Multi-channel cardiovascular monitoring method, device and system and storage medium

Abstract

The invention discloses a multi-channel cardiovascular monitoring method, device and system and a storage medium. The multi-channel cardiovascular monitoring method comprises the following steps: S1, acquiring multi-band impedance data of human skin and chest; s2, performing fusion processing on the multi-band impedance data, and constructing a high-frequency deep human body impedance model; s3, on the basis of the fused multi-band impedance data, equivalent impedance parameters are calculated through a high-frequency impedance model, the impedance parameters are input into a customized neural network model improved for impedance signal characteristics, and cardiovascular parameters of the heart rate, the respiratory rate, the pulsation volume and the pulsation index are output through the neural network model; and S4, constructing a cardiovascular abnormal probability model based on the cardiovascular parameters, and outputting real-time health state evaluation and multi-level risk early warning signals. By adopting the technical scheme of the invention, comprehensive acquisition and intelligent analysis of cardiovascular related physiological parameters can be realized, and cross-terminal remote health monitoring and risk early warning are supported.

Description

Technical Field

[0001] The present invention belongs to the field of medical health monitoring technology, and in particular relates to a multi-channel cardiovascular monitoring method and device, system, and storage medium. Background Art

[0002] The functional status of the cardiovascular system directly reflects a person's health. Early screening and real-time monitoring of cardiovascular disease are key to preventing sudden death and reducing the burden of healthcare. Research has shown that cardiovascular parameters are closely related to autonomic nervous system activity and are susceptible to factors such as mood swings and physiological stress. Non-invasive and continuous acquisition of cardiovascular-related physiological signals and the establishment of high-precision prediction models have significant clinical value in enabling personalized health management and improving disease early warning capabilities.

[0003] Bioelectrical impedance analysis (BIA) technology indirectly obtains physiological information such as blood flow and fluid distribution by measuring the impedance response of human tissue to alternating current. Because human tissue is a complex coupled system of resistance and capacitance, factors considered include but are not limited to myocardial conductance, blood resistance, and lung capacitance effects. Its impedance characteristics are highly correlated with cardiovascular function. For example, myocardial ischemia can lead to increased local tissue resistivity, while changes in blood volume can cause fluctuations in the imaginary part of impedance.

[0004] Traditional cardiovascular monitoring technologies mainly rely on electrocardiogram, ultrasound Doppler or optical plethysmography, which have the following problems:

[0005] 1. Invasiveness and high cost: For example, cardiac catheterization is invasive and CT / MRI equipment is expensive;

[0006] 2. Single biomarker reliance: Existing BIA devices often rely on a single parameter and lack specificity;

[0007] 3. Poor real-time performance: Traditional methods make it difficult to achieve continuous monitoring and immediate warning.

[0008] Although bioelectrical impedance-based monitoring technology has the advantages of being non-invasive and low-cost, existing methods have the following shortcomings:

[0009] 1. Only supports single-band, single-channel measurement: Traditional BIA uses a single frequency band and a single channel to collect impedance data. It cannot distinguish between the skin capacitance effect and deep tissue impedance. Therefore, fluctuations in skin contact impedance and environmental electromagnetic interference can easily affect data stability, resulting in significant errors in hemodynamic parameters.

[0010] 2. Time series modeling flaws: Traditional RNN, LSTM, and other models have low feature extraction efficiency for long-time series impedance data, making it difficult to capture the non-stationary and nonlinear characteristics of high-frequency physiological signals, and the accuracy of deep tissue parameter extraction is insufficient.

[0011] 3. Lack of remote networking capabilities: Existing equipment lacks multi-terminal collaborative early warning and cloud data fusion mechanisms, making it difficult to support large-scale health management needs.

[0012] In summary, the existing cardiovascular parameter monitoring technology still needs to be improved in terms of comprehensive data collection, intelligent analysis, and remote connectivity. A new technical solution is urgently needed to overcome the above shortcomings and realize continuous monitoring, intelligent analysis, and remote early warning of multiple physiological parameters. Summary of the Invention

[0013] The technical problem to be solved by the present invention is to provide a multi-channel cardiovascular monitoring method and device, system, and storage medium to solve the problems of single monitoring parameters, insufficient data analysis, and lack of remote connection in the existing technology. It can realize the comprehensive collection and intelligent analysis of cardiovascular-related physiological parameters, and support remote health monitoring and risk warning across terminals.

