Static electrocardiogram analysis method combined with noninvasive cardiac output monitoring
By combining static electrocardiogram analysis with non-invasive cardiac output monitoring, a model of electrocardiogram-hemodynamic coupling was established. Deep neural networks were used to assess the risk of cardiac abnormalities and generate personalized reports. This solved the problems of real-time performance and accuracy in traditional electrocardiogram analysis, and enabled precise assessment and personalized treatment of cardiac health.
Patent Information
- Application Number
- CN202511530584.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-24
- Publication Date
- 2026-02-27
AI Technical Summary
Traditional static electrocardiogram analysis methods lack real-time performance and accuracy, fail to fully capture complex physiological changes related to cardiac output, and rely on human experience, leading to inconsistent diagnoses and poor reliability, thus limiting the widespread use and accuracy of non-invasive cardiac output monitoring.
By combining static electrocardiogram analysis with non-invasive cardiac output monitoring, this method acquires patients' historical electrocardiogram data, establishes an electrocardiogram characteristic baseline, constructs a cardiac-hemodynamic coupling model, uses deep neural networks to assess the risk of cardiac abnormalities, generates personalized analysis reports, and supports remote consultations and intelligent decision-making for multi-center medical teams through Internet of Things (IoT) technology.
It achieves deep coupling between electrocardiogram and hemodynamic parameters, providing accurate and personalized cardiac health assessment, reducing human error, improving diagnostic accuracy and treatment efficiency, and supporting early detection and personalized treatment of heart disease.
Smart Images
Figure CN121570181A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent medical technology, specifically to a static electrocardiogram analysis method that combines non-invasive cardiac output monitoring. Background Technology
[0002] Noninvasive cardiac output monitoring typically relies on traditional resting electrocardiogram (ECG) analysis methods to assess cardiac function; however, these methods have several significant limitations. First, traditional resting ECG analysis cannot reflect dynamic changes in the heart in real time, resulting in a lack of real-time and accurate assessment of cardiac status. Second, resting ECG cannot fully capture complex physiological changes directly related to cardiac output, such as cardiac load and hemodynamics; therefore, the assessment results may deviate from the patient's actual cardiac function. Furthermore, resting ECG analysis methods largely depend on human experience, lacking automation and intelligence, and are easily influenced by the analyst's subjective factors, leading to inconsistent and unreliable diagnoses. In practical clinical applications, these technical deficiencies limit the widespread use and accuracy of noninvasive cardiac output monitoring. Therefore, further optimization of ECG analysis methods is needed to improve the accuracy, real-time performance, and intelligence of monitoring to achieve more precise noninvasive cardiac output assessment. Summary of the Invention
[0003] To address the aforementioned technical issues, a static electrocardiogram analysis method combining non-invasive cardiac output monitoring is provided. This technical solution resolves the problems of lack of automation and intelligence, and susceptibility to subjective factors of the analyst.
[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0005] Static electrocardiogram analysis methods combined with noninvasive cardiac output monitoring include:
[0006] Acquire historical electrocardiogram (ECG) data of patients at different times, analyze the trend of changes in ECG waveform characteristics, and establish a baseline of ECG characteristics for patients.
[0007] Based on non-invasive sensors, real-time hemodynamic parameters of the patient's heart are acquired. Combined with the patient's electrocardiogram characteristic baseline, a patient electrocardiogram-hemodynamic coupling relationship model is constructed to identify abnormal electrocardiogram curves of the patient's heart.
[0008] For patients with abnormal electrocardiogram curves, a deep neural network model is trained to assess the risk level of the abnormal electrocardiogram and generate a personalized intelligent electrocardiogram analysis report for the patient.
[0009] Based on the patient's personalized intelligent electrocardiogram analysis report, combined with IoT technology, it is uploaded to the cloud platform to support remote consultation and intelligent auxiliary decision-making for multi-center medical teams.
