A method for early warning of heart failure decompensation

Abstract

The application provides a heart failure decompensation early warning method, which comprises the following steps: continuously collecting time sequence signals of heart rate, blood pressure and respiration of a patient, dynamically fitting a low-order virtual closed-loop control model representing cardiovascular regulation function, injecting a virtual test signal into the model, calculating a response of the model to real-time calculate a stability margin index, continuously tracking changes of the index and detecting a rapid narrowing trend or a mutation point of the index, and generating an early warning signal when an abnormality meeting a preset mutation mode is detected. The scheme can be deployed on a general computing device through a software program, or can be integrated into a special hardware system, so that an integrated solution is provided for hospitals and families, and the technical application threshold is reduced.

Description

Technical Field

[0001] This invention relates to the field of medical early warning, and more particularly to an early warning method for heart failure decompensation. Background Technology

[0002] Acute decompensation is a leading cause of readmission and death in heart failure patients after discharge. Current home monitoring technologies primarily rely on static threshold alarms for single physiological parameters, resulting in significant drawbacks such as delayed warnings, high false alarm rates, and data silos. Delayed warnings occur only after obvious clinical symptoms appear, missing the early intervention window. High false alarm rates stem from a failure to consider dynamic correlations between parameters and individual differences, making them susceptible to random interference and leading to decreased patient compliance and clinical trust. Data silos refer to heterogeneous data from different devices with varying formats and asynchronous timing, making it difficult to integrate and form a complete physiological time-series map for in-depth analysis. While existing technologies address heart failure risk control, their risk stratification remains based on static thresholds, lacking analysis of multi-parameter time-series correlations and dynamic trends, thus failing to address these fundamental problems. Therefore, a new method is urgently needed to provide earlier, more accurate, and more intelligent warnings of heart failure decompensation risk. Summary of the Invention

[0003] This invention proposes an early warning method for heart failure decompensation, comprising:

[0004] S1. Continuously collect time-series signals of at least two physiological parameters in the target patient, namely heart rate, blood pressure, and respiration;

[0005] S2. Based on the real-time or near-real-time data of the time-series signal, dynamically fit a low-order virtual closed-loop control model characterizing the cardiovascular autonomic nervous regulation function. The low-order virtual closed-loop control model is a heart rate-blood pressure coupling transfer function model characterizing the cardiovascular autonomic nervous regulation function.

[0006] S3. Inject a preset virtual test signal into the low-order virtual closed-loop control model. The virtual test signal is a small step signal simulating blood pressure changes or a sine wave signal of a specific frequency generated in a computer simulation environment.

[0007] S4. Calculate the response of the low-order virtual closed-loop control model to the virtual test signal, and deduce the stability margin index of the current model in real time based on the response.

[0008] S5. Continuously track the change of the stability margin index over time, calculate the rate of change of the stability margin index within the sliding time window, and determine the rapid narrowing trend when the rate of change continuously exceeds the first threshold, or determine the abrupt change point when the stability margin index crosses the second threshold within a preset short period of time.

[0009] S6. When a rapid narrowing or mutation that conforms to a preset mutation pattern is detected in the stability margin index, an early warning signal indicating that the cardiovascular regulatory system is on the verge of instability is generated.

[0010] Furthermore, the model parameters of the low-order virtual closed-loop control model include at least one or more of the following: reflection gain, delay time, and integral time constant. The model parameters are dynamically estimated from the timing signal using recursive least squares or Kalman filtering.

[0011] Furthermore, the stability margin index includes phase margin and / or gain margin; the stability margin index is calculated by performing frequency domain analysis or eigenvalue analysis on the transfer function of the low-order virtual closed-loop control model.

[0012] Furthermore, the early warning method for heart failure decompensation also includes a model confidence assessment step:

[0013] Real-time evaluation of the quality of the time-series signal and the confidence interval of the parameter estimation of the low-order virtual closed-loop control model;

[0014] When the signal quality is lower than a preset threshold or the confidence interval width exceeds a preset range, the generation of the early warning signal is paused or the generation weight of the early warning signal is reduced.

[0015] Furthermore, the preset mutation pattern is predefined by analyzing the changes in stability margin indicators of patients before an acute decompensated event in historical data.

