Electrocardiogram monitor early warning method and system based on tensor modeling and chaos analysis

Abstract

The application discloses an electrocardio monitor early warning method and system based on tensor modeling and chaos analysis, and relates to the field of intelligent medical monitoring. The early warning system comprises an electrocardio signal acquisition module, a tensor modeling module for constructing an electrocardio signal tensor model, a feature processing module for constructing a dominant feature vector, a chaos analysis module for calculating a first early warning index, a second early warning index and a real-time early warning index based on a chaos analysis method, a dynamic early warning threshold generation module for calculating a preset dynamic early warning threshold, an electrocardio signal feature classification module for identifying an abnormal feature possibility mark, and an early warning output module for outputting the first early warning index, the second early warning index, the real-time early warning index and the abnormal feature possibility mark and issuing an electrocardio signal abnormality early warning. The application combines tensor modeling, a chaos analysis method and a dynamic early warning threshold, thereby improving the early electrocardio abnormality capturing capability, the early warning accuracy and the adaptation to target objects.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical monitoring technology, specifically to an early warning method and system for electrocardiogram monitors based on tensor modeling and chaos analysis. Background Technology

[0002] Cardiovascular disease is one of the leading causes of death worldwide, and early identification and timely intervention are crucial for reducing mortality. Electrocardiogram (ECG) monitors, which collect real-time electrocardiogram signals, detect abnormalities such as arrhythmias and myocardial ischemia and trigger warnings, have become indispensable medical devices in clinical practice. However, existing ECG monitors still face significant technical bottlenecks in clinical applications. Their warning accuracy is affected by multiple factors, seriously impacting the efficiency of clinical decision-making and the safety of patient care.

[0003] Current mainstream ECG analysis techniques rely on statistical features such as the RR interval (the time interval between adjacent R wave peaks on an ECG), QRS width, and heart rate variability to construct judgment models. These methods essentially extract linear patterns from signals. However, cardiac electrical activity is regulated by multiple systems, including the autonomic nervous system and myocardial cell metabolism, exhibiting complex nonlinear dynamic behavior. Analysis based solely on linear statistical features fundamentally contradicts the complex nature of cardiac electrical activity. Numerous studies have confirmed that healthy human ECG signals possess typical chaotic attractor characteristics, with waveforms maintaining dynamic equilibrium amidst seemingly random fluctuations. This nonlinear dynamic characteristic directly reflects normal cardiac autonomic nervous system regulation. However, in pathological states such as myocardial infarction, atrial fibrillation, and ventricular tachycardia, the chaotic attractor structure of ECG signals undergoes significant distortion, even exhibiting characteristic changes indicating attractor collapse. Unfortunately, traditional linear analysis methods struggle to capture these crucial pathological signals, leading to high false alarm rates and difficulty in responding to early ECG abnormalities. Summary of the Invention

[0004] The purpose of this invention is to provide an early warning method and system for electrocardiogram (ECG) monitors based on tensor modeling and chaos analysis, so as to solve the problem of high false alarm rate in existing ECG monitoring early warning methods mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] An ECG monitor early warning system based on tensor modeling and chaos analysis includes:

[0007] The electrocardiogram (ECG) signal acquisition module is used to acquire multi-lead raw ECG signals of the target object; the multi-lead raw ECG signal refers to the cardiac electrical activity potential change data synchronously acquired by multiple body surface electrodes of the ECG monitor at preset standard acquisition positions on the human body;

[0008] The tensor modeling module is used to construct an electrocardiogram signal tensor model based on the original multi-lead electrocardiogram signal.

[0009] The feature processing module is used to perform noise reduction and dimensionality reduction processing on the electrocardiogram signal tensor model, and to construct a dominant feature vector that reflects the dynamic characteristics of the electrocardiogram signal.

[0010] The chaos analysis module is used to process the dominant feature vector based on the chaos analysis method and calculate the first early warning index, the second early warning index, and the real-time early warning index.

[0011] A dynamic early warning threshold generation module is used to collect the respiratory rate and blood oxygen saturation of the target object, and calculate a preset dynamic early warning threshold based on the respiratory rate and blood oxygen saturation.

[0012] The ECG signal feature classification module is used to determine whether the ECG signal of the target object is abnormal based on the real-time warning index and the preset dynamic warning threshold. If an abnormality is determined, the module identifies the probability identifier of the abnormal features of the ECG signal.

[0013] Early warning output module: used to output the first early warning indicator, the second early warning indicator and the real-time early warning indicator. When it is determined that the electrocardiogram signal of the target object is abnormal, an early warning of abnormal electrocardiogram signal is issued and the probability identifier of the abnormal feature is output.