[0014] To achieve the above object, the present invention adopts the following technical solutions:

[0015] A multi-channel cardiovascular monitoring method, comprising:

[0016] Step S1, obtaining multi-band impedance data of human skin and chest cavity;

[0017] Step S2: fusing the multi-band impedance data to construct a high-frequency deep human body impedance model;

[0018] Step S3: Based on the fused multi-band impedance data, equivalent impedance parameters are calculated using a high-frequency impedance model, the impedance parameters are input into a customized neural network model improved for impedance signal characteristics, and the neural network model is used to output cardiovascular parameters such as heart rate, respiratory rate, stroke volume, and pulsatility index;

[0019] Step S4: constructing a cardiovascular abnormality probability model based on the cardiovascular parameters, and outputting a real-time health status assessment and a multi-level risk warning signal.

[0020] Preferably, in step S2, the multi-band impedance data is subjected to fusion processing, including dynamic baseline calibration, frequency domain noise suppression and cross-band feature extraction, to establish a high-frequency deep human body impedance model.

[0021] Preferably, in step S3, the neural network model is constructed based on state space modeling, including a spectrum-time domain feature fusion embedding mechanism and a sparse feedforward enhancement layer.

[0022] The present invention also provides a multi-channel cardiovascular monitoring device, comprising:

[0023] The first processing module is used to obtain multi-band impedance data of human skin and chest cavity;

[0024] A second processing module is used to fuse the multi-band impedance data to construct a high-frequency deep human body impedance model;

[0025] a third processing module, configured to calculate equivalent impedance parameters using a high-frequency impedance model based on the fused multi-band impedance data, input the impedance parameters into a customized neural network model improved for impedance signal characteristics, and use the neural network model to output cardiovascular parameters such as heart rate, respiratory rate, stroke volume, and pulsatility index;

[0026] The fourth processing module is used to construct a cardiovascular abnormality probability model based on the cardiovascular parameters and output a real-time health status assessment and a multi-level risk warning signal.

[0027] Preferably, the second processing module performs fusion processing on the multi-band impedance data, including dynamic baseline calibration, frequency domain noise suppression and cross-band feature extraction, to establish a high-frequency deep human body impedance model.

[0028] Preferably, the neural network model is constructed based on state space modeling, and includes a spectrum-time domain feature fusion embedding mechanism and a sparse feedforward enhancement layer.

[0029] The present invention also provides a multi-channel cardiovascular monitoring system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes the multi-channel cardiovascular monitoring method when executed by the processor.

[0030] The present invention also provides a storage medium having a computer program stored thereon, and the computer program executes the multi-channel cardiovascular monitoring method when running.

[0031] The present invention collects raw impedance data through multi-channel and multi-band excitation, establishes a high-frequency impedance model, and then introduces an improved state-space neural network model customized for the characteristics of physiological impedance signals to achieve efficient modeling of sparse signals of abnormal data. The neural network model adopts a spectrum-time domain feature fusion mechanism, a bidirectional state space unit and a sparse feedforward enhancement structure, which significantly improves the detection capability of abnormal physiological parameters and the real-time risk warning performance. The device integrates data acquisition and wireless communication functions, and the client supports data processing, model reasoning, dynamic waveform display and multi-user network monitoring. It supports multi-user network monitoring and is suitable for active health management of families, communities and medical institutions, providing an efficient solution for early screening of cardiovascular diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0033] Figure 1 This is a flow chart of a multi-channel cardiovascular monitoring method according to an embodiment of the present invention;

[0034] Figure 2 This is the block diagram of the impedance acquisition system;

[0035] Figure 3 This is a schematic diagram of multi-channel electrode attachment;

[0036] Figure 4 is the linear equivalent circuit model of skin impedance;

[0037] Figure 5 Schematic diagram of the fitting curve for high-frequency deep human body impedance modeling;

[0038] Figure 6 Flowchart for cardiovascular status assessment;

[0039] Figure 7 A diagram of a custom neural network architecture;

[0040] Figure 8 This is a schematic diagram of remote networking technology. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] Example 1:

[0044] like Figure 1 As shown, an embodiment of the present invention provides a multi-channel cardiovascular monitoring method, comprising:

[0045] Step 1: Obtain raw impedance data of human skin and chest cavity through a multi-frequency, multi-channel bioelectrical impedance acquisition device. The data includes the real and imaginary impedance parts. The data is transmitted to a remote terminal through a wireless communication module, and the real-time waveform is dynamically displayed on the client.