[0016] Furthermore, the generated early warning signals are used to trigger immediate clinical review, initiate intensive diuresis regimens, or advise patients to seek immediate medical attention.

[0017] Furthermore, the present invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the early warning method for heart failure decompensation as described in any of the preceding claims.

[0018] Furthermore, this invention also proposes an early warning system for heart failure decompensation based on virtual closed-loop stability testing, comprising:

[0019] A multi-parameter synchronous acquisition module is used to continuously and synchronously acquire heart rate, blood pressure, and respiratory signals;

[0020] The virtual closed-loop model dynamic fitting module is used to fit and update the low-order virtual closed-loop control model in real time. The low-order virtual closed-loop control model is a heart rate-blood pressure coupling transfer function model that characterizes the cardiovascular autonomic nervous regulation function.

[0021] The virtual stability test module is used to inject virtual test signals into the low-order virtual closed-loop control model and calculate the stability margin index. The virtual test signal is a small step signal simulating blood pressure changes or a sine wave signal of a specific frequency generated in a computer simulation environment.

[0022] The mutation point detection and early warning module is used to track changes in stability margin. It detects rapid narrowing trends or mutation points by calculating the rate of change of the stability margin index within a sliding time window, and generates an early warning signal when a rapid narrowing or mutation that conforms to a preset mutation pattern is detected.

[0023] Furthermore, the virtual closed-loop model dynamic fitting module dynamically estimates the model parameters using recursive least squares or Kalman filtering. The model parameters include at least one or more of reflection gain, delay time, and integral time constant.

[0024] Furthermore, the stability margin index includes phase margin or gain margin, and the virtual stability test module calculates the stability margin index by performing frequency domain analysis or eigenvalue analysis on the transfer function of the low-order virtual closed-loop control model.

[0025] Compared with existing technologies, this invention offers significant advantages. By constructing and dynamically testing a virtual closed-loop control model characterizing cardiovascular autonomic nervous system regulation, it can detect engineering signs of decreased system stability before patients exhibit obvious clinical symptoms, enabling early warning and saving valuable time for clinical intervention. Employing multi-parameter time-series correlation analysis, such as a heart rate and blood pressure coupling model, combined with a sliding window dual-threshold and confidence assessment mechanism, it effectively filters out random fluctuations and noise interference, reducing false alarm rates and improving the clinical reliability and operability of the warning signals. This invention solves the standardization and time alignment problems of multi-source heterogeneous physiological data, forming a complete automated pipeline from data acquisition, model fitting, and virtual testing to risk decision-making. Warning signals can directly trigger preset clinical response processes, such as generating medical orders to notify healthcare professionals, achieving seamless integration from technical alarms to clinical action. The entire analysis process is performed on the virtual model in the digital domain, evaluating stability through virtual test signals. No real physical stimulation is required on the patient, making it safe, non-invasive, and suitable for long-term continuous monitoring. This solution can be deployed on various general-purpose computing devices via software programs or integrated into dedicated hardware systems, providing an integrated solution for hospitals and homes and lowering the technical application threshold. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating an early warning method for heart failure decompensation proposed in this invention. Detailed Implementation

[0027] refer to Figure 1This invention proposes an early warning method for heart failure decompensation, comprising:

[0028] S1. Continuously collect time-series signals of at least two physiological parameters in the target patient, namely heart rate, blood pressure, and respiration;

[0029] S2. Based on the real-time or near-real-time data of the time-series signal, dynamically fit a low-order virtual closed-loop control model that characterizes the cardiovascular autonomic nervous system regulation function. The low-order virtual closed-loop control model is a heart rate-blood pressure coupling transfer function model that characterizes the cardiovascular autonomic nervous system regulation function.

[0030] S3. Inject a preset virtual test signal into the low-order virtual closed-loop control model. The virtual test signal is a small step signal or a sine wave signal of a specific frequency that simulates blood pressure changes and is generated in a computer simulation environment.

[0031] S4. Calculate the response of the low-order virtual closed-loop control model to the virtual test signal, and deduce the stability margin index of the current model in real time based on the response.