[0014] The principle of the ECG monitor early warning system based on tensor modeling and chaotic analysis of this invention is as follows: The ECG signal acquisition module collects multi-lead raw ECG signals of the target object; the tensor modeling module constructs an ECG signal tensor model based on the multi-lead raw ECG signals to integrate multi-dimensional information of the raw ECG signals; the feature processing module performs noise reduction and dimensionality reduction on the ECG signal tensor model, filters out interference signals and extracts core information, constructing a dominant feature vector reflecting the dynamic characteristics of the ECG signal, focusing on key analysis objects; the chaotic analysis module uses chaotic analysis methods to process the dominant feature vector, quantifies the chaotic characteristics of the ECG signal, calculates a first early warning index, a second early warning index, and a real-time early warning index after adaptive fusion of the first and second early warning indices, realizing a quantitative assessment of the degree of abnormality in the target object's ECG signal. This system enhances the ability to capture abnormal ECG signals. Simultaneously, the dynamic warning threshold generation module collects the target subject's respiratory rate and blood oxygen saturation to calculate a preset dynamic warning threshold adapted to the target subject's physiological state, improving the targeting and adaptability of the warning threshold setting, thereby increasing the accuracy of abnormal warnings from the ECG monitor. The ECG signal feature classification module compares real-time warning indicators with the dynamic warning thresholds to determine whether the ECG signal is abnormal, and identifies the probability identifiers of abnormal ECG signal characteristics when an abnormality is determined, providing a reference for clinicians and improving the efficiency of clinical intervention. The warning output module continuously outputs three types of warning indicators and immediately issues an ECG abnormality warning when an abnormal ECG signal is determined, outputting the probability identifiers of abnormal ECG signal characteristics, providing more comprehensive and interpretable reference information for clinical intervention.

[0015] Preferably, in order to capture effective information from multi-lead raw electrocardiogram (ECG) signals and improve the comprehensiveness of effective information capture, overcoming the problem of limited information from a single feature dimension, the expression of the ECG signal tensor model is as follows:

[0016] ;

[0017] ;

[0018] ;

[0019] ;

[0020] In the formula, X is the electrocardiogram signal tensor model. The set of real numbers indicates that all elements of the ECG signal tensor model are real numbers. T, C, and F are the time dimension, lead dimension, and feature dimension of the ECG signal tensor model, respectively, and t and c are the time sequence number and lead sequence number of the original multi-lead ECG signal, respectively. , and These are the elements in the electrocardiogram (ECG) signal tensor model with time dimension t, lead dimension c, and feature dimensions of 1, 2, and 3, respectively. This is the raw electrocardiogram signal of lead c with time sequence number t. The first-order difference of the raw ECG signal in lead c with time sequence number t. The frequency domain energy characteristics of the original ECG signal in lead c with time sequence number t are given. In the ECG signal tensor model X, the first feature dimension is used to retain the original ECG signals acquired from each lead, the second feature dimension is used to record the time domain variation characteristics of the original ECG signal, and the third feature dimension is used to record the frequency domain energy characteristics of the original ECG signal. By constructing the ECG signal tensor model, the multi-dimensional information of the ECG signal in the time domain, spatial domain, and frequency domain is retained, effectively reducing the problem of overemphasizing or seriously neglecting the ECG signal characteristics by single-dimensional analysis methods.

[0021] Preferably, to effectively reflect the dynamic characteristics of electrocardiogram signals, the dominant feature vector is obtained in the following ways:

[0022] The ECG signal tensor model is decomposed using the Tucker decomposition algorithm to obtain the core tensor, time dimension factor matrix, lead dimension factor matrix, and feature dimension factor matrix. The decomposition formula of the Tucker decomposition algorithm is as follows:

[0023] ;

[0024] In the formula, G is the core tensor. For time dimension factor matrix, For the lead dimension factor matrix, For feature dimension factor matrix, , and These represent tensor multiplications in the 1st, 2nd, and 3rd dimensions, respectively.

[0025] Based on the core tensor and the lead dimension factor matrix, calculate the translead enhancement core tensor;

[0026] The transconductance enhancement core tensor is reduced in dimensionality to obtain the dominant feature vector.

[0027] Preferably, the transconductance enhancement core tensor is obtained by means of:

[0028] Based on the attention mechanism, the contribution weight of each lead is calculated according to the core tensor and the lead dimension factor matrix. The calculation formula is as follows:

[0029] ;

[0030] ;

[0031] In the formula, Let S(c) be the contribution weight of lead c, S(c) be the significance score of lead c, i be the lead index, and C be the total number of leads. This represents the c-th column vector of the lead dimension factor matrix. It is an L2 norm;

[0032] Based on the core tensor and the contribution weight of each lead, the translead enhancement core tensor is calculated using the following formula:

[0033] ;

[0034] In the formula, To enhance the core tensor across leads, This represents a weighted product along the lead dimension. The contribution weight is calculated based on the significance score of each lead, which can effectively distinguish between effective leads and noisy leads. Thus, the weighted product highlights the core dynamic information corresponding to the effective leads in the core tensor and reduces the interference of noisy leads on the effective dynamic features in the subsequently constructed dominant feature vector.

[0035] Preferably, the real-time early warning indicators are obtained through the following methods:

[0036] The dominant feature vector is reconstructed in phase space to obtain the reconstructed phase space trajectory, so as to restore the chaotic dynamic state of the electrocardiogram signal;

[0037] The dimensions of the reconstructed phase space trajectory are adjusted to obtain a three-dimensional reconstructed phase space trajectory, which is adapted to the three-dimensional characteristics of the Lorentz equations.

[0038] Based on the Lorentz equations and the three-dimensional reconstructed phase space trajectory, the current Lorentz system parameters are calculated to quantify the chaotic state of the electrocardiogram signal.

