Intelligent cardiovascular health non-inductive monitoring method, device and system

Abstract

The invention relates to the field of heart rate detection, and discloses an intelligent cardiovascular health non-inductive monitoring method, device and system. On the basis of motion analysis of the whole heart in the cardiac cycle, a basic model between two types of one-dimensional representations of the body surface is established, a sign signal preprocessing method and an electrocardiosignal reconstruction method fusing mechanism modeling and a deep learning method are provided, and therefore an electrocardiogram is synthesized, and the electrocardiosignal reconstruction accuracy is improved. The electrophysiological function information, the biological difference and the pathological information of the heart can be effectively reserved; the function of multi-dimensional non-inductive heart physiology evaluation on single equipment is realized; meanwhile, based on BCG signals collected in a non-contact mode, the detection mode is more convenient and safer; compared with non-contact equipment for acquiring other modal data, the device is small in interference and easy to charge; real-time electrocardiogram data of the heart can be captured without direct contact with the skin, and the defect that only coarse-grained heart activity monitoring can be carried out in horizontal heart monitoring is overcome.

Description

Technical Field

[0001] The present invention relates to the field of heart rate detection technology, and in particular to an intelligent cardiovascular health non-contact monitoring method, device and system. Background Art

[0002] Ballistocardiogram (BCG) refers to the mechanical movement signal of the human chest, which is mainly caused by the contraction and relaxation of the heart and the breathing of the lungs. When the heart contracts, the contraction of the heart muscle causes changes in blood flow and pressure, which causes tiny mechanical vibrations in the chest. These vibration signals can be captured to form BCG signals. BCG signals are non-invasive, convenient, repeatable, and high-resolution physiological signals with many current applications.

[0003] The traditional electrocardiogram (ECG) monitoring method requires sticking or clamping traditional contact electrodes on the skin surface. Usually, in order to obtain good electrophysiological information, electrolytes need to be applied to the skin surface. As shown in patent CN210990273U, this ECG acquisition method still requires wearing electrodes, which is not suitable for patients with skin damage due to burns and other reasons. At the same time, the wearing method based on electrodes is not suitable for long-term use in daily life. For some detection devices that do not require skin contact, there is a disadvantage of inaccurate detection. As shown in patent CN106404452B, only coarse-grained cardiac activity (such as heart rate, bed, heartbeat interval, etc.) can be monitored and evaluated. Summary of the invention

[0004] In order to solve at least one aspect of the above problems, the present invention first provides an intelligent cardiovascular health non-sensing monitoring method, comprising the following steps:

[0005] Signal collection: Lay the flexible detection belt under the chest of the person to be monitored, and use the characteristics of the flexible detection belt to produce micro-deformation when under pressure to collect the mechanical movement signal of the chest of the person to be monitored;

[0006] Signal preprocessing: Use Fourier transform to analyze the frequency distribution of the collected ECG signal and select specific frequencies for filtering; use variational mode decomposition (VMD) technology to decompose and reconstruct the collected BCG signal, and obtain a stable representation of the overall mechanical movement of the heart through pattern matching;

[0007] Feature acquisition: Use the Pan-Tompkins algorithm to retrieve the R wave in the long-term ECG signal. According to the biological mechanism, use the start and end points of each heart beat ECG as the time calibration to segment the synchronous BCG signal and determine the position of each BCG characteristic wave.

[0008] Establishing a model: Establishing a biological mechanism and blood circulation feedback model, which includes a myocardial cell excitation-contraction coupling equivalent model, a heart overall composite network model, and an electrocardiogram-ballistocardiogram coupling model. The electrocardiogram-ballistocardiogram coupling model uses a mutual supervision detection and positive feedback correction algorithm based on a three-layer iterative regularization algorithm TSI-DTW to accelerate the convergence speed of the network model and improve the accuracy of waveform detection;

[0009] Establish a non-contact ECG signal reconstruction model: Establish a multi-domain generative adversarial network MultiGAN to achieve end-to-end reconstruction mapping from BCG to ECG;

[0010] Heart assessment: The synthetic ECG and measured BCG are combined to evaluate the speed of changes in cardiac contractility, ventricular valve movement, cardiac contractility, and fluctuations in cardiac output CO related to the aortic blood pressure difference.

[0011] Optionally, in the signal preprocessing step, the collected BCG signal is processed specifically as follows: the BCG signal is decomposed into independent eigenmode functions of different center frequencies using a VMD method, and each mode expression is:

[0012] u k (t) = A k (t)cos(φ k (t)) t=1,2,3,...,k

[0013] In the formula, A k (t) represents the envelope, φk(t) represents the phase, and the VMD method is used to solve the following expression in combination with the Lagrange multiplier method. In the following expression, the unconstrained variational problem is adaptively determined to determine u k (t) and center frequency ω k :

[0014]

[0015] Where δ(t) is the Dirac function, * is the convolution operator, the penalty term α ensures the accuracy of decomposition and reconstruction in the presence of Gaussian noise, and the Lagrangian penalty term λ(t) can ensure strict constraints on the decomposition process;

[0016] Using the alternating direction multiplier method ADMM, the optimization problem with multiple variables is transformed into an optimization problem about u k (t), w k And the three optimization sub-problems of λ(t), construct the augmented Lagrangian function according to the constraints of the sub-problems and the objective function, and then solve the optimal solutions of the three optimization sub-problems respectively.