[0032] S5. Continuously track the change of the stability margin index over time, calculate the rate of change of the stability margin index within the sliding time window, and determine the rapid narrowing trend when the rate of change continuously exceeds the first threshold, or determine the abrupt change point when the stability margin index crosses the second threshold within a preset short period of time.

[0033] S6. When a rapid narrowing or mutation that conforms to a preset mutation pattern is detected in the stability margin index, an early warning signal indicating that the cardiovascular regulatory system is on the verge of instability is generated.

[0034] Specifically, this method targets patients clinically diagnosed with chronic heart failure who are under outpatient or general ward monitoring. In practice, the patient's physiological parameters, such as heart rate, blood pressure, and respiration, are continuously collected using medical equipment, forming a time-series signal sequence sampled at fixed time intervals. Before fitting the model, the time-series signal undergoes preprocessing, such as filtering, to remove motion artifacts and other interference. The acquisition equipment is also periodically calibrated to ensure signal accuracy. Next, based on this real-time data, a differential equation or state-space equation of mathematical order no more than third order is dynamically fitted. This equation serves as a low-order virtual closed-loop control model characterizing the cardiovascular autonomic nervous system's regulatory function. The core principle lies in abstracting the human regulatory system as an engineering closed-loop feedback control system, using the collected system output signals to back-estimate the dynamic characteristics of the internal controller. Then, virtual test signals generated in a computer simulation environment are injected into this virtual model, such as small steps simulating blood pressure changes or sinusoidal signals of a specific frequency, to perform a digital stress test on the virtual model. By calculating the response of a computational model to test signals, stability margin indices such as phase margin or gain margin can be derived in real time. These indices are used to quantify the safety margin of a dynamic system from its instability boundary. This method replaces real stimuli with virtual testing, greatly improving the safety of monitoring. The changes in stability margin indices over time are then continuously tracked, and their patterns are observed by comparing them with the patient's historical baseline. When the stability of the virtual model declines due to deterioration in the real physiological condition, it manifests as a rapid narrowing or abrupt change in the margin indices. The system uses typical change patterns summarized from historical decompensation event data as preset mutation patterns and checks whether the current index changes conform to these patterns. Finally, when a rapid narrowing or abrupt change conforming to the preset mutation pattern is detected, an early warning signal indicating that the cardiovascular regulatory system is on the verge of instability is generated. Simultaneously with the generation of the warning, all raw data and intermediate calculation results within a period before and after the triggering time are recorded for clinical review. This method transforms complex physiological states into quantifiable engineering stability indices, providing a non-invasive, continuous, and objective early warning method. It improves the objectivity and sensitivity of the warning, helps to detect the trend of disease deterioration earlier, and buys valuable time for clinical intervention.

[0035] Furthermore, the model parameters of the low-order virtual closed-loop control model include at least one or more of the following: reflection gain, delay time, and integral time constant. The model parameters are dynamically estimated from the timing signal using recursive least squares or Kalman filtering.

[0036] Specifically, the low-order virtual closed-loop control model is a heart rate-blood pressure coupling model, a dynamic mathematical model that takes blood pressure changes as input and heart rate changes as output. In addition to key parameters such as reflection gain, delay time, and integral time constant, this model can also include parameters describing peripheral vascular resistance. Reflection gain is a coefficient in the model representing the magnitude of pressure reflex sensitivity; a larger absolute value indicates stronger reflex regulation. Delay time refers to the physiological interval between the occurrence of a blood pressure change and the resulting reflexive change in heart rate. The integral time constant describes the speed of regulation, reflecting the system's ability to eliminate steady-state errors. As the core mechanism of cardiovascular regulation, the physiological process of pressure reflex is concretized into this classic control model structure containing gain, delay, and integral terms. Recursive least squares can be used to dynamically estimate these parameters. This method continuously updates parameters by minimizing the sum of squared errors between the model-predicted heart rate and the actual acquired heart rate. In practice, it is usually implemented with a forgetting factor to adapt to the time-varying nature of physiological parameters. Kalman filtering is another alternative approach, treating the problem as a state estimation problem. Before implementation, the covariance matrix of process noise and measurement noise needs to be pre-defined to optimally estimate the hidden model parameter states while considering system dynamics and noise. Both algorithms can achieve online dynamic fitting of the model, enabling the virtual model to track the patient's pressure reflex function in real time. The parameter estimation process also includes verifying the reasonableness of the results; for example, the reflex gain should be negative, and a positive value is considered a failure, with the valid parameters from the previous time step being used. By clarifying the physiological meaning and key parameters of the model, the assessment becomes more physiologically targeted. Specifying mature dynamic parameter estimation algorithms ensures the real-time performance and robustness of the model fitting, providing an accurate and reliable model foundation for subsequent stability analysis.