[0039] Based on the current Lorentz system parameters and the Lorentz equations, a fitted three-dimensional reconstructed phase space trajectory point set A is constructed, and a healthy baseline trajectory point set B is constructed. The first early warning index is calculated based on the global difference between the fitted three-dimensional reconstructed phase space trajectory point set A and the healthy baseline trajectory point set B. By fitting the fitted three-dimensional reconstructed phase space trajectory point set A using the Lorentz equations, the original chaotic information of the electrocardiogram signal can be preserved while removing noise and interference trajectory points. The global difference in the point set refers to the difference in overall geometric shape, spatial distribution, or dynamic characteristics between two trajectory point sets in three-dimensional phase space, which is distinct from the local distance difference between single points. By quantifying the global difference between the fitted three-dimensional reconstructed phase space trajectory point set A and the healthy baseline trajectory point set B, the overall morphological difference between the chaotic electrocardiogram trajectory of the target object and the chaotic trajectory of the healthy baseline can be intuitively reflected, thus providing a data basis for judging whether the electrocardiogram signal of the target object is abnormal.

[0040] Based on the degree of deviation between the current Lorenz system parameters and the preset health benchmark Lorenz system parameters, a second early warning indicator is calculated; by quantifying the degree of deviation of the Lorenz system parameters, early abnormal signals in the target object's electrocardiogram signal that are morphologically normal but have abnormal chaotic system dynamics are captured, thereby improving the accuracy of monitoring early electrocardiogram abnormalities.

[0041] Real-time early warning indicators are calculated based on the first and second early warning indicators. The real-time early warning indicators integrate two types of indicators: those reflecting abnormal surface morphology of chaotic electrocardiogram (ECG) signals and those reflecting abnormal dynamic mechanisms of chaotic systems. This improves the comprehensiveness, accuracy, and robustness of ECG abnormality monitoring.

[0042] Preferably, to reduce the impact of the magnitude difference in Lorentz system parameters on the accuracy and comparability of the calculation results of the second early warning indicator, the calculation formula for the second early warning indicator is as follows:

[0043] ;

[0044] In the formula, The second early warning indicator is defined as follows: σ, ρ, and β are parameters in the current Lorenz system parameters that reflect the uniformity of electrocardiographic conduction, rhythm stability, and energy dissipation characteristics of the target object's electrocardiogram signal. , and These are the parameters in the preset healthy benchmark Lorenz system parameters that reflect the uniformity of electrocardiogram conduction, rhythm stability, and energy dissipation characteristics of healthy electrocardiogram signals. , and These are the weighting coefficients for deviations in electrocardiographic conduction uniformity, rhythm stability, and energy dissipation characteristics, respectively.

[0045] Preferably, the calculation formula for the real-time early warning indicator is as follows:

[0046] ;

[0047] ;

[0048] In the formula, For real-time early warning indicators, For adaptive fusion coefficients, The first early warning indicator, This is the second early warning indicator. and These are the minimum and maximum threshold values ​​for the adaptive fusion coefficients, respectively, and k is the adjustment sensitivity coefficient. The statistical maximum value of the second early warning indicator for healthy ECG signals is used. By introducing an adaptive fusion coefficient, the fusion weights of the first and second early warning indicators are adaptively allocated based on the abnormal scenarios of different ECG signals, which further improves the accuracy and robustness of the real-time early warning indicators.

[0049] Preferably, to improve the adaptability of the preset dynamic warning threshold to the physiological state of the target object and reduce the false positive alarm rate of the electrocardiogram monitor due to different movement states of the target object, the calculation formula for the preset dynamic warning threshold is as follows:

[0050] ;

[0051] In the formula, To preset dynamic early warning thresholds, This is the weighting coefficient for the respiratory rate term. This is the respiratory rate sensitivity coefficient. This is the weighting coefficient for the blood oxygen saturation term. For threshold correction term, The respiratory rate of the target subject. This is the baseline value for normal respiratory rate. The target is the target's blood oxygen saturation. By introducing respiratory rate and blood oxygen saturation terms, the preset dynamic warning threshold can be adjusted in real time according to the target's exercise and physiological state, thereby improving the accuracy of ECG abnormality warnings.

[0052] Preferably, the identification of abnormal features in the electrocardiogram signal specifically includes:

[0053] Based on the current Lorenz system parameters, the health baseline Lorenz system parameters, and the second early warning indicator, a Lorenz parameter feature vector F is constructed, specifically as follows: ;

[0054] Based on the Lorentz parameter feature vector F and a preset lightweight classifier, the possible abnormal features of the electrocardiogram signal are identified. The possible abnormal features include: possible myocardial ischemia, possible ventricular arrhythmia, possible sinus arrhythmia, and possible other abnormal states. Using the preset lightweight classifier, based on the Lorentz parameter which reflects the abnormal dynamic mechanism of the chaotic system, possible cardiac abnormalities of the target object are identified. On the one hand, this can improve the abnormality identification rate and adapt to the low computing power of embedded devices. On the other hand, it can also provide more reference information for clinicians, thereby improving the rate of clinical intervention.

[0055] This invention also provides an early warning method for an electrocardiogram monitor early warning system based on tensor modeling and chaos analysis, comprising the following steps:

[0056] S1: Collect the multi-lead raw electrocardiogram signal of the target object in real time according to the first preset sampling frequency, and collect the respiratory rate and blood oxygen saturation of the target object according to the second preset sampling frequency. Calculate the preset dynamic warning threshold based on the respiratory rate and blood oxygen saturation. The second preset sampling frequency is less than the first preset sampling frequency to reduce the power consumption of the respiratory rate and blood oxygen saturation acquisition unit and the fluctuation of the preset dynamic warning threshold.

[0057] S2: Based on the original multi-lead ECG signal and the preset number of sampling points, construct an ECG signal tensor model; when the number of sampling points of the original multi-lead ECG signal reaches the preset number of sampling points, the system automatically constructs an ECG signal tensor model to improve the system's real-time processing efficiency of ECG signals.