[0017] Optionally, the VMD parameters are optimized, and the specific optimization steps are as follows: first determine the range of the parameters, then perform VMD decomposition on the original signal and calculate the relative entropy under each parameter combination, iteratively update the parameters and calculate the relative entropy, and finally select the parameters corresponding to the minimum relative entropy as the optimal solution; preset the cardiac pattern matching score threshold, discard the data segments whose distance from the best template matching result LDTW is higher than the threshold, and reduce the impact of body motion interference of continuous data on the BCG signal.

[0018] Optionally, the best influencing parameter combination of VMD is searched, and the penalty parameter and the number of components are selected according to the minimum relative entropy. First, the original signal is decomposed into k narrowband modal functions, which are divided into low-frequency modes and high-frequency modes according to the size of the center frequency. The initial value of the decomposition layer K is 5, the step size is 1, and the range is [3, 20]. The initial value of the penalty parameter a is 2000, the step size is 50, the range is [100, 3000], and the convergence tolerance t is 1. 0 The size is 1*.

[0019] Optionally, the specific method for establishing the myocardial cell excitation-contraction coupling equivalent model is as follows:

[0020] At the i-th time point in a cardiac cycle, the electrical signal is transmitted sequentially along the excitation conduction system to the n-th myocardial cell, and the cardiac electrical activity e(t i ), mechanical activity m(t i +δt i ) as the observed value, a two-port network is established, and the s-domain expression of the electrical-mechanical transport signal relationship of the “empty heart” at the i-th time point of the cardiac cycle is:

[0021]

[0022] In the formula, G n (·) represents the transfer function of the ECC mechanism, represents the absolute refractory period ERP, represents the mechanical activity of the heart, e(s) represents the electrical activity of the heart, ECG n (s) indicates the electrocardiogram data;

[0023]

[0024] In the formula, f m (·) represents the mechanical transmission function, Represents force sensor results.

[0025] Optionally, the specific method for establishing the whole heart composite network model is as follows:

[0026] At the jth time point in a cardiac cycle (j∈ pumping event), the electrical signal is transmitted sequentially along the excitation conduction system to the kth myocardial cell; when the pumping event occurs, the electromechanical movement of the entire heart is the composite port network T j ; Then at the jth time point of the cardiac cycle, the s-domain expression of the electrical-mechanical motion signal relationship of the entire heart is:

[0027]

[0028] In the formula, F k (·) represents the transfer function of the electromechanical motion of the whole heart, represents the absolute refractory period ERP, represents the mechanical activity of the heart, e(s) represents the electrical activity of the heart, ECG k (s) indicates the electrocardiogram data;

[0029] In summary, the s-domain expression of the electromechanical relationship of the heart as a whole during the entire cardiac cycle is:

[0030]

[0031] Where j∈ pumping event, T represents the length of cardiac cycle, G n (·) represents the transfer function of the ECC mechanism, F k (·) represents the transfer function of the electromechanical motion of the whole heart.

[0032] Optionally, the specific method for establishing the electrocardiogram-ballistocardiogram coupling model is as follows:

[0033] Taking the whole cycle movement of the heart as the observation object, starting from the sinoatrial node, the myocardial cells that sequentially depolarize and repolarize along the excitation conduction system can be equivalent to the potential changes of different phases of the sinoatrial node electrical activity, that is: (e 1 (s),e 2 (s),…,e T (s)) = ECG (t); By introducing the blood circulation feedback system, the relationship between the ballistocardiogram and the electrocardiogram is described;

[0034] The electrocardiogram-ballistocardiogram transfer function represented by the electrical and mechanical activity of the heart is:

[0035]

[0036] In the formula, F BCG (t) is the system output, which is used to describe the mechanical observation signal and its unit is Newton (N); I ECG(t) is the system input, which is used to describe the electrical observation signal in ampere (A); G(s) is the transfer function of excitation-contraction coupling; H(s) is the transfer function of blood circulation; u(t–t 0 ) is a unit step function, which means that from t 0 The blood circulation events occur after the moment t. This formula describes the change of BCG over time, including the open-loop response of the myocardial cell two-port network system and the positive feedback of blood circulation at t. 0 The impact of time superposition:

[0037] A mutually supervised BCG detection method is proposed, including feedback correction, detection algorithm constraints and a three-layer iterative time dynamic warping algorithm TSI-DTW. The waveform detection results are used to feed back DTW template training. The mutually supervised learning method of DTW template for feature extraction and labeling can accelerate the convergence speed of the network model. Moreover, the DTW template corrected by mutual supervised learning can correct the shortcomings of manual experience, reduce poorly marked heartbeats, and further improve the accuracy of waveform detection.

[0038] Optionally, the specific method of establishing the non-contact ECG signal reconstruction model is as follows:

[0039] The multi-domain generative adversarial network Mixed-GAN includes a generator network G, a time domain discriminator D, and a frequency domain discriminator D. 2 , the input of generator G is the measured ECG and BCG, and the output is the synthesized X GenECG , the input of the time domain discriminator D is X GenECG and X RealECG , frequency domain discriminator D 2 The input is the spectrum amplitude of the real ECG synthesized after Fourier transform, denoted as S GenECG and S RealECG

[0040] The generator is used to generate synthetic data similar to real ECG signals from noise; the generator includes five fully connected layers, the first four fully connected layers perform batch normalization BN and ReLU activation function to enhance the nonlinear representation ability of the model; the last fully connected layer uses the Tanh function to ensure that the generated ECG signal falls within the range of [-1, 1];

[0041] In generative adversarial training, based on the consistency of the time domain waveform, the objective function of updating the time domain discriminator D is expressed as follows:

[0042]

