Dynamic electrocardiogram rapid pre-judgment and analysis system and method thereof

Abstract

The invention provides a dynamic electrocardiogram rapid pre-judgment analysis system and method, and the system comprises a cloud platform which is used for storing the electrocardiogram data of a patient, and the electrocardiogram data comprises a dynamic electrocardiogram; and the pre-judgment analysis module is used for pre-judging and analyzing the dynamic electrocardiogram stored in the cloud platform. The method comprises the following steps: preprocessing a dynamic electrocardiogram through a pre-judgment analysis module to obtain a heart beat signal, judging whether the heart beat signal is abnormal or not through an analysis model, detecting an abnormal heart beat signal, standardizing the abnormal heart beat signal, extracting time domain characteristics, frequency domain characteristics and morphological characteristics from the abnormal heart beat signal and the abnormal standardized signal, and analyzing the time domain characteristics, the frequency domain characteristics and the morphological characteristics. According to the method and the device, the combined features are obtained through integration, and the abnormal features of the combined features are rapidly detected according to the comparison of the combined features and the features of normal signals, so that the feature points of the abnormal electrocardiogram of the patient are accurately obtained, and the problem of inaccurate judgment in the automatic detection and analysis process of the dynamic electrocardiogram is avoided.

Description

Technical Field

[0001] The present invention relates to the technical field of ambulatory electrocardiogram analysis, and more specifically, to a rapid pre-judgment analysis system and method for ambulatory electrocardiogram. Background Art

[0002] Electrocardiogram examination mainly reflects the electrical activity of cardiac excitation. Therefore, it has a definite value for the diagnosis and analysis of various arrhythmias and conduction blocks. Conventional electrocardiogram examinations can only detect the electrocardiogram data of the patient's current state. For occasional palpitations, or arrhythmias caused by factors such as electrolyte disorders, surgery, drug treatment, infection, acid-base imbalance, pregnancy, etc. to varying degrees, it is not easy to detect with a resting electrocardiograph or ordinary ambulatory electrocardiogram. Or, for arrhythmias that have been detected by conventional examinations but whose nature or potential risks are not yet fully clear, it is also not easy to detect. To solve this problem, portable wearable electrocardiogram monitoring devices can continuously monitor the electrocardiogram data of patients and can effectively detect occasional arrhythmias.

[0003] The portable wearable electrocardiogram monitoring device uploads the continuously monitored electrocardiogram data to the cloud platform for storage. It is necessary to analyze the ambulatory electrocardiogram stored in the cloud platform to determine whether the ambulatory electrocardiogram is abnormal. The ambulatory electrocardiogram continuously records the entire process of the patient's electrocardiogram activity in the daily life state of the patient, and the amount of information it contains is much larger than that of a conventional electrocardiogram. In addition, due to the diversity and complexity of human physiological signals, there are problems of inaccurate judgment in the process of automatically detecting and analyzing the ambulatory electrocardiogram. Therefore, the present invention provides a rapid pre-judgment analysis system and method for ambulatory electrocardiogram to solve the above problems. Summary of the Invention

[0004] The purpose of the present invention is to provide a rapid pre-judgment analysis system and method for ambulatory electrocardiogram to solve the problems raised in the above background art.

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

[0006] A rapid pre-judgment analysis system for ambulatory electrocardiogram, comprising:

[0007] A cloud platform for storing the electrocardiogram data of patients, where the electrocardiogram data includes ambulatory electrocardiograms;

[0008] A pre-judgment analysis module for pre-judging and analyzing the ambulatory electrocardiograms stored in the cloud platform, determining whether the ambulatory electrocardiograms are abnormal, and analyzing the abnormal characteristics of the detected ambulatory electrocardiograms to output a result report. The pre-judgment analysis module is arranged in a terminal device, and the ambulatory electrocardiogram data in the cloud platform is imported into the pre-judgment analysis module of the terminal device for analysis;

[0009] A workstation for viewing or printing the result report of dynamic electrocardiogram detection. After the pre-judgment and analysis module outputs the result report, it is uploaded to the workstation for viewing or printing.

[0010] A preferred technical solution of the present application: The dynamic electrocardiogram analysis process of the pre-judgment and analysis module includes data preprocessing, input into the analysis model, abnormality judgment, feature extraction, and detection of abnormal features.