[0058] S3: Denoising and dimensionality reduction are performed on the ECG signal tensor model to obtain the dominant feature vector;

[0059] S4: Process the dominant feature vector based on the chaotic analysis method to obtain the first early warning index, the second early warning index, and the real-time early warning index;

[0060] S5: Output the first early warning indicator, the second early warning indicator, and the real-time early warning indicator;

[0061] S6: Based on the real-time warning indicator and the preset dynamic warning threshold, determine whether the ECG signal of the target object is abnormal; if it is determined that the ECG signal of the target object is not abnormal, then directly execute step S7; otherwise, issue an ECG signal abnormality warning, identify the abnormal feature probability identifier of the ECG signal, and output the abnormal feature probability identifier.

[0062] S7: Determine whether to continue the monitoring task. If to continue, return to step S1; otherwise, do not perform the operation.

[0063] One or more technical solutions provided by this invention have at least the following technical effects or advantages:

[0064] 1. By working collaboratively through multiple modules of the ECG monitor early warning system, the ECG abnormalities of the target object are quantified based on tensor modeling and chaos analysis methods. Combined with dynamic early warning thresholds adapted to the physiological state of the target object, the system realizes real-time automated monitoring and early warning indicator output of ECG abnormalities, thereby improving the accuracy and interpretability of ECG abnormality early warning.

[0065] 2. By tensor modeling the original ECG signals from multiple leads, the original information, first-order difference, and frequency domain energy characteristics of the ECG signals are integrated, overcoming the one-sidedness of information from a single feature dimension and improving the comprehensiveness of ECG signal information capture.

[0066] 3. By decomposing the tensor model using Tucker, and based on the lead dimension factor matrix and attention mechanism, the tensor decomposition results are correlated with the lead contribution, thereby enhancing the core dynamic information corresponding to the effective leads in the core tensor, while reducing the interference of noisy leads on the effective dynamic features in the constructed dominant feature vector;

[0067] 4. By reconstructing the phase space trajectory and fitting it with the Lorentz equations, a three-dimensional trajectory point set is obtained, reducing noise and interference points in the reconstructed phase space trajectory point set; the real-time warning indicators integrate the first warning indicator of quantified trajectory morphology abnormality and the second warning indicator of quantified chaotic system dynamic mechanism abnormality, improving the comprehensiveness and robustness of the real-time warning indicators and enhancing the ability to capture early electrocardiogram abnormalities.

[0068] 5. By introducing an adaptive fusion coefficient that can be adjusted in real time according to the magnitude of the second early warning indicator, the fusion weights of the two types of early warning indicators can be adaptively allocated according to different abnormalities in the phase space trajectory point set of the three-dimensional reconstruction of the electrocardiogram signal, thereby improving the scientificity and accuracy of the real-time early warning indicator calculation results.

[0069] 6. The dynamic early warning threshold is adaptively calculated based on the physiological state of the target object, which improves the adaptability of the ECG monitor early warning system to the target object and reduces the false positive alarm rate caused by the mismatch between the fixed early warning threshold and the specific target object;

[0070] 7. Using a pre-defined lightweight classifier, the system identifies possible abnormal features of electrocardiogram (ECG) signals based on the Lorenz parameter feature vector, enabling rapid classification of abnormal ECG signals, providing clinicians with more reference information, and improving the efficiency of clinical intervention.

[0071] 8. By real-time output of the first warning indicator for abnormal trajectory morphology, the second warning indicator for abnormal dynamic mechanism of chaotic system, and the adaptive fusion of real-time warning indicators, the system provides monitoring personnel with real-time and highly interpretable ECG monitoring data. When issuing an ECG signal abnormality warning, the system also adds an abnormal feature probability identifier of the output abnormal ECG to accelerate the speed of clinical judgment and intervention. Attached Figure Description

[0072] The accompanying drawings, which are provided to further illustrate embodiments of the invention and constitute a part of this invention, are not intended to limit the scope of the invention.

[0073] Figure 1 This is a schematic diagram of the composition of the ECG monitor early warning system based on tensor modeling and chaos analysis in this invention;

[0074] Figure 2 It is the fitting curve of the preset dynamic early warning threshold in this invention;

[0075] Figure 3This is a flowchart illustrating the early warning method for electrocardiogram monitors based on tensor modeling and chaos analysis in this invention. Detailed Implementation

[0076] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other.

[0077] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0078] Example 1

[0079] Please refer to Figure 1-2 This invention provides an early warning system for an electrocardiogram (ECG) monitor based on tensor modeling and chaotic analysis, comprising: an ECG signal acquisition module, a tensor modeling module, a feature processing module, a chaotic analysis module, a dynamic early warning threshold generation module, an ECG signal feature classification module, and an early warning output module. The ECG signal acquisition module is used to acquire multi-lead raw ECG signals of the target object in real time according to a first preset sampling frequency, supporting 5-lead or 12-lead expansion, with synchronous acquisition of ECG signals between leads. The ECG signal acquisition module includes: acquisition electrodes, a signal conditioning unit, and a data conversion and transmission unit. The acquisition electrodes are Ag / AgCl disposable ECG electrodes, and the signal conditioning unit uses the ECG-specific chip ADS1298, integrating an instrumentation amplifier, a high-pass filter unit, a low-pass filter unit, and a 50Hz notch filter unit. The high-pass filter unit has a cutoff frequency of 0.5Hz to suppress baseline drift interference in the ECG signal, the low-pass filter unit has a cutoff frequency of 35Hz to filter out electromyographic noise interference, and the 50Hz notch filter unit is used to specifically suppress power frequency interference to obtain a high-fidelity ECG signal. The data conversion and transmission unit is equipped with a 16-bit high-precision analog-to-digital converter module with a signal sampling accuracy of ≤1μV, which can realize the accurate quantization of the conditioned ECG signal. A serial peripheral interface is used to establish a data transmission link with the STM32H743 main control chip of the ECG monitor's early warning system, and the data transmission rate is set to 1Mbps to improve the real-time performance and reliability of the transmission of multi-lead raw ECG signals from the acquisition end to the processing end. The first preset sampling frequency is set to 250Hz to balance the amount of data and the acquisition accuracy.

