Heart rhythm analysis method and device for cardio-pulmonary resuscitation compression period

Abstract

The invention discloses a heart rhythm analysis method and device for a cardiopulmonary resuscitation compression period, and the method specifically comprises the following steps: S1, obtaining historical CPR data, stored in a server, during the cardiopulmonary resuscitation compression period, and carrying out the preprocessing of the historical CPR data, so as to obtain preprocessed data; s2, performing primary heart rhythm analysis on the preprocessed data to obtain a primary analysis result which comprises a first classification result and a second classification result; s3, repeating S1 to S2 to obtain a plurality of analysis results, and adopting a voting mechanism to give a result report of suggested electric shock or non-suggested electric shock; and S4, according to the result report, prompting to carry out electric shock defibrillation or keep external chest compression.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and relates to a method and device for analyzing heart rhythm during cardiopulmonary resuscitation (CPR) compression. Background Art

[0002] In the first aid of out-of-hospital cardiac arrest patients, it is generally necessary to actively perform cardiopulmonary resuscitation (CPR), that is, external chest compression and artificial ventilation, to improve the patient's chance of survival. At the same time, it is necessary to correctly use an AED (Automated External Defibrillator) device for defibrillation shock.

[0003] Currently, almost all actual AED shock advice algorithms (SAA) need to pause external chest compression to collect a clean electrocardiogram for analysis and make a reliable shock advice, or confirm an unreliable shock advice during compression. However, pausing external chest compression will reduce the patient's chance of survival or miss the best first aid time. Therefore, developing a reliable method and device for continuous heart rhythm analysis during external chest compression can significantly reduce the compression pause interval and has great practical application value. Summary of the Invention

[0004] Aiming at the technical problem of too long compression stop time during heart rhythm analysis during cardiopulmonary resuscitation in the prior art, the present invention proposes a method and device for analyzing heart rhythm during cardiopulmonary resuscitation compression.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A method for analyzing heart rhythm during cardiopulmonary resuscitation compression specifically includes the following steps:

[0007] S1: Obtain historical CPR data during cardiopulmonary resuscitation compression stored in the server, and perform preprocessing to obtain preprocessed data;

[0008] S2: Perform a first heart rhythm analysis on the preprocessed data to obtain a first analysis result, and the analysis result includes a first classification result and a second classification result;

[0009] S3: Repeat S1-S2 to obtain several analysis results, and adopt a voting mechanism to give a result report of suggesting shock or not suggesting shock;

[0010] S4: According to the result report, prompt to perform defibrillation shock or maintain external chest compression.

[0011] Preferably, in the step S1, the historical CPR data includes an electrocardiogram signal and a thoracic impedance signal.

[0012] Preferably, the step S2 includes:

[0013] S2-1: Determine whether there is external chest compression according to the thoracic impedance signal in the preprocessed data; if yes, enter S2-2, if not, enter S2-3;

[0014] S2-2: Generate a reference signal according to the thoracic impedance signal in the preprocessed data; extract a feature set from the electrocardiogram signal, input the feature set into a first classifier, and output a first classification result;

[0015] S2-3: Extract a fourth feature set from the preprocessed data and input it into a second classifier to obtain a second classification result.

[0016] Preferably, the step S2-2 includes:

[0017] S2-2-1: Obtain the period information of the thoracic impedance signal in the preprocessed data;

[0018] S2-2-2: Generate a filter reference signal according to the obtained period information;

[0019] S2-2-3: Filter the electrocardiogram signal in the preprocessed data twice respectively to obtain corresponding first-filtered electrocardiogram signal and second-filtered electrocardiogram signal;

[0020] S2-2-4: Decompose the first-filtered electrocardiogram signal to obtain decomposition coefficients, and then reconstruct the decomposition coefficients to obtain a reconstructed electrocardiogram signal;

[0021] S2-2-5: Extract the normalized spectral index and the normalized time index from the decomposition coefficients;

[0022] S2-2-6: Extract a first feature set according to the decomposition coefficients and the reconstructed electrocardiogram signal;

[0023] S2-2-7: Extract a second feature set according to the normalized spectral index and the normalized time index;

[0024] S2-2-8: Extract a third feature set according to the second-filtered electrocardiogram signal;

[0025] S2-2-9: Connect the first feature set, the second feature set, and the third feature set into a feature vector, input it into a first classifier, and obtain a first classification result.

[0026] Preferably, in the step S2-2-2, the reference signal is:

[0027] Φ(t)=[cos(ω i0 t),sin(ω i0t),..., cos(N·ω i0 t), sin(N·ω i0 t)] T (1)

[0028] In formula (1), Φ(t) represents the filter reference signal; ω i0 represents the fundamental angular frequency of the signal segment between two adjacent peak points of the thoracic impedance signal; t is time; N represents the filter order; T represents the transpose matrix.