[0011] A preferred technical solution of the present application: The process of the data preprocessing is as follows:

[0012] Let the original dynamic electrocardiogram signal be x(t). The process of removing baseline drift and high-frequency noise from the original dynamic electrocardiogram signal is as follows:

[0013] High-pass filtering to remove baseline drift:

[0014] x[n] = x(t = nT s )

[0015] y f [n] = x[n] - x[n - 1] + β·y f [n - 1]

[0016] Where x(t) represents the original continuous-time dynamic electrocardiogram signal, x[n] represents the discrete-time signal obtained by sampling the continuous-time signal x(t), where n is the discrete-time index, T s is the sampling interval in seconds, T s represents the time interval between two consecutive sampling points, nT s represents the time corresponding to the nth sampling point, x[n] is the sampling value of x(t) at the discrete-time point t = nT s , β is the filtering coefficient, controlling the attenuation characteristic of the filter, y f [n] is the signal after removing the filtering;

[0017] Discrete wavelet transform to remove high-frequency noise:

[0018]

[0019] Where ψ j,k [n] is the discrete wavelet basis function, N is the length of the signal, that is, n = [0, 1, 2,... N - 1], j is the scale parameter, k is the translation parameter, W[j, k] is the wavelet coefficient, and threshold processing is performed according to the detail coefficients in the wavelet coefficient W[j, k] to remove the high-frequency noise in the filtered signal y f [n], and the signal y d [n] after removing the noise is obtained;

[0020] For the signal y after removing the noised [n]Segment the signal, where the signal segmentation includes R-wave detection and heartbeat segmentation;

[0021] The process of the R-wave detection is as follows:

[0022] Calculate the derivative signal d[n]:

[0023]

[0024] Calculate the squared signal s[n]:

[0025] s[n] = (d[n]) 2

[0026] Calculate the sliding window integral signal I[n]:

[0027]

[0028] where M is the window size in the sliding window integral and m is the summation variable in the sliding window integral;

[0029] Detect the R-wave peak and record the R-wave position r i , r i represents the time position of the i-th R-wave, and determine the start and end positions of each heartbeat through r i ;

[0030] The process of the heartbeat segmentation is as follows:

[0031] With the R-wave position r i as the center, intercept the signal segment of each heartbeat to obtain the heartbeat signal z i [n] as follows:

[0032] z i [n] = y d [r i -Δt:r i +Δt]

[0033] where z i [n] is the signal segment of the i-th heartbeat, Δt is the time offset, and the length of each heartbeat signal segment is L, where

[0034] A preferred technical solution of this application: The process of the input analysis model is as follows:

[0035] Construct an analysis model, which includes a dynamic depth convolutional layer, an adaptive dilation convolutional layer, and a fully connected layer;

[0036] The expression of the dynamic depth convolutional layer is as follows:

[0037]

[0038] Among them, V c [p c , q c is the convolution output of the dynamic depth convolution layer, K c is the size of the convolution kernel in the dynamic depth convolution, p c represents the height index of the output feature map, q c represents the width index of the output feature map, j c represents the width index of the convolution kernel, H(·) is the dynamic convolution kernel generation function, and the input signal of the dynamic depth convolution layer is the heartbeat signal z i [n], convert the one-dimensional signal z i [n] into a two-dimensional feature map

[0039] Channel attention mechanism:

[0040]

[0041] Weighted feature:

[0042] V a [p a , q a = a c ·V c [p c , q c

[0043] Pointwise convolution:

[0044]

[0045] Among them, a c is the channel attention weight, σ c is the activation function, P c is the dimension of the intermediate layer, W c is the attention weight matrix, V a [p a , q a is the weighted feature map, g o [q o , r o is the convolution kernel, q o is the input channel index, r o is the output channel index, C o is the number of input channels of the pointwise convolution, w o [q o , r o is the output feature;

[0046] The expression of the adaptive dilation convolution layer is as follows: ​

[0047]

[0048] where y t [t o is the convolution output of the adaptive dilation convolution layer, t o is the time index, K t is the size of the convolution kernel in the adaptive dilation convolution, k t is the index of the product kernel, w t is the convolution kernel weight of the adaptive dilation convolution, d t is the dilation factor, and the input signal of the adaptive dilation convolution layer is the heartbeat signal z i [n], z i [t o -d t ·k t represents the value of the input heartbeat signal z i [n] at time t o -d t ·k t ;

