Method and system for predicting risk of cardiovascular and cerebrovascular diseases

Abstract

The invention provides a cardiovascular and cerebrovascular disease risk prediction method and system, and relates to the technical field of data processing.The method comprises the steps that real-time physiological parameters of a target user are obtained, and the real-time physiological parameters comprise the ambulatory blood pressure fluctuation rate, the serum lipoprotein level, the heart rate variability frequency domain index and the sleep apnea hypopnea index; the method comprises the following steps: preprocessing real-time physiological parameters, extracting instantaneous waveform features of blood pressure signals by adopting wavelet transform, and extracting features related to health conditions, including blood pressure level, blood fat level, blood sugar level, electrocardiogram abnormal indexes and cardiac ultrasound abnormal indexes, to form a feature vector set; according to the method, the real-time physiological parameters are acquired, the feature vector set is formed through preprocessing, the krill swarm algorithm is used for optimization, the three-level fusion model is constructed, finally, the cardiovascular and cerebrovascular disease risk degree of the target user is accurately evaluated, early warning information is output, and the accuracy and timeliness of cardiovascular and cerebrovascular disease risk prediction are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a method and system for predicting the risk of cardiovascular and cerebrovascular diseases. Background Art

[0002] The monitoring of cardiovascular and cerebrovascular diseases depends on limited physiological parameters for risk prediction, such as static indicators like blood pressure, blood lipid, and blood glucose. However, these indicators can only reflect the physical condition at a certain moment and are difficult to comprehensively and dynamically reflect the risk of cardiovascular and cerebrovascular diseases.

[0003] For example, only focusing on the absolute value of blood pressure while ignoring the dynamic fluctuations of blood pressure, such as the circadian rhythm change of blood pressure and the instantaneous volatility of blood pressure, these dynamic information is also important for evaluating the risk of cardiovascular and cerebrovascular diseases.

[0004] In addition, traditional technologies mainly detect the overall blood lipid levels such as total serum cholesterol and triglycerides, and cannot accurately reflect the levels of different lipoprotein subclasses, such as low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), etc. These lipoprotein subclasses are closely related to the occurrence risk of cardiovascular and cerebrovascular diseases. Only focusing on some single features, such as blood pressure and blood lipid levels, while ignoring the interaction and influence between these features. For example, there may be a complex interaction between blood pressure and blood lipid levels, jointly affecting the risk of cardiovascular and cerebrovascular diseases. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method and system for predicting the risk of cardiovascular and cerebrovascular diseases, providing a basis for the early prediction of cardiovascular and cerebrovascular diseases.

[0006] To solve the above technical problem, the technical solution of the present invention is as follows:

[0007] In the first aspect, a method for predicting the risk of cardiovascular and cerebrovascular diseases, the method includes:

[0008] Obtain the real-time physiological parameters of the target user, including dynamic blood pressure volatility, serum lipoprotein level, heart rate variability frequency domain index, and sleep apnea hypopnea index;

[0009] Preprocess the real-time physiological parameters, extract the instantaneous waveform features of the blood pressure signal by wavelet transform, and extract the features related to the health status, including blood pressure level, blood lipid level, blood glucose level, electrocardiogram abnormality index, and cardiac ultrasound abnormality index, to form a feature vector set;

[0010] Use the krill herd algorithm to perform feature selection and optimization on the feature vector set, and automatically search for the key feature subset for evaluating the cardiovascular health status by simulating the foraging behavior of the krill herd to obtain an optimized feature vector set;

[0011] According to the optimized feature vector set, a three-level fusion model including spatio-temporal coding of physiological signals, embedding of clinical knowledge graphs, and dynamic reinforcement learning is constructed through a probabilistic timed Petri net;

[0012] Input the feature vector of the target user into the three-level fusion model. According to the inference rules of the Petri net, obtain the risk degree of the target user's health status, and output the risk assessment result and generate a warning message according to the risk degree of the target user's health status.

[0013] Furthermore, preprocess the real-time physiological parameters, use wavelet transform to extract the instantaneous waveform features of blood pressure signals, and extract the features related to health status, including blood pressure level, blood lipid level, blood glucose level, electrocardiogram abnormality index, and cardiac ultrasound abnormality index, to form a feature vector set, including:

[0014] Use the wavelet transform method to extract the instantaneous waveform features of blood pressure signals to obtain the dynamic features of blood pressure changes;

[0015] Extract the features related to health status from the data processed by wavelet transform, including blood pressure level, blood lipid level, blood glucose level, electrocardiogram abnormality index, and cardiac ultrasound abnormality index, to form a feature vector set.

[0016] Furthermore, use the krill herd algorithm to perform feature selection and optimization on the feature vector set. By simulating the foraging behavior of the krill herd, automatically search for the key feature subset for cardiovascular health status assessment to obtain the optimized feature vector set, including:

[0017] Determine the size of the krill herd, that is, the number of candidate features participating in feature selection, and assign an initial position to each krill;

[0018] Set the perception range, movement step size, maximum number of iterations of the krill, and define a contribution degree function to evaluate the contribution degree of the current position of the krill to cardiovascular health status assessment;

[0019] For each krill, extract the corresponding features according to the current position to construct a prediction model;

[0020] Use the sample data related to cardiovascular health status to train the prediction model, and calculate the contribution degree value of each krill according to the contribution degree function;

[0021] Assign the contribution degree value to the corresponding krill as the performance evaluation in the current iteration, and for each krill, calculate the neighbor set within the perception range, and determine the learning object of the current krill in the neighbor set;

[0022] Update the current position of the krill according to the learning object, the position difference between the current krill, and the preset movement step size.

[0023] Repeat the processes of feature extraction, model construction, training, and contribution degree calculation until the preset maximum number of iterations is reached, and finally obtain a set of optimized feature combinations.

[0024] According to the optimized feature combination, extract the corresponding features from the original feature vector set to form an optimized feature vector set.

[0025] Furthermore, according to the contribution degree function, calculate the contribution degree of the current position of the krill to the assessment of cardiovascular health status, including:

[0026] When constructing a prediction model with the feature subset corresponding to the current position of the krill, calculate the number of samples correctly predicted as positive examples and the number of samples correctly predicted as negative examples, as well as the number of samples incorrectly predicted as positive examples and the number of samples incorrectly predicted as negative examples; calculate the accuracy index value according to the number of positive examples, the number of negative examples, the number of positive examples, and the number of negative examples;

[0027] Determine the size of the feature subset corresponding to the current position of the krill, that is, the number of features included in the feature subset, and calculate the feature subset factor index value according to the number of features included in the feature subset;

[0028] Calculate the entropy of the entire data set and the entropy of the data subsets corresponding to different values of each feature. For each feature, calculate the weighted average of the entropies of the data subsets corresponding to all its values; calculate the difference between the entropy of the entire data set and the weighted average, and calculate the ratio of the difference to the maximum value of the information gain among all features; calculate the information gain influence value according to the ratio;

[0029] Fuse the accuracy index value, the feature subset factor index value, and the information gain influence value to obtain the contribution degree of the current position of the krill to the assessment of cardiovascular health status.

[0030] Furthermore, according to the optimized feature vector set, construct a three-level fusion model including physiological signal spatio-temporal coding, clinical knowledge graph embedding, and dynamic reinforcement learning through a probabilistic timed Petri net, including:

[0031] Initialize the construction environment of the probabilistic timed Petri net, including defining the state, transitions, arcs, related probabilities, and time parameters of the net;

[0032] Encode and fuse the blood pressure waveform features and heart rate variability physiological signal data in the optimized feature vector set to form an intermediate feature representation including the dynamic changes, spatial relationships, and clinical knowledge of physiological signals;

[0033] Through a dynamic reinforcement learning algorithm, the intermediate feature representation containing the dynamic changes of physiological signals, spatial relationships, and clinical knowledge is used as input data for learning and optimization. After multiple iterative trainings, a three-level fusion model containing physiological signal spatio-temporal encoding, clinical knowledge graph embedding, and dynamic reinforcement learning is formed.