[0043] In the formula, E(·) represents expectation; D 1 (X GenECG ) represents the time domain discriminator D 1To generate the output of ECG; D 1 (X RealECG ) means D 1 Output of real ECG;

[0044] Based on frequency domain consistency, update the frequency domain discriminator D 2 The objective function is expressed as follows:

[0045]

[0046] In the formula, E(·) represents expectation; D 2 (S GenECG ) represents the frequency domain discriminator D 2 To generate the output of ECG spectrum amplitude; D 2 (S RealECG ) means D 2 Output of real ECG spectrum amplitude;

[0047] The loss function for updating the generator G is expressed as follows:

[0048]

[0049] Based on the deep learning network architecture that has learned cross-domain mapping between cardiac mechanical activity and cardiac electrical activity, the cardiac mechanical activity data extracted at the current moment is input and the ECG measurement result at the current moment is output, thereby realizing non-contact electrocardiogram monitoring; after machine training on 25 healthy users' experimental data, the Pearson correlation coefficient of this method on the test set data is 0.963, and the root mean square error is 0.058.

[0050] Compared with the prior art, the intelligent cardiovascular health non-contact monitoring method in the present invention proposes a domain mapping algorithm for a specified one-dimensional mechanical signal of the heart, which can obtain an electrocardiogram based on the one-dimensional mechanical signal of the heart. Based on the electrocardiogram, a fine-grained evaluation of the heart can be achieved, and real-time electrocardiogram data of the heart can be captured without direct contact with the skin, which solves the deficiency that recumbent heart monitoring can only perform coarse-grained heart activity monitoring.

[0051] In addition, the present invention provides an intelligent cardiovascular health non-sensing monitoring device, comprising:

[0052] A flexible detection belt is used to collect mechanical motion signals of the chest of a human body to be monitored. The flexible detection belt includes a piezoelectric sensor, a piezoresistive sensor, a dynamic range controllable data acquisition circuit, a feedback control module, an AI autonomous data analysis module, a second data analysis module, a minute data analysis module, a daily analysis module, a data storage module, a WIFI Internet of Things protocol module, a router, a cloud server, a background management, and a user terminal;

[0053] The main processor is used for data processing and comprehensive evaluation according to the intelligent cardiovascular health non-contact monitoring method as described above.

[0054] Compared with the prior art, the advantages of the intelligent cardiovascular health non-contact monitoring device described in the present invention are that the non-contact acquisition of key life characteristics signals is invisible to the patient and does not require connecting objects to the patient; it is small in size, light and thin, can be rolled up, and is easy to carry; it has a simple structure and is easy to install.

[0055] In addition, the present invention provides an intelligent cardiovascular health non-contact monitoring system, including an electronic terminal, on which is provided a WIFI module, which is wirelessly connected to a flexible detection belt, and the electronic terminal stores instructions that can be executed by the main processor, and the instructions are executed by the main processor to implement the intelligent cardiovascular health non-contact monitoring method as described above.

[0056] Compared with the prior art, the advantages of the intelligent cardiovascular health non-contact monitoring system described in the present invention are that it adopts an open design concept, equipment and cloud service, and has flexible expansion capabilities; it adopts the idea of ​​both long and short storage, and has a hardware data backtracking function for delays / missed transmissions caused by various situations: the equipment has a network disconnection backhaul function, which can recover the previous 60 seconds of second data and the previous 30 minutes of minute data; it adopts local daily storage for up to 15 days to eliminate unexpected interference and ensure that data is not lost. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A flowchart of an intelligent cardiovascular health non-sensing monitoring method according to an embodiment of the present invention;

[0058] Figure 2 It is a schematic diagram of a multi-module hardware sensor;

[0059] Figure 3 This is a device architecture diagram for integrating multi-module sensor intelligent fusion modules;

[0060] Figure 4 is the filtering result and the spectrum of ECG before and after filtering;

[0061] Figure 5 Flow chart of optimization calculation of VMD parameters;

[0062] Figure 6 This is the BCG pretreatment result;

[0063] Figure 7 It is the result of signal segmentation and feature extraction;

[0064] Figure 8 The ECC mechanism of the cell and its equivalent circuit;

[0065] Fig. 9It is the overall complex network model of the heart;

[0066] Fig.10 It is the blood circulation feedback system;

[0067] Fig.11 It is the pattern matching result of TSI-DTW mutual supervision algorithm;

[0068] Fig.12 Generate adversarial network architecture graphs for multiple domains;

[0069] Fig.13 The results of the model reconstruction based on non-contact ECG signals;

[0070] Fig.14 The extended application results of synthetic ECG and measured BCG;

[0071] Fig.15 Design ideas for the communication layer of the intelligent cardiovascular health non-contact monitoring system;

[0072] Fig.16 It is the mobile application interface. DETAILED DESCRIPTION

[0073] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0074] In the description of the present invention, it should be understood that the terms "upper", "lower" and the like indicate positions or location relationships that are based on the positions or location relationships when the product is normally used.

[0075] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features.