[0049] Temporal attention mechanism:

[0050]

[0051] Weighted feature:

[0052] y a [t a = a t ·y t [t o

[0053] Lightweight residual connection:

[0054] y r [t r = z i [t o + y a [t a

[0055] where, W t is the attention weight, σ t is the activation function, a t is the temporal attention weight, y a [t a is the weighted feature, z i [t o represents the value of the input heartbeat signal z i [n] at time t o , y r [t r ​​is the residual connection output, and the context feature vector c is obtained by summarizing the residual connection output o ;

[0056] The expression of the fully connected layer is as follows:

[0057] Linear transformation:

[0058] z o = W o ·c o + b o

[0059] Probability mapping:

[0060]

[0061] Among them, W o is the weight matrix of the fully connected layer, b o is the bias vector of the fully connected layer, z o is the output of the linear transformation, P represents the predicted class probability, y p is the true class label, l represents the class, L p is the total number of classes, k p represents the class index.

[0062] A preferred technical solution of this application: The process of the abnormality judgment is as follows:

[0063] According to the output P(y p = l|c o ), the predicted class The expression is as follows:

[0064]

[0065] Among them, argmax represents selecting the class with the highest probability. When , it means the heartbeat signal is normal. When , it means the heartbeat signal is abnormal.

[0066] A preferred technical solution of this application: The abnormal heartbeat signal z i [n] is standardized as follows:

[0067]

[0068] Among them, c[n] is the abnormal standardized signal, μ is the mean of the heartbeat signal z i [n], and σ is the standard deviation of the heartbeat signal zi[n].

[0069] A preferred technical solution of this application: The process of feature extraction is as follows:

[0070] Extract the abnormal heartbeat signal z i The time-domain features of [n], where the time-domain features include the RR interval and the QRS wave width;

[0071] The expression for extracting the RR interval feature is as follows:

[0072] RR i = r i+1 - r i

[0073] The expression for extracting the QRS wave width feature is as follows:

[0074] Q i = r end - r sta

[0075] where, r sta represents the start time of the i-th QRS wave, and r end represents the end time of the i-th QRS wave;

[0076] Extract the frequency-domain features of the abnormal normalized signal c[n] as follows:

[0077]

[0078] where, X[v] is the frequency-domain signal representing the complex value of the signal at frequency v, v is the frequency index representing the v-th frequency component in the frequency domain, where v = [0, 1, 2, ……, N - 1], q represents the imaginary unit, satisfying q 2 = -1, and e is the base of the natural logarithm;

[0079] Extract the morphological features of the abnormal normalized signal c[n], where the morphological features include the waveform area and the waveform derivative;

[0080] The expression for extracting the waveform area feature is as follows:

[0081]

[0082] The expression for extracting the waveform derivative feature is as follows:

[0083] D i [n] = c[n] - c[n - 1]

[0084] where, A i is the waveform area feature of the i-th heartbeat, and D i [n] is the waveform derivative feature of the i-th heartbeat;

[0085] Combine the above-extracted features to form a joint feature F, then F = [RR i , Q i,X[v],A i ,D i [n]]。

[0086] A preferred technical solution of the present application: The abnormal feature detection is to detect the abnormal features in the combined feature F by comparing the extracted combined feature F with the features of the normal signal.

[0087] The present application also provides a method for rapid pre-judgment and analysis of dynamic electrocardiogram, including the following steps:

[0088] S1. Download the electrocardiogram data from the cloud platform and import the patient's dynamic electrocardiogram in the electrocardiogram data into the pre-judgment analysis module;

[0089] S2. Analyze the dynamic electrocardiogram through the pre-judgment analysis module. First, preprocess the original dynamic electrocardiogram signal to remove the baseline drift and high-frequency noise in the original dynamic electrocardiogram signal;

[0090] S3. Then segment the signal after removing the noise. First, detect the R wave, record the position of the R wave, determine the start and end positions of each heartbeat, and then perform heartbeat segmentation. Take the position of the R wave as the center and intercept the signal segment of each heartbeat to obtain the heartbeat signal;

[0091] S4. Input the heartbeat signal into the analysis model to determine whether the heartbeat signal is abnormal and detect the abnormal heartbeat signal;

[0092] S5. Perform standardization processing on the abnormal heartbeat signal to obtain the abnormal standardized signal, and then perform feature extraction;

[0093] S6. Extract the time-domain features from the abnormal heartbeat signal, extract the frequency-domain features and morphological features from the abnormal standardized signal, and integrate the extracted features to obtain the combined feature;

[0094] S7. According to the extracted combined feature, perform feature comparison to detect the abnormal features in the combined feature, so as to detect the abnormal features of the dynamic electrocardiogram;

[0095] S8. Generate a result report for analyzing the dynamic electrocardiogram, check whether the result report is correct, and then upload the result report to the workstation for viewing or printing through the workstation.