[0010] Preferably, historical electrocardiogram (ECG) data of patients at different periods are obtained, the Symlets wavelet basis is selected, and the DWT discrete wavelet transform algorithm is used to associate and pair historical ECG data of patients at different periods with the Symlets wavelet basis. The historical ECG data of patients at different periods are decomposed layer by layer, and the local features of the historical ECG data of patients at different periods are extracted to obtain the low-pass filter and high-pass filter of each layer in the historical ECG data.
[0011] Based on the low-pass and high-pass filters of each layer of historical electrocardiogram data, the approximation coefficient and detail coefficient are calculated, the low-frequency filter is decomposed step by step, and iterative updates are performed until convergence to obtain the low-frequency and high-frequency coefficients of higher-level historical electrocardiogram data.
[0012] For the low-frequency and high-frequency coefficients of the historical electrocardiogram data obtained at a higher level, the median absolute value of the high-frequency coefficients of the historical electrocardiogram data is calculated, and noise in the high-frequency coefficients is removed by dynamically adjusting the threshold; the threshold includes: hard threshold and soft threshold.
[0013] Preferably, the inverse discrete wavelet transform algorithm is used to merge the low-frequency and high-frequency coefficients of higher-level historical electrocardiogram data to reconstruct the historical electrocardiogram data;
[0014] Based on the reconstructed historical electrocardiogram data, data preprocessing was performed.
[0015] Two sets of historical electrocardiogram time series data were selected. Using the Euclidean distance formula, the distance between each pair of time points in the two sets of historical electrocardiogram time series data was randomly calculated to construct a historical electrocardiogram distance matrix. The minimum distance from each point to the starting point was calculated using a recursive formula to obtain the best alignment path of the historical electrocardiogram.
[0016] Based on historical electrocardiogram data, the Fridericia formula was used to correct the QT interval for heart rate, eliminate the influence of heart rate changes, measure the difference in QT interval, and calculate the QTc dispersion.
[0017] Based on historical electrocardiogram data, the PR interval sequence of 30 consecutive heartbeats was extracted, timestamps were aligned, the time mean and amplitude mean of the PR interval were calculated, the trend of PR interval over time was fitted by the least squares method, and the slope of the PR interval was calculated.
[0018] By fusing QTc dispersion and PR interval slope, time-domain features of historical electrocardiograms can be obtained.
[0019] Using the wavelet packet energy entropy algorithm, historical electrocardiogram data is decomposed into multiple frequency band sub-signals. Multiple frequency bands are obtained recursively layer by layer. The energy of each frequency band sub-signal is calculated, and the energy ratio of each frequency band sub-signal is obtained. Using the entropy formula, the complexity of the energy distribution of each frequency band sub-signal is quantified, and the frequency domain features of historical electrocardiograms are extracted.
[0020] By combining the time-domain and frequency-domain characteristics of historical electrocardiograms, comprehensive features of historical electrocardiograms are obtained, the trend of changes in patient electrocardiogram waveform characteristics is analyzed, and a baseline of patient electrocardiogram characteristics is established.
[0021] Preferably, based on non-invasive sensors, real-time hemodynamic parameters of the patient's heart are acquired, and combined with the patient's electrocardiogram data, the data is preprocessed.
[0022] Using an event-driven algorithm, corresponding peaks in blood flow signals and electrocardiogram signals are identified, timestamps are aligned, and synchronous multimodal data is used as the output, as shown in the following formula:
[0023]
[0024] in, Let k be the k-th synchronization timestamp, where k is the time step. To minimize time t, Hemodynamic signals The first derivative, The first derivative of the electrocardiogram signal E(t) is... The threshold for the significance of the rate of change of hemodynamic signals. The threshold for the significance of the rate of change of the electrocardiogram signal. For logical difference;
[0025] Based on synchronous multimodal data, a Granger causal model is trained according to the optimal lag order of the Bayesian information criterion. The causal relationship between hemodynamic parameters and electrocardiogram features is analyzed, the influence of hemodynamic parameters on electrocardiogram features is quantified, and regularization methods are used to prevent overfitting, thus constructing a patient electrocardiogram-hemodynamic coupling relationship model.