[0076] The embodiment of the present invention provides an intelligent cardiovascular health non-sensing monitoring method, including: Figure 1The steps shown are as follows: using flexible sensors to collect mechanical motion signals related to the heart on the body surface; performing signal preprocessing on the collected signals, including signal filtering, segmentation and feature extraction; proposing an electrocardiogram-ballistocardiogram coupling model based on biological mechanisms such as myocardial cells, overall heart activity, and blood circulation; proposing a mutual supervision detection and positive feedback correction algorithm based on the three-layer iterative regularization algorithm TSI-DTW: the preprocessed BCG signal is pattern matched to obtain a stable representation synthesis module of the overall mechanical motion of the heart; proposing an improved generative adversarial network (Mixed-GAN) to achieve end-to-end reconstruction mapping from BCG to ECG; finally, based on the deep learning network architecture that has learned the cross-domain mapping of cardiac mechanical activity and cardiac electrical activity, the cardiac mechanical activity data extracted at the current moment is input, and the reconstructed ECG at the current moment is output, thereby realizing non-contact electrocardiogram monitoring.

[0077] Signal acquisition:

[0078] like Figure 2 and 3 As shown in the figure, a multi-mode body surface motion sensor is used to capture tiny movements of the human heart, breathing, etc. The sensor runs the AI ​​adaptive algorithm locally and autonomously, automatically adjusting the optimal working parameters according to the user and environment, thereby avoiding adaptability problems under different people and different usage conditions; using the fast-response autonomous IoT protocol, the sensor to the server and then to the terminal is transmitted without delay in seconds; adopting an open design concept, the sensor and cloud service have flexible expansion capabilities; the sensor has a network disconnection return function, adopts the idea of ​​both long and short storage, and has a hardware data backtracking function for delays / missing transmissions caused by various situations, which can recover the first 60 seconds of second data and the first 30 minutes of minute data; using local daily storage for up to 15 days to eliminate unexpected interference and ensure that data will not be lost.

[0079] ECG preprocessing:

[0080] Using Fourier transform to analyze the frequency distribution of ECG signals, it is observed that the frequency of ECG signals is mainly below 45Hz. Since the useful information of ECG signals is below 30Hz, the ECG is filtered using a Butterworth bandpass filter from 1 to 30Hz. The filtering results and the spectrum of ECG before and after filtering are shown in Figure 2. Figure 4 shown.

[0081] BCG preprocessing of cardiac shock signal:

[0082] The observed BCG mainly contains respiratory signals, noise and BCG itself. After spectrum analysis, it was found that the respiratory signal appears as low-frequency harmonics in the 0.2-0.9Hz frequency band, corresponding to a mode with a lower center frequency; while the energy of the cardiac impact signal is mainly concentrated in the 1-10Hz frequency band. Therefore, this application uses a variational mode decomposition VMD method that can adaptively determine the center frequency and bandwidth to separate the respiratory signal and noise interference in the observed BCG. After decomposition by the VMD method, the original signal is decomposed into independent intrinsic mode functions with different center frequencies, and the expressions of each mode are:

[0083] u k (t) = A k (t)cos(φ k (t)) t=1,2,3,...,k

[0084] In the formula, A k (t) represents the envelope, and φk(t) represents the phase. Combined with the Lagrange multiplier method, the VMD algorithm adaptively determines u by solving the following unconstrained variational problem: k (t) and center frequency ω k .

[0085]

[0086] Where δ(t) is the Dirac function, * is the convolution operator, the penalty term α ensures the accuracy of decomposition and reconstruction in the presence of Gaussian noise, and the Lagrangian penalty term λ(t) can ensure strict constraints on the decomposition process.

[0087] Based on the alternating direction multiplier method ADMM, the optimization problem with multiple variables is transformed into an optimal solution for u k (t), w k Specifically, the augmented Lagrangian function is constructed according to the constraints and objective functions of the sub-problems, and then the optimal solutions of the three optimizers are solved respectively.

[0088] Furthermore, the optimization calculation process of VMD parameters is as follows: Figure 5 As shown in the figure. When optimizing VMD parameters, the range of parameters is first determined, and then the original signal is decomposed by VMD and the relative entropy under each parameter combination is calculated. The parameters are iteratively updated and the relative entropy is calculated, and finally the parameters corresponding to the minimum relative entropy are selected as the optimal solution. Considering the influence of body motion interference of continuous data on BCG signals, the heart pattern matching score threshold is pre-set to discard data segments whose distance from the best template matching result LDTW is higher than the threshold.

[0089] The optimal influencing parameter combination of VMD is searched, and the penalty parameter and the number of components are selected according to the minimum relative entropy. Specifically, the original signal is first decomposed into k narrowband modal functions, which are divided into low-frequency modes and high-frequency modes according to the size of the center frequency. The initial value of the decomposition layer K is 5, the step size is 1, and the range is [3, 20]; the initial value of the penalty parameter a is 2000, the step size is 50, and the range is [100, 3000]. Convergence tolerance t 0 The size is 1*10 -7 .like Figure 6 As shown, it is the result of BCG decomposition and reconstruction, which is consistent with its spectrum distribution law.

[0090] Feature extraction:

[0091] The Pan-Tompkins algorithm is used to retrieve the R wave in the long-term ECG signal and then according to the biological mechanism, the ECG start and end points of each heart beat are used as the time calibration to segment the synchronous BCG signal. Figure 7 As shown, the position of each characteristic wave of BCG is determined according to the generation mechanism of each BCG wave and the duration of cardiac cycle events.

[0092] Equivalent model of myocardial cell excitation-contraction coupling:

[0093] At the i-th time point in a cardiac cycle, the electrical signal is transmitted sequentially along the excitation conduction system to the n-th myocardial cell. The ECC mechanism of the cell and its equivalent circuit are as follows: Figure 8 shown.