[0034] Furthermore, the three-level fusion model includes:

[0035] The first level is the physiological signal spatio-temporal encoding module, which performs spatio-temporal encoding on the physiological signals in the optimized feature vector set to obtain the temporal relationship and spatial features between the signals;

[0036] The second level is the clinical knowledge graph embedding module, which fuses clinical knowledge with feature vectors and transforms clinical knowledge into structured information;

[0037] The third level is the dynamic reinforcement learning module, which performs dynamic learning based on the outputs of the physiological signal spatio-temporal encoding module and the clinical knowledge graph embedding module, and finally outputs the health status risk degree of the target user.

[0038] Furthermore, the feature vector of the target user is input into the three-level fusion model. According to the inference rules of Petri nets, the health status risk degree of the target user is obtained, and based on the health status risk degree of the target user, a risk assessment result and a warning message are output, including:

[0039] The feature vector of the target user is used as input data and transmitted to the first level of the three-level fusion model, that is, the physiological signal spatio-temporal encoding module;

[0040] In the physiological signal spatio-temporal encoding module, the temporal relationship and spatial features between physiological signals are extracted to form an intermediate feature representation;

[0041] The intermediate feature representation is transmitted to the second level, that is, the clinical knowledge graph embedding module, where clinical knowledge is fused with the intermediate feature representation to generate a fused feature representation;

[0042] The fused feature representation is transmitted to the third level, that is, the dynamic reinforcement learning module, which performs dynamic learning and optimization based on the fused feature representation. Through continuous adjustment, the health status risk degree of the target user is output;

[0043] According to the health status and risk degree of the target user, a corresponding risk assessment report is generated, including risk level, risk factor analysis, and recommended measures;

[0044] Set a warning threshold according to the health status and risk level of the target user, compare the health status and risk level of the target user with the warning threshold, and when the health status and risk level of the target user ≥ the warning threshold, automatically generate a warning message according to the preset warning rules, including the risk level, the main risk factors, and the recommended emergency actions.

[0045] Further, transmit the fused feature representation to the third level, that is, the dynamic reinforcement learning module, perform dynamic learning and optimization according to the fused feature representation, and through continuous adjustment, output the health status risk level of the target user, including:

[0046] Initialize the dynamic reinforcement learning model, including setting the structure, parameters, and learning rate of the dynamic reinforcement learning model;

[0047] The dynamic reinforcement learning model receives the fused feature representation and performs dynamic learning according to the preset learning algorithm;

[0048] In each stage of learning, predict the health status risk level of a target user according to the current feature representation, compare it with the actual health status risk level, and calculate the difference between the prediction result and the actual result;

[0049] Repeat the learning and optimization, and when the preset learning stop condition is reached, output the health status risk level of the target user.

[0050] In the second aspect, a cardiovascular disease risk prediction system, including:

[0051] A data acquisition module for acquiring the physiological signal data of the target user;

[0052] A feature extraction module for extracting features from the physiological signal data and forming a feature vector set;

[0053] A feature selection and optimization module for using the krill herd algorithm to perform feature selection and optimization on the feature vector set, and automatically searching for and determining the feature subset that is crucial for cardiovascular health status assessment;

[0054] A three-level fusion module for performing deep-level and multi-dimensional fusion processing on the optimized feature vector set to obtain a fused feature representation;

[0055] A risk prediction module for predicting the health status risk level of the target user according to the feature representation output by the three-level fusion module, obtaining a risk prediction result, and automatically generating a warning message when the health status risk level of the target user exceeds the preset threshold.

[0056] In a third aspect, a computer-readable storage medium stores a program which, when executed by a processor, implements the method described above.

[0057] The above solution of the present invention has at least the following beneficial effects:

[0058] By using wavelet transform to extract the instantaneous waveform features of blood pressure signals, it is possible to capture the subtle and important change information in the blood pressure signals, and at the same time extract various features related to health conditions, such as blood pressure level, blood lipid level, blood glucose level, electrocardiogram abnormality index, and cardiac ultrasound abnormality index, etc., to form a feature vector set. This preprocessing method not only retains the key information in the physiological parameters, but also removes noise and redundant data, improving the quality and usability of the data. Using the krill herd algorithm to perform feature selection and optimization on the feature vector set, by simulating the foraging behavior of the krill herd, automatically search for the feature subset that is crucial for the assessment of cardiovascular health conditions. This method avoids the subjectivity and limitations of manual feature selection, and can quickly and accurately find the most predictive feature combination, improving the efficiency and accuracy of feature selection.

[0059] The feature vector set optimized by the krill herd algorithm removes the features that contribute less to the assessment of cardiovascular health conditions, retains the key features, reduces the number and complexity of features, reduces the computational cost of the model, and at the same time improves the generalization ability and prediction accuracy of the model. By constructing a three-level fusion model including spatio-temporal encoding of physiological signals, clinical knowledge graph embedding, and dynamic reinforcement learning using a probabilistic timed Petri net, the spatio-temporal information of physiological signals, clinical knowledge, and the optimization ability of reinforcement learning are organically combined. Spatio-temporal encoding of physiological signals can capture the dynamic changes of physiological signals in time and space, clinical knowledge graph embedding introduces rich medical domain knowledge, and dynamic reinforcement learning improves the adaptability and accuracy of the model through continuous learning and optimization. This three-level fusion model can make full use of information from different sources to more comprehensively and accurately assess the risk degree of cardiovascular diseases of the target user. Compared with traditional models, it can better process complex physiological data and medical knowledge, improving the accuracy and reliability of risk prediction.

[0060] Input the feature vector of the target user into the three - level fusion model. According to the inference rules of Petri nets, obtain the risk level of the target user's health condition. This personalized assessment method can fully consider the unique physiological characteristics and health conditions of each user, providing more accurate and targeted risk assessment results for users. According to the risk level of the target user's health condition, output the risk assessment result and generate a warning message. When the risk level is high, issue a warning in a timely manner to remind the user to take corresponding preventive and treatment measures, which helps to detect the risk of cardiovascular and cerebrovascular diseases early, reduce the incidence and mortality of the diseases, and improve the user's health level and quality of life. This method provides a scientific and objective risk prediction tool for doctors, which can help doctors more accurately assess the patient's condition and risk, formulate more reasonable treatment plans, and improve the clinical treatment effect. The research and application of this method contribute to a deeper understanding of the pathogenesis and influencing factors of cardiovascular and cerebrovascular diseases, providing new ideas and methods for medical research and promoting the development of the field of prevention and treatment of cardiovascular and cerebrovascular diseases. Brief Description of the Drawings

[0061] Figure 1 It is a schematic flowchart of the method for predicting the risk of cardiovascular and cerebrovascular diseases provided by the embodiment of the present invention.

[0062] Figure 2 It is a schematic diagram of the system for predicting the risk of cardiovascular and cerebrovascular diseases provided by the embodiment of the present invention. Detailed Embodiment

[0063] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be completely conveyed to those skilled in the art.

[0064] As Figure 1 shown, the embodiment of the present invention proposes a method for predicting the risk of cardiovascular and cerebrovascular diseases, and the method includes the following steps:

[0065] Step 1, obtain the real - time physiological parameters of the target user, including dynamic blood pressure volatility, serum lipoprotein level, heart rate variability frequency - domain index, and sleep apnea - hypopnea index;

[0066] Step 2, pre - process the real - time physiological parameters, use wavelet transform to extract the instantaneous waveform features of the blood pressure signal, and extract the features related to the health condition, including blood pressure level, blood lipid level, blood glucose level, electrocardiogram abnormality index, and cardiac ultrasound abnormality index, to form a feature vector set;

[0067] Step 3: Use the krill herd algorithm to perform feature selection and optimization on the feature vector set. By simulating the foraging behavior of the krill herd, automatically search for the feature subset that is crucial for the assessment of cardiovascular health status, so as to obtain the optimized feature vector set;

[0068] Step 4: According to the optimized feature vector set, construct a three-level fusion model that includes spatio-temporal coding of physiological signals, clinical knowledge graph embedding, and dynamic reinforcement learning through a probabilistic timed Petri net;

[0069] Step 5: Input the feature vector of the target user into the three-level fusion model. According to the inference rules of the Petri net, obtain the risk degree of the health status of the target user, and according to the risk degree of the health status of the target user, output the risk assessment result and generate a warning message.

