extract the indicators of each node and construct a corresponding fuzzy membership function for each indicator.

[0098] In this optional embodiment, the formula of the fuzzy membership function is:

[0099]

[0100] Where μ A (x) represents the membership value of indicator x in level A; x represents the indicator value to be evaluated; a represents the threshold parameter of the membership function; b represents the threshold parameter of the membership function; c represents the threshold parameter of the membership function; d represents the threshold parameter of the membership function.

[0101] S42. Using the constructed fuzzy membership function, calculate the membership value of each node at different levels and obtain the fuzzy representation matrix of the node with respect to various indicators.

[0102] S43. Normalize the obtained fuzzy representation matrix, and use the entropy weight method to calculate the information entropy and corresponding weight of each indicator to generate a weighted decision matrix that integrates the fuzzy membership and entropy weight information.

[0103] S44. Based on the information reflected in the weighted decision matrix, the monitoring priority weight of each node in the causal relationship diagram is evaluated and determined, the causal relationship diagram is optimized by adjusting the importance weight of the node, and a multi-objective optimization function for node monitoring is established.

[0104] S45. Use a multi-objective evolutionary algorithm to solve the multi-objective optimization function, obtain the Pareto optimal solution set, and obtain the hierarchical monitoring strategy corresponding to each solution.

[0105] It should be noted that the specific embodiment of integrating the fuzzy membership function and the entropy weight algorithm, adjusting the nodes in the causal relationship graph, and combining the multi-objective evolutionary algorithm to generate a hierarchical monitoring strategy is as follows:

[0106] First, based on the cardiovascular risk evolution causal relationship graph (containing 45 nodes and 112 causal edges), key indicators (such as the frequency of sudden increases in blood pressure and left ventricular wall thickness) were extracted, and a fuzzy membership function was constructed by combining the trapezoidal membership function and the expert experience method (for example, the morning systolic blood pressure of 130-160 mmHg corresponds to the "high-risk" level with membership parameters a=130, b=145, and c=160); then, the fuzzy statistical method was used to calculate the membership of the node indicators under the five risk levels, generating a 32-dimensional × 5-level fuzzy representation matrix (for example, the arterial plaque thickness indicator has a membership of 0.87 at the "very high-risk" level); then, the range method was used to normalize the matrix, and the entropy weight method was used to calculate the matrix. Information entropy (such as the entropy value of blood lipid indicators of 0.32) and weights (such as the weight of low-density lipoprotein of 0.18) were used to form a weighted decision matrix that integrated the risk level membership and indicator importance. A multi-objective optimization function was constructed based on the matrix data (the objectives included monitoring cost ≤ 3,000 yuan / month and risk coverage ≥ 95%). The Pareto front was generated through 200 iterations of the NSGA-II algorithm (the HV value increased by 19.3%), and 16 groups of optimal solutions including hierarchical monitoring strategies were screened out (such as optimizing the dynamic monitoring frequency of high-risk nodes to 15 minutes / time), which increased the sensitivity of asymptomatic lesion warning to 89.7% and reduced the consumption of redundant monitoring resources by 35%.

[0107] In summary, with the help of the above technical solutions of the present invention, by collecting and preprocessing the multi-source cardiovascular data of patients and introducing conditional generative adversarial networks to generate high-quality and diverse synthetic data, the data scale is effectively expanded, the balance and representativeness of the data distribution are improved, and thus the accuracy and robustness of disease prediction are improved. By analyzing the extracted feature data through the causal discovery algorithm and constructing a graph structure that characterizes the causal relationship between feature variables and cardiovascular events, it not only helps to reveal the potential mechanism of disease evolution, but also provides scientific support and an explainable theoretical basis for the formulation of personalized diagnosis and treatment plans. By combining the fuzzy membership function with the entropy weight algorithm, the nodes in the causal graph are dynamically quantitatively evaluated, and a multi-objective evolutionary algorithm is used to generate a personalized hierarchical monitoring strategy, thereby fully considering the differences between individual patients and improving the flexibility and scientificity of the monitoring plan.