[0094] In order to observe the mapping relationship between cardiac electrical and mechanical signals without discussing the process, we take the cardiac electrical activity e(t i ), mechanical activity m(t i +δt i ) as the observed value, and a two-port network is established. Then, at the i-th time point of the cardiac cycle, the s-domain expression of the electrical-mechanical signal relationship of the “empty heart” is:

[0095]

[0096] In the formula, G n (·) represents the transfer function of the ECC mechanism, represents the absolute refractory period ERP, represents the mechanical activity of the heart, e(s) represents the electrical activity of the heart, ECG n (s) indicates the electrocardiogram reading.

[0097]

[0098] In the formula, f m (·) represents the mechanical transmission function, Represents force sensor results.

[0099] The overall complex network model of the heart:

[0100] like Fig. 9 As shown in Figure 1, at the jth time point (j∈ pumping event) in a cardiac cycle, the electrical signal is transmitted sequentially along the excitation conduction system to the kth myocardial cell. When a pumping event occurs, the electromechanical motion of the entire heart is the composite port network T j Then at the jth time point of the cardiac cycle, the s-domain expression of the electrical-mechanical transport signal relationship of the entire heart is:

[0101]

[0102] In the formula, F k (·) represents the transfer function of the electromechanical motion of the whole heart, represents the absolute refractory period ERP, represents the mechanical activity of the heart, e(s) represents the electrical activity of the heart, ECG k (s) indicates the electrocardiogram reading.

[0103] In summary, the s-domain expression of the electromechanical relationship of the heart as a whole during the entire cardiac cycle is:

[0104]

[0105] Where j∈ pumping event, T represents the length of cardiac cycle, G n (·) represents the transfer function of the ECC mechanism, F k (·) represents the transfer function of the electromechanical motion of the whole heart.

[0106] ECG-Ballistocardiogram coupling model:

[0107] In order to obtain an accurate and complete electrocardiogram, it is necessary to know the mechanical motion state of the heart. However, the mechanical motion state of the heart cannot be directly monitored. The mechanical signal ballistocardiogram at a specific monitoring position on the body surface is selected as the observation signal of mechanical motion, and the electrocardiogram is used as the observation signal of electrical activity.

[0108] like Fig.10 As shown, the whole cycle movement of the heart is taken as the observation object. Starting from the sinoatrial node, the myocardial cells that sequentially depolarize and repolarize along the excitation conduction system can be equivalent to the potential changes of different phases of the sinoatrial node electrical activity, that is: (e 1 (s),e 2 (s),…,e T (s)) = ECG (t). Similarly, the cardiac mechanical activity can be divided into piecewise functions. Further, this application introduces Fig.10The blood circulation feedback system shown describes the relationship between the ballistocardiogram and the electrocardiogram.

[0109] Note: G is the electrical excitation and mechanical response of the heart, H is the product of the transfer function of each blood circulation event, u is the delay switch of each blood circulation action, t 0 It represents the phase difference between BCG and ECG signals after all delays are added together.

[0110] The electrocardiogram-ballistocardiogram transfer function represented by the electrical and mechanical activity of the heart is:

[0111]

[0112] In the formula, F BCG (t) is the system output, describing the mechanical observation signal, and its unit is Newton (N); I ECG (t) is the system input, describing the electrical observation signal, in amperes (A). G(s) is the transfer function of excitation-contraction coupling; H(s) is the transfer function of blood circulation; u(t–t 0 ) is a unit step function, which means that from t 0 Blood circulation events occur after the moment begins.

[0113] This formula describes the change of BCG over time, which includes the open-loop response of the myocardial cell two-port network system and the positive feedback of blood circulation at t 0 The impact of moment superposition: On the one hand, the positive feedback of blood circulation in the excitation-contraction coupling mechanism will make the overall response of the system stronger; but on the other hand, due to the intensification of oscillations, it will also increase the instability of the system.

[0114] Cardiac Pattern Matching Model:

[0115] BCG and ECG signals are typical biological signals in the form of time series, which are random and unstable. However, continuous cardiac signals have stable periodicity. For the same subject, the electro-mechanical relationship described by continuous ECG and BCG has stable periodicity. A mutually supervised BCG detection method is proposed, including feedback correction, detection algorithm constraints and a three-layer iterative time dynamic warping algorithm TSI-DTW. Fig.11 The results of pattern matching are shown. The corrected non-overlapping segments have morphological regularity, and the characteristic waves are consistent with the typical waveform.

[0116] Table 1 Comparison of heart rate calculation results based on BCG

[0117]

[0118] Table 1 summarizes the results of TSI-DTW, CEEMDAN, and PDA algorithms for extracting heart rate from BCG on public datasets. Through the comparison, it can be found that under the same public dataset, the mutual supervision learning method of using waveform detection results to feed back DTW template training and DTW templates to perform feature extraction and annotation can accelerate the convergence speed of the network model, and the DTW template corrected by mutual supervision learning can correct the shortcomings of manual experience and reduce poorly labeled heartbeats. Excellent training templates will further improve the accuracy of waveform detection.

[0119] Non-contact ECG signal reconstruction model:

[0120] Algorithm architecture:

[0121] Fig.13 The structure diagram of the multi-domain generative adversarial network is shown. Specifically, Mixed-GAN includes a generator network G and a temporal discriminator D. 1 and frequency domain discriminator D 2 The input of the generator G is the measured ECG and BCG, and the output is the synthesized X GenECG . Time domain discriminator D 1 The input is X GenECG and X RealECG , frequency domain discriminator D 2 The input is the spectrum amplitude of the real ECG synthesized after Fourier transform, denoted as S GenECG and S RealECG .