[0029] Preferably, in S2-2-3, the specific iterative filtering process of the electrocardiogram signal through the filter is as follows:

[0030]

[0031] In formula (2), r e (t) is an intermediate variable, represents the first exponential forgetting factor at time t; Φ(t) represents the reference signal; T represents the transpose matrix; P(t) and P(t - 1) respectively represent the iterative matrices of the filter at time t and t - 1, k(t) represents the filter gain, S ecg1 (t) represents the first filtered electrocardiogram signal at time t; S cor (t) represents the electrocardiogram signal at time t in the preprocessed data; Θ(t - 1) represents the coefficients of the filter at time t - 1; Θ(t) represents the coefficients of the filter at time t; represents the reciprocal of r e (t), e(t) represents the estimated residual of the filter at time t, q 1 represents the first noise level parameter; I 2N represents the identity matrix of order 2N.

[0032] Preferably, in S2-2-4, the first filtered electrocardiogram signal S ecg1 is subjected to 7-layer wavelet decomposition using stationary wavelet transform, and the decomposition coefficients obtained are W f = [D 1 ; D 2 ; D 3 ; D 4 ; D 5 ; D 6 ; D 7 ; A 7 , D 1 represents the first wavelet coefficient, D 2 represents the second wavelet coefficient, D 3 represents the third wavelet coefficient, D 4 represents the fourth wavelet coefficient, D 5 represents the fifth wavelet coefficient, D 6 represents the sixth wavelet coefficient, D7 Represents the seventh wavelet coefficient, A 7 Represents the approximation coefficient.

[0033] After denoising the decomposition coefficients using the soft threshold method, from D 3 , D 4 , D 5 , D 6 , D 7 Perform wavelet reconstruction to obtain the reconstructed electrocardiogram signal S rec .

[0034] Preferably, in the S2-2-5, the method for extracting the normalized spectral index and the normalized time index is:

[0035] E f (a, b) = H(L(a) · W f (a, b)) (3)

[0036]

[0037] In formulas (3), (4), and (5), E f (a, b) represents the enhanced decomposition coefficient, a is the scale factor, b is the translation amount; L and H are measurable functions; W f is the decomposition coefficient; F(a) is the central frequency at scale a; T(b) is the time of translation amount b; NSI(b) represents the normalized spectral index; NTI(a) represents the normalized time index.

[0038] The present invention also provides a heart rhythm analysis device for use during cardiopulmonary resuscitation compression, including a data acquisition module 1, a preprocessing module 2, a heart rhythm analysis module 3, and an output module 4;

[0039] The data acquisition module 1 is used to obtain historical CPR data stored in the server during cardiopulmonary resuscitation compression;

[0040] The preprocessing module 2 is used to preprocess the historical CPR data to obtain preprocessed data;

[0041] The heart rhythm analysis module 3 is used to perform heart rhythm analysis on the preprocessed data to obtain an analysis result, and the analysis result includes a first classification result and a second classification result;

[0042] The output module 4 is used to output an operation prompt according to the analysis result, and the operation prompt includes defibrillation or maintaining external chest compression.

[0043] Preferably, the heart rhythm analysis module 3 includes:

[0044] A judgment unit 301, used to judge whether there is external chest compression according to the chest impedance signal in the preprocessed data;

[0045] A period acquisition unit 302, configured to acquire the period information of the thoracic impedance signal;

[0046] A reference signal generation unit 303, configured to generate a reference signal according to the period information;

[0047] A filter unit 304, configured to filter the electrocardiogram signals in the preprocessed data respectively according to the reference signal to obtain a first filtered electrocardiogram signal and a second filtered electrocardiogram signal;

[0048] A decomposition and reconstruction unit 305, configured to decompose the first filtered electrocardiogram signal to obtain decomposition coefficients, and then reconstruct the decomposition coefficients to obtain a reconstructed electrocardiogram signal;

[0049] A normalized index extraction unit 306, configured to extract a normalized spectral index NSI and a normalized time index NTI from the decomposition coefficients;

[0050] A first feature extraction unit 307, configured to extract a first feature set from the decomposition coefficients and the reconstructed electrocardiogram signal;

[0051] A second feature extraction unit 308, configured to extract a second feature set according to the normalized spectral index NSI and the normalized time index NTI;

[0052] A third feature extraction unit 309, configured to extract a third feature set from the second filtered electrocardiogram signal;

[0053] A first classification unit 310, configured to output a first classification result according to the first feature set, the second feature set, and the third feature set;

[0054] A fourth feature extraction unit 311, configured to extract a fourth feature set according to the preprocessed data;

[0055] A second classification unit 312, configured to output a second classification result according to the fourth feature set.

[0056] In summary, due to the adoption of the above technical solutions, compared with the prior art, the present invention has at least the following beneficial effects:

[0057] The present technical solution uses an advanced extended recursive least squares (ExRLS) filter to remove the artifacts of the CPR data. At the same time, a dual-filter joint analysis technology is designed according to the characteristics of the filtering residuals, and new targeted features are designed according to the characteristics of each filter, which improves the accuracy and robustness of the heart rhythm analysis algorithm, and achieves a balance between the filtering effect and the calculation efficiency, which is beneficial to applying the present technical solution to general AED devices.