[0070] In the embodiment of the present invention, by acquiring real-time physiological parameters such as dynamic blood pressure volatility, serum lipoprotein level, heart rate variability frequency domain index, and sleep apnea hypopnea index, the physiological state of the target user can be comprehensively and continuously monitored. The monitoring of real-time physiological parameters helps to timely detect abnormal changes in physiological indicators, so as to realize early warning of potential health risks and provide the possibility for timely intervention and treatment.

[0071] Step 2: Through preprocessing operations (such as denoising, normalization, etc.), the quality of physiological parameter data can be improved, and the influence of noise and interference on subsequent analysis can be reduced. Adopt methods such as wavelet transform to extract the instantaneous waveform features of blood pressure signals and features related to health status (such as blood pressure level, blood lipid level, etc.), which can more accurately depict the physiological state of the target user. Organize the extracted features into a feature vector set for subsequent feature selection and optimization as well as model construction.

[0072] Step 3: The krill herd algorithm can automatically search for the feature subset that is crucial for the assessment of cardiovascular health status by simulating the foraging behavior of the krill herd, reduce manual intervention and subjective bias, and improve the objectivity and accuracy of feature selection. Through feature selection and optimization, a set of feature combinations that contribute the most to the assessment of cardiovascular health status can be obtained, forming the optimized feature vector set.

[0073] Step 4: The three-level fusion model constructed through a probabilistic timed Petri net can fuse multi-source information such as spatio-temporal coding of physiological signals, clinical knowledge graph embedding, and dynamic reinforcement learning, and more comprehensively evaluate the health status of the target user. The three-level fusion model can comprehensively consider the dynamic changes of physiological signals, the prior information of clinical knowledge, and the dynamic adjustment of individual health status, thereby improving the prediction performance and generalization ability of the model.

[0074] Step 5: Input the feature vector of the target user into the three-level fusion model. According to the inference rules of Petri nets, the health status risk level of the target user can be obtained, providing strong support for personalized medicine. Based on the health status risk level of the target user, output the risk assessment result and generate a warning message, which helps to timely detect potential health risks and take corresponding intervention measures to reduce the incidence and mortality of diseases. By providing accurate risk assessment and warning information, it can help doctors better understand the health status of patients and formulate more reasonable treatment plans, thus improving the quality and efficiency of medical services.

[0075] In a preferred embodiment of the present invention, the above step 1 of obtaining the real-time physiological parameters of the target user, including dynamic blood pressure volatility, serum lipoprotein level, heart rate variability frequency domain index, and sleep apnea hypopnea index, may include:

[0076] In an embodiment of the present invention, medical devices are selected to monitor the physiological parameters of the target user. For example, a dynamic blood pressure monitor is used to measure blood pressure volatility, a blood analyzer is used to detect serum lipoprotein levels, an electrocardiograph is used to record heart rate variability data, and a polysomnograph is used to evaluate the sleep apnea hypopnea index. According to the operation manuals of the devices and clinical requirements, the devices are correctly configured and calibrated to ensure the accuracy and reliability of the collected data.

[0077] Connect the devices to the target user and collect data according to the operation guides of the devices. For example, attach a cuff to the blood pressure monitor and start continuous blood pressure monitoring; attach electrode patches to the electrocardiograph and start recording electrocardiogram data; let the target user wear a polysomnograph for nocturnal sleep monitoring. During the data collection process, ensure the stable operation of the devices to avoid interference and errors. At the same time, record key information such as the collection time, location, and device model. Transmit the collected physiological data to the data processing center for storage and processing in real time. Wired or wireless connection methods can be used, such as USB, Bluetooth, Wi-Fi, etc. During the data transmission process, ensure the security and privacy of the data. Use encryption technology to protect the data to prevent data leakage and tampering.

[0078] In a preferred embodiment of the present invention, the above step 2 of preprocessing the real-time physiological parameters, using wavelet transform to extract the instantaneous waveform features of the blood pressure signal, and extracting the features related to the health status, including blood pressure level, blood lipid level, blood glucose level, electrocardiogram abnormality index, and cardiac ultrasound abnormality index, to form a feature vector set, may include:

[0079] Step 221: Use the wavelet transform method to extract the instantaneous waveform features of the blood pressure signal to obtain the dynamic features of blood pressure changes;

[0080] Step 222: Extract features related to health status from the data processed by wavelet transform, including blood pressure level, blood lipid level, blood glucose level, electrocardiogram abnormality indicators, and cardiac ultrasound abnormality indicators, to form a feature vector set.

[0081] In an embodiment of the present invention, continuous blood pressure monitoring data of a target user is obtained, and the time interval and sampling rate of the data are ensured to meet the requirements of wavelet transform. The data is denoised to remove possible interference and noise, and the signal-to-noise ratio of the signal is improved. According to the characteristics of the blood pressure signal, a suitable wavelet basis function, the Daubechies wavelet, is selected. The decomposition level of wavelet transform is determined, which depends on the complexity of the signal and the required feature details. The blood pressure signal is subjected to wavelet transform to obtain the approximation coefficients and detail coefficients at different scales. The instantaneous waveform features in the detail coefficients, such as the peak value, valley value, slope, etc. of the waveform, are extracted, and the extracted instantaneous waveform features are organized into a feature vector.

[0082] Step 222: Calculate statistical features such as the average value and standard deviation of blood pressure to reflect the overall level and fluctuation of blood pressure. According to clinical standards, the blood pressure level is divided into categories such as normal, prehypertension, and hypertension. Indexes such as total cholesterol, triglyceride, high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C) are extracted from the blood analysis report. According to clinical standards, it is evaluated whether the blood lipid level is normal and whether there are risks such as hyperlipidemia. Indexes such as fasting blood glucose and postprandial blood glucose are extracted from the blood glucose monitoring data. According to clinical standards, it is evaluated whether the blood glucose level is normal and whether there are risks such as diabetes. The electrocardiogram data is automatically analyzed to identify abnormal waveforms such as arrhythmia and myocardial ischemia, and characteristic parameters such as the QRS complex and T wave of the electrocardiogram are extracted. The cardiac ultrasound images are automatically analyzed to identify abnormal cardiac structures and abnormal cardiac functions. Indexes such as left ventricular ejection fraction (LVEF) and left ventricular end-diastolic diameter (LVEDD) of the cardiac ultrasound are extracted to evaluate the cardiac health status. The extracted features are organized into a feature vector set.

[0083] Suppose a continuous blood pressure monitoring data of a target user, as well as the corresponding blood analysis report and electrocardiogram and cardiac ultrasound examination results are received. The following steps are taken for processing:

[0084] Perform wavelet transform on the blood pressure signal to extract the instantaneous waveform features of blood pressure, such as the peak value and slope of the waveform. Extract indicators such as total cholesterol, triglyceride, HDL-C, and LDL-C from the blood analysis report and evaluate whether the blood lipid level is normal. Extract indicators such as fasting blood glucose and postprandial blood glucose from the blood glucose monitoring data and evaluate whether the blood glucose level is normal. Automatically analyze the electrocardiogram data to identify abnormal waveforms such as arrhythmia and myocardial ischemia, and extract characteristic parameters such as QRS complex and T wave. Automatically analyze the cardiac ultrasound image to identify abnormal cardiac structures and abnormal cardiac functions, etc., and extract indicators such as LVEF and LVEDD. Organize the extracted features into a feature vector set.

[0085] Using methods such as wavelet transform to extract the instantaneous waveform features of physiological signals can more precisely depict the dynamic change process of physiological signals and improve the accuracy of feature extraction. Automatically extracting features related to health status, such as blood pressure level, blood lipid level, blood glucose level, etc., improves the efficiency of feature extraction. By extracting a variety of features related to health status to form a feature vector set, this helps to more accurately evaluate the health status of the target user and discover potential health risks. The health status risk prediction based on the feature vector set can provide personalized medical advice and treatment plans for each target user, which helps to promote the development of personalized medicine and improve the pertinence and effectiveness of medical services.