[0122] Note: G: Generative network, D 1 : Time domain discrimination network, D 2 : Frequency domain discrimination network. X RealECG : real ECG time series; X GenECG : Synthesized ECG time series; S GenECG : real ECG frequency domain amplitude; S RealECG : Synthesized ECG frequency domain amplitude.

[0123] Generator and Discriminator:

[0124] The generator model is the core part of the whole algorithm, responsible for generating synthetic data similar to real ECG signals from noise. The generator model proposed in this application consists of five fully connected layers. The first four fully connected layers perform batch normalization BN and ReLU activation functions to enhance the nonlinear representation ability of the model. The last layer uses the Tanh function to ensure that the generated ECG signal falls within the range of [-1, 1].

[0125] Loss function:

[0126] In generative adversarial training, based on the consistency of the time domain waveform, the time domain discriminator D is updated1 The objective function is expressed as follows:

[0127]

[0128] In the formula, E(·) represents expectation; D 1 (X GenECG ) represents the time domain discriminator D 1 To generate the output of ECG; D 1 (X RealECG ) means D 1 Output for real ECG.

[0129] Based on frequency domain consistency, update the frequency domain discriminator D 2 The objective function is expressed as follows:

[0130]

[0131] In the formula, E(·) represents expectation; D 2 (S GenECG ) represents the frequency domain discriminator D 2 To generate the output of ECG spectrum amplitude; D 2 (S RealECG ) means D 2 Output of the true ECG spectrum amplitude.

[0132] The loss function for updating the generator G is expressed as follows:

[0133]

[0134] Based on the deep learning network architecture that has learned the cross-domain mapping between cardiac mechanical activity and cardiac electrical activity, the cardiac mechanical activity data extracted at the current moment is input and the output is as follows Fig.14 The ECG measurement result at the current moment is shown, thereby realizing non-contact ECG monitoring. After training with 25 healthy users' experimental data, the method has a Pearson correlation coefficient of 0.963 and a root mean square error of 0.058 on the test set data.

[0135] Expanded applications:

[0136] This application uses the following parameters as an example to describe the expanded application potential of synthetic ECG and measured BCG.

[0137] The following parameters are calculated for the synthetic ECG and measured BCG. The calculation results are as follows: Fig.15 shown.

[0138]

[0139] Compared with the prior art, the intelligent cardiovascular health non-contact monitoring method in the present invention proposes a domain mapping algorithm for a specified one-dimensional mechanical signal of the heart, which can obtain an electrocardiogram based on the one-dimensional mechanical signal of the heart. Based on the electrocardiogram, a fine-grained evaluation of the heart can be achieved, and real-time electrocardiogram data of the heart can be captured without direct contact with the skin, which solves the deficiency that recumbent heart monitoring can only perform coarse-grained heart activity monitoring.

[0140] Another embodiment of the present invention provides an intelligent cardiovascular health non-contact monitoring device, comprising:

[0141] A flexible detection belt is used to collect mechanical motion signals of the chest of a human body to be monitored. The flexible detection belt includes a piezoelectric sensor, a piezoresistive sensor, a dynamic range controllable data acquisition circuit, a feedback control module, an AI autonomous data analysis module, a second data analysis module, a minute data analysis module, a daily analysis module, a data storage module, a WIFI Internet of Things protocol module, a router, a cloud server, a background management, and a user terminal;

[0142] The main processor is used for data processing and comprehensive evaluation according to the intelligent cardiovascular health non-contact monitoring method as described above.

[0143] Compared with the prior art, the advantages of the intelligent cardiovascular health non-contact monitoring device described in the present invention are that the non-contact acquisition of key life characteristics signals is invisible to the patient and does not require connecting objects to the patient; it is small in size, light and thin, can be rolled up, and is easy to carry; it has a simple structure and is easy to install.

[0144] Another embodiment of the present invention provides an intelligent cardiovascular health non-contact monitoring system, including an electronic terminal, on which is provided a WIFI module, which is wirelessly connected to a flexible detection belt, and the electronic terminal stores instructions that can be executed by the main processor, and the instructions are executed by the main processor to implement the intelligent cardiovascular health non-contact monitoring method as described above.

[0145] Heart monitoring system:

[0146] Further, based on the algorithm results of the above method and device, based on Fig.15 The communication architecture shown in the figure submits the result to the terminal shown in the figure. The system also includes Fig.16 The mobile phone application shown and the Wi-Fi module in conjunction with the local gateway can additionally access the integrated management system.

[0147] Compared with the prior art, the advantages of the intelligent cardiovascular health non-contact monitoring system described in the present invention are that it adopts an open design concept, equipment and cloud service, and has flexible expansion capabilities; it adopts the idea of ​​both long and short storage, and has a hardware data backtracking function for delays / missed transmissions caused by various situations: the equipment has a network disconnection backhaul function, which can recover the previous 60 seconds of second data and the previous 30 minutes of minute data; it adopts local daily storage for up to 15 days to eliminate unexpected interference and ensure that data is not lost.

[0148] This application is based on the analysis of the relationship between the two types of movements of the entire heart during the cardiac cycle, and establishes a basic model between two types of one-dimensional representations of the body surface; proposes a device that integrates a multi-module sensor intelligent fusion module; a series of two-sign signal preprocessing methods; and a method for reconstructing ECG signals that integrates mechanism modeling and deep learning methods. The electrocardiogram synthesized based on this method can effectively retain cardiac electrophysiological function information, biological differences, and pathological information; it breaks through the dilemma of the conflict between collection and evaluation of existing cardiovascular mobile assessment equipment, and realizes multi-dimensional non-sensing assessment of cardiac physiological function on a single device; the model is based on non-contact BCG signals, and its detection method is more convenient and safe; at the same time, compared with non-contact devices that collect other modal data, it is less subject to interference and easy to charge.