[0037] It should be noted that the low-order virtual closed-loop control model is a transfer function model with a mathematical order of no more than three, which takes blood pressure changes as input and heart rate changes as output, and is used to simulate the cardiovascular pressure reflex closed-loop regulation loop.

[0038] Specifically, the virtual test signals include two forms. One is a small step signal, which refers to the simulated blood pressure input value being instantaneously increased or decreased by a small fixed value from the baseline level at the zero point of the simulation time. This amplitude is set far below the threshold that might cause a real physiological response. The other is a sinusoidal signal of a specific frequency, i.e., a simulated blood pressure input signal with a constant amplitude and a periodic variation according to a sinusoidal law. During implementation, multiple different frequency points are usually selected for frequency sweep testing to obtain more comprehensive frequency response characteristics. Before the simulation injection, the signal is also subjected to amplitude limiting to prevent numerical calculation overflow. The entire injection and response calculation process runs in an independent simulation thread to ensure that it does not affect the main thread of real-time data acquisition and model fitting. Furthermore, the step signal test is used to examine the transient response and recovery capability of the system to sudden disturbances, and the response pattern can reflect the system's damping and stability. The sinusoidal signal test is used to obtain the response amplitude and phase of the system under disturbances at different frequencies, which is the basis for frequency domain stability analysis. These tests are performed entirely on the virtual model in the digital domain, and the response is obtained by numerically solving the model's differential equations or calculating the transfer function. Because it does not involve any real energy or matter acting on the human body, this method ensures absolute non-invasiveness and safety in principle. By defining these two classic test signals, a safe and effective virtual testing method is provided, capable of fully stimulating the dynamic characteristics of the model and providing rich information for stability assessment. By explicitly avoiding the application of any actual physical stimulation to the patient, the possibility of risks arising from the monitoring itself is completely eliminated, making this method suitable for long-term continuous monitoring of critically ill patients.

[0039] In this embodiment, the virtual test signal exists only in the digital simulation environment and is a small step signal simulating blood pressure changes or a sinusoidal signal of a specific frequency, without applying any actual physical stimulation to the patient.

[0040] Furthermore, the stability margin index includes phase margin and / or gain margin; the stability margin index is calculated by performing frequency domain analysis or eigenvalue analysis on the transfer function of the low-order virtual closed-loop control model.

[0041] Specifically, the stability margin indices include phase margin and gain margin. Phase margin is defined as the difference between the phase distance and the negative 180 degrees on the open-loop frequency response curve when the system gain is one. Gain margin is defined as the number of decibels by which the gain falls below one at the frequency point where the phase reaches negative 180 degrees. Calculating these indices requires first obtaining the mathematical expression of the model. The transfer function is the ratio of the system output to the Laplace transform of the input under zero initial conditions. Eigenvalues ​​are the eigenvalues ​​that make the determinant of the system state matrix zero; they are also called system poles. When using frequency domain analysis, the model transfer function must first be represented in the complex frequency domain, and then a Nyquist plot or Bode plot must be plotted to directly read the phase margin or gain margin. When using eigenvalue analysis, the eigenvalues ​​of the model state matrix need to be solved, and the stability margin is evaluated based on the distance of the eigenvalues ​​from the imaginary axis in the left half of the complex plane or their angle with the negative real axis. The calculation process usually also includes smoothing filtering of the margin indices to eliminate noise caused by fluctuations in model parameter estimation. Furthermore, phase margin and gain margin are key quantitative indicators for measuring system stability in classical control theory. Phase margin reflects the system's tolerable additional phase lag, while gain margin reflects the system's tolerable additional gain amplification. After obtaining the model through virtual testing, it can be transformed into a transfer function for frequency domain analysis or converted into state-space form to solve for eigenvalues ​​for evaluation. These two methods provide a rigorous mathematical measure of the stability of the virtual model from different perspectives. By introducing precisely quantified engineering stability assessment indicators, and applying mature concepts such as phase margin and gain margin to the evaluation of physiological systems, the abstract concept of stability becomes calculable and traceable. This reveals the inherent dynamic instability risk of the system more directly and profoundly than simply observing the absolute numerical changes of traditional indicators.