[0046] Step 2: performing fusion processing on the multi-band impedance data, including dynamic baseline calibration, frequency domain noise suppression and cross-band feature extraction, to establish a high-frequency deep human body impedance model;

[0047] Step 3: Based on the fused multi-frequency data, equivalent impedance parameters are calculated using a high-frequency impedance model. These impedance parameters are then input into a customized neural network model optimized for impedance signal characteristics. The neural network model is constructed based on state-space modeling and includes a spectral-temporal feature fusion embedding mechanism and a sparse feedforward enhancement layer. The neural network model is used to output cardiovascular parameters such as heart rate, respiratory rate, stroke volume, and pulsatility index.

[0048] Step 4: Build a cardiovascular abnormality probability model based on the physiological parameters, output real-time health status assessment and multi-level risk warning signals, and realize cloud data synchronization and collaborative warning through a multi-terminal, cross-scenario networking monitoring architecture to complete visual interaction.

[0049] As an implementation of the embodiment of the present invention, step S1 includes the following steps:

[0050] Step 1.1: Use a distributed electrode array with no fewer than three channels to synchronously collect bioelectrical impedance data, including but not limited to, data from the chest, neck, and limbs. The excitation frequency range should cover a logarithmically distributed frequency range of 100 Hz to 300 kHz, with at least three frequency bands.

[0051] Step 1.2: Configure a dynamic gain amplifier to match the change in skin contact impedance in real time. The raw data, including the real and imaginary impedance parts, modulus, and phase angle, is transmitted to the client via the wireless communication module and dynamically displays the multi-channel time-frequency waveform.

[0052] As an implementation of an embodiment of the present invention, step S2 includes the following steps:

[0053] Step 2.1: Construct a high-frequency deep human body impedance model to suppress the capacitance effect of the epidermal stratum corneum;

[0054] Step 2.2: Based on the real and imaginary impedance data at multiple frequency points, perform a linear fit on the impedance spectrum using the least squares method. The modeling formula is:

[0055] R=k·X+R0

[0056] Where R is the real part of the impedance, X is the imaginary part of the impedance, k is the fitting slope, and R0 is the intercept of the fitting line with the real axis, which represents the high-frequency deep human body resistance value.

[0057] As an implementation of an embodiment of the present invention, step S3 includes the following steps:

[0058] Step 3.1: Data preprocessing: The impedance model fitting results obtained in step 2, i.e., the slope k and intercept R0 of the high-frequency deep impedance model, are subjected to frequency-domain wavelet denoising and smoothing to improve signal stability. Furthermore, cross-band feature fusion is used to further enhance the robustness and consistency of the deep impedance features.

[0059] Step 3.2: Temporal feature extraction:

[0060] A state-space neural network modified for physiological impedance characteristics was used, with the fitting parameters slope k and intercept R0 obtained in step 3.1 as the input sequence. The spectral-temporal fusion embedding layer (STE-Conv1D) was used to extract composite features, the bidirectional state-space unit (B-SSMU) was used to model long-term and short-term sequence dependencies, and the sparse enhanced feedforward layer (SEE-FFN) was used to highlight abnormal data points, resulting in the output of time series features. The network structure met the following requirements:

[0061] h t =SEE-FFN(B-SSMU(STE-Conv1D(k,R0)))

[0062] Where k is the slope obtained by high-frequency impedance spectrum fitting, R0 is the deep human body resistance value obtained by high-frequency fitting, STE-Conv1D represents the spectrum-time domain embedding convolution operation, B-SSMU is the bidirectional state space modeling unit, and SEE-FFN is the sparse anomaly enhancement feedforward layer;

[0063] Step 3.3: Physiological parameter prediction: The time series feature vector output from step 3.2 is fed into the regression prediction layer to output physiological parameters, including key cardiovascular physiological indicators such as heart rate, respiratory rate, cardiac output per stroke, and blood flow pulsatility index, thereby achieving effective monitoring and early warning of abnormal cardiovascular functional states.