[0086] In a preferred embodiment of the present invention, in step 3 above, the krill herd algorithm is used to perform feature selection and optimization on the feature vector set. By simulating the foraging behavior of the krill herd, automatically search for the key feature subset for cardiovascular health status assessment to obtain the optimized feature vector set, which may include:

[0087] Step 331: Determine the size of the krill herd, that is, the number of candidate features participating in feature selection, and assign an initial position to each krill;

[0088] Step 332: Set the perception range, moving step size, and maximum number of iterations of the krill, and define a contribution degree function to evaluate the contribution degree of the current position of the krill to cardiovascular health status assessment;

[0089] Step 333: For each krill, extract the corresponding features according to the current position to construct a prediction model;

[0090] Step 334: Use the sample data related to cardiovascular health status to train the prediction model, and calculate the contribution degree value of each krill according to the contribution degree function;

[0091] Step 335: Assign the contribution degree value to the corresponding krill as the performance evaluation in the current iteration. For each krill, calculate the set of neighbors within the perception range, and determine the learning object of the current krill in the neighbor set.

[0092] Step 336: Update the position of the current krill according to the learning object, the position difference between the current krill and the preset moving step size.

[0093] Step 337: Repeat the processes of feature extraction, model construction, training, and contribution degree calculation until the preset maximum number of iterations is reached, and finally obtain a set of optimized feature combinations.

[0094] Step 338: Extract the corresponding features from the original feature vector set according to the optimized feature combination to form an optimized feature vector set.

[0095] In the embodiment of the present invention, first read the dimension of the feature vector set, which represents the number of candidate features. Determine the size of the krill swarm according to this number. For example, if the feature vector set contains 100 features, then initialize 100 krills. Each krill is assigned an initial position in the feature space, which can be a random vector representing a point in the feature space. The selection of the initial position can be a uniform random distribution to ensure that the krill swarm widely explores in the feature space.

[0096] Step 332: Set the perception range (such as 5 unit distances), moving step size (such as 0.1 unit distances), and maximum number of iterations (such as 100 times) for the krill swarm. These parameters determine the exploration ability and search efficiency of the krill in the feature space. Define a function to evaluate the contribution degree of the current position of the krill to the assessment of cardiovascular health status, and this function can be constructed based on the accuracy of the prediction model.

[0097] Step 333: For each krill, extract the corresponding feature subset from the feature vector set according to its current position. For example, if the feature indices corresponding to the current position of the krill are [2, 5, 7], then extract these three features. Use the extracted feature subset to construct a prediction model for predicting cardiovascular health status.

[0098] Step 334: Use the sample data related to cardiovascular health status to train the prediction model. The training data should include feature vectors and corresponding health status labels. According to the performance of the trained model, use the contribution degree function to calculate the contribution degree value of each krill. This value reflects the importance of the current position of the krill (i.e., the feature subset) to the assessment of cardiovascular health status.

[0099] Step 335: Assign the contribution degree value to the corresponding krill as its performance evaluation in the current iteration. For each krill, calculate the set of neighbors within its perception range. The krill in the neighbor set are the objects that the current krill can "learn" from. When calculating the neighbor set, the distance between krill (such as Euclidean distance) and the perception range can be considered.

[0100] Step 336: In the neighbor set, select the neighbor with the highest contribution degree as the learning object of the current krill. Update the position of the current krill according to the position difference between the learning object and the current krill and the preset moving step size. The update formula can be expressed as: Repeat the process of steps 333 to 336 until the preset maximum number of iterations is reached. In each iteration, the krill swarm gradually converges to a set of feature subsets that contribute the most to the evaluation of cardiovascular health status. After multiple iterations, a set of optimized feature combinations are obtained, and these features have the highest contribution degree to the evaluation of cardiovascular health status.

[0101] Step 338: According to the optimized feature combination, extract the corresponding features from the original feature vector set, and organize the extracted features into an optimized feature vector set.

[0102] Suppose there is a feature vector set containing 100 features for evaluating cardiovascular health status. Initialize 100 krills, and each krill represents a point in the feature space. Set the perception range of the krill to 5 unit distances, the moving step size to 0.1 unit distance, and the maximum number of iterations to 100 times.

[0103] In the first iteration, each krill extracts the corresponding feature subset according to its initial position and constructs a prediction model. After training the model with the training data, calculate the contribution degree value of each krill. Then, each krill calculates the set of neighbors within its perception range and selects the neighbor with the highest contribution degree as the learning object. Update the position of the current krill according to the position difference between the learning object and the current krill. After repeating the process 100 times, the krill swarm gradually converges to a set of feature subsets that contribute the most to the evaluation of cardiovascular health status, and these features are extracted from the original feature vector set to form an optimized feature vector set.

[0104] Automatically searching for the key feature subset for cardiovascular health status assessment through the krill herd algorithm can remove redundant and irrelevant features and improve the accuracy of the prediction model. The optimized feature vector set contains the most important features for cardiovascular health status assessment, which helps the model capture the relationship between physiological state and health status more accurately. The optimized feature vector set reduces the number and complexity of features, reduces the risk of model overfitting, and thus enhances the stability of the model. A stable model performs more consistently on different data sets, improving the reliability of the prediction results. Reducing the number of features can reduce the complexity of the model and reduce the consumption of computing resources and time. In real-time or large-scale data processing scenarios, the improvement of computing efficiency is particularly important. Conducting cardiovascular health status assessment based on the optimized feature vector set can provide personalized medical advice and treatment plans for each patient. The feature vectors of different patients may be different, and the optimized feature vector set can better reflect the personalized physiological state of patients. The krill herd algorithm is an automated feature selection method that reduces manual intervention and subjective bias. Automated feature selection improves the objectivity and accuracy of feature selection, making the feature selection process more scientific and reliable.

[0105] In a preferred embodiment of the present invention, according to the contribution degree function, calculating the contribution degree of the current position of the krill to the cardiovascular health status assessment includes:

[0106] When constructing a prediction model using the feature subset corresponding to the current position of the krill, calculating the number of samples correctly predicted as positive examples, the number of samples correctly predicted as negative examples, the number of samples incorrectly predicted as positive examples, and the number of samples incorrectly predicted as negative examples; calculating the accuracy index value according to the number of positive examples, the number of negative examples, the number of positive examples, and the number of negative examples;

[0107] Determining the size of the feature subset corresponding to the current position of the krill, that is, the number of features included in the feature subset, and calculating the feature subset factor index value according to the number of features included in the feature subset;

[0108] Calculating the entropy of the entire data set and the entropy of the data subsets corresponding to different values of each feature, and for each feature, calculating the weighted average of the entropies of the data subsets corresponding to all values; calculating the difference between the entropy of the entire data set and the weighted average, and calculating the ratio of the difference to the maximum value of the information gain among all features; calculating the information gain influence value according to the ratio;

[0109] Fusing the accuracy index value, the feature subset factor index value, and the information gain influence value to obtain the contribution degree of the current position of the krill to the cardiovascular health status assessment.

[0110] In the embodiment of the present invention, using the feature subset F corresponding to the current position of the krill iBuild a prediction model and make predictions through this model. Count the number of samples in the following four cases:

[0111] The number of samples correctly predicted as positive examples (TP i );

[0112] The number of samples correctly predicted as negative examples (TN i );

[0113] The number of samples incorrectly predicted as positive examples (FP i );

[0114] The number of samples incorrectly predicted as negative examples (FN i ).

[0115] According to the counted number of samples, calculate the accuracy index value Determine the size of the feature subset corresponding to the current position of the krill, that is, the number of features included in the feature subset (|F i |). Calculate the feature subset factor index value Here, □ is a small constant used to avoid a zero denominator, and calculate the entropy E(D) of the entire dataset D. For each feature f in the feature subset, calculate the entropy E(D v ) of the data subset corresponding to all its values v ∈ V(f); where v is the value of feature f; V(f) is the set of all values of feature f. For each feature f, calculate the weighted average of the entropies of the data subsets corresponding to all its values v Calculate the difference between the entropy of the entire dataset and the weighted average Calculate the ratio of the difference to the maximum information gain I among all features. For each feature f, calculate its information gain influence value max Perform the above calculations for all features in the feature subset and sum them to obtain the total information gain influence value. Perform a weighted sum of the accuracy index, the feature subset factor index, and the information gain influence value

[0116] where α, β, and γ are weight coefficients used to adjust the proportion of each part in the total contribution degree, and output the contribution degree C(P i ) of the current position of the krill to the assessment of cardiovascular health status.