[0042] Specifically, the process for detecting rapid narrowing trends or abrupt change points is as follows: First, a sliding time window is defined, which is a time interval that moves forward over time for local data analysis. Its length is, for example, four, eight, or twelve hours, and can be configured according to the requirements of warning sensitivity and specificity. Within this window, the rate of change of the stability margin index, i.e., the average slope of the index value changing over time, can be calculated using linear regression or the difference method. The first threshold is a critical rate of change derived from a large amount of stable-phase patient data, used to distinguish between normal fluctuations and abnormal rapid declines. When the rate of change continuously exceeds the first threshold, it is determined to be a rapid narrowing trend. Abrupt change point detection focuses on the absolute collapse of the index level. The second threshold is the absolute safety lower limit of the index determined based on historical decompensation event data, usually slightly higher than the clinically observed critical value. A short period of time refers to a preset short time interval, such as thirty minutes to two hours. When the index drops sharply within this time and crosses the second threshold, it is determined to be an abrupt change point. The judgment logic can also use majority voting or state machine management for the results of multiple consecutive time windows to improve robustness. Rapid narrowing trend detection uses a sliding window to filter out short-term noise and capture the rate of sustained deterioration. Exceeding the first threshold indicates accelerated stability decay, consistent with early warning positioning. Mutation point detection focuses on the absolute collapse of the indicator; falling below the second threshold indicates potential system instability, requiring immediate intervention. These two strategies complement each other, forming a gradient-based early warning mechanism. A clearly defined and operable mutation detection algorithm is provided. By defining the sliding window, rate of change, and dual thresholds, qualitative descriptions are transformed into precise programmable logic rules, reducing the difficulty of system implementation and making the triggering conditions clear and interpretable, facilitating clinical understanding and trust.

[0043] Furthermore, the early warning method for heart failure decompensation also includes a model confidence assessment step: real-time assessment of the quality of the time-series signal and the confidence interval of the parameter estimation of the low-order virtual closed-loop control model; when the signal quality is lower than a preset threshold or the confidence interval width exceeds a preset range, pausing the generation of the early warning signal or reducing the generation weight of the early warning signal.

[0044] Specifically, this step is a crucial step in quantitatively evaluating the reliability of the input data and model output results. Signal quality refers to the proportion of effective information components in the original physiological signal. The evaluation specifically calculates the signal-to-noise ratio, detects the signal loss rate, and identifies large artifacts caused by motion or poor contact. The confidence interval is the range of values ​​that the true values ​​of the model parameters may fall into under a given probability. The wider the interval, the more uncertain the estimate; its width is calculated using the covariance matrix of the parameter estimation algorithm. The system sets graded thresholds for signal quality and confidence interval width, such as excellent, acceptable, and poor. Differentiated strategies are adopted for different levels: excellent signals generate normal warnings; acceptable signals generate warnings but with an additional low-confidence marker; and poor signals pause warnings and only record data. The generation weight is a coefficient used to adjust the urgency or priority of the final warning signal, and can be dynamically adjusted according to the confidence level. The reliability of any data analysis model depends on the input quality and estimation accuracy. This step constructs a dynamic quality control closed loop, monitoring in real time whether the raw data is qualified and whether the model processing is accurate. When excessive data noise, discontinuous signals, or highly uncertain parameter estimates are detected, the system can automatically recognize that the current results are unreliable. At this point, by pausing or reducing the weight of alerts, misleading alarms can be proactively avoided when the data model is unreliable. This significantly enhances the system's robustness and credibility. By introducing confidence assessment, the system possesses self-error detection and degradation capabilities, effectively preventing false alarms caused by temporary poor data or transient misfitting, reducing clinical interference, and making medical staff more willing to trust and adopt the system's alert information.