[0064] As an implementation of the embodiment of the present invention, step S4 includes the following steps:

[0065] Step 4.1: Deploy a cloud server to achieve real-time synchronization of data from multiple devices, supporting multi-level data sharing among medical institutions, communities, and families;

[0066] Step 4.2: The client integrates a three-level dynamic early warning module for high risk, medium risk, and low risk, dynamically displaying a risk heat map and a historical trend comparison chart;

[0067] Step 4.3: Abnormal events trigger cross-terminal collaborative alarms, including SMS, APP push, and medical end pop-up windows.

[0068] Example 2:

[0069] An embodiment of the present invention provides a multi-channel cardiovascular monitoring method, comprising:

[0070] Step 1: Multi-frequency impedance measurement and signal modeling

[0071] Step 1.1: Use a multi-channel multi-frequency bioimpedance measurement module to perform impedance acquisition.

[0072] like Figure 2 As shown, the impedance measurement module of the present invention includes an electrode array, an excitation source, a measurement circuit, an analog-to-digital converter and a wireless module.

[0073] The distributed electrode array preferably adopts a configuration of more than 3 channels and is attached to different parts of the human body at the same time to introduce weak AC current and measure the voltage response of different pathways. The excitation source provides a multi-frequency sinusoidal signal current with a frequency ranging from 100Hz to 300kHz at logarithmic intervals to fully characterize the impedance characteristics of the tissue at different depths. The measurement circuit amplifies and synchronously demodulates the voltage signal of each channel to extract the real and imaginary components of the impedance respectively. Figure 3 and as shown in Table 1.

[0074] Table 1

[0075] name Chest adhesive position F 4th intercostal space on the left side of the sternum B The back corresponds to the posterior wall of the heart L 5th intercostal space at the left midaxillary line R 5th intercostal space at the right midaxillary line U 2nd intercostal space at the left midclavicular line (base of the heart) D Lower edge of the left costal arch (apex of the heart)

[0076] To adapt to the dynamic changes in human skin resistance, an automatic gain control mechanism is incorporated into the measurement circuit to match the contact impedance and ensure the signal remains within the optimal measurement range. The raw impedance data measured on each channel, including parameters such as the real and imaginary impedance parts, amplitude, and phase, is time-segmented by a multiplexer to achieve multi-channel synchronous scanning. This involves high-speed polling, sequentially measuring each channel at each specified frequency, to quickly acquire an equivalent, synchronized multi-channel, multi-frequency impedance spectrum. The collected impedance data is transmitted to a host computer via a wireless module, enabling remote, real-time monitoring.

[0077] Step 1.2: Perform multi-frequency data fusion and high-frequency impedance modeling

[0078] In order to extract reliable physiological features from multi-band impedance data, the present invention proposes a deep impedance modeling method based on high-frequency band fitting.

[0079] First, the linear equivalent circuit model of skin impedance is introduced, such as Figure 4 shown.

[0080] The linear resistance R sTo simulate the impedance of deep skin tissue, linear capacitance C p and linear resistance R p The parallel combination is used to simulate the impedance of the stratum corneum of the skin epidermis.

[0081] The complex impedance expression of this impedance model is:

[0082]

[0083] As the frequency increases, the skin surface capacitance effect gradually weakens, and the impedance model gradually tends to a pure resistance state. Therefore, the present invention selects high-frequency band data for impedance spectrum fitting and modeling to effectively remove the capacitance effect interference of the epidermal stratum corneum.

[0084] Secondly, dynamic baseline correction is performed on the impedance data of each channel to eliminate slow drift introduced by electrode polarization and ambient temperature changes. Subsequently, filtering and wavelet denoising are applied in the frequency domain to suppress the impact of narrowband noise, such as power frequency interference, on the impedance at each frequency point. After preprocessing, the impedance response in the high-frequency band is focused to reduce the interference of the skin surface capacitance effect. The real and imaginary impedance values ​​of several high-frequency points are selected, and a least-squares method is used for linear fitting to establish a deep equivalent impedance model.

[0085] The modeling formula is expressed as:

[0086] R=k·X+R0

[0087] Among them, R is the real part of the impedance, X is the imaginary part of the impedance, k is the fitting slope, and R0 is the intercept of the fitting line and the real axis, that is, the real part value when the imaginary part tends to zero. d It is the equivalent resistance value of the deep tissue of the human body at high frequency, reflecting the pure resistance component after removing the influence of the epidermal capacitance.