[0096] Adopting the technical solution provided by the present invention, compared with the prior art, it has the following beneficial effects:

[0097] The present invention pre - processes the ambulatory electrocardiogram through a pre - judgment analysis module to obtain heartbeat signals, then uses an analysis model to determine whether the heartbeat signals are abnormal, detects abnormal heartbeat signals, standardizes the abnormal heartbeat signals, extracts time - domain features, frequency - domain features, and morphological features from the abnormal heartbeat signals and the abnormal standardized signals, integrates them to obtain combined features, and compares the combined features with the features of normal signals to quickly detect the features with abnormal combined features, thereby accurately obtaining the characteristic points of the patient's electrocardiogram abnormality and avoiding the problem of inaccurate judgment in the process of automatic detection and analysis of ambulatory electrocardiograms. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 It is a system block diagram of the present invention;

[0099] Figure 2 It is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0100] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. The present invention will be further described below with reference to the embodiments.

[0101] Please refer to Figure 1 - Figure 2 , the embodiments of the present application provide a rapid pre - judgment analysis system for ambulatory electrocardiogram, including:

[0102] A cloud platform for storing the electrocardiogram data of patients, where the electrocardiogram data includes ambulatory electrocardiograms;

[0103] A pre - judgment analysis module for pre - judging and analyzing the ambulatory electrocardiograms stored in the cloud platform, determining whether the ambulatory electrocardiograms are abnormal, and analyzing and detecting the abnormal features of the ambulatory electrocardiograms to output a result report. The pre - judgment analysis module is installed in the terminal device, and the ambulatory electrocardiogram data in the cloud platform is imported into the pre - judgment analysis module of the terminal device for analysis;

[0104] A workstation for viewing or printing the result report of the ambulatory electrocardiogram detection. After the pre - judgment analysis module outputs the result report, it is uploaded to the workstation for viewing or printing.

[0105] In this embodiment, the ambulatory electrocardiogram analysis process of the pre - judgment analysis module includes data pre - processing, input into the analysis model, abnormality judgment, feature extraction, and detection of abnormal features.

[0106] In this embodiment, the process of data pre - processing is as follows:

[0107] Let the original ambulatory electrocardiogram signal be \(x(t)\). The process of removing baseline drift and high-frequency noise from the original ambulatory electrocardiogram signal is as follows:

[0108] Removing baseline drift by high-pass filtering:

[0109] \(x[n]=x(t = nT s )

[0110] y f [n]=x[n]-x[n - 1]+\beta\cdot y f [n - 1]

[0111] where \(x(t)\) represents the original continuous-time ambulatory electrocardiogram signal, where \(t\) is the continuous-time variable in seconds, \(x[n]\) represents the discrete-time signal obtained by sampling the continuous-time signal \(x(t)\), where \(n\) is the discrete-time index. Specifically, \(n\) is usually an integer, \(T s is the sampling interval in seconds, \(T s represents the time interval between two consecutive sampling points, \(nT s represents the time corresponding to the \(n\)th sampling point, \(x[n]\) is the sampling value of \(x(t)\) at the discrete-time point \(t = nT s \), \(\beta\) is the filter coefficient that controls the attenuation characteristic of the filter, \(y f [n]\) is the signal after removing the filtering;

[0112] Removing high-frequency noise by discrete wavelet transform:

[0113]

[0114] where \(\psi j,k [n]\) is the discrete wavelet basis function, \(n\) is the discrete-time index, \(N\) is the length of the signal, i.e., \(n=[0,1,2,\cdots,N - 1]\), \(j\) is the scale parameter used to control the stretching of the wavelet, \(k\) is the translation parameter used to control the position of the wavelet, \(W[j,k]\) is the wavelet coefficient. By performing threshold processing on the detail coefficients in the wavelet coefficient \(W[j,k]\), the high-frequency noise in the signal \(y f [n]\) after removing the filtering is removed to obtain the signal \(y d [n]\);