[0117] The contribution degree function comprehensively considers the accuracy of the prediction model (calculated through TP i , TN i , FP i , FN i ) and the size of the feature subset (through |Fi |Calculation) and the information gain of features (calculated through entropy and information gain). This comprehensive consideration helps to select the feature subset with the most predictive ability for cardiovascular health status assessment, thereby improving the accuracy of the prediction model. By selecting the feature subset with the highest contribution degree, the number and complexity of features can be reduced, and the risk of model overfitting can be lowered. A stable model performs more consistently on different datasets, improving the reliability of the prediction results. Reducing the number of features can reduce the complexity of the model, as well as the consumption of computing resources and time. The optimized feature vector set can better reflect the personalized physiological state of patients, providing strong support for personalized medicine. The feature vectors of different patients may vary. Based on the optimized feature vector set, the cardiovascular health status assessment can provide personalized medical advice and treatment plans for each patient. The krill herd algorithm combined with the contribution degree function realizes an automated feature selection method, reducing manual intervention and subjective bias. The automated feature selection improves the objectivity and accuracy of feature selection, making the feature selection process more scientific and reliable.

[0118] In a preferred embodiment of the present invention, in step 4 above, according to the optimized feature vector set, a three-level fusion model including physiological signal spatio-temporal coding, clinical knowledge graph embedding, and dynamic reinforcement learning is constructed through a probabilistic timed Petri net, which may include:

[0119] Step 441, initialize the construction environment of the probabilistic timed Petri net, including defining the states, transitions, arcs, related probabilities, and time parameters of the net;

[0120] Step 442, perform encoding and fusion processing on the blood pressure waveform features and heart rate variability physiological signal data in the optimized feature vector set to form an intermediate feature representation including the dynamic changes, spatial relationships, and clinical knowledge of physiological signals;

[0121] Step 443, through the dynamic reinforcement learning algorithm, use the intermediate feature representation including the dynamic changes, spatial relationships, and clinical knowledge of physiological signals as input data for learning and optimization. After multiple iterative trainings, a three-level fusion model including physiological signal spatio-temporal coding, clinical knowledge graph embedding, and dynamic reinforcement learning is formed.

[0122] In the embodiment of the present invention, determine the set of states in the probabilistic timed Petri net. These states represent the possible configurations or states of the system at different time points. Assign a unique identifier to each state, and determine the events or actions that can trigger state transitions. These events or actions are called transitions. Define the triggering conditions and execution effects for each transition, including the probability and time delay of state transitions.

[0123] Determine the connection relationships between states and transitions, which are called arcs. Define the weight for each arc, representing the intensity or probability of state transition when the transition is triggered. Set the trigger probability for each transition, representing the likelihood of the transition being triggered in a given state, and set the time delay for each transition, representing the time required for state transition after the transition is triggered.

[0124] Step 442: Extract the blood pressure waveform features in the optimized feature vector set, such as the peak value, slope, duration, etc. of the waveform. Use a suitable coding method (such as binary coding, real number coding, etc.) to convert the blood pressure waveform features into a processable format. Extract the heart rate variability physiological signal data in the optimized feature vector set, such as the RR interval, heart rate variability index, etc., and also use the coding method to convert the heart rate variability physiological signal data into a processable format. Perform fusion processing on the coded blood pressure waveform features and heart rate variability physiological signal data to form an intermediate feature representation that includes the dynamic changes, spatial relationships, and clinical knowledge of the physiological signals. The fusion processing can adopt methods such as weighted average, splicing, convolution, etc., and select a suitable method according to the specific application scenario.

[0125] Step 333: Select a suitable dynamic reinforcement learning algorithm, such as Q-learning, Deep Q-Network (DQN), etc. Initialize the parameters in the algorithm, such as the learning rate, discount factor, exploration rate, etc. Use the intermediate feature representation obtained in Step 332 as the input data and input it into the dynamic reinforcement learning algorithm. The dynamic reinforcement learning algorithm learns the optimal policy through multiple iterative trainings according to the input data. In each iteration, the algorithm selects an action based on the current state, observes the new state and reward after executing the action, and updates the policy. Through continuous iterative training, the algorithm gradually converges to the final policy, forming a three-level fusion model that includes the spatio-temporal coding of physiological signals, the embedding of the clinical knowledge graph, and dynamic reinforcement learning.

[0126] Suppose physiological signal data and a clinical knowledge graph of a target user are received, and the processing is carried out according to the following steps:

[0127] Initialize the construction environment of the probabilistic timed Petri net, and define states, transitions, arcs, and related probability and time parameters. Perform coding and fusion processing on the blood pressure waveform features and heart rate variability physiological signal data in the physiological signal data to form an intermediate feature representation. Use the dynamic reinforcement learning algorithm to learn and optimize with the intermediate feature representation as the input data. Through multiple iterative trainings, the algorithm gradually converges to the final policy, forming a three-level fusion model. When new physiological signal data is input, the model can make predictions and decisions according to the learned policy, providing accurate health status risk assessment and early warning information for doctors.

[0128] By constructing a probabilistic timed Petri net, the model can make full use of the spatio-temporal encoding of physiological signals, clinical knowledge graph embedding, and dynamic reinforcement learning mechanism to improve the model's expressiveness and generalization ability. This enables the model to more accurately predict the health status risk of the target user and provide more reliable decision-making support for doctors. The three-level fusion model can perform personalized learning and optimization based on the physiological signal data and clinical knowledge graph of each target user, which helps to promote the development of personalized medicine and provide customized health risk assessment and early warning services for each target user. By automatically processing and analyzing physiological signal data and clinical knowledge graphs, the three-level fusion model can greatly improve the efficiency and quality of medical services. Doctors can obtain accurate health status risk assessment results more quickly and take corresponding treatment measures, thereby improving the treatment effect and satisfaction of patients.

[0129] In another preferred embodiment of the present invention, the three-level fusion model may include:

[0130] The first level is the physiological signal spatio-temporal encoding module, which performs spatio-temporal encoding on the physiological signals in the optimized feature vector set to obtain the temporal relationship and spatial features between the signals;

[0131] The second level is the clinical knowledge graph embedding module, which fuses clinical knowledge with feature vectors and transforms clinical knowledge into structured information;

[0132] The third level is the dynamic reinforcement learning module, which performs dynamic learning based on the outputs of the physiological signal spatio-temporal encoding module and the clinical knowledge graph embedding module, and finally outputs the health status risk level of the target user.

[0133] In the embodiment of the present invention, physiological signal data in the feature vector set, including blood pressure, heart rate, electrocardiogram, etc., are received, the temporal relationship between physiological signals is analyzed, such as the correlation between blood pressure fluctuations and heart rate changes, and time series analysis methods (such as LSTM) are used to encode the temporal relationship to extract time features. For physiological signals with spatial features (such as the lead signals of electrocardiogram), spatial encoding methods (such as convolutional neural network CNN) are used to extract spatial features, and the spatial features are fused with the temporal features to form a comprehensive physiological signal feature representation.

[0134] Second level: Clinical Knowledge Graph Embedding Module. Collect clinical knowledge related to cardiovascular and cerebrovascular diseases, including disease symptoms, risk factors, treatment plans, etc. Construct a clinical knowledge graph to represent clinical knowledge in a structured form. Use graph embedding methods (such as TransE, GraphSAGE, etc.) to transform the clinical knowledge graph into a representation in a low-dimensional vector space. Fuse the physiological signal feature vector with the clinical knowledge graph embedding vector to form a feature representation containing rich clinical information. Design appropriate fusion strategies (such as weighted average, concatenation, etc.) to fuse the physiological signal features with the clinical knowledge graph embedding features, ensuring that the fused features can fully reflect the association between physiological signals and clinical knowledge.

[0135] Third level: Dynamic Reinforcement Learning Module. Select appropriate dynamic reinforcement learning algorithms (such as DQN, A3C, etc.) and initialize model parameters, including learning rate, discount factor, exploration rate, etc. Use the outputs of the physiological signal spatio-temporal encoding module and the clinical knowledge graph embedding module as state representations, and design a state encoding method to ensure that the state can accurately reflect the current physiological state and clinical knowledge. According to the current state, use the policy network to select actions (such as predicting the risk level of the target user's health status). The action selection process needs to consider the balance between exploration and exploitation. Design an appropriate reward mechanism to give rewards or punishments according to the difference between the prediction result and the true result. The reward mechanism needs to be able to guide the model towards a more accurate risk prediction direction. Use the reinforcement learning algorithm to train the model, and optimize the model parameters through multiple iterations. During the training process, continuously adjust parameters such as the learning rate and discount factor to improve the convergence speed and accuracy of the model. After training, use the model to predict new physiological signal data and output the risk level of the target user's health status.