[0088] In order to calculate the minimum residual sum of squares between the predicted value and the measured value and solve the optimal fitting parameters, the objective function is defined as follows:

[0089]

[0090] Where: N is the total number of multi-frequency measurement data; X i is the measured value of the imaginary part of the impedance at the i-th frequency point; R i is the measured value of the real part of the impedance at the i-th frequency point. By taking the partial derivatives of the objective function with respect to k and R0 and setting them to zero, the normal equations are obtained:

[0091]

[0092] The solution is:

[0093]

[0094]

[0095] like Figure 5 As shown, the real and imaginary components of the high-frequency impedance are approximately linear in the complex plane. The intercept, extrapolated to zero imaginary component, corresponds to the limit of the body's deep impedance. This parameter is closely related to physiological states such as thoracic blood volume and tissue fluid conductivity, and serves as a key characteristic for subsequent cardiovascular parameter calculations.

[0096] At the same time, the slope k of the multi-frequency fitting reflects the impedance dispersion characteristics and can be used to evaluate the capacitance effect at the skin-tissue interface.

[0097] The above multi-band data fusion and modeling process realizes the frequency domain feature fusion extraction, that is, using the information of multiple frequencies to more accurately separate the deep effective impedance signal, which greatly improves the sensitivity and specificity of physiological changes compared with single frequency measurement.

[0098] Step 2: Database construction and algorithm training

[0099] To improve the accuracy and robustness of the model, the present invention constructed a context-aware physiological signal database and used it to train and optimize the algorithm model offline. This database collects human body impedance and related physiological parameter data in various situations to support the training of deep learning models and the establishment of risk assessment models.

[0100] Step 2.1: Multi-scenario physiological data collection and database construction

[0101] The data sources of the database include multi-channel bioelectrical impedance signals and corresponding reference physiological parameters of subjects with diseases such as heart failure in different states such as rest, exercise, emotional fluctuations, and whether they have participated in previous treatment. In order to enrich the diversity of the data, the experiment selected volunteers of different age groups and genders to record for a long time by wearing the monitoring device of the present invention. In terms of scenario design, it includes basic data in a quiet and relaxed state, as well as physical activities such as aerobic exercise and posture changes, and scenes such as watching videos that induce tension / excitement and performing cognitive tasks that cause mental stress. In addition, there are also positive sample data collected from healthy people under the above conditions.

[0102] In each scenario, the subject's raw impedance data, as well as reference heart rate, respiratory rate, and other indicators, are synchronously recorded. These reference values ​​can be obtained by medical-grade equipment such as electrocardiographs and respiratory belts, and are annotated with the current scenario type and subjective state description. After data collection and preprocessing, a comprehensive database of multiple scenarios and long time series containing more than 20 subjects was constructed. Each data record includes: multi-channel multi-frequency impedance waveform sequence, extracted feature parameters, corresponding physiological indicator true values, and metadata such as scenario labels. The context-aware database established in this way covers a wide range of physiological change patterns, ensuring that the model training has a sufficiently diverse sample base.

[0103] Step 2.2: Model training and validation

[0104] Based on the multi-channel impedance database of real patients, the improved state-space neural network model is trained with offline multi-task supervision. The training uses high-frequency impedance fitting parameters as input and clinically measured heart rate, respiratory rate, SV, BPI, etc. as supervision labels. In the specific implementation, the convergence effect is improved through staged training: first, the basic physiological cycle is preliminarily captured with the state-space unit, and then the spectrum-time domain feature embedding layer and the sparse feedforward enhancement layer are used for joint optimization training. The present invention uses cross-validation and independent test sets to verify the model performance to ensure good generalization ability of the model. The results show that the improved neural network has significant accuracy advantages in the prediction of multi-task physiological parameters in complex situations and noisy environments.

[0105] Step 2.3: Context-aware optimization and continuous improvement

[0106] Utilize data with scene labels to further optimize the risk assessment mechanism for contextual awareness. For example, adjust the risk threshold based on the normal physiological response of the subject after exercise to avoid misjudging the increase in heart rate caused by exercise as abnormal; for temporary heart rate changes that occur in emotional stress scenarios, combine the scene labels to determine it as a short-term stress response, thereby reducing unnecessary alarms. After deployment, the model can also be continuously improved through cloud updates: the data of new users can be anonymously aggregated into the database after obtaining consent, continuously enriching the model's cognition of different groups of people and different situations. Through the above-mentioned database construction and training methods, the algorithm model of the present invention has the ability to perceive situational changes. It not only performs well under standard conditions, but also can adjust the interpretation of physiological signals according to the user's situation, improving the intelligence and reliability of monitoring and early warning.