[0026] Preferably, based on the constructed electrocardiogram-hemodynamic coupling model, the changes in the patient's cardiac hemodynamic parameters are monitored in real time, and combined with the patient's electrocardiogram characteristic baseline, the real-time hemodynamic parameters are compared with the electrocardiogram characteristic baseline data to identify abnormal electrocardiogram curves in the patient's heart.
[0027] As a further detail, the abnormal electrocardiogram curves include: arrhythmias, myocardial ischemia, and heart failure.
[0028] Preferably, for patients with abnormal electrocardiogram curves, an abnormal electrocardiogram feature dataset is constructed, and the data is preprocessed.
[0029] By combining the time-domain and frequency-domain features of historical electrocardiograms as input, convolutional layers are used to extract electrocardiogram waveform features. Multiple convolutional layers are used to extract waveform features from low to high levels in the electrocardiogram. Pooling operations are used to reduce the dimensionality of the convolutional layer output features, thereby learning deep features in the electrocardiogram waveform.
[0030] By utilizing the LSTM (Long Short-Term Memory) network and learning deep features from electrocardiogram (ECG) waveforms, we can capture the long-term dependencies of ECG changes over time.
[0031] Preferably, a fully connected layer is designed to splice the deep features in the electrocardiogram waveform and the long-term dependence of the electrocardiogram on changes over time. The deep features in the electrocardiogram waveform and the long-term dependence of the electrocardiogram on changes over time are used as input. The patient's abnormal electrocardiogram is classified by weighted summation and activation function, and the patient's abnormal electrocardiogram risk level score is output.
[0032] As a further detail, the risk levels include: low, medium, and high.
[0033] By integrating basic patient information, baseline ECG characteristics, abnormal ECG curves, and risk level scores, a personalized intelligent ECG analysis report is generated for each patient.
[0034] Preferably, based on the patient's personalized intelligent electrocardiogram analysis report, the data is uploaded to the cloud platform using IoT technology to achieve real-time transmission and sharing of electrocardiogram data;
[0035] Based on the storage and management of uploaded electrocardiogram (ECG) data, a global model is initialized. Using ECG data uploaded by various medical centers as training samples, a federated learning algorithm is adopted to achieve collaborative training and optimization of ECG data from multiple centers.
[0036] The intelligent decision support module is designed and embedded in the cloud platform to receive and analyze electrocardiogram data uploaded by various medical centers. Based on the clinical knowledge base, it automatically provides preliminary diagnostic suggestions and treatment plan recommendations, supporting remote consultations and intelligent decision support for multi-center medical teams.
[0037] Compared with the prior art, the beneficial effects of the present invention are as follows:
[0038] This invention proposes a static electrocardiogram (ECG) analysis scheme combining non-invasive cardiac output monitoring (NIVMP). By integrating NIVMP with static ECG analysis, a deep coupling between ECG and hemodynamic parameters is achieved, providing a more accurate and personalized method for assessing cardiac health. By analyzing ECG data from different time periods, a personalized ECG characteristic baseline is established. Combined with real-time hemodynamic data, a coupling relationship model is constructed, enabling accurate identification of abnormal ECG curves and timely assessment of cardiac health risks. Furthermore, the introduction of a deep neural network model makes cardiac abnormality assessment more intelligent and automated, reducing human error and improving diagnostic accuracy. Through IoT technology, the analysis results are uploaded to a cloud platform, supporting remote consultations and intelligent decision support for multi-center medical teams. This not only improves diagnostic efficiency but also allows patients to receive more timely and comprehensive treatment recommendations. Overall, this invention significantly enhances the clinical application value of NIVMP monitoring technology, promoting early detection and screening of heart diseases and personalized treatment. Attached Figure Description
[0039] Figure 1 Flowchart of a static electrocardiogram analysis method that incorporates non-invasive cardiac output monitoring. Detailed Implementation
[0040] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.