[0080] The tensor modeling module is used to construct an ECG signal tensor model based on the original multi-lead ECG signal and a preset number of sampling points (250). The expression for the ECG signal tensor model is as follows:

[0081] ;

[0082] ;

[0083] ;

[0084] ;

[0085] In the formula, X is the electrocardiogram signal tensor model. Let T be the set of real numbers, C be the time dimension, C be the lead dimension, and F be the feature dimension of the ECG signal tensor model, respectively. T = 1s, C = 5 or 12, F = 3, and t and c be the time sequence number and lead sequence number of the original multi-lead ECG signal, respectively. , and These are the elements in the electrocardiogram (ECG) signal tensor model with time dimension t, lead dimension c, and feature dimensions of 1, 2, and 3, respectively. This is the raw ECG signal of lead c with time sequence number t, in mV. This is the first-order difference of the raw ECG signal in lead c at time sequence t, reflecting the rate of change of the ECG signal, with units of mV / sampling point. The frequency domain energy characteristics of the original electrocardiogram signal in lead c with time sequence number t;

[0086] in, The method for obtaining the frequency domain energy characteristics of lead c at time t can be as follows: Perform Morlet wavelet transform on the original ECG signal of lead c at time t to obtain wavelet coefficients at different scales. Sum the squares of the absolute values ​​of all Morlet wavelet coefficients at all scales to obtain the frequency domain energy characteristics of lead c at time t. The calculation formula is:

[0087] ;

[0088] In the formula, m is the wavelet scale index, and M is the total number of scales. The wavelet coefficients of the original ECG signal in lead c with time sequence number t at scale m.

[0089] The feature processing module is used for noise reduction and dimensionality reduction of the ECG signal tensor model to construct a dominant feature vector reflecting the dynamic characteristics of the ECG signal; wherein the dominant feature vector is obtained in the following ways:

[0090] The ECG signal tensor model is decomposed using the Tucker decomposition algorithm to obtain the core tensor, time dimension factor matrix, lead dimension factor matrix, and feature dimension factor matrix. The decomposition formula of the Tucker decomposition algorithm is as follows:

[0091] ;

[0092] In the formula, G is the core tensor, and its dimension is set to ( ), For time dimension factor matrix, For the lead dimension factor matrix, For feature dimension factor matrix, , and These represent tensor multiplications in the 1st, 2nd, and 3rd dimensions, respectively.

[0093] Based on the attention mechanism, the contribution weight of each lead is calculated according to the core tensor and the lead dimension factor matrix. The calculation formula is as follows:

[0094] ;

[0095] ;

[0096] In the formula, Let S(c) be the contribution weight of lead c, S(c) be the significance score of lead c, i be the lead index, and C be the total number of leads. This represents the c-th column vector of the lead dimension factor matrix. It is an L2 norm;

[0097] Based on the core tensor and the contribution weight of each lead, the translead enhancement core tensor is calculated using the following formula:

[0098] ;

[0099] In the formula, To enhance the core tensor across leads, This represents a weighted product along the lead dimension;

[0100] The transconductance enhancement core tensor Flatten the vector along the time-feature dimension, and then use principal component analysis or singular value decomposition to reduce the dimension to a 1-dimensional vector of length 120, thus obtaining the dominant feature vector.

[0101] The chaos analysis module is used to process the dominant feature vector based on chaos analysis methods, and calculate the first early warning indicator, the second early warning indicator, and the real-time early warning indicator, specifically including the following steps:

[0102] The dominant feature vector is reconstructed in phase space using a time-delay embedding method to obtain a reconstructed phase space trajectory. The point set of the reconstructed phase space trajectory is as follows:

[0103] ;

[0104] In the formula, The dominant feature vector, The dominant feature vector is the first The feature values ​​of each sampling point, where j is the dominant feature vector. The effective sampling point index is i, the time delay offset index is τ, the time delay parameter is τ is selected by mutual information method, and n is the embedding dimension is determined by false nearest neighbor method.

[0105] Adjusting the dimension of the reconstructed phase space trajectory yields a three-dimensional reconstructed phase space trajectory, the point set of which is:

[0106] ;

[0107] In the formula, a, b, and c are the time delay offset indices of the first, second, and third dimension sampling points in the three-dimensional reconstructed phase space trajectory, respectively;

[0108] Based on the Lorentz equations and the reconstructed phase space trajectory, the current Lorentz system parameters are calculated. The Lorentz equations are as follows:

[0109] ;

[0110] ;

[0111] ;

[0112] The values ​​of the first, second, and third dimensions of each trajectory point in the reconstructed three-dimensional phase space trajectory are respectively mapped to x, y, and z in the Lorentz equations, i.e. , , The parameters σ, ρ, and β of the current Lorentz system are obtained by fitting using the nonlinear least squares method.