[0107] Step 3: Physiological parameter extraction and state perception modeling

[0108] like Figure 6 As shown, the present invention deploys an improved time series neural network model customized for the characteristics of bioelectrical impedance signals on the host computer client to perform real-time physiological parameter extraction and status evaluation on continuously collected multi-channel impedance data.

[0109] Figure 7 The following schematically shows the composition and working principle of the customized neural network architecture of the present invention. Figure 7 As shown, the network first converts multi-frequency impedance parameters into fused feature representations through a spectrum-time domain feature fusion module, then bidirectionally models the feature sequence through forward and backward state space units, and finally extracts high-order features through a sparse feedforward layer and outputs cardiovascular parameter prediction results. Compared to existing general neural network architectures, the model of the present invention has been specially customized in terms of structure, input, and training methods to better fit the characteristics of bioelectrical impedance signals, thereby significantly improving the accuracy and reliability of multi-channel cardiovascular monitoring. The specific implementation method steps are as follows:

[0110] Step 3.1: Spectral-temporal fusion embedding feature extraction:

[0111] For the acquired impedance time series signal, the host computer first performs necessary time domain preprocessing, such as removing baseline drift and bandpass filtering to separate the frequency changes caused by heartbeat and respiration. The impedance waveform obtained after processing has the superposition of rapid oscillations mainly caused by heartbeat and slow fluctuations mainly caused by respiratory movement. The model uses the slope k and intercept R obtained by high-frequency impedance fitting. o The spectral feature vector is constructed by parameters and then fused with the time domain features, so that the spectral characteristics and instantaneous change characteristics of the impedance signal can be captured simultaneously, forming a high-dimensional embedded feature sequence with complete spectral and time domain information, thereby improving the characterization ability of complex biological signals.

[0112] Step 3.2: Bidirectional state space model timing modeling:

[0113] The improved neural network model includes a customized bidirectional state space modeling unit (B-SSMU). The B-SSMU processes the feature sequence through forward and backward state space equations respectively, which can be formally expressed as:

[0114]

[0115] in, and Represent the forward and backward state variables, u tis the fused feature input. This unit processes the aforementioned embedded impedance feature sequence in both forward and backward directions, which is equivalent to a bidirectional sequence modeling structure that can capture the global contextual dependencies of the impedance signal over the time series, thereby more accurately extracting long-term dependent features such as the heart rate cycle and respiratory cycle. This structure achieves deep modeling of the temporal characteristics of physiological signals while ensuring the miniaturization of the model, and is suitable for real-time operation on client devices. After being trained with a large amount of physiological data, the model can automatically learn rules that traditional algorithms need to set manually, such as identifying the heart rate cycle based on rapid perturbations of the chest impedance waveform, determining the respiratory cycle based on slow fluctuations, and inferring hemodynamic changes by combining synchronous changes in signals from different channels.

[0116] Step 3.3: Establishment of sparse abnormal event enhancement module:

[0117] To effectively handle a small number of significant but sparse abnormal events in impedance data, such as sudden impedance changes caused by patient movement or transient electrode changes, the present invention introduces a sparse event enhancement (SEE) feedforward layer. This layer activates and strengthens only the connection weights that are highly correlated with abnormal changes, highlighting and enhancing the response strength of abnormal data, thereby improving the model's sensitivity and generalization ability to sparse abnormal trends.

[0118] Step 3.4: Calculation of cardiovascular physiological parameters and status assessment

[0119] Based on the customized structure of steps 3.1 through 3.3 above, the output of the neural network model of the present invention includes real-time estimates of multiple cardiovascular physiological parameters. Based on the amplitude and frequency of the instantaneous impedance drop caused by each heartbeat in the impedance waveform, the current heart rate (HR) and relative changes in stroke volume (SV) are calculated; based on the cyclical fluctuations of the impedance baseline with respiration, the respiratory rate (RR) is calculated; and by integrating multi-band impedance amplitude changes and multi-channel signal characteristics, the blood pulse index (BPI) is estimated to reflect the strength of the peripheral circulation. By jointly analyzing these parameters, the model can provide a quantitative assessment of the current cardiovascular function status. For example, detecting an accelerated heart rate and a decreased BPI may indicate poor cardiac output; decreased heart rate variability may indicate autonomic nervous system dysregulation. By utilizing frequency-domain and time-domain information fusion and bidirectional modeling, the model can more reliably extract cardiac and respiratory cycle features from the impedance signal. Furthermore, through the sparse feedforward layer, the model maintains stable output performance even in the face of abnormal impedance fluctuations.