[0058] This technical solution uses wavelet analysis technology to decompose the estimated electrocardiogram signal. At the same time, pseudo differential like operators are used to enhance the wavelet coefficients, and then the normalized spectral index (NSI) and the normalized time index (NTI) signals are extracted for feature extraction and analysis, further improving the accuracy of the heart rhythm analysis algorithm.

[0059] By designing a dual-filter joint analysis technology, new targeted features, and a new heart rhythm analysis algorithm structure, this technical solution can significantly reduce the chest compression pause interval and the confirmation rate of unreliable heart rhythm analysis during compression compared with the existing methods. At the same time, this technical solution also has high accuracy and reliability. Therefore, it has obvious advantages compared with the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 FIG. is a schematic diagram of a heart rhythm analysis method during cardiopulmonary resuscitation compression according to an exemplary embodiment of the present invention.

[0061] Figure 2 FIG. is a schematic diagram of the output process of the first classification result according to an exemplary embodiment of the present invention.

[0062] Figure 3 FIG. is a schematic diagram of the output process of the second classification result according to an exemplary embodiment of the present invention.

[0063] Figure 4 FIG. is a schematic diagram of a heart rhythm analysis device during cardiopulmonary resuscitation compression according to an exemplary embodiment of the present invention.

[0064] Figure 5 FIG. is a schematic diagram of the heart rhythm analysis module according to an exemplary embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0065] The present invention will be further described in detail below with reference to the embodiments and specific implementation manners. However, it should not be understood that the scope of the above-mentioned subject matter of the present invention is limited to the following embodiments. Any technology implemented based on the content of the present invention belongs to the scope of the present invention.

[0066] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "longitudinal", "transverse", "up", "down", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0067] Such asFigure 1 As shown, the present invention provides a method for analyzing heart rhythm during cardiopulmonary resuscitation (CPR) compression, specifically including the following steps:

[0068] S1: Obtain the historical CPR data during cardiopulmonary resuscitation compression stored in the server, and perform preprocessing to obtain preprocessed data.

[0069] In this embodiment, the CPR data includes electrocardiogram (ECG) signals and thoracic impedance signals (TTI).

[0070] In this embodiment, the preprocessing at least includes denoising. For example, perform band-pass filtering on the electrocardiogram signal with a frequency band of 0.7 - 30 Hz, and perform band-pass filtering on the thoracic impedance signal with a frequency band of 0.7 - 10 Hz. The filtering frequency band and filtering method can also be selected according to the actual situation, and no specific limitation is made in this embodiment.

[0071] S2: Perform a first heart rhythm analysis on the preprocessed data to obtain a first analysis result, and the analysis result includes a first classification result and a second classification result.

[0072] In this embodiment, a continuous heart rhythm analysis algorithm is used to reliably analyze the electrocardiogram signal containing CPR artifacts, and suggestions for shock or no shock are made according to the analysis result. The specific steps include:

[0073] S2-1: Determine whether there is external chest compression according to the thoracic impedance signal (TTI) in the preprocessed data; if yes, enter S2-2, if not, enter S2-3.

[0074] In this embodiment, it is determined by setting a thoracic impedance compression detection threshold. If the thoracic impedance signal is greater than or equal to the thoracic impedance compression detection threshold, it is determined that there is external chest compression. If the thoracic impedance signal is less than the thoracic impedance compression detection threshold, it is determined that there is no external chest compression.

[0075] S2-2: Generate a reference signal according to the thoracic impedance signal (TTI) in the preprocessed data; and extract a feature set from the electrocardiogram signal, input the feature set into the constructed first classifier, and output the first classification result, as Figure 2 shown.

[0076] S2-2-1: Obtain the period information of the thoracic impedance signal in the preprocessed data.

[0077] In this embodiment, the period information of the thoracic impedance signal is detected using the peak points of the signal: Let the thoracic impedance signal sequence be TTI t , t = 1, 2,..., L; the peak points are p i , i = 1, 2,..., k; where t is time, L is the signal length, and k is the number of peak points; then the period of the thoracic impedance signal is p i+1 -p i。

[0078] In this embodiment, the peak point detection is performed by using the findpeaks method of matlab, and other methods can also be used according to the actual situation. This embodiment does not make specific limitations.

[0079] S2-2-2: Generate a filter reference signal according to the obtained period information.

[0080] In this embodiment, let the sampling frequency of the filter be fs, the reference signal be Φ(t), which is modeled as the Fourier series of the fundamental angular frequency of each press. Let the fundamental angular frequency of the signal segment between two adjacent peak points of the thoracic impedance signal be ω i0 ,then Also, let the filter order be N, then the reference signal is modeled as:

[0081] Φ(t) = [cos(ω i0 t), sin(ω i0 t),..., cos(N·ω i0 t), sin(N·ω i0 t)] T (1)

[0082] In formula (1), Φ(t) represents the filter reference signal; ω i0 represents the fundamental angular frequency of the signal segment between two adjacent peak points of the thoracic impedance signal; t is time; N represents the filter order; T represents the transpose matrix.

[0083] S2-2-3: Use two filters to filter the electrocardiogram signal in the preprocessed data twice respectively to obtain the corresponding first filtered electrocardiogram signal and second filtered electrocardiogram signal.