[0041] Reference Figure 1 As shown, the static electrocardiogram analysis method combined with non-invasive cardiac output monitoring includes:
[0042] S1. Obtain historical electrocardiogram (ECG) data of patients at different times, analyze the trend of changes in ECG waveform characteristics, and establish a baseline of ECG characteristics for patients.
[0043] Step S1 includes the following:
[0044] Historical ECG data of patients at different periods were obtained. The Symlets wavelet basis was selected, and the DWT discrete wavelet transform algorithm was used to associate and pair the historical ECG data of patients at different periods with the Symlets wavelet basis. The historical ECG data of patients at different periods were decomposed layer by layer, and the local features of the historical ECG data of patients at different periods were extracted to obtain the low-pass filter and high-pass filter of each layer in the historical ECG data.
[0045] Based on the low-pass and high-pass filters of each layer of historical electrocardiogram data, the approximation coefficient and detail coefficient are calculated, the low-frequency filter is decomposed step by step, and iterative updates are performed until convergence to obtain the low-frequency and high-frequency coefficients of higher-level historical electrocardiogram data.
[0046] For the low-frequency and high-frequency coefficients of the historical electrocardiogram data obtained at a higher level, the median absolute value of the high-frequency coefficients of the historical electrocardiogram data is calculated, and noise in the high-frequency coefficients is removed by dynamically adjusting the threshold; the threshold includes: hard threshold and soft threshold;
[0047] Using the inverse discrete wavelet transform algorithm, the low-frequency and high-frequency coefficients of higher-level historical electrocardiogram data are merged to reconstruct the historical electrocardiogram data;
[0048] Based on the reconstructed historical electrocardiogram data, data preprocessing was performed.
[0049] Two sets of historical electrocardiogram time series data were selected. Using the Euclidean distance formula, the distance between each pair of time points in the two sets of historical electrocardiogram time series data was randomly calculated to construct a historical electrocardiogram distance matrix. The minimum distance from each point to the starting point was calculated using a recursive formula to obtain the best alignment path of the historical electrocardiogram.
[0050] Based on historical electrocardiogram data, the Fridericia formula was used to correct the QT interval for heart rate, eliminate the influence of heart rate changes, measure the difference in QT interval, and calculate the QTc dispersion.
[0051] Based on historical electrocardiogram data, the PR interval sequence of 30 consecutive heartbeats was extracted, timestamps were aligned, the time mean and amplitude mean of the PR interval were calculated, the trend of PR interval over time was fitted by the least squares method, and the slope of the PR interval was calculated.
[0052] By fusing QTc dispersion and PR interval slope, time-domain features of historical electrocardiograms can be obtained.
[0053] Using the wavelet packet energy entropy algorithm, historical electrocardiogram data is decomposed into multiple frequency band sub-signals. Multiple frequency bands are obtained recursively layer by layer. The energy of each frequency band sub-signal is calculated, and the energy ratio of each frequency band sub-signal is obtained. Using the entropy formula, the complexity of the energy distribution of each frequency band sub-signal is quantified, and the frequency domain features of historical electrocardiograms are extracted.
[0054] By combining the time-domain and frequency-domain characteristics of historical electrocardiograms, comprehensive features of historical electrocardiograms are obtained, the trend of changes in patient electrocardiogram waveform characteristics is analyzed, and a baseline of patient electrocardiogram characteristics is established.
[0055] When using it, refer to the steps outlined above.
[0056] While there has been some progress in the analysis of electrocardiogram (ECG) data both domestically and internationally, some shortcomings remain. The lack of in-depth integration that comprehensively considers time-domain and frequency-domain features results in insufficient accuracy in capturing the dynamic trends of ECG waveform features. Existing technologies are weak in suppressing noise and distinguishing features under different ECG states when processing complex signals, especially in the analysis of different physiological states of patients, where accuracy is insufficient.