[0149] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which are equivalent modifications and improvements made to the above embodiments based on the essential technology of the present invention, and all of them belong to the protection scope of the present invention.

Claims

1. An intelligent cardiovascular health non-sensing monitoring method, characterized in that: The following steps are involved: Signal collection: Lay the flexible detection belt under the chest of the person to be monitored, and use the characteristics of the flexible detection belt to produce micro-deformation when under pressure to collect the mechanical movement signal of the chest of the person to be monitored; Signal preprocessing: Use Fourier transform to analyze the frequency distribution of the collected ECG signal and select specific frequencies for filtering; use variational mode decomposition (VMD) technology to decompose and reconstruct the collected BCG signal, and obtain a stable representation of the overall mechanical movement of the heart through pattern matching; Feature acquisition: Use the Pan-Tompkins algorithm to retrieve the R wave in the long-term ECG signal. According to the biological mechanism, use the start and end points of each heart beat ECG as the time calibration to segment the synchronous BCG signal and determine the position of each BCG characteristic wave. Establishing a model: Establishing a biological mechanism and blood circulation feedback model, which includes a myocardial cell excitation-contraction coupling equivalent model, a heart overall composite network model, and an electrocardiogram-ballistocardiogram coupling model. The electrocardiogram-ballistocardiogram coupling model uses a mutual supervision detection and positive feedback correction algorithm based on a three-layer iterative regularization algorithm TSI-DTW to accelerate the convergence speed of the network model and improve the accuracy of waveform detection; Establish a non-contact ECG signal reconstruction model: Establish a multi-domain generative adversarial network MultiGAN to achieve end-to-end reconstruction mapping from BCG to ECG; Heart assessment: The synthetic ECG and measured BCG are combined to evaluate the speed of changes in cardiac contractility, ventricular valve movement, cardiac contractility, and fluctuations in cardiac output CO related to the aortic blood pressure difference.

2. The intelligent cardiovascular health non-sensing monitoring method according to claim 1, characterized in that: In the signal preprocessing step, the collected BCG signal is processed specifically as follows: the BCG signal is decomposed into independent eigenmode functions of different center frequencies using the VMD method, and the expressions of each mode are: u k (t)=A k (t)cos(φ k (t)) t=1,2,3,...,k In the formula, A k (t) represents the envelope, φ k (t) represents the phase. Combined with the Lagrange multiplier method, the VMD method is used to solve the following expression. In the following expression, there is no constrained variational problem. u is determined adaptively. k (t) and center frequency ω k : Where δ(t) is the Dirac function, * is the convolution operator, the penalty term α ensures the accuracy of decomposition and reconstruction in the presence of Gaussian noise, and the Lagrangian penalty term λ(t) can ensure strict constraints on the decomposition process; Using the alternating direction multiplier method ADMM, the optimization problem with multiple variables is transformed into an optimization problem about u k (t), w k And the three optimization sub-problems of λ(t), construct the augmented Lagrangian function according to the constraints of the sub-problems and the objective function, and then solve the optimal solutions of the three optimization sub-problems respectively.

3. The intelligent cardiovascular health non-sensing monitoring method according to claim 2 is characterized in that: The VMD parameters are optimized. The specific optimization steps are as follows: first determine the range of the parameters, then perform VMD decomposition on the original signal and calculate the relative entropy under each parameter combination, iteratively update the parameters and calculate the relative entropy, and finally select the parameters corresponding to the minimum relative entropy as the optimal solution; The cardiac pattern matching score threshold is preset, and the data segments whose distance from the best template matching result LDTW is higher than the threshold are discarded to reduce the influence of body motion interference of continuous data on BCG signals.

4. The intelligent cardiovascular health non-sensing monitoring method according to claim 3 is characterized in that: The optimal influencing parameter combination of VMD is searched, and the penalty parameter and the number of components are selected according to the minimum relative entropy. First, the original signal is decomposed into k narrowband modal functions, which are divided into low-frequency modes and high-frequency modes according to the size of the center frequency. The initial value of the decomposition layer K is 5, the step size is 1, and the range is [3, 20]. The initial value of the penalty parameter a is 2000, the step size is 50, the range is [100, 3000], and the convergence tolerance t0 is 1*.

5. The intelligent cardiovascular health non-sensing monitoring method according to claim 1, characterized in that: The specific method for establishing the myocardial cell excitation-contraction coupling equivalent model is as follows: At the i-th time point in a cardiac cycle, the electrical signal is transmitted sequentially along the excitation conduction system to the n-th myocardial cell, and the cardiac electrical activity e(t i ), mechanical activity m(t i +δt i ) as the observed value, a two-port network is established. At the i-th time point of the cardiac cycle, the s-domain expression of the electrical-mechanical transport signal relationship of the "empty heart" is: In the formula, G n (·) represents the transfer function of the ECC mechanism, represents the absolute refractory period ERP, represents the mechanical activity of the heart, e(s) represents the electrical activity of the heart, ECG n (s) indicates the electrocardiogram data; In the formula, f m (·) represents the mechanical transmission function, Represents force sensor results.