[0084] In this embodiment, two extended recursive least squares (ExRLS) filters are used to filter the CPR artifacts of the electrocardiogram signal in the preprocessed data, and the first filtered electrocardiogram signal and the second filtered electrocardiogram signal are obtained respectively. It should be noted that other adaptive filters such as the least mean square algorithm (LMS), recursive least squares algorithm (RLS), and Kalman filter can also be selected for the filter. This embodiment does not make specific limitations.

[0085] In this embodiment, let the first filter and the second filter be ExRLS1 and ExRLS2 respectively, then the filter coefficients are:

[0086] Θ(t) = [a 1 (t), b 1 (t),..., a N (t), b N (t)] T ; t = 1, 2,..., L (2)

[0087] In formula (2), Θ(t) represents the coefficients of the filter at time t; a N (t) and b N (t) respectively represent the filter coefficient components corresponding to the reference signal components.

[0088] Let the electrocardiogram signal (including artifacts) in the preprocessed data be S cor , and the first filtered electrocardiogram signal obtained after S cor passes through the first filter ExRLS1 is S ecg1 , the first exponential forgetting factor is β 1 , the first noise level parameter is q 1 , and the first filter is initialized as Θ(0) = 0, P(0) = I 2N , where I 2N is the identity matrix of order 2N, and P(0) represents the initial iteration matrix of the filter.

[0089] For t = 1, 2,..., L, the specific iterative filtering process of the electrocardiogram signal in the preprocessed data through the first filter is as follows:

[0090]

[0091] In formula (3), r e (t) is an intermediate variable, represents the first exponential forgetting factor at time t; Φ(t) represents the reference signal; T represents the transpose matrix; P(t) and P(t - 1) respectively represent the iteration matrices of the filter at time t and t - 1, k(t0 represents the filter gain, S ecg1 (t) represents the first filtered electrocardiogram signal at time t; S cor (t) represents the electrocardiogram signal at time t in the preprocessed data; Θ(t - 1) represents the coefficients of the filter at time t - 1; Θ(t) represents the coefficients of the filter at time t; represents the reciprocal of r e (t), e(t) represents the estimated residual of the filter at time t, q 1 represents the first noise level parameter; I 2N represents the identity matrix of order 2N.

[0092] Similarly, the filtering process of the second filter ExRLS2 is the same as above, and the second filtered electrocardiogram signal output by the second filter ExRLS2 is S ecg2 , the second exponential forgetting factor is β 2 , and the second noise level parameter is q 2 .

[0093] S2 - 2 - 4: For the first filtered electrocardiogram signal Secg1 Decompose to obtain decomposition coefficients, and then reconstruct the decomposition coefficients to obtain the reconstructed electrocardiogram signal.

[0094] In this embodiment, the stationary wavelet transform (SWT) is used to perform 7-layer wavelet decomposition on the first filtered electrocardiogram signal S ecg1 to obtain the decomposition coefficients W f = [D 1 ; D 2 ; D 3 ; D 4 ; D 5 ; D 6 ; D 7 ; A 7 , where D 1 represents the first wavelet coefficient, D 2 represents the second wavelet coefficient, D 3 represents the third wavelet coefficient, D 4 represents the fourth wavelet coefficient, D 5 represents the fifth wavelet coefficient, D 6 represents the sixth wavelet coefficient, D 7 represents the seventh wavelet coefficient, and A 7 represents the approximation coefficient.

[0095] In this embodiment, after denoising the decomposition coefficients using the soft threshold method, the reconstructed electrocardiogram signal S 3 is obtained by reconstructing D 4 , D 5 , D 6 , D 7 . D1 and D2 are regarded as noise and do not participate in the reconstruction. rec .

[0096] In this embodiment, other decomposition methods can also be used to decompose the first filtered electrocardiogram signal S ecg1 , such as the continuous wavelet transform (CWT), variational mode decomposition (VMD), etc., which are not specifically limited in this embodiment.

[0097] S2-2-5: Enhance the decomposition coefficients using a pseudo-differential operator-like method, and extract the normalized spectrum index NSI (Normalized Spectrum Index) and the normalized time index NTI (Normalized Time Index). Specifically as follows:

[0098] Let the enhanced decomposition coefficient be E f , where L and H are measurable functions, a is the scale factor, b is the translation amount, F(a) is the center frequency corresponding to the scale, and T(b) is the time corresponding to the translation amount. Then the processing is as follows:

[0099] E f(a, b) = H(L(a)·W f (a, b)) (8)

[0100]

[0101] In formulas (8), (9), and (10), E f (a, b) represents the enhanced decomposition coefficient, a is the scale factor, b is the translation amount; L and H are measurable functions; W f is the decomposition coefficient; F(a) is the central frequency at scale a; T(b) is the time corresponding to the translation amount b; NSI(b) represents the normalized spectral index; NTI(a) represents the normalized time index.

[0102] In this embodiment, L and H are specifically taken as L(a) = 1, H(·) = |·| 1 / 2 . It should be noted that L and H can also be selected as other measurable functions according to the actual situation, and this embodiment does not make specific restrictions.