[0108] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A cardiovascular data monitoring method for cardiology, characterized in that: The method includes: S1. Collect and preprocess multi-source cardiovascular data of patients and build a cardiovascular dataset using a conditional generative adversarial network. S2. Based on the established cardiovascular dataset, a cardiovascular network framework is constructed, and the cardiovascular network parameters are optimized using a meta-learning algorithm to capture the characteristic data in the cardiovascular dataset; S3. Use causal discovery algorithms to analyze the causal relationship between feature data and cardiovascular events, and construct a causal relationship diagram based on the analysis results; S4. Fuzzy membership function and entropy weight algorithm are integrated to adjust the nodes in the causal relationship graph, and a hierarchical monitoring strategy is generated by combining the multi-objective evolutionary algorithm.

2. A cardiovascular data monitoring method for cardiology according to claim 1, characterized in that: The acquisition and preprocessing of multi-source cardiovascular data of patients and the establishment of a cardiovascular dataset in combination with a conditional generative adversarial network include: S11. Collect multi-source cardiovascular data of the patient and use time synchronization technology and feature alignment technology to align the multi-source cardiovascular data in time and feature dimensions; S12, using a bandpass filter to denoise the aligned multi-source cardiovascular data; S13. Based on the denoised multi-source cardiovascular data, construct a conditional vector and convert the format of the conditional vector using embedded coding technology; S14. Use the format-converted conditional vector as input and use the conditional generative adversarial network to generate a cardiovascular dataset.

3. A cardiovascular data monitoring method for cardiology according to claim 1, characterized in that: The cardiovascular network framework is constructed based on the established cardiovascular dataset, and the cardiovascular network parameters are optimized using a meta-learning algorithm to capture the characteristic data in the cardiovascular dataset, including: S21. Based on the generated cardiovascular dataset, build a cardiovascular feature modeling network that integrates long short-term memory network and attention mechanism, and initialize the cardiovascular network parameters; S22. Combine meta-learning algorithms to build a multi-task meta-learning framework, and use the multi-task meta-learning framework to perform gradient updates and parameter aggregation for each task to optimize cardiovascular network parameters; S23. Based on the optimized cardiovascular network parameters, a deep learning algorithm is used to capture the characteristic data in the cardiovascular dataset.

4. A cardiovascular data monitoring method for cardiology according to claim 1, characterized in that: The method of using a causal discovery algorithm to analyze the causal relationship between characteristic data and cardiovascular events and constructing a causal relationship diagram based on the analysis results includes: S31. Based on the extracted feature data, an adaptive dynamic time window mechanism is constructed. Combined with a multi-scale causal convolutional network, the feature data is modeled at different time scales to capture the lag effect and cross-temporal dependency of the feature data and extract temporal dynamic features. S32. Based on the extracted temporal dynamic features, a constraint-based causal discovery algorithm is used to identify the potential causal structure between the feature data and cardiovascular events, and to generate an initial set of causal hypotheses; S33. Based on the generated initial causal hypothesis set, a hybrid verification framework including statistical testing and logical reasoning is constructed. In combination with the conditional independence test algorithm, the initial causal hypotheses are gradually verified and screened to generate a candidate causal relationship set. S34. Based on the generated set of candidate causal relationships, a causal relationship graph representing the evolution process of cardiovascular risk is constructed with feature data as graph nodes and causal relationships as directed edges.