[0045] Furthermore, the preset mutation pattern is predefined by analyzing the changes in stability margin indicators of patients before an acute decompensated event in historical data.

[0046] Specifically, the pre-defined mutation patterns are derived from the analysis of historical data. Historical data refers to datasets of heart failure patients collected in the past, containing complete monitoring records and clear clinical outcomes. This data typically originates from long-term monitoring data and electronic medical records stored in hospital information systems. Acute decompensated events are clinical deterioration events requiring urgent medical intervention, such as intravenous medication or admission to the intensive care unit. The analysis process includes data cleaning, aligning event occurrence time points, and extracting stability margin indicator sequences within a specific time window prior to the event. The change characteristics are the regularities exhibited by these indicator sequences in terms of morphology, speed, and amplitude. Through cluster analysis or expert rule induction, typical mutation patterns such as precipitous drops, step-like declines, and accelerated declines can be summarized from the sequences. These patterns are parameterized and stored in a pattern library for real-time detection modules to call and match. The pre-defined mutation patterns are essentially early warning signal templates strongly correlated with adverse outcomes, learned from historical experience. By retrospectively analyzing real-world data, common abnormal changes in indicators before the event in patients ultimately experiencing decompensation are identified and abstracted into standard templates. In real-time monitoring, the current patient's indicator change curve is compared with a preset template using similarity calculation or feature matching. A successful match indicates an ominous sign similar to that of a historically critically ill patient, triggering an early warning. This is an early warning logic based on historical experience evidence. This provides the early warning model with a solid clinical data foundation, improving its clinical relevance and specificity. The mutation pattern defined based on real historical events is more objective and reliable than thresholds set solely by theoretical deduction or expert experience, making the system alerts more closely linked to actual clinical risks and providing higher clinical guidance value.

[0047] Furthermore, the generated early warning signals are used to trigger immediate clinical review, initiate intensive diuresis regimens, or advise patients to seek immediate medical attention.

[0048] Specifically, the generated early warning signals automatically generate to-do tasks through the hospital information system interface and push them to the responsible physician's mobile terminal. Immediate clinical review refers to medical staff quickly conducting a face-to-face assessment of the patient after receiving the warning. Triggering this action simultaneously notifies the nurse to re-measure vital signs and contacts the attending physician. Strengthening the diuretic regimen refers to temporarily increasing the dosage or frequency of diuretics based on the existing medication regimen and the pre-planned protocol. This can be prompted within the doctor's preset electronic order set or automatically generated for the doctor's confirmation. Advising the patient to seek immediate medical attention involves issuing a clear instruction to the patient and their family to seek emergency medical assistance at a medical institution. This is usually done through an emergency notification sent via the patient's application and can provide one-click emergency call or navigation to the nearest hospital. The exit points and action points of the warning signals are clearly defined, completing a closed loop from technical warning to clinical action. The value of the warning signals lies in driving subsequent behavior and binding them to the preset clinical response process. Once a warning is generated, it automatically transforms into a series of executable clinical tasks or patient instructions, rather than just a screen prompt. This ensures that technological discoveries can be seamlessly integrated into existing workflows, promptly translating into interventions that may alter patient outcomes and preventing a disconnect between warnings and action. It also ensures the practicality and effectiveness of the early warning system. By directly linking technical warning outputs to specific clinical interventions, early warnings move beyond mere monitoring and become triggers for proactive management processes. This significantly shortens the time from risk detection to responsive action, truly realizing the clinical value of early warnings and potentially improving patient outcomes.

[0049] Furthermore, the present invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the early warning method for heart failure decompensation.