[0103] S2-2-6: Extract the first feature set according to the decomposition coefficient and the reconstructed electrocardiogram signal S rec Extract the first feature set.

[0104] In this embodiment, according to D in the decomposition coefficient 3 , D 4 , D 5 , D 6 , D 7 and the reconstructed electrocardiogram signal S rec 8 features are extracted, specifically:

[0105] The features extracted according to the reconstructed electrocardiogram signal include: VF filter leakage VFleak, slope feature bCP, amplitude feature bWT, center frequency of the power spectral density in the range of 4 - 30 Hz hPSDfc;

[0106] The features extracted according to the detail coefficients include: the first quartile FQR(D7) of D7, the kurtosis Kurt(D7) of D7, the impulse factor ImpF(D4) of D4, and the Hjorth complexity index Hcmp(D4) of D4.

[0107] In this embodiment, let the duration of the signal segment for which features are to be extracted be L s seconds, with a value of x i , i = 1, 2,..., N (the same applies to the following feature calculations unless otherwise specified), then the specific features in the first feature set are:

[0108]

[0109] In formula (11), VFleak represents the VF filter leakage; x irepresents the i-th value; x i-1 represents the (i - 1)-th value; N represents the total number of values; T represents the average period of the signal;

[0110]

[0111] In formula (12), bCP represents the slope feature; t() is the time length that satisfies the condition; is the square of the signal slope; th s is the given threshold; L s represents the duration of the signal segment from which the feature is to be extracted;

[0112]

[0113] In formula (13), bWT represents the amplitude feature; the signal first passes through a band - pass filter of 6.5 - 30 Hz, is the value of the given upper percentile point, is the value of the given lower percentile point;

[0114]

[0115] In formula (14), hPSDfc represents the center frequency of the power spectral density from 4 - 30 Hz; hp i is the power spectral density of the signal from 4 - 30 Hz; hf i is the corresponding frequency;

[0116] FQR(D7) = prctile(x,0.25) (15)

[0117] In formula (15), FQR(D7) represents the first quartile of the detail coefficient D7; prctile() is the function for calculating the percentile;

[0118]

[0119] In formula (16), Kurt(D7) represents the kurtosis of D7, E() is the function for calculating the mathematical expectation, μ is the signal mean, and σ is the signal standard deviation;

[0120]

[0121] In formula (17), ImpF(D4) represents the impulse factor of the detail coefficient D4; max() is the function for calculating the maximum value;

[0122]

[0123] In formula (18), Hcmp(D4) represents the Hjorth complexity index of the detail coefficient D4; σ 0is the standard deviation of the signal, σ 1 is the standard deviation of the first-order difference of the signal, σ 2 is the standard deviation of the second-order difference of the signal.

[0124] It should be noted that some of the sub-feature sets in the first feature set can also be replaced by other feature sets, and this embodiment does not make specific restrictions.

[0125] S2-2-7: Extract the second feature set according to the normalized spectral index NSI and the normalized time index NTI.

[0126] In this embodiment, 2 features are extracted according to the normalized spectral index NSI and the normalized time index NTI, which are respectively: the signal mean mNSI of the normalized spectral index, and the signal mean mNTI of the normalized time index. It should be noted that some of the sub-feature sets in the second feature set can also be replaced by other feature sets, and this embodiment does not make specific restrictions.

[0127] The features are specifically:

[0128]

[0129] In formula (19), mNSI represents the signal mean of the normalized spectral index; NSI i is the normalized spectral index; mNTI represents the signal mean of the normalized time index; NTI i is the normalized time index.

[0130] S2-2-8: Extract the third feature set according to the second filtered electrocardiogram signal S ecg2 Extract the third feature set.

[0131] In this embodiment, 3 features are extracted according to the second filtered electrocardiogram signal S ecg2 which are respectively: the mean lPSDm of the power spectral density of 1-4Hz, the standard deviation lPSDs, and the center frequency lPSDfc. It should be noted that some of the sub-feature sets in the third feature set can also be replaced by other feature sets, and this embodiment does not make specific restrictions.

[0132] The features are specifically:

[0133]

[0134] In formula (20), lPSDm represents the mean of the power spectral density of 1-4Hz; lp i represents the power spectral density of the signal at 1-4Hz;

[0135]

[0136] In formula (21), lPSDs represents the standard deviation; lp iis the power spectral density of 1 - 4Hz, μ lp is the mean value of the power spectral density;

[0137]

[0138] In formula (22), lPSDfc represents the center frequency; lp i is the power spectral density of 1 - 4Hz, lf i is the corresponding frequency.

[0139] S2 - 2 - 9: Connect the first feature set, the second feature set, and the third feature set into a feature vector, and send it to the first classifier for classification to obtain the first classification result.

[0140] In this embodiment, the first classifier is trained and classified using a support vector machine (SVM). The first classification result includes two categories: recommended shock and non - recommended shock. Other machine learning algorithms can also be used as classifiers, such as random forest, logistic regression, etc. This embodiment does not make specific limitations.