[0113] Based on the current Lorentz system parameters and the Lorentz equations, a fitted three-dimensional reconstructed phase space trajectory point set A is constructed. A healthy baseline trajectory point set B is constructed based on the electrocardiogram signals of healthy individuals. According to the global difference between the fitted three-dimensional reconstructed phase space trajectory point set A and the healthy baseline trajectory point set B, a first warning index is calculated. Optionally, the first warning index is the Hausdorff distance between the fitted three-dimensional reconstructed phase space trajectory point set A and the healthy baseline trajectory point set B. The formula for calculating the Hausdorff distance is:

[0114] ;

[0115] In the formula, The Hausdorff distance is the distance between the fitted 3D reconstructed phase space trajectory point set A and the health baseline trajectory point set B. Let A be a point in the set of points A representing the fitted three-dimensional reconstructed phase space trajectory. A point in the set of health baseline trajectory points B;

[0116] Based on the degree of deviation between the current Lorenz system parameters and the preset health benchmark Lorenz system parameters, a second early warning index is calculated, wherein the calculation formula for the second early warning index is:

[0117] ;

[0118] In the formula, As the second early warning indicator, , and These are the parameters in the preset healthy benchmark Lorenz system parameters that reflect the uniformity of electrocardiogram conduction, rhythm stability, and energy dissipation characteristics of healthy electrocardiogram signals. , , , , and These are the weighting coefficients for the deviation terms of electrocardiographic conduction homogeneity, rhythm stability, and energy dissipation characteristics, respectively. , , ;

[0119] The real-time early warning indicator is calculated based on the first early warning indicator and the second early warning indicator. The calculation formula for the real-time early warning indicator is as follows:

[0120] ;

[0121] ;

[0122] In the formula, For real-time early warning indicators, For adaptive fusion coefficients, The first early warning indicator, This is the second early warning indicator. and These are the minimum and maximum threshold values ​​for the adaptive fusion coefficients, respectively. , k is the sensitivity adjustment coefficient, k=5. The statistical maximum value of the second warning indicator for healthy electrocardiogram signals. .

[0123] The dynamic warning threshold generation module is used to collect the respiratory rate and blood oxygen saturation of the target object according to a second preset sampling frequency, and calculate a preset dynamic warning threshold based on the respiratory rate and blood oxygen saturation; wherein the calculation formula of the preset dynamic warning threshold is:

[0124] ;

[0125] In the formula, To preset dynamic early warning thresholds, This is the weighting coefficient for the respiratory rate term. , This is the respiratory rate sensitivity coefficient. , This is the weighting coefficient for the blood oxygen saturation term. , For threshold correction term, In practical applications , , and Can be calibrated by clinical testing. The respiratory rate of the target subject. This is the baseline value for normal respiratory rate. , The target's blood oxygen saturation; Figure 2 For the target object's blood oxygen saturation At that time, preset dynamic early warning threshold Respiratory rate of the target subject and threshold correction term The fitted curve;

[0126] The dynamic warning threshold generation module uses an integrated sensor MAX30102 to synchronously collect the respiratory rate and blood oxygen saturation of the target object at a sampling frequency of 1Hz, and takes the 10-second moving average value as the respiratory rate of the target object for calculating the preset dynamic warning threshold. and blood oxygen saturation .

[0127] The ECG signal feature classification module is used to determine whether the ECG signal of the target object is abnormal based on the real-time warning indicators and the preset dynamic warning threshold. If an abnormality is determined, it identifies the probability identifier of the abnormal features of the ECG signal. The rule for determining whether the ECG signal of the target object is abnormal is: real-time warning indicators... If the value exceeds the preset dynamic early warning threshold for three consecutive monitoring periods, i.e., for three consecutive seconds. ;

[0128] Specifically, the identification of abnormal features in electrocardiogram (ECG) signals includes:

[0129] Based on the current Lorenz system parameters, the health baseline Lorenz system parameters, and the second early warning indicator, a Lorenz parameter feature vector F is constructed, specifically as follows: ;

[0130] Based on the Lorenz parameter feature vector F and a preset lightweight classifier, the abnormal feature probability identifiers of the electrocardiogram signal are identified; the abnormal feature probability identifiers include: possible myocardial ischemia, possible ventricular arrhythmia, possible sinus arrhythmia, and possible other abnormal states;

[0131] The preset lightweight classifier uses a random forest classifier, which includes 10 decision trees. The maximum depth of a single decision tree is 8 layers, and the minimum number of samples in a leaf node is 5. The MIT-BIH public ECG dataset is used as the training dataset. The training dataset includes four types of labeled samples: myocardial ischemia, ventricular arrhythmia, sinus arrhythmia, and other ECG abnormalities. The total number of samples is ≥10,000. The dataset is divided into training and test sets in a 7:3 ratio. The model parameters are optimized using 10-fold cross-validation. The preset lightweight classifier is obtained after training.

[0132] The early warning output module includes an OLED screen, a buzzer alarm, and a red LED light. The OLED screen displays the target object's first early warning indicator, second early warning indicator, real-time early warning indicator, heart rate, and blood oxygen saturation in real time. When the target object's electrocardiogram (ECG) signal is found to be abnormal, it further displays an indication of the likelihood of the abnormality. The buzzer alarm and the red LED light are used to emit a buzzer alarm sound and a red flashing alarm signal, respectively, when the target object's ECG signal is found to be abnormal, to remind medical staff to pay attention to the target object's ECG signal and determine whether further medical examination is needed, thereby intervening in the target object's possible cardiac abnormalities as early as possible.