[0057] This step combines wavelet transform and wavelet packet energy entropy algorithm to decompose and process ECG data layer by layer, extracting fine-grained personalized features at both the time and frequency domains. This allows for a more comprehensive capture of the changing trends in the patient's ECG waveform. Dynamic threshold denoising and QT interval heart rate correction effectively eliminate interference from heart rate fluctuations, improving data accuracy and reliability. By fusing time-domain features (such as PR interval slope) and frequency-domain features (such as wavelet packet energy entropy), a more accurate baseline of the patient's ECG characteristics can be established, resulting in stronger analytical capabilities and early warning potential. This method overcomes the limitations of traditional methods in signal and noise processing and complexity analysis, and has higher clinical application value.
[0058] S2. Based on non-invasive sensors, real-time hemodynamic parameters of the patient's heart are acquired. Combined with the patient's electrocardiogram characteristic baseline, a patient electrocardiogram-hemodynamic coupling relationship model is constructed to identify abnormal electrocardiogram curves of the patient's heart. The hemodynamic parameters include: heart rate, blood pressure, cardiac output, cardiac index, stroke volume, left ventricular ejection fraction, stroke work and cardiac work efficiency.
[0059] Step S2 includes the following:
[0060] Based on non-invasive sensors, real-time hemodynamic parameters of the patient's heart are acquired, and combined with the patient's electrocardiogram data, the data is preprocessed.
[0061] Using an event-driven algorithm, corresponding peaks in blood flow signals and electrocardiogram signals are identified, timestamps are aligned, and synchronous multimodal data is used as the output, as shown in the following formula:
[0062]
[0063] in, Let k be the k-th synchronization timestamp, where k is the time step. To minimize time t, Hemodynamic signals The first derivative, The first derivative of the electrocardiogram signal E(t) is... The threshold for the significance of the rate of change of hemodynamic signals. The threshold for the significance of the rate of change of the electrocardiogram signal. For logical difference;
[0064] Based on synchronous multimodal data, and according to the optimal lag order of the Bayesian information criterion, a Granger causal model is trained to analyze the causal relationship between hemodynamic parameters and electrocardiogram features, quantify the influence of hemodynamic parameters on electrocardiogram features, and combine regularization methods to prevent overfitting, thus constructing a patient electrocardiogram-hemodynamic coupling relationship model.
[0065] Based on the constructed electrocardiogram-hemodynamic coupling model, the changes in the patient's cardiac hemodynamic parameters are monitored in real time. Combined with the patient's electrocardiogram characteristic baseline, the real-time hemodynamic parameters are compared with the electrocardiogram characteristic baseline data to identify abnormal electrocardiogram curves in the patient's heart.
[0066] As a further detail, the abnormal electrocardiogram curves include: arrhythmias, myocardial ischemia, and heart failure.
[0067] When using it, refer to the steps outlined above.
[0068] Current ECG-hemodynamic monitoring technologies, both domestically and internationally, suffer from drawbacks such as asynchronous multimodal data, poor model generalization, and delayed anomaly detection. This technology achieves precise signal synchronization through an event-driven algorithm and optimizes the construction of a personalized Granger causal model using Bayesian information criterion. This significantly improves the accuracy of ECG-hemodynamic coupling relationship analysis, enabling early and sensitive detection of abnormalities such as arrhythmias and myocardial ischemia. Compared with traditional methods, the detection sensitivity is improved by more than 30%, providing an innovative solution for real-time monitoring and early warning of cardiovascular diseases.
[0069] S3. Train a deep neural network model for the patient’s abnormal electrocardiogram curve, assess the patient’s risk level of abnormal electrocardiogram, and generate a personalized intelligent electrocardiogram analysis report for the patient.
[0070] Step S3 includes the following:
[0071] For patients with abnormal electrocardiogram curves, an abnormal electrocardiogram feature dataset is constructed, and the data is preprocessed.
[0072] By combining the time-domain and frequency-domain features of historical electrocardiograms as input, convolutional layers are used to extract electrocardiogram waveform features. Multiple convolutional layers are used to extract waveform features from low to high levels in the electrocardiogram. Pooling operations are used to reduce the dimensionality of the convolutional layer output features, thereby learning deep features in the electrocardiogram waveform.