[0120] Compared to traditional threshold detection or filtering methods, the algorithm of this invention more reliably separates heartbeat and respiratory signals in noisy environments, accurately capturing subtle physiological changes. This ensures the accuracy and real-time nature of physiological parameter extraction, providing high-quality input for subsequent health risk assessments.

[0121] Step 4: Establish remote networking and risk warning mechanisms

[0122] like Figure 8 As shown, the present invention fully utilizes the Internet of Things technology to build a multi-terminal remote networking architecture and designs a three-level risk warning mechanism to achieve cloud-based sharing of monitoring data and timely warning of cardiovascular events. The specific implementation steps are as follows:

[0123] Step 4.1: Synchronize cloud data and collaborate with multiple terminals:

[0124] The monitoring client synchronizes the locally acquired raw impedance data and the analyzed physiological parameters to the cloud health management platform through a secure network connection. Multiple scattered monitoring devices can be connected to the cloud platform at the same time to form a telemedicine Internet of Things (IoMT) network to achieve centralized data management and cross-terminal access. Authorized users can access the cardiovascular data and historical trends of relevant individuals in real time by logging into the cloud platform or mobile application. This cloud networking method breaks the geographical restrictions, allowing medical professionals to remotely monitor changes in patients' indicators; at the same time, family members can also participate in care and understand the health status of the protected persons. When the monitoring equipment of an elderly person living alone detects an abnormality, children and community doctors can receive timely notifications and intervene to deal with it, thereby greatly improving the timeliness of responding to sudden cardiovascular events. The remote networking function expands the scope of application of the device of the present invention, making it not only a personal health device, but also an active health monitoring system that connects families and medical services.

[0125] Step 4.2: Visualization of cardiovascular status and risk level display

[0126] The client application provides an intuitive interface to display real-time monitoring data and health assessment results. The interface includes real-time dynamic drawing of multi-channel impedance waveforms, as well as trend charts and risk level indicators generated based on the analyzed physiological parameters. In particular, the present invention introduces a visualization method of risk heat maps, which displays the comprehensive health risk level over a period of time in a chart with varying color depths, intuitively indicating the distribution of different levels such as normal, warning, and dangerous on the time axis. At the same time, the interface displays historical curves of key parameters such as heart rate, respiratory rate, and BPI, making it convenient for users and medical staff to observe the changing trajectories of these parameters over time. Through the combination of heat maps and trend curves, users can quickly identify the time periods and patterns of abnormalities, and assist in judging whether the current cardiovascular state is gradually deteriorating or a short-term fluctuation. Such a visual design improves the efficiency of information transmission, allowing non-professional users to understand changes in their own health and seek help in a timely manner when needed.

[0127] Step 4.3: Establish a three-level risk warning mechanism:

[0128] Based on extracted physiological parameters and assessed health status indices, the system establishes a three-level risk warning mechanism. When indicators are within a safe range, the interface displays a normal state. When certain parameters exceed warning thresholds or exhibit abnormal trends, the system enters a medium-risk warning state, prompting the user through the interface and suggesting measures. If parameters deviate significantly from normal and meet the high-risk criteria, the system triggers a high-risk alert. The high-risk alert prompts the user to seek medical attention promptly with distinct visual and audible signals locally, while automatically sending an alert notification to pre-defined emergency contacts and medical service providers via the cloud platform. The entire warning triggering and notification process has a latency of less than 100 milliseconds, ensuring that relevant personnel in emergency situations receive alerts and take action in near-real time. Through this three-level early warning strategy, the system can implement progressive response measures based on the degree of risk, avoiding excessive alerts that disturb the user while ensuring that important warnings are not missed in critical moments. This embodies a design concept that balances safety and humanization.