[0050] Specifically, a computer-readable storage medium is a physical carrier capable of storing data in magneto-optical, photoelectric, or semiconductor manner and read by a computer device, such as a server hard drive, embedded device flash memory chip, or mobile device memory card. The computer program stored on it is an executable file or script consisting of a series of processor-executable instruction codes, containing multiple code modules corresponding to data acquisition instructions, filtering preprocessing instructions, model fitting algorithm instructions, virtual testing simulation instructions, stability calculation instructions, mutation detection logic instructions, and early warning generation and communication instructions. The processor refers to a computing chip capable of executing instructions, such as a central processing unit, microcontroller, or graphics processing unit. When the program is executed by the processor, these instructions are loaded and run sequentially, controlling the processor to work collaboratively with sensors, memory, and communication modules to completely reproduce all steps of the method. What is protected is the software implementation carrier of the method. The aforementioned method flow is converted into specific program code using a software programming language and solidified on the storage medium. When a computing device containing a processor reads and runs the program, the processor executes all operations step by step according to the code logic, including data acquisition, model fitting, virtual testing, stability analysis, mutation detection, and early warning generation. Media and programs are the necessary tools and manifestations for the automatic and repeated implementation of a method on general-purpose or special-purpose hardware. By tangibly protecting the software implementation of the aforementioned early warning method for heart failure decompensation, the method can be easily copied, distributed, and deployed on various computing devices such as cloud servers, edge computing gateways, or smart monitors, greatly promoting the application and commercialization of the technology.

[0051] Furthermore, the present invention also proposes an early warning system for heart failure decompensation based on virtual closed-loop stability testing, comprising: a multi-parameter synchronous acquisition module for continuously and synchronously acquiring heart rate, blood pressure and respiratory signals;

[0052] The virtual closed-loop model dynamic fitting module is used to fit and update the low-order virtual closed-loop control model in real time. The low-order virtual closed-loop control model is a heart rate-blood pressure coupling transfer function model that characterizes the cardiovascular autonomic nerve regulation function.

[0053] The virtual stability test module is used to inject virtual test signals into the low-order virtual closed-loop control model and calculate the stability margin index. The virtual test signal is a small step signal simulating blood pressure changes or a sine wave signal of a specific frequency generated in a computer simulation environment.

[0054] The mutation point detection and early warning module is used to track changes in stability margin. It detects rapid narrowing trends or mutation points by calculating the rate of change of the stability margin index within a sliding time window, and generates an early warning signal when a rapid narrowing or mutation that conforms to a preset mutation pattern is detected.

[0055] Furthermore, the virtual closed-loop model dynamic fitting module dynamically estimates the model parameters using recursive least squares or Kalman filtering. The model parameters include at least one or more of reflection gain, delay time, and integral time constant.

[0056] Furthermore, the stability margin index includes phase margin or gain margin, and the virtual stability test module calculates the stability margin index by performing frequency domain analysis or eigenvalue analysis on the transfer function of the low-order virtual closed-loop control model.

[0057] Specifically, the system includes a multi-parameter synchronous acquisition module, a collection of hardware circuits and sensors responsible for acquiring physiological signals from the human body and converting them into digital signals. This module includes ECG electrodes, a photoelectric pulse wave sensor, an impedance respiration sensor, and corresponding analog-to-digital conversion circuits. The virtual closed-loop model dynamic fitting module is a computational unit that performs specific mathematical operations to determine model parameters, consisting of parameter estimation algorithm software running in a digital signal processor or general-purpose processor. The virtual stability testing module is a computational unit that performs simulation and frequency domain or time domain analysis. The mutation point detection and early warning module is a control unit that performs logical judgments and signal generation. The latter two modules can be integrated in the same or different processors, exchanging data via an internal bus or shared memory. The system also includes a power management module to power each module, and a network communication module for receiving instructions and sending early warning information. Its working principle is that the system is a hardware embodiment of the method. The multi-parameter synchronous acquisition module continuously acquires raw physiological data as a sensing organ. The virtual closed-loop model dynamic fitting module, as the first core computational unit, processes data and constructs a virtual physiological model. The virtual stability testing module, as the second core computational unit, performs digital stability stress testing on the model. The mutation point detection and early warning module, acting as the decision-making and output unit, assesses risk and issues alarms based on test results. These four modules are interconnected via data flow, forming a complete automated pipeline for signal perception, information processing, model evaluation, and risk decision-making, jointly achieving continuous intelligent early warning without human intervention. This provides an integrated, dedicated early warning device or system solution, integrating disparate algorithmic steps into a unified hardware framework, improving system reliability and operational efficiency. This system can be deployed as a standalone monitoring device in wards or homes, providing users with out-of-the-box early warning services, lowering the barrier to entry and reducing technical complexity.