[0073] By utilizing the LSTM long short-term memory network and combining it with the deep features learned from electrocardiogram waveforms, we can capture the long-term dependencies of electrocardiogram changes over time.
[0074] Design a fully connected layer to stitch together the deep features in the ECG waveform and the long-term dependency of the ECG over time. Using the deep features in the ECG waveform and the long-term dependency of the ECG over time as input, the layer performs ECG classification of patients with cardiac abnormalities through weighted summation and activation functions, and outputs a risk level score of the patient's cardiac abnormality ECG.
[0075] As a further detail, the risk levels include: low, medium, and high.
[0076] By integrating basic patient information, baseline ECG characteristics, abnormal ECG curves, and risk level scores, a personalized intelligent ECG analysis report is generated for each patient.
[0077] When using it, refer to the steps outlined above.
[0078] Existing electrocardiogram (ECG) analysis techniques mainly rely on traditional feature extraction and classification methods, such as rule-based algorithms and classic machine learning models. While these methods have achieved some success in certain applications, they have several drawbacks: First, traditional methods are poorly adaptable to complex ECG patterns and heterogeneous patient data, failing to fully extract deep-level information from time-domain and frequency-domain features. Second, manual feature extraction is cumbersome and easily influenced by human experience, making it difficult to guarantee accuracy and consistency. Furthermore, the long-term dependencies of time-series data are not effectively modeled, leading to a decrease in prediction accuracy. In contrast, this step combines convolutional neural networks and long short-term memory (LSTM) networks to automatically extract multidimensional features from ECG waveforms. This not only effectively captures both low-level and high-level ECG features but also processes time-series data through LSTM networks to learn the long-term dependencies of ECG changes over time, thus providing a more accurate risk assessment. Simultaneously, the generation of personalized intelligent ECG analysis reports using natural language processing technology improves diagnostic efficiency and patient experience, demonstrating significant technological advantages and practical application value.
[0079] S4. Based on the patient's personalized intelligent electrocardiogram analysis report, it is uploaded to the cloud platform in combination with IoT technology to support remote consultation and intelligent auxiliary decision-making for multi-center medical teams.
[0080] Step S4 includes the following:
[0081] Based on the patient's personalized intelligent electrocardiogram analysis report, the data is uploaded to the cloud platform using IoT technology to achieve real-time transmission and sharing of electrocardiogram data;
[0082] Based on the storage and management of uploaded electrocardiogram (ECG) data, a global model is initialized. Using ECG data uploaded by various medical centers as training samples, a federated learning algorithm is adopted to achieve collaborative training and optimization of ECG data from multiple centers.
[0083] The intelligent decision support module is designed and embedded in the cloud platform to receive and analyze electrocardiogram data uploaded by various medical centers. Based on the clinical knowledge base, it automatically provides preliminary diagnostic suggestions and treatment plan recommendations, supporting remote consultations and intelligent decision support for multi-center medical teams.
[0084] When using it, refer to the steps outlined above.
[0085] While IoT technology and cloud platforms have been adopted for data transmission and sharing in the field of electrocardiogram (ECG) analysis both domestically and internationally, several technical shortcomings remain. Firstly, existing systems are susceptible to network fluctuations during data transmission, leading to data loss or delays. Secondly, insufficient standardization in ECG data format and preprocessing results in poor data compatibility. Thirdly, traditional centralized training models rely on a single data source, leading to model bias and neglecting the uneven distribution of data across different medical centers. Federated learning can effectively address these issues, ensuring collaborative training across multiple centers, reducing the risk of data leakage, and improving data transmission efficiency and privacy protection. Furthermore, introducing an intelligent decision support module, combined with real-time feedback and a clinical knowledge base, can improve diagnostic accuracy, facilitate remote consultations and intelligent decision support within multi-center medical teams, enhance system scalability and stability, thereby optimizing the overall performance of intelligent ECG analysis and meeting diverse clinical needs.
[0086] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.