[0136] Suppose physiological signal data and clinical medical record of a target user are received, and the processing is carried out according to the following steps:

[0137] Clean and preprocess the received physiological signal data, extract the temporal relationship and spatial features of the physiological signals using time series analysis methods and spatial encoding methods, fuse the extracted features to form a comprehensive representation of physiological signal features. Construct a clinical knowledge graph related to cardiovascular and cerebrovascular diseases based on clinical medical records, and use the graph embedding method to transform the clinical knowledge graph into a representation in a low-dimensional vector space. Fuse the physiological signal feature vector and the clinical knowledge graph embedding vector to form a feature representation containing rich clinical information. Initialize a dynamic reinforcement learning model, and set appropriate parameters and state representation methods. Use the fused features as the state input, select actions according to the policy network (predict the health status risk level of the target user), give rewards or punishments according to the difference between the prediction result and the real result, train and optimize the model. After training is completed, use the model to predict new physiological signal data and output the health status risk level of the target user.

[0138] Through the spatio-temporal encoding of physiological signals and the embedding of clinical knowledge graphs, the model can make full use of the correlation information between physiological signals and clinical knowledge to improve the accuracy of risk prediction. The dynamic reinforcement learning module can perform dynamic learning and optimization based on new physiological signal data and clinical knowledge, enhancing the generalization ability of the model, which enables the model to adapt to the physiological states and clinical situations of different target users and improves the applicability of the model. The three-level fusion model can perform personalized learning and prediction based on the physiological signal data and clinical medical records of each target user, which helps to promote the development of personalized medicine and provide customized cardiovascular and cerebrovascular disease risk assessment and early warning services for each target user.

[0139] In a preferred embodiment of the present invention, in step 5 above, input the feature vector of the target user into the three-level fusion model, and according to the inference rules of the Petri net, obtain the health status risk level of the target user, and according to the health status risk level of the target user, output the risk assessment result and generate a warning message, which may include:

[0140] Step 552, take the feature vector of the target user as input data and transfer it to the first level of the three-level fusion model, that is, the physiological signal spatio-temporal encoding module;

[0141] Step 553, in the physiological signal spatio-temporal encoding module, extract the temporal relationship and spatial features between physiological signals to form an intermediate feature representation;

[0142] Step 554, transfer the intermediate feature representation to the second level, that is, the clinical knowledge graph embedding module, and fuse the clinical knowledge with the intermediate feature representation to generate a fused feature representation;

[0143] Step 555: Transmit the fused feature representation to the third level, i.e., the dynamic reinforcement learning module, perform dynamic learning and optimization based on the fused feature representation, and through continuous adjustment, output the risk level of the target user's health status;

[0144] Step 556: Generate a corresponding risk assessment report according to the health status and risk level of the target user, including risk level, risk factor analysis, and recommended measures;

[0145] Step 557: Set a warning threshold according to the health status and risk level of the target user, compare the health status and risk level of the target user with the warning threshold, and when the health status and risk level of the target user ≥ the warning threshold, automatically generate warning information according to the preset warning rules, including risk level, main risk factors, and recommended emergency actions.

[0146] In the embodiment of the present invention, physiological signal data is obtained from the target user through medical devices (such as electrocardiographs, sphygmomanometers, etc.) to ensure the accuracy and continuity of data collection. According to the characteristics of physiological signals, relevant features are extracted, such as time series features (such as mean, standard deviation, peak value, etc.), spatial features (such as the relationship between leads in electrocardiogram), and the extracted features are encoded to form a feature vector set.

[0147] Step 552: Design the input interface of the three-level fusion model to ensure that it can receive the feature vector set as input data, and transmit the feature vector set to the physiological signal spatio-temporal encoding module through the input interface.

[0148] Step 553: Use time series analysis methods (such as LSTM, GRU, etc.) to extract the temporal relationship between physiological signals, analyze the changing trend of physiological signals over time, and extract time features. For physiological signals with spatial features (such as electrocardiogram), use spatial encoding methods (such as CNN) to extract spatial features, analyze the distribution and relationship of physiological signals in space, and extract spatial features. The extracted temporal relationship and spatial features are fused to form a comprehensive intermediate feature representation, ensuring that the fused features can fully reflect the temporal and spatial characteristics of physiological signals.

[0149] Step 554: Construct a clinical knowledge graph related to cardiovascular and cerebrovascular diseases, including disease symptoms, risk factors, treatment plans, etc., to ensure the accuracy and integrity of the knowledge graph. Use graph embedding methods to transform the clinical knowledge graph into a representation in a low-dimensional vector space, ensuring that the embedded vectors can accurately reflect the semantic and structural information of clinical knowledge. The intermediate feature representation is fused with the clinical knowledge graph embedding vectors, and appropriate fusion strategies (such as weighted average, concatenation, etc.) are designed to ensure that the fused features can fully reflect the association between physiological signals and clinical knowledge.

[0150] Step 555: Initialize the dynamic reinforcement learning model, set appropriate parameters and state representation methods to ensure that the model can receive the fused feature representation as input data. Use the fused feature representation as the state input and select an action according to the policy network (such as predicting the risk level of the health condition). During the learning process, give rewards or punishments based on the difference between the predicted result and the true result, and adjust the model parameters. Through multiple iterative trainings, continuously optimize the model parameters to improve the prediction accuracy, and adjust parameters such as the learning rate and discount factor according to the actual situation to accelerate the convergence speed and improve the model performance. After the training is completed, use the model to predict new physiological signal data and output the risk level of the target user's health condition.

[0151] Step 556: According to the preset risk level classification criteria (such as low, medium, high, extremely high), map the risk level of the target user's health condition to the corresponding risk level to ensure that the risk level is scientifically and reasonably classified and can accurately reflect the risk level of the target user's health condition. Conduct an in-depth analysis of the fused feature representation, identify the main factors affecting the target user's health condition, and combine clinical knowledge to interpret and explain the risk factors to provide strong support for the risk assessment report. Based on the risk level and the analysis results of the risk factors, develop personalized recommended measures for the target user. Integrate the risk level, risk factor analysis, and recommended measures into a complete risk assessment report.

[0152] Step 557: Set a reasonable warning threshold according to the actual needs and safety considerations to ensure that the warning threshold can accurately reflect the critical level of the health condition risk and avoid false alarms and missed alarms. Compare the risk level of the target user's health condition with the warning threshold. When the risk level ≥ warning threshold, trigger the warning mechanism, and automatically generate warning information according to the preset warning rules. The warning information includes content such as the risk level, main risk factors, and recommended emergency actions to ensure that the information is comprehensive and accurate.

[0153] Suppose an electrocardiogram data and clinical medical record of a target user are received, and the processing is carried out according to the following steps:

[0154] Obtain electrocardiogram data from the target user through an electrocardiograph, and perform cleaning and standardization processing. Extract features of the electrocardiogram, such as R-wave peak, QT interval, etc., to form a feature vector set. Transmit the feature vector set to the physiological signal spatio-temporal encoding module. The module uses the LSTM method to extract the temporal relationship of the electrocardiogram data and the CNN method to extract spatial features. Integrate the extracted temporal relationship and spatial features to form an intermediate feature representation, and construct a clinical knowledge graph related to cardiovascular and cerebrovascular diseases, including disease symptoms, risk factors, etc. Use the graph embedding method to transform the knowledge graph into a representation in a low-dimensional vector space, and integrate the intermediate feature representation with the clinical knowledge graph embedding vector to generate a fused feature representation. Transmit the fused feature representation to the dynamic reinforcement learning module. The module performs dynamic learning and optimization based on the fused feature representation. By continuously adjusting the model parameters, improve the prediction accuracy. After training is completed, use the model to predict new electrocardiogram data and output the risk level of the target user's health status.

[0155] According to the health status and risk level of the target user, classify them into the "high-risk" level. Analyze the fused feature representation, identify the main risk factors as hypertension and bad living habits, and formulate recommended measures for the target user, such as improving living habits and regularly monitoring blood pressure. Integrate the above information into a risk assessment report and send it to the target user. Set the warning threshold as the "medium-risk" level. Compare the health status risk level of the target user with the warning threshold and find that the risk level > warning threshold. According to the preset warning rules, automatically generate warning information, including the risk level, main risk factors (hypertension and bad living habits), and recommended emergency actions (seek medical attention immediately and adjust living habits), and promptly publish the warning information to the target user and the medical team for taking emergency actions.