[0141] S2 - 3: Extract the fourth feature set from the pre - processed data and input it into the second classifier to obtain the second classification result, as Figure 3 shown.

[0142] S2 - 3 - 1: Extract the fourth feature set from the pre - processed data.

[0143] In this embodiment, 7 features are extracted from the pre - processed data, which are: VF filter leakage VFleak, 4 - 8Hz spectral energy X4, root mean square of absolute amplitude StdAbs, slope feature bCP, center frequency of 1 - 4Hz power spectral density lPSDfc, center frequency of 4 - 30Hz power spectral density hPSDfc, phase space reconstruction PSR.

[0144] It should be noted that some sub - feature sets in the fourth feature set can also be replaced by other feature sets. This embodiment does not make specific limitations.

[0145] Specifically:

[0146] The calculations of VFleak, bCP, lPSDfc, and hPSDfc are the same as above;

[0147]

[0148] In formula (23), X4 represents the 4 - 8Hz spectral energy; A i represents the square of the amplitude of the 4 - 8Hz signal Fourier transform;

[0149]

[0150] In formula (24), StdAbs represents the mean square deviation of the absolute amplitude; μ is the mean value of the absolute value of the signal;

[0151]

[0152] In formula (25), PSR represents phase space reconstruction; the signal is first normalized to the interval (0, 40), and then the phase space diagram (x(t), x(t + τ)) is drawn, where τ = 0.5 s is the delay time; N τ is the number of grids passed by the phase point trajectory in the phase space diagram.

[0153] S2-3-2: Send the fourth feature set into the second classifier for classification to obtain the second classification result.

[0154] In this embodiment, the second classifier is trained and classified using a support vector machine (SVM), and the second classification result includes two categories: recommended shock and non-recommended shock. Other machine learning algorithms can also be used as classifiers, such as random forest, logistic regression, etc., which are not specifically limited in this embodiment.

[0155] S3: Repeat S1 - S2 to obtain several analysis results, and use a voting mechanism to give a result report of recommended shock or non-recommended shock.

[0156] In this embodiment, the analysis is repeated three times to obtain three analysis results. When all three analysis results are recommended shock, a result report of recommended shock is given, and in other cases, a result report of non-recommended shock is given. It should be noted that the number of analyses and the conditions for giving recommendations can also be determined by comprehensive consideration according to the actual situation, which are not specifically limited in this embodiment.

[0157] S4: According to the result report, prompt for defibrillation by electric shock or maintain external chest compression.

[0158] If the result report is recommended shock, prompt and perform defibrillation by electric shock after the end of this compression. If the result report is non-recommended shock, prompt to maintain external chest compression after the end of this compression to reduce unnecessary compression interruptions.

[0159] As Figure 4 shown, based on the above-mentioned method for analyzing heart rhythm during cardiopulmonary resuscitation compression, the present invention also provides a device for analyzing heart rhythm during cardiopulmonary resuscitation compression, including a data acquisition module 1, a preprocessing module 2, a heart rhythm analysis module 3, and an output module 4; the output end of the data acquisition module 1 is connected to the input end of the preprocessing module 2, the output end of the preprocessing module 2 is connected to the input end of the heart rhythm analysis module 3, and the output end of the heart rhythm analysis module 3 is connected to the input end of the output module 4.

[0160] A data acquisition module 1 for obtaining historical CPR data during cardiopulmonary resuscitation compression stored from a server;

[0161] A preprocessing module 2 for preprocessing the historical CPR data to obtain preprocessed data;

[0162] A heart rhythm analysis module 3 for performing heart rhythm analysis on the preprocessed data to obtain analysis results, where the analysis results include a first classification result and a second classification result; the first classification result includes recommended defibrillation and non-recommended defibrillation; the second classification result includes recommended defibrillation and non-recommended defibrillation; by synthesizing several analysis results, a result report of recommended defibrillation or non-recommended defibrillation is given.

[0163] An output module 4 for outputting an operation prompt according to the result report, where the operation prompt includes defibrillation or maintaining external chest compression.

[0164] In this embodiment, as Figure 5 shown, the heart rhythm analysis module 3 includes a judgment unit 301, a period acquisition unit 302, a reference signal generation unit 303, a filter unit 304, a decomposition and reconstruction unit 305, a normalization index extraction unit 306, a first feature extraction unit 307, a second feature extraction unit 308, a third feature extraction unit 309, a first classification unit 310, a fourth feature extraction unit 311, and a second classification unit 312;

[0165] The output end of the judgment unit 301 is respectively connected to the input ends of the period acquisition unit 302 and the fourth feature extraction unit 311, and the output end of the fourth feature extraction unit 311 is connected to the input end of the second classification unit 312; the output end of the period acquisition unit 302 is connected to the input end of the reference signal generation unit 303, the output end of the reference signal generation unit 303 is connected to the input end of the filter unit 304, the output end of the filter unit 304 is respectively connected to the input ends of the decomposition and reconstruction unit 305 and the third feature extraction unit 309, the output end of the decomposition and reconstruction unit 305 is respectively connected to the input ends of the first feature extraction unit 307 and the normalization index extraction unit 306, the output end of the normalization index extraction unit 306 is connected to the input end of the second feature extraction unit 308, and the output ends of the first feature extraction unit 307, the second feature extraction unit 308, and the third feature extraction unit 309 are connected and then connected to the input end of the first classification unit 310.