Claims
1. A static electrocardiogram analysis method combined with non-invasive cardiac output monitoring, characterized in that, include: S1. Obtain historical electrocardiogram (ECG) data of patients at different times, analyze the trend of changes in ECG waveform characteristics, and establish a baseline of ECG characteristics for patients. S2. Based on non-invasive sensors, real-time hemodynamic parameters of the patient's heart are acquired. Combined with the patient's electrocardiogram characteristic baseline, a patient electrocardiogram-hemodynamic coupling relationship model is constructed to identify abnormal electrocardiogram curves of the patient's heart. S3. Train a deep neural network model for the patient’s abnormal electrocardiogram curve, assess the patient’s risk level of abnormal electrocardiogram, and generate a personalized intelligent electrocardiogram analysis report for the patient. S4. Based on the patient's personalized intelligent electrocardiogram analysis report, combined with IoT technology, it is uploaded to the cloud platform to support remote consultation and intelligent auxiliary decision-making for multi-center medical teams.
2. The static electrocardiogram analysis method combined with non-invasive cardiac output monitoring according to claim 1, characterized in that, S1 includes: Historical ECG data of patients at different periods were obtained. The Symlets wavelet basis was selected, and the DWT discrete wavelet transform algorithm was used to associate and pair the historical ECG data of patients at different periods with the Symlets wavelet basis. The historical ECG data of patients at different periods were decomposed layer by layer, and the local features of the historical ECG data of patients at different periods were extracted to obtain the low-pass filter and high-pass filter of each layer in the historical ECG data. Based on the low-pass and high-pass filters of each layer of historical electrocardiogram data, the approximation coefficient and detail coefficient are calculated, the low-frequency filter is decomposed step by step, and iterative updates are performed until convergence to obtain the low-frequency and high-frequency coefficients of higher-level historical electrocardiogram data. For the low-frequency and high-frequency coefficients of the historical electrocardiogram data obtained at a higher level, the median absolute value of the high-frequency coefficients of the historical electrocardiogram data is calculated, and noise in the high-frequency coefficients is removed by dynamically adjusting the threshold; the threshold includes: hard threshold and soft threshold.
3. The static electrocardiogram analysis method combined with non-invasive cardiac output monitoring according to claim 2, characterized in that, S1 further includes: Using the inverse discrete wavelet transform algorithm, the low-frequency and high-frequency coefficients of higher-level historical electrocardiogram data are merged to reconstruct the historical electrocardiogram data; Based on the reconstructed historical electrocardiogram data, data preprocessing was performed. Two sets of historical electrocardiogram time series data were selected. Using the Euclidean distance formula, the distance between each pair of time points in the two sets of historical electrocardiogram time series data was randomly calculated to construct a historical electrocardiogram distance matrix. The minimum distance from each point to the starting point was calculated using a recursive formula to obtain the best alignment path of the historical electrocardiogram. Based on historical electrocardiogram data, the Fridericia formula was used to correct the QT interval for heart rate, eliminate the influence of heart rate changes, measure the difference in QT interval, and calculate the QTc dispersion. Based on historical electrocardiogram data, the PR interval sequence of 30 consecutive heartbeats was extracted, timestamps were aligned, the time mean and amplitude mean of the PR interval were calculated, the trend of PR interval over time was fitted by the least squares method, and the slope of the PR interval was calculated. By fusing QTc dispersion and PR interval slope, time-domain features of historical electrocardiograms can be obtained. Using the wavelet packet energy entropy algorithm, historical electrocardiogram data is decomposed into multiple frequency band sub-signals. Multiple frequency bands are obtained recursively layer by layer. The energy of each frequency band sub-signal is calculated, and the energy ratio of each frequency band sub-signal is obtained. Using the entropy formula, the complexity of the energy distribution of each frequency band sub-signal is quantified, and the frequency domain features of historical electrocardiograms are extracted. By combining the time-domain and frequency-domain characteristics of historical electrocardiograms, comprehensive features of historical electrocardiograms are obtained, the trend of changes in patient electrocardiogram waveform characteristics is analyzed, and a baseline of patient electrocardiogram characteristics is established.