6. The intelligent cardiovascular health non-sensing monitoring method according to claim 1, characterized in that: The specific method for establishing the whole heart composite network model is as follows: At the jth time point in a cardiac cycle (j∈ pumping event), the electrical signal is transmitted sequentially along the excitation conduction system to the kth myocardial cell; when the pumping event occurs, the electromechanical movement of the entire heart is the composite port network T j ; Then at the jth time point of the cardiac cycle, the s-domain expression of the electrical-mechanical motion signal relationship of the entire heart is: In the formula, F k (·) represents the transfer function of the electromechanical motion of the whole heart, represents the absolute refractory period ERP, represents the mechanical activity of the heart, e(s) represents the electrical activity of the heart, ECG k (s) indicates the electrocardiogram data; In summary, the s-domain expression of the electromechanical relationship of the heart as a whole during the entire cardiac cycle is: Where j∈ pumping event, T represents the length of cardiac cycle, G n (·) represents the transfer function of the ECC mechanism, F k (·) represents the transfer function of the electromechanical motion of the whole heart.

7. The intelligent cardiovascular health non-sensing monitoring method according to claim 1, characterized in that: The specific method for establishing the electrocardiogram-ballistocardiogram coupling model is as follows: Taking the whole cycle movement of the heart as the observation object, starting from the sinoatrial node, the myocardial cells that sequentially depolarize and repolarize along the excitation conduction system can be equivalent to the potential changes of different phases of the sinoatrial node electrical activity, that is: (e1(s), e2(s),…, e T (s)) = ECG (t); By introducing the blood circulation feedback system, the relationship between the ballistocardiogram and the electrocardiogram is described; The electrocardiogram-ballistocardiogram transfer function represented by the electrical and mechanical activity of the heart is: In the formula, F BCG (t) is the system output, which is used to describe the mechanical observation signal and its unit is Newton (N); I ECG (t) is the system input, used to describe the electrical observation signal, and the unit is ampere (A); G(s) is the transfer function of excitation-contraction coupling; H(s) is the transfer function of blood circulation; u(t–t0) is the unit step function, indicating the occurrence of blood circulation events starting from time t0; This formula describes the change of BCG over time, including the open-loop response of the myocardial fine two-port network system and the superposition of the positive feedback of the blood circulation at time t0: A mutually supervised BCG detection method is proposed, including feedback correction, detection algorithm constraints and a three-layer iterative time dynamic warping algorithm TSI-DTW. The waveform detection results are used to feed back DTW template training. The mutually supervised learning method of DTW template for feature extraction and labeling can accelerate the convergence speed of the network model. Moreover, the DTW template corrected by mutual supervised learning can correct the shortcomings of manual experience, reduce poorly marked heartbeats, and further improve the accuracy of waveform detection.

8. The intelligent cardiovascular health non-sensing monitoring method according to claim 1, characterized in that: The specific method for establishing the non-contact ECG signal reconstruction model is as follows: The multi-domain generative adversarial network Mixed-GAN includes a generator network G, a time domain discriminator D1, and a frequency domain discriminator D2. The input of the generator G is the measured ECG and BCG, and the output is the synthesized X GenECG , the input of the time domain discriminator D1 is X GenECG and X RealECG The input of the frequency domain discriminator D2 is the spectrum amplitude of the synthesized and real ECG after Fourier transform, denoted as S GenECG and S RealECG ; The generator is used to generate synthetic data similar to real ECG signals from noise; the generator includes five fully connected layers, the first four fully connected layers perform batch normalization BN and ReLU activation function to enhance the nonlinear representation ability of the model; the last fully connected layer uses the Tanh function to ensure that the generated ECG signal falls within the range of [-1, 1]; In generative adversarial training, based on the consistency of the time domain waveform, the objective function of updating the time domain discriminator D1 is expressed as follows: In the formula, E(·) represents the expectation; D1(X GenECG ) represents the output of the time domain discriminator D1 for generating ECG; D1(X RealECG ) represents the output of D1 for the real ECG; Based on frequency domain consistency, the objective function of updating the frequency domain discriminator D2 is expressed as follows: In the formula, E(·) represents the expectation; D2(S GenECG ) represents the output of the frequency domain discriminator D2 for generating the ECG spectrum amplitude; D2(S RealECG ) represents the output of D2 for the real ECG spectrum amplitude; The loss function for updating the generator G is expressed as follows: Based on the deep learning network architecture that has learned the cross-domain mapping between cardiac mechanical activity and cardiac electrical activity, the cardiac mechanical activity data extracted at the current moment is input and the ECG measurement results at the current moment are output, thereby realizing non-contact electrocardiogram monitoring; After machine training on experimental data of 25 healthy users, the Pearson correlation coefficient of this method on the test set data was 0.963 and the root mean square error was 0.

058.

9. An intelligent cardiovascular health non-sensing monitoring device, characterized in that: include: A flexible detection belt is used to collect mechanical motion signals of the chest of a human body to be monitored. The flexible detection belt includes a piezoelectric sensor, a piezoresistive sensor, a dynamic range controllable data acquisition circuit, a feedback control module, an AI autonomous data analysis module, a second data analysis module, a minute data analysis module, a daily analysis module, a data storage module, a WIFI Internet of Things protocol module, a router, a cloud server, a background management, and a user terminal; A general processor is used for performing data processing and comprehensive evaluation according to the intelligent cardiovascular health non-contact monitoring method as described in any one of claims 1-8.

10. An intelligent cardiovascular health non-sensing monitoring system, characterized in that: It comprises an electronic terminal, on which a WIFI module is provided, which is wirelessly connected to a flexible detection belt, and the electronic terminal stores instructions for implementing the intelligent cardiovascular health non-sensing monitoring method according to any one of claims 1 to 8.

Images