[0133] Please refer to Figure 3 The early warning method for an electrocardiogram monitor early warning system based on tensor modeling and chaos analysis provided by this invention includes the following steps:

[0134] S1: The ECG signal acquisition module acquires the multi-lead raw ECG signal of the target object in real time at a sampling frequency of 250Hz. The multi-lead raw ECG signal can be selected as a 5-lead raw ECG signal or a 12-lead raw ECG signal. The dynamic warning threshold generation module synchronously acquires the respiratory rate and blood oxygen saturation of the target object at a sampling frequency of 1Hz, and calculates the preset dynamic warning threshold based on the 10-second moving average of the respiratory rate and blood oxygen saturation.

[0135] S2: The tensor modeling module constructs a 250×5×3 or 250×12×3-dimensional electrocardiogram signal tensor model based on the 5-lead or 12-lead original electrocardiogram signal and a preset number of 250 sampling points.

[0136] S3: The feature processing module performs noise reduction and dimensionality reduction on the ECG signal tensor model to obtain a 1-dimensional dominant feature vector with 120 sampling points;

[0137] S4: The chaos analysis module processes the dominant feature vector based on the chaos analysis method to obtain the first early warning index, the second early warning index, and the real-time early warning index.

[0138] S5: The early warning output module outputs the first early warning indicator, the second early warning indicator, and the real-time early warning indicator;

[0139] S6: The ECG signal feature classification module determines whether the ECG signal of the target object is abnormal based on the real-time warning index and the preset dynamic warning threshold. If the ECG signal of the target object is not abnormal, the ECG monitor warning system directly executes step S7. Otherwise, the warning output module issues an ECG signal abnormality warning, the ECG signal feature classification module calls the preset lightweight classifier to identify the abnormal feature probability identifier of the ECG signal, and the warning output module outputs the abnormal feature probability identifier.

[0140] S7: The ECG monitor's early warning system determines whether to continue the monitoring task. If it continues, it returns to step S1; otherwise, it does not perform the operation and the system enters standby mode.

[0141] This invention collects multi-lead raw electrocardiogram (ECG) signals from target subjects, processes these signals using tensor modeling and chaos analysis, extracts abnormal features reflecting the dynamic characteristics of the ECG signals, and provides early warnings to alert medical staff to ECG signal abnormalities in target subjects. This provides clinicians with a reference for determining whether further medical examinations are necessary. The core function of this invention is to assist in the detection of abnormal changes at the ECG signal level, rather than directly determining or diagnosing any heart disease. Ultimately, timely medical intervention reduces the health risks caused by the failure to promptly detect ECG signal abnormalities.