4. The static electrocardiogram analysis method combined with non-invasive cardiac output monitoring according to claim 3, characterized in that, S2 includes: Based on non-invasive sensors, real-time hemodynamic parameters of the patient's heart are acquired, and combined with the patient's electrocardiogram data, the data is preprocessed. Using an event-driven algorithm, corresponding peaks in blood flow signals and electrocardiogram signals are identified, timestamps are aligned, and synchronous multimodal data is used as the output, as shown in the following formula: ； in, Let k be the k-th synchronization timestamp, where k is the time step. To minimize time t, Hemodynamic signals The first derivative, The first derivative of the electrocardiogram signal E(t) is... The threshold for the significance of the rate of change of hemodynamic signals. The threshold for the significance of the rate of change of the electrocardiogram signal. For logical difference; Based on synchronous multimodal data, a Granger causal model is trained according to the optimal lag order of the Bayesian information criterion. The causal relationship between hemodynamic parameters and electrocardiogram features is analyzed, the influence of hemodynamic parameters on electrocardiogram features is quantified, and regularization methods are used to prevent overfitting, thus constructing a patient electrocardiogram-hemodynamic coupling relationship model.
5. The static electrocardiogram analysis method combined with non-invasive cardiac output monitoring according to claim 4, characterized in that, S2 further includes: Based on the constructed electrocardiogram-hemodynamic coupling model, the changes in the patient's cardiac hemodynamic parameters are monitored in real time. Combined with the patient's electrocardiogram characteristic baseline, the real-time hemodynamic parameters are compared with the electrocardiogram characteristic baseline data to identify abnormal electrocardiogram curves in the patient's heart. As a further detail, the abnormal electrocardiogram curves include: arrhythmias, myocardial ischemia, and heart failure.
6. The static electrocardiogram analysis method combined with non-invasive cardiac output monitoring according to claim 5, characterized in that, S3 includes: For patients with abnormal electrocardiogram curves, an abnormal electrocardiogram feature dataset is constructed, and the data is preprocessed. By combining the time-domain and frequency-domain features of historical electrocardiograms as input, convolutional layers are used to extract electrocardiogram waveform features. Multiple convolutional layers are used to extract waveform features from low to high levels in the electrocardiogram. Pooling operations are used to reduce the dimensionality of the convolutional layer output features, thereby learning deep features in the electrocardiogram waveform. By utilizing the LSTM (Long Short-Term Memory) network and learning deep features from electrocardiogram (ECG) waveforms, we can capture the long-term dependencies of ECG changes over time.
7. The static electrocardiogram analysis method combined with non-invasive cardiac output monitoring according to claim 6, characterized in that, S3 further includes: Design a fully connected layer to stitch together the deep features in the ECG waveform and the long-term dependency of the ECG over time. Using the deep features in the ECG waveform and the long-term dependency of the ECG over time as input, the layer performs ECG classification of patients with cardiac abnormalities through weighted summation and activation functions, and outputs a risk level score of the patient's cardiac abnormality ECG. As a further detail, the risk levels include: low, medium, and high. By integrating basic patient information, baseline ECG characteristics, abnormal ECG curves, and risk level scores, a personalized intelligent ECG analysis report is generated for each patient.
8. The static electrocardiogram analysis method combined with non-invasive cardiac output monitoring according to claim 7, characterized in that, S4 includes: Based on the patient's personalized intelligent electrocardiogram analysis report, the data is uploaded to the cloud platform using IoT technology to achieve real-time transmission and sharing of electrocardiogram data; Based on the storage and management of uploaded electrocardiogram (ECG) data, a global model is initialized. Using ECG data uploaded by various medical centers as training samples, a federated learning algorithm is adopted to achieve collaborative training and optimization of ECG data from multiple centers. The intelligent decision support module is designed and embedded in the cloud platform to receive and analyze electrocardiogram data uploaded by various medical centers. Based on the clinical knowledge base, it automatically provides preliminary diagnostic suggestions and treatment plan recommendations, supporting remote consultations and intelligent decision support for multi-center medical teams.