[0129] Example 3:

[0130] An embodiment of the present invention further provides a multi-channel cardiovascular monitoring device, comprising:

[0131] The first processing module is used to obtain multi-band impedance data of human skin and chest cavity;

[0132] A second processing module is used to fuse the multi-band impedance data to construct a high-frequency deep human body impedance model;

[0133] a third processing module, configured to calculate equivalent impedance parameters using a high-frequency impedance model based on the fused multi-band impedance data, input the impedance parameters into a customized neural network model improved for impedance signal characteristics, and use the neural network model to output cardiovascular parameters such as heart rate, respiratory rate, stroke volume, and pulsatility index;

[0134] The fourth processing module is used to construct a cardiovascular abnormality probability model based on the cardiovascular parameters and output a real-time health status assessment and a multi-level risk warning signal.

[0135] As an implementation method of an embodiment of the present invention, the second processing module performs fusion processing on the multi-band impedance data, including dynamic baseline calibration, frequency domain noise suppression and cross-band feature extraction, to establish a high-frequency deep human body impedance model.

[0136] As an implementation method of an embodiment of the present invention, the neural network model is constructed based on state space modeling, and includes a spectrum-time domain feature fusion embedding mechanism and a sparse feedforward enhancement layer.

[0137] Example 4:

[0138] An embodiment of the present invention further provides a multi-channel cardiovascular monitoring system, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a multi-channel cardiovascular monitoring method when executed by the processor.

[0139] Example 5:

[0140] An embodiment of the present invention further provides a storage medium having a computer program stored thereon, and the computer program executes the multi-channel cardiovascular monitoring method when running.

[0141] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A multi-channel cardiovascular monitoring method, characterized in that: include: Step S1, obtaining multi-band impedance data of human skin and chest cavity; Step S2: fusing the multi-band impedance data to construct a high-frequency deep human body impedance model; Step S3: Based on the fused multi-band impedance data, equivalent impedance parameters are calculated using a high-frequency impedance model, the impedance parameters are input into a customized neural network model improved for impedance signal characteristics, and the neural network model is used to output cardiovascular parameters such as heart rate, respiratory rate, stroke volume, and pulsatility index; Step S4: constructing a cardiovascular abnormality probability model based on the cardiovascular parameters, and outputting a real-time health status assessment and a multi-level risk warning signal.

2. The multi-channel cardiovascular monitoring method according to claim 1, wherein: In step S2, the multi-band impedance data is fused and processed, including dynamic baseline calibration, frequency domain noise suppression and cross-band feature extraction, to establish a high-frequency deep human body impedance model.

3. The multi-channel cardiovascular monitoring method according to claim 2, wherein: In step S3, the neural network model is constructed based on state space modeling, including a spectrum-time domain feature fusion embedding mechanism and a sparse feedforward enhancement layer.

4. A multi-channel cardiovascular monitoring device, characterized in that: include: The first processing module is used to obtain multi-band impedance data of human skin and chest cavity; A second processing module is used to fuse the multi-band impedance data to construct a high-frequency deep human body impedance model; a third processing module, configured to calculate equivalent impedance parameters using a high-frequency impedance model based on the fused multi-band impedance data, input the impedance parameters into a customized neural network model improved for impedance signal characteristics, and use the neural network model to output cardiovascular parameters such as heart rate, respiratory rate, stroke volume, and pulsatility index; The fourth processing module is used to construct a cardiovascular abnormality probability model based on the cardiovascular parameters and output a real-time health status assessment and a multi-level risk warning signal.

5. The multi-channel cardiovascular monitoring device according to claim 4, wherein: The second processing module performs fusion processing on the multi-band impedance data, including dynamic baseline calibration, frequency domain noise suppression and cross-band feature extraction, to establish a high-frequency deep human body impedance model.

6. The multi-channel cardiovascular monitoring device according to claim 5, wherein: The neural network model is constructed based on state-space modeling, including a spectrum-time domain feature fusion embedding mechanism and a sparse feedforward enhancement layer.

7. A multi-channel cardiovascular monitoring system, characterized in that: include: A memory and a processor, wherein the memory stores a computer program executed by the processor, and when the computer program is executed by the processor, the multi-channel cardiovascular monitoring method according to any one of claims 1 to 3 is executed.

8. A storage medium, characterized in that: The storage medium stores a computer program, which, when run, executes the multi-channel cardiovascular monitoring method according to any one of claims 1 to 3.

Images