[0166] Among them,

[0167] The judgment unit 301 is used to judge whether there is external chest compression according to the thoracic impedance signal (TTI) in the preprocessed data;

[0168] A period acquisition unit 302, configured to acquire the period information of the thoracic impedance signal;

[0169] A reference signal generation unit 303, configured to generate a reference signal according to the period information;

[0170] A filter unit 304, including a first filtering subunit and a second filtering subunit, configured to filter the electrocardiogram signal in the preprocessed data respectively according to the reference signal to obtain a first filtered electrocardiogram signal and a second filtered electrocardiogram signal;

[0171] A decomposition and reconstruction unit 305, configured to decompose the first filtered electrocardiogram signal to obtain decomposition coefficients, and then reconstruct the decomposition coefficients to obtain a reconstructed electrocardiogram signal;

[0172] A normalization index extraction unit 306, configured to extract a normalized spectral index NSI and a normalized time index NTI from the decomposition coefficients;

[0173] A first feature extraction unit 307, configured to extract a first feature set from the decomposition coefficients and the reconstructed electrocardiogram signal S rec Extract the first feature set;

[0174] A second feature extraction unit 308, configured to extract a second feature set according to the normalized spectral index NSI and the normalized time index NTI;

[0175] A third feature extraction unit 309, configured to extract a third feature set from the second filtered electrocardiogram signal;

[0176] A first classification unit 310, configured to output a first classification result according to the first feature set, the second feature set, and the third feature set;

[0177] A fourth feature extraction unit 311, configured to extract a fourth feature set from the preprocessed data;

[0178] A second classification unit 312, configured to output a second classification result according to the fourth feature set.

[0179] The significance of the present invention lies in that, aiming at the problem that most current AEDs do not have continuous and reliable heart rhythm analysis during CPR compression, a new method for reliable continuous heart rhythm analysis during CPR compression is proposed. This method has high analysis accuracy and reliability, and is simple to implement. It can significantly reduce the interruption frequency and time of external chest compressions during CPR, reduce unnecessary potential heart rhythm confirmation, increase the survival chance of cardiac arrest patients, and contribute to the implementation of high-quality first aid. At the same time, compared with existing similar methods, this method has the advantages of high analysis accuracy and simple implementation, and can be easily applied to existing general AED devices.

[0180] Those of ordinary skill in the art can understand that the above embodiments are specific examples for implementing the present invention, and in actual applications, various changes can be made to them in form and details without departing from the spirit and scope of the present invention.

Claims

1. A method for analyzing heart rhythm during cardiopulmonary resuscitation, characterized in that: The specific steps include: S1: Acquire historical CPR data during cardiopulmonary resuscitation stored in the server, and perform preprocessing to obtain preprocessed data; S2: Perform a heart rhythm analysis on the preprocessed data to obtain an analysis result, which includes a first classification result and a second classification result; S3: Repeat S1-S2 to obtain several analysis results, and use a voting mechanism to give a report of whether to recommend electric shock or not; S4: Based on the result report, either perform electric shock defibrillation or maintain chest compressions.

2. A method for analyzing heart rhythm during cardiopulmonary resuscitation according to claim 1, characterized in that: In S1, the historical CPR data includes an electrocardiogram signal and a chest impedance signal.

3. A method for analyzing heart rhythm during cardiopulmonary resuscitation according to claim 1, characterized in that: The S2 includes: S2-1: Determine whether chest compression is performed based on the chest impedance signal in the preprocessed data; if yes, proceed to S2-2; otherwise, proceed to S2-3; S2-2: Generate a reference signal according to the chest impedance signal in the preprocessed data; and extract a feature set from the electrocardiogram signal, input the feature set into a first classifier, and output a first classification result; S2-3: Extracting a fourth feature set from the preprocessed data and inputting it into a second classifier to obtain a second classification result.

4. A method for analyzing heart rhythm during cardiopulmonary resuscitation according to claim 1, characterized in that: The S2-2 includes: S2-2-1: Obtaining period information of the chest impedance signal in the preprocessed data; S2-2-2: Generate a filter reference signal according to the acquired period information; S2-2-3: filtering the ECG signal in the preprocessed data twice to obtain a corresponding first filtered ECG signal and a second filtered ECG signal; S2-2-4: Decomposing the first filtered ECG signal to obtain decomposition coefficients, and then reconstructing the decomposition coefficients to obtain a reconstructed ECG signal; S2-2-5: extracting normalized spectral index and normalized time index from the decomposition coefficient; S2-2-6: extracting a first feature set according to the decomposition coefficients and the reconstructed ECG signal; S2-2-7: extracting a second feature set according to the normalized spectral index and the normalized time index; S2-2-8: extracting a third feature set according to the second filtered ECG signal; S2-2-9: Connect the first feature set, the second feature set, and the third feature set into a feature vector, input the vector into the first classifier, and obtain a first classification result.