[0156] Through spatio-temporal encoding of physiological signals and embedding of clinical knowledge graphs, the model can fully utilize the correlation information between physiological signals and clinical knowledge, improving the accuracy and reliability of risk prediction. The dynamic reinforcement learning module can perform dynamic learning and optimization based on new physiological signal data and clinical knowledge, enhancing the adaptability and generalization ability of the model. This enables the model to adapt to the physiological states and clinical conditions of different target users, improving the practicality and value of the model. The three-level fusion model can perform personalized learning and prediction based on the physiological signal data and clinical medical records of each target user, which helps to promote the development of personalized medicine and health management, providing customized health status risk assessment and early warning services for each target user, and improving the efficiency and quality of medical services. The generation of risk assessment reports and early warning information can enhance the target users' awareness and attention to health status risks, which helps target users to take proactive preventive measures to reduce health status risks. When the risk level reaches or exceeds the early warning threshold, the timely release of early warning information can prompt the medical team or relevant personnel to take emergency actions promptly, which helps to avoid the deterioration of health status and ensure the safety of the target users' lives. The generation of risk assessment reports and early warning information can provide strong support for the rational allocation of medical resources, which helps to achieve the precise allocation of medical resources and improve the quality and efficiency of medical services.

[0157] In another preferred embodiment of the present invention, step 555, performing dynamic learning and optimization based on the fused feature representation, and outputting the health status risk degree of the target user through continuous adjustment, may include:

[0158] Step 5551, initializing the dynamic reinforcement learning model, including setting the structure, parameters, and learning rate of the dynamic reinforcement learning model;

[0159] Step 5552, the dynamic reinforcement learning model receives the fused feature representation and performs dynamic learning according to a preset learning algorithm;

[0160] Step 5553, in each stage of learning, predicting the health status risk degree of a target user based on the current feature representation, comparing it with the actual health status risk degree, and calculating the difference between the prediction result and the actual result;

[0161] Step 5554, repeating the learning and optimization, and when the preset learning stop condition is reached, outputting the health status risk degree of the target user.

[0162] In an embodiment of the present invention, a suitable dynamic reinforcement learning algorithm (such as DQN, DDPG, etc.) is selected according to the task requirements, and the network structure of the model is designed, including an input layer, a hidden layer, and an output layer. The input layer is used to receive the fused feature representation, the hidden layer is used to process and learn features, and the output layer is used to output the predicted health status risk level. Parameters such as the number of neurons and activation functions in each layer are determined. The weights and biases of the model are initialized to ensure that the initialized parameters enable the model to have reasonable performance in the initial stage of training. A suitable learning rate is set to control the step size of parameter update during the training process. The selection of the learning rate needs to be adjusted according to the task complexity and data scale to ensure that the model can converge within a reasonable time.

[0163] Step 5552: The fused feature representation is used as input data and passed to the dynamic reinforcement learning model, ensuring that the format and dimension of the input data match the requirements of the model's input layer. According to the preset learning algorithm (such as Q-learning, policy gradient, etc.), the model processes and learns the input data. During the learning process, the model selects a strategy based on the current state and action and updates the model parameters to optimize the prediction performance.

[0164] Step 5553: The input data is predicted using the current model parameters to obtain the health status risk level of the target user, ensuring the rationality and accuracy of the prediction result, reflecting the actual health status of the target user. The prediction result is compared with the actual health status risk level, and the difference between the prediction result and the actual result is calculated as the basis for model performance evaluation. A loss function (such as mean square error) is calculated according to the difference to measure the goodness of the model's prediction performance.

[0165] Step 5554: According to the loss function and the learning algorithm, the model parameters are continuously updated for repeated learning. During the learning process, the value of the loss function is gradually reduced to improve the model's prediction performance. A suitable learning stop condition (such as reaching the maximum number of iterations) is set. When the learning stop condition is met, the model training is stopped to avoid overfitting or excessive training time. The trained model is used to predict new input data, and the health status risk level of the target user is output, ensuring the rationality and accuracy of the output result to provide support for medical decision-making.

[0166] Assume that the fused feature representation has been received, and now it is necessary to use the dynamic reinforcement learning model to predict the health status risk level. The following is the execution process:

[0167] Select the DDPG algorithm as the dynamic reinforcement learning model, design the corresponding network structure, initialize parameters such as the weights and biases of the model, and set the learning rate to 0.001. Receive the fused feature representation, including physiological signal features, clinical knowledge graph embedding vectors, etc., and ensure that the format and dimensions of the input data match the requirements of the model input layer. Process and learn the input data according to the DDPG algorithm. During the learning process, select a strategy based on the current state and action, and update the model parameters to optimize the prediction performance. Use the current model parameters to predict the input data to obtain the health status risk level of the target user, compare the prediction result with the actual health status risk level, and calculate the loss function value. Continuously update the model parameters according to the loss function and the learning algorithm, and perform repeated learning. When the loss function value converges or reaches the maximum number of iterations, stop the model training. Use the trained model to predict new input data, output the health status risk level of the target user, and feedback the prediction result to the medical system or the user to provide support for medical decision-making.

[0168] The dynamic reinforcement learning model can perform dynamic learning and optimization based on the fused feature representation, improving the prediction accuracy. Through continuous iterative training, the model can gradually approximate the real situation and provide a more reliable basis for medical decision-making. The dynamic reinforcement learning model can adapt to different physiological signal data and clinical knowledge graph embedding vectors, enhancing the adaptability of the model, which enables the model to be applied in different scenarios and provide personalized health status risk assessment services for more users. Accurate prediction of the health status risk level can provide strong support for medical decision-making. Doctors can formulate more scientific and reasonable treatment plans based on the prediction results, improving the treatment effect and patient satisfaction. The application of the dynamic reinforcement learning model in the prediction of the health status risk level promotes the development of the medical and health field, which helps to achieve the rational allocation and efficient utilization of medical resources and improve the quality and efficiency of medical services.

[0169] As Figure 2 shown, the embodiment of the present invention also provides a cardiovascular disease risk prediction system, including:

[0170] A data acquisition module for acquiring physiological signal data of a target user;

[0171] A feature extraction module for extracting features from the physiological signal data and forming a feature vector set;

[0172] A feature selection and optimization module for using the krill herd algorithm to perform feature selection and optimization on the feature vector set, automatically searching for and determining the key feature subset for cardiovascular health status assessment;

[0173] A three - level fusion module, which is used to perform deep - level and multi - dimensional fusion processing on the optimized feature vector set to obtain a fused feature representation;

[0174] A risk prediction module, which is used to predict the risk degree of the health status of the target user according to the feature representation output by the three - level fusion module, obtain a risk prediction result, and according to the risk prediction result, when the risk degree of the health status of the target user exceeds a preset threshold, automatically generate a warning message.

[0175] It should be noted that this system corresponds to the above - mentioned method. All implementation manners in the above - mentioned method embodiments are applicable to this embodiment and can also achieve the same technical effects.

[0176] An embodiment of the present invention also provides a computer - readable storage medium storing instructions. When the instructions run on a computer, the computer is made to execute the method as described above. All implementation manners in the above - mentioned method embodiments are applicable to this embodiment and can also achieve the same technical effects.

[0177] The above is the preferred implementation manner of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for predicting cardiovascular and cerebrovascular disease risk, characterized in that: The method comprises: Obtain the target user's real-time physiological parameters, including dynamic blood pressure fluctuation rate, serum lipoprotein level, heart rate variability frequency domain index and sleep apnea hypopnea index; Preprocess the real-time physiological parameters, use wavelet transform to extract the instantaneous waveform features of the blood pressure signal, and extract features related to health conditions, including blood pressure level, blood lipid level, blood sugar level, electrocardiogram abnormality index and cardiac ultrasound abnormality index, to form a feature vector set; The krill swarm algorithm is used to select and optimize the feature vector set. By simulating the foraging behavior of krill swarms, the feature subsets that are key to the assessment of cardiovascular health status are automatically searched to obtain the optimized feature vector set. According to the optimized feature vector set, a three-level fusion model including spatiotemporal encoding of physiological signals, clinical knowledge graph embedding and dynamic reinforcement learning is constructed through the probabilistic temporal Petri network. The feature vector of the target user is input into the three-level fusion model, and the health risk degree of the target user is obtained according to the inference rules of the Petri net. According to the health risk degree of the target user, the risk assessment result is output and the warning information is generated.