5. A cardiovascular data monitoring method for cardiology according to claim 4, characterized in that: Based on the extracted feature data, an adaptive dynamic time window mechanism is constructed, and combined with a multi-scale causal convolutional network, the feature data is modeled at different time scales to capture the lag effect and cross-time dependency of the feature data. The extracted time dynamic features include: S311. Based on the volatility of the feature data, an adaptive time window mechanism based on a sliding window and dynamic threshold adjustment is constructed to adjust the length and step size of the time window in real time; S312. Using the constructed adaptive time window mechanism, segment the feature data, extract the local dynamic features within each time window, and identify the changing trend in the feature data; S313. Combining the changing trends in feature data with a multi-scale causal convolutional network, by adjusting the convolution kernel scale and dilation rate, we can capture the lag effect and cross-temporal dependency between feature data to simulate the dynamic evolution of feature data. S314. According to the simulation results of the dynamic evolution process, the evolution rules of the feature data at different time scales are identified, and time series feature data are extracted based on the identification results.

6. A cardiovascular data monitoring method for cardiology according to claim 5, characterized in that: The method uses a constraint-based causal discovery algorithm based on the extracted temporal dynamic features to identify the potential causal structure between the feature data and cardiovascular events, and generates an initial set of causal hypotheses including: S321. Establish a set of conditional variables based on the extracted temporal dynamic features, and gradually expand the set of conditional variables through a causal discovery algorithm to construct an undirected skeleton graph between the conditional variables; S322. Based on the constructed undirected skeleton graph, a separation criterion is used to identify collision structures, and based on the identified collision structures, the edge direction propagation rule is used to infer the direction of directed edges between conditional variables to determine the causal path between the temporal dynamic features; S323. Embed the preset constraints into the causal reasoning process to eliminate unreasonable causal paths that do not meet the constraints of the loyalty assumption; S324. Based on the elimination results, the causal paths after elimination are scored using the Bayesian Information Criterion, and causal paths with high scores are prioritized to generate an initial set of causal hypotheses that characterize the evolution of cardiovascular risk.

7. A cardiovascular data monitoring method for cardiology according to claim 6, characterized in that: The constraints include: time prior constraints, medical domain knowledge constraints, data type and variable attribute constraints, loyalty assumption constraints and structural topology constraints.

8. A cardiovascular data monitoring method for cardiology according to claim 7, characterized in that: The calculation formula for scoring the causal path after elimination using the Bayesian information criterion is: Where BIC represents the Bayesian Information Criterion; G represents the causal path to be scored; m represents the number of time-dynamic feature variables to be evaluated; i represents the index value; represents the maximum likelihood function value of the i-th node under the given parent node condition; θ i represents the parameters of the i-th node; D represents the time dynamic characteristics; d i represents the parameter dimension of the i-th node; n represents the total number of time windows of the time series.

9. A cardiovascular data monitoring method for cardiology according to claim 1, characterized in that: The fusion of fuzzy membership function and entropy weight method, adjusting the nodes in the causal network, and combining the multi-objective evolutionary algorithm to generate a hierarchical monitoring strategy include: S41. Based on the constructed causal relationship graph, extract the indicators of each node and construct a corresponding fuzzy membership function for each indicator; S42. Using the constructed fuzzy membership function, calculate the membership value of each node at different levels, and obtain the fuzzy representation matrix of the node with respect to various indicators; S43, normalizing the obtained fuzzy representation matrix, and using the entropy weight method to calculate the information entropy and corresponding weight of each indicator, to generate a weighted decision matrix that integrates the fuzzy membership and entropy weight information; S44. Based on the information reflected in the weighted decision matrix, evaluate and determine the monitoring priority weight of each node in the causal relationship graph, optimize the causal relationship graph by adjusting the importance weight of the node, and establish a multi-objective optimization function for node monitoring; S45. Use a multi-objective evolutionary algorithm to solve the multi-objective optimization function, obtain the Pareto optimal solution set, and obtain the hierarchical monitoring strategy corresponding to each solution.

10. A cardiovascular data monitoring method for cardiology according to claim 9, characterized in that: The formula of the fuzzy membership function is: Where μ A (x) represents the membership value of indicator x in level A; x represents the indicator value to be evaluated; a represents the threshold parameter of the membership function; b represents the threshold parameter of the membership function; c represents the threshold parameter of the membership function; d represents the threshold parameter of the membership function.

Images