[0058] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for early warning of heart failure decompensation, characterized in that, include: S1. Continuously collect time-series signals of at least two physiological parameters in the target patient, namely heart rate, blood pressure, and respiration; S2. Based on the real-time or near-real-time data of the time-series signal, dynamically fit a low-order virtual closed-loop control model characterizing the cardiovascular autonomic nervous regulation function. The low-order virtual closed-loop control model is a heart rate-blood pressure coupling transfer function model characterizing the cardiovascular autonomic nervous regulation function. S3. Inject a preset virtual test signal into the low-order virtual closed-loop control model. The virtual test signal is a small step signal simulating blood pressure changes or a sine wave signal of a specific frequency generated in a computer simulation environment. S4. Calculate the response of the low-order virtual closed-loop control model to the virtual test signal, and deduce the stability margin index of the current model in real time based on the response. S5. Continuously track the change of the stability margin index over time, calculate the rate of change of the stability margin index within the sliding time window, and determine the rapid narrowing trend when the rate of change continuously exceeds the first threshold, or determine the abrupt change point when the stability margin index crosses the second threshold within a preset short period of time. S6. When a rapid narrowing or mutation that conforms to a preset mutation pattern is detected in the stability margin index, an early warning signal indicating that the cardiovascular regulatory system is on the verge of instability is generated.

2. The early warning method for heart failure decompensation as described in claim 1, characterized in that, The model parameters of the low-order virtual closed-loop control model include at least one or more of the following: reflection gain, delay time, and integral time constant. The model parameters are dynamically estimated from the time-series signal using recursive least squares or Kalman filtering.

3. The early warning method for heart failure decompensation as described in claim 1, characterized in that, The stability margin index includes phase margin and / or gain margin; the stability margin index is calculated by performing frequency domain analysis or eigenvalue analysis on the transfer function of the low-order virtual closed-loop control model.

4. The early warning method for heart failure decompensation as described in claim 1, characterized in that, The aforementioned early warning method for heart failure decompensation also includes a model confidence assessment step: Real-time evaluation of the quality of the time-series signal and the confidence interval of the parameter estimation of the low-order virtual closed-loop control model; When the signal quality is lower than a preset threshold or the confidence interval width exceeds a preset range, the generation of the early warning signal is paused or the generation weight of the early warning signal is reduced.

5. The early warning method for heart failure decompensation as described in claim 1, characterized in that, The preset mutation pattern is predefined by analyzing the changes in stability margin indicators of patients before an acute decompensated event in historical data.

6. The early warning method for heart failure decompensation as described in claim 1, characterized in that, The early warning signals are used to trigger immediate clinical review, initiate an intensive diuretic regimen, or advise the patient to seek immediate medical attention.

7. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the early warning method for heart failure decompensation as described in any one of claims 1 to 6.

8. An early warning system for heart failure decompensation based on virtual closed-loop stability testing, characterized in that, include: A multi-parameter synchronous acquisition module is used to continuously and synchronously acquire heart rate, blood pressure, and respiratory signals; The virtual closed-loop model dynamic fitting module is used to fit and update the low-order virtual closed-loop control model in real time. The low-order virtual closed-loop control model is a heart rate-blood pressure coupling transfer function model that characterizes the cardiovascular autonomic nerve regulation function. The virtual stability test module is used to inject virtual test signals into the low-order virtual closed-loop control model and calculate the stability margin index. The virtual test signal is a small step signal simulating blood pressure changes or a sine wave signal of a specific frequency generated in a computer simulation environment. The mutation point detection and early warning module is used to track changes in stability margin. It detects rapid narrowing trends or mutation points by calculating the rate of change of the stability margin index within a sliding time window, and generates an early warning signal when a rapid narrowing or mutation that conforms to a preset mutation pattern is detected.

9. The system as described in claim 8, characterized in that, The virtual closed-loop model dynamic fitting module dynamically estimates the model parameters using recursive least squares or Kalman filtering. The model parameters include at least one or more of the following: reflection gain, delay time, and integral time constant.

10. The system as described in claim 8, characterized in that, The stability margin index includes phase margin or gain margin, and the virtual stability test module calculates the stability margin index by performing frequency domain analysis or eigenvalue analysis on the transfer function of the low-order virtual closed-loop control model.

Images