5. A method for analyzing heart rhythm during cardiopulmonary resuscitation according to claim 4, characterized in that: In S2-2-2, the reference signal is: Φ(t)=[cos(ω i0 t),sin(ω i0 t),...,cos(N·ω i0 t),sin(N·ω i0 (t)] T (1) In formula (1), Φ(t) represents the filter reference signal; ω i0 It represents the basic angular frequency of the signal segment between two adjacent peak points of the chest impedance signal; t is time; N represents the filter order; T represents the transposed matrix.

6. A method for analyzing heart rhythm during cardiopulmonary resuscitation according to claim 4, characterized in that: In S2-2-3, the specific iterative filtering process of the ECG signal through the filter is as follows: In formula (2), r e (t) is the intermediate variable, represents the first exponential forgetting factor at time t; Φ(t) represents the reference signal; T represents the transposed matrix; P(t) and P(t-1) represent the iterative matrix of the filter at time t and time t-1 respectively, k(t) represents the filter gain, S ecg1 (t) represents the first filtered ECG signal at time t; S cor (t) represents the ECG signal at time t in the preprocessed data; Θ(t-1) represents the coefficient of the filter at time t-1; Θ(t) represents the coefficient of the filter at time t; Represents r e The inverse of (t), e(t) represents the estimated residual of the filter at time t, and q1 represents the first noise level parameter; 2N represents the identity matrix of rank 2N.

7. A method for analyzing heart rhythm during cardiopulmonary resuscitation according to claim 4, characterized in that: In S2-2-4, the first filtered ECG signal S is transformed by stationary wavelet transform. ecg1 Perform 7-layer wavelet decomposition, and the decomposition coefficient is W f =[D1; D2; D3; D4; D5; D6; D7; A7], D1 represents the first wavelet coefficient, D2 represents the second wavelet coefficient, D3 represents the third wavelet coefficient, D4 represents the fourth wavelet coefficient, D5 represents the fifth wavelet coefficient, D6 represents the sixth wavelet coefficient, D7 represents the seventh wavelet coefficient, and A7 represents the approximate coefficient. After the decomposition coefficients are denoised using the soft threshold method, the reconstructed ECG signal S is obtained by wavelet reconstruction using D3, D4, D5, D6, and D7. rec .

8. A method for analyzing heart rhythm during cardiopulmonary resuscitation according to claim 4, characterized in that: In S2-2-5, the method for extracting the normalized spectral index and the normalized time index is: E f (a,b)=H(L(a)·W f (a,b)) (3) In formulas (3), (4) and (5), E f (a,b) represents the enhanced decomposition coefficient, a is the scale factor, b is the translation; L and H are measurable functions; W f is the decomposition coefficient; F(a) is the center frequency of scale a; T(b) is the time of translation b; NSI(b) represents the normalized spectral index; NTI(a) represents the normalized time index.

9. A cardiac rhythm analysis device for use during cardiopulmonary resuscitation compression based on the method according to any one of claims 1 to 8, characterized in that: It comprises a data acquisition module (1), a preprocessing module (2), a heart rhythm analysis module (3) and an output module (4); A data acquisition module (1) is used to acquire historical CPR data stored during cardiopulmonary resuscitation from a server; A preprocessing module (2), used for preprocessing the historical CPR data to obtain preprocessed data; A heart rhythm analysis module (3), used for performing heart rhythm analysis on the pre-processed data to obtain analysis results, wherein the analysis results include a first classification result and a second classification result; The output module (4) is used to output an operation prompt according to the analysis result, the operation prompt including electric shock defibrillation or maintaining chest compression.

10. A cardiac rhythm analysis device for use during cardiopulmonary resuscitation compression according to claim 9, characterized in that: The heart rhythm analysis module (3) comprises: A judgment unit (301), used for judging whether chest compression exists according to the chest impedance signal in the pre-processed data; A cycle acquisition unit (302), used for acquiring cycle information of a chest impedance signal; A reference signal generating unit (303), configured to generate a reference signal according to the period information; A filter unit (304) is used to filter the electrocardiogram signals in the preprocessed data according to the reference signal to obtain a first filtered electrocardiogram signal and a second filtered electrocardiogram signal; A decomposition and reconstruction unit (305), used for decomposing the first filtered ECG signal to obtain decomposition coefficients, and then reconstructing the decomposition coefficients to obtain a reconstructed ECG signal; A normalized index extraction unit (306), used for extracting a normalized spectrum index NSI and a normalized time index NTI from the decomposition coefficient; A first feature extraction unit (307), used for decomposing coefficients and reconstructing the electrocardiographic signal to extract a first feature set; A second feature extraction unit (308), configured to extract a second feature set according to the normalized spectral index NSI and the normalized time index NTI; A third feature extraction unit (309), used for extracting a third feature set from the second filtered electrocardiogram signal; A first classification unit (310), configured to output a first classification result according to the first feature set, the second feature set, and the third feature set; A fourth feature extraction unit (311), used for extracting a fourth feature set based on the preprocessed data; The second classification unit (312) is used to output a second classification result according to the fourth feature set.

Images