2. The method for predicting cardiovascular and cerebrovascular disease risk according to claim 1, characterized in that: Preprocess the real-time physiological parameters, use wavelet transform to extract the instantaneous waveform characteristics of the blood pressure signal, and extract the characteristics related to the health status, including blood pressure level, blood lipid level, blood sugar level, electrocardiogram abnormality index and cardiac ultrasound abnormality index, to form a feature vector set, including: Wavelet transform method is used to extract instantaneous waveform features of blood pressure signals to obtain dynamic characteristics of blood pressure changes; Features related to health conditions, including blood pressure levels, blood lipid levels, blood sugar levels, electrocardiogram abnormality indicators, and cardiac ultrasound abnormality indicators, are extracted from the data processed by wavelet transform to form a feature vector set.

3. The method for predicting cardiovascular and cerebrovascular disease risk according to claim 2, characterized in that: The krill swarm algorithm is used to select and optimize the feature vector set. By simulating the foraging behavior of krill swarms, the feature subsets that are key to the assessment of cardiovascular health status are automatically searched to obtain the optimized feature vector set, including: Determine the size of the krill population, i.e., the number of candidate features participating in feature selection, and assign an initial position to each krill; Set the krill's sensing range, moving step length, and maximum number of iterations, and define a contribution function. According to the contribution function, calculate the contribution of the krill's current position to the assessment of cardiovascular health status. For each krill, extract the corresponding features according to the current position to build a prediction model; The prediction model is trained using sample data related to cardiovascular health status, and the contribution value of each krill is calculated based on the contribution function; Assign the contribution value to the corresponding krill as the performance evaluation in the current iteration, and for each krill, calculate the neighbor set within the perception range, and determine the learning object of the current krill in the neighbor set; Update the current krill position according to the learning object, the current krill position difference and the preset moving step length; The process of feature extraction, model building, training, and contribution calculation is repeated until the preset maximum number of iterations is reached, and finally a set of optimized feature combinations is obtained. According to the optimized feature combination, the corresponding features are extracted from the original feature vector set to form an optimized feature vector set.

4. The method for predicting cardiovascular and cerebrovascular disease risk according to claim 3, characterized in that: According to the contribution function, the contribution of the current location of krill to the assessment of cardiovascular health status is calculated, including: When constructing a prediction model based on the feature subset corresponding to the current position of the krill, the number of samples correctly predicted as positive examples, the number of samples correctly predicted as negative examples, the number of samples incorrectly predicted as positive examples, and the number of samples incorrectly predicted as negative examples are calculated; the accuracy index value is calculated based on the number of positive examples, the number of negative examples, the number of positive examples, and the number of negative examples; Determine the size of the feature subset corresponding to the current position of the krill, that is, the number of features contained in the feature subset, and calculate the feature subset factor index value according to the number of features contained in the feature subset; Calculate the entropy of the entire data set and the entropy of the data subset corresponding to different values ​​of each feature. For each feature, calculate the weighted average of the entropy of the data subset corresponding to all values; calculate the difference between the entropy of the entire data set and the weighted average, and calculate the ratio between the difference and the maximum value of information gain among all features; calculate the information gain impact value based on the ratio; The accuracy index value, feature subset factor index value and information gain influence value are fused to obtain the contribution of the current position of krill to the assessment of cardiovascular health status.

5. The method for predicting cardiovascular and cerebrovascular disease risk according to claim 4, characterized in that: According to the optimized feature vector set, a three-level fusion model including spatiotemporal encoding of physiological signals, clinical knowledge graph embedding and dynamic reinforcement learning is constructed through the probabilistic temporal Petri network, including: Initialize the construction environment of the probabilistic time-Petri network, including defining the network's states, transitions, arcs, and related probabilities and time parameters; The blood pressure waveform features and heart rate variability physiological signal data in the optimized feature vector set are encoded and fused to form an intermediate feature representation that includes the dynamic changes of physiological signals, spatial relationships and clinical knowledge; Through the dynamic reinforcement learning algorithm, the intermediate feature representation containing the dynamic changes of physiological signals, spatial relationships and clinical knowledge is used as input data for learning and optimization. After multiple iterative training, a three-level fusion model is formed, which includes spatiotemporal encoding of physiological signals, embedding of clinical knowledge graphs and dynamic reinforcement learning.

6. The method for predicting cardiovascular and cerebrovascular disease risk according to claim 5, characterized in that: Three-level fusion model, including: The first level is the physiological signal spatiotemporal coding module, which performs spatiotemporal coding on the physiological signals in the optimized feature vector set to obtain the temporal relationship and spatial characteristics between the signals; The second level is the clinical knowledge graph embedding module, which integrates clinical knowledge with feature vectors and transforms clinical knowledge into structured information; The third level is the dynamic reinforcement learning module, which performs dynamic learning based on the outputs of the physiological signal spatiotemporal encoding module and the clinical knowledge graph embedding module, and ultimately outputs the health risk level of the target user.

7. The method for predicting cardiovascular and cerebrovascular disease risk according to claim 6, characterized in that: The feature vector of the target user is input into the three-level fusion model. According to the inference rules of the Petri network, the health risk level of the target user is obtained. According to the health risk level of the target user, the risk assessment result is output and the warning information is generated, including: The feature vector of the target user is used as input data and passed to the first level of the three-level fusion model, i.e., the physiological signal spatiotemporal encoding module; In the physiological signal spatiotemporal encoding module, the temporal relationship and spatial features between physiological signals are extracted to form an intermediate feature representation; The intermediate feature representation is passed to the second level, i.e., the clinical knowledge graph embedding module, where the clinical knowledge is fused with the intermediate feature representation to generate a fused feature representation. The fused feature representation is passed to the third level, i.e., the dynamic reinforcement learning module, which performs dynamic learning and optimization based on the fused feature representation and outputs the health risk level of the target user through continuous adjustment. Generate corresponding risk assessment reports based on the health status and risk level of target users, including risk level, risk factor analysis, and recommended measures; According to the health status and risk level of the target user, a warning threshold is set, and the health status and risk level of the target user are compared with the warning threshold. When the health status and risk level of the target user ≥ the warning threshold, warning information is automatically generated according to the preset warning rules, including risk level, main risk factors, and recommended emergency actions.

8. The method for predicting cardiovascular and cerebrovascular disease risk according to claim 7, characterized in that: Dynamic learning is performed based on the outputs of the physiological signal spatiotemporal encoding module and the clinical knowledge graph embedding module, and the health risk level of the target user is finally output, including: Initialize the dynamic reinforcement learning model, including setting the structure, parameters, and learning rate of the dynamic reinforcement learning model; The dynamic reinforcement learning model receives the fused feature representation and performs dynamic learning according to the preset learning algorithm; In each stage of learning, the health risk level of a target user is predicted based on the current feature representation, and compared with the actual health risk level, and the difference between the predicted result and the actual result is calculated; Learning and optimization are repeated, and when the preset learning stop condition is reached, the health risk level of the target user is output.

9. A cardiovascular and cerebrovascular disease risk prediction system, the system implementing the method according to any one of claims 1 to 8, characterized in that: include: A data acquisition module, used to obtain physiological signal data of the target user; A feature extraction module, used to extract features from the physiological signal data and form a feature vector set; A feature selection and optimization module, which uses a krill swarm algorithm to select and optimize feature vector sets, automatically searching for and determining feature subsets that are critical to assessing cardiovascular health status; The three-level fusion module is used to perform deep and multi-dimensional fusion processing on the optimized feature vector set to obtain the fused feature representation; The risk prediction module is used to predict the health risk level of the target user based on the feature representation output by the three-level fusion module, obtain the risk prediction result, and automatically generate warning information based on the risk prediction result when the health risk level of the target user exceeds a preset threshold.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, which implements the method according to any one of claims 1 to 8 when executed by a processor.

Images