[0142] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0143] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An early warning system for electrocardiogram monitors based on tensor modeling and chaotic analysis, characterized in that, include: The electrocardiogram (ECG) signal acquisition module is used to acquire multi-lead raw ECG signals from the target object. The tensor modeling module is used to construct an electrocardiogram (ECG) signal tensor model based on the original multi-lead ECG signal. The expression of the ECG signal tensor model is as follows: ； ； ； ； In the formula, X is the electrocardiogram signal tensor model. Let T be the set of real numbers, C be the time dimension, C be the lead dimension, and F be the feature dimension of the ECG signal tensor model, respectively, and t be the time sequence number and lead sequence number of the original multi-lead ECG signal, respectively. , and These are the elements in the electrocardiogram (ECG) signal tensor model with time dimension t, lead dimension c, and feature dimensions of 1, 2, and 3, respectively. This is the raw electrocardiogram signal of lead c with time sequence number t. The first-order difference of the raw ECG signal in lead c with time sequence number t. The frequency domain energy characteristics of the original electrocardiogram signal in lead c with time sequence number t; The feature processing module is used for noise reduction and dimensionality reduction processing of the ECG signal tensor model to construct a dominant feature vector reflecting the dynamic characteristics of the ECG signal. The dominant feature vector is obtained in the following ways: The ECG signal tensor model is decomposed using the Tucker decomposition algorithm to obtain the core tensor, time dimension factor matrix, lead dimension factor matrix, and feature dimension factor matrix. The decomposition formula of the Tucker decomposition algorithm is as follows: ； In the formula, G is the core tensor. For time dimension factor matrix, For the lead dimension factor matrix, For feature dimension factor matrix, , and These represent tensor multiplications in the 1st, 2nd, and 3rd dimensions, respectively. Based on the core tensor and the lead dimension factor matrix, the translead enhancement core tensor is calculated. The translead enhancement core tensor is obtained by means of: Based on the attention mechanism, the contribution weight of each lead is calculated according to the core tensor and the lead dimension factor matrix. The calculation formula is as follows: ； ； In the formula, Let S(c) be the contribution weight of lead c, S(c) be the significance score of lead c, i be the lead index, and C1 be the total number of leads. This represents the c-th column vector of the lead dimension factor matrix. It is an L2 norm; Based on the core tensor and the contribution weight of each lead, the translead enhancement core tensor is calculated using the following formula: ； In the formula, To enhance the core tensor across leads, This represents a weighted product along the lead dimension; The transconductance enhancement core tensor is reduced in dimensionality to obtain the dominant feature vector; The chaos analysis module is used to process the dominant feature vector based on chaos analysis methods, and calculate a first early warning indicator, a second early warning indicator, and a real-time early warning indicator. The real-time early warning indicator is obtained in the following ways: The dominant feature vector is reconstructed in phase space to obtain the reconstructed phase space trajectory; Adjusting the dimensions of the reconstructed phase space trajectory yields a three-dimensional reconstructed phase space trajectory; Based on the Lorentz equations and the three-dimensional reconstructed phase space trajectory, calculate the current Lorentz system parameters; Based on the current Lorentz system parameters and the Lorentz equations, a set of fitted three-dimensional reconstructed phase space trajectory points A is constructed, and a set of healthy baseline trajectory points B is constructed. Based on the global difference between the set of fitted three-dimensional reconstructed phase space trajectory points A and the set of healthy baseline trajectory points B, a first early warning index is calculated. Based on the degree of deviation between the current Lorenz system parameters and the preset health benchmark Lorenz system parameters, a second early warning index is calculated. The calculation formula for the second early warning index is as follows: ； In the formula, The second early warning indicator is defined as follows: σ, ρ, and β are parameters in the current Lorenz system parameters that reflect the uniformity of electrocardiographic conduction, rhythm stability, and energy dissipation characteristics of the target object's electrocardiogram signal. , and These are the parameters in the preset healthy benchmark Lorenz system parameters that reflect the uniformity of electrocardiogram conduction, rhythm stability, and energy dissipation characteristics of healthy electrocardiogram signals. , and These are the weighting coefficients for the deviation terms of electrocardiographic conduction homogeneity, rhythm stability, and energy dissipation characteristics, respectively. The real-time early warning indicator is calculated based on the first early warning indicator and the second early warning indicator. The calculation formula for the real-time early warning indicator is as follows: ； ； In the formula, For real-time early warning indicators, For adaptive fusion coefficients, The first early warning indicator, This is the second early warning indicator. and These are the minimum and maximum threshold values ​​for the adaptive fusion coefficients, respectively, and k is the adjustment sensitivity coefficient. The statistical maximum value of the second warning indicator for healthy electrocardiogram signals; A dynamic early warning threshold generation module is used to collect the respiratory rate and blood oxygen saturation of the target object, and calculate a preset dynamic early warning threshold based on the respiratory rate and blood oxygen saturation. The calculation formula for the preset dynamic early warning threshold is as follows: ； In the formula, To preset the dynamic early warning threshold, α is the weighting coefficient of the respiratory rate term. This is the respiratory rate sensitivity coefficient. This is the weighting coefficient for the blood oxygen saturation term. For threshold correction term, The respiratory rate of the target subject. This is the baseline value for normal respiratory rate. The target's blood oxygen saturation; The ECG signal feature classification module is used to determine whether the ECG signal of the target object is abnormal based on the real-time warning index and the preset dynamic warning threshold. If an abnormality is determined, the module identifies the probability identifier of the abnormal features of the ECG signal. Early warning output module: used to output the first early warning indicator, the second early warning indicator and the real-time early warning indicator. When it is determined that the electrocardiogram signal of the target object is abnormal, an early warning of abnormal electrocardiogram signal is issued and the probability identifier of the abnormal feature is output.

2. The ECG monitor early warning system based on tensor modeling and chaotic analysis according to claim 1, characterized in that, The specific features identified by the abnormal characteristics of the electrocardiogram signal include: Based on the current Lorenz system parameters, the health baseline Lorenz system parameters, and the second early warning indicator, a Lorenz parameter feature vector F1 is constructed, specifically as follows: In the formula, σ, ρ, and β are parameters in the current Lorentz system parameters that reflect the uniformity of electrocardiographic conduction, rhythm stability, and energy dissipation characteristics of the target object's electrocardiogram signal, respectively. , and These are the parameters in the preset healthy benchmark Lorenz system parameters that reflect the uniformity of electrocardiogram conduction, rhythm stability, and energy dissipation characteristics of healthy electrocardiogram signals. This is the second early warning indicator; Based on the Lorenz parameter feature vector F1 and the preset lightweight classifier, the abnormal feature probability identifiers of the electrocardiogram signal are identified; the abnormal feature probability identifiers include: possible myocardial ischemia, possible ventricular arrhythmia, possible sinus arrhythmia, and possible other abnormal states.

3. An early warning method for an electrocardiogram monitor early warning system based on tensor modeling and chaos analysis as described in claim 1 or 2, characterized in that, Includes the following steps: S1: Collect the multi-lead raw electrocardiogram signal of the target object in real time according to the first preset sampling frequency, collect the respiratory rate and blood oxygen saturation of the target object according to the second preset sampling frequency, and calculate the preset dynamic warning threshold based on the respiratory rate and blood oxygen saturation. S2: Construct an ECG signal tensor model based on the original multi-lead ECG signal and the preset number of sampling points; S3: Denoising and dimensionality reduction are performed on the ECG signal tensor model to obtain the dominant feature vector; S4: Process the dominant feature vector based on the chaotic analysis method to obtain the first early warning index, the second early warning index, and the real-time early warning index; S5: Output the first early warning indicator, the second early warning indicator, and the real-time early warning indicator; S6: Based on the real-time warning indicators and the preset dynamic warning threshold, determine whether the electrocardiogram signal of the target object is abnormal; If it is determined that the target object's electrocardiogram signal is not abnormal, then step S7 is executed directly; otherwise, an abnormal electrocardiogram signal warning is issued, the abnormal feature probability identifier of the electrocardiogram signal is identified, and the abnormal feature probability identifier is output. S7: Determine whether to continue the monitoring task. If to continue, return to step S1; otherwise, do not perform the operation.

Images