[0115] Segmenting the signal \(y d [n]\) after removing the noise. The signal segmentation includes R-wave detection and heartbeat segmentation;

[0116] The process of R-wave detection is as follows:

[0117] Calculating the derivative signal \(d[n]\):

[0118]

[0119] Calculate the squared signal s[n]:

[0120] s[n] = (d[n]) 2

[0121] Calculate the sliding window integration signal I[n]:

[0122]

[0123] where M is the window size in the sliding window integration, representing the number of sampling points for the smoothed signal, n is the discrete-time index, and m is the summation variable in the sliding window integration, representing the index of each sampling point within the window;

[0124] Detect the R-wave peak and record the R-wave position r i , r i represents the time position of the i-th R-wave, used to mark the position of each R-wave. Determine the start and end positions of each heartbeat through r i ;

[0125] The process of heartbeat segmentation is as follows:

[0126] Centered on the R-wave position r i , intercept the signal segment of each heartbeat to obtain the heartbeat signal z i [n] as follows:

[0127] z i [n] = y d [r i -Δt:r i +Δt]

[0128] where z i [n] is the signal segment of the i-th heartbeat, Δt is the time offset, and the length of each heartbeat signal segment is L, where

[0129] In this embodiment, the process of inputting into the analysis model is as follows:

[0130] Construct an analysis model, which includes a dynamic depth convolutional layer, an adaptive dilation convolutional layer, and a fully connected layer;

[0131] The expression of the dynamic depth convolutional layer is as follows:

[0132]

[0133] where V c [p c , q c is the convolutional output of the dynamic depth convolutional layer, K c is the size of the convolutional kernel in the dynamic depth convolution, pc Denotes the height index of the output feature map, q c Denotes the width index of the output feature map, j c Denotes the width index of the convolutional kernel, H(·) is the dynamic convolutional kernel generation function, and the input signal of the dynamic depth convolutional layer is the heartbeat signal z i [n], converting the one-dimensional signal z i [n] into a two-dimensional feature map

[0134] Channel attention mechanism:

[0135]

[0136] Weighted feature:

[0137] V a [p a ,q a =a c ·V c [p c ,q c

[0138] Pointwise convolution:

[0139]

[0140] Among them, a c Is the channel attention weight, σ c Is the activation function, P c Is the dimension of the intermediate layer, W c Is the attention weight matrix, V a [p a ,q a is the weighted feature map, g o [q o ,r o is the convolutional kernel, q o Is the input channel index, r o Is the output channel index, C o Is the number of input channels of the pointwise convolution, w o [q o ,r o is the output feature;

[0141] The expression of the adaptive dilation convolutional layer is as follows:

[0142]

[0143] Among them, y t [t o is the convolution output of the adaptive dilation convolutional layer, t o ​The time index represents the position of the signal on the time axis, K t is the size of the convolutional kernel in the adaptive dilation convolution, k t is the index of the product kernel, representing the k t th weight in the convolutional kernel, w t is the convolutional kernel weight of the adaptive dilation convolution, d t is the dilation factor, used to control the sampling interval of the convolutional kernel. The input signal of the adaptive dilation convolution layer is the heartbeat signal z i [n], z i [t o -d t ·k t represents the value of the input heartbeat signal z i [n] at time t o -d t ·k t ;

[0144] Time attention mechanism:

[0145]

[0146] Weighted feature:

[0147] y a [t a = a t ·y t [t o

[0148] Lightweight residual connection:

[0149] y r [t r = z i [t o + y a [t a

[0150] Among them, W t is the attention weight, σ t is the activation function, a t is the time attention weight, y a [t a is the weighted feature, z i [t o represents the value of the input heartbeat signal z i [n] at time t o ; y r [t r is the output of the residual connection. Summarizing the output of the residual connection gives the context feature vector c o ;

[0151] The expression of the fully connected layer is as follows:​​

[0152] Linear transformation:

[0153] z o = W o ·c o + b o

[0154] Probability mapping:

[0155]

[0156] where W o is the weight matrix of the fully connected layer, b o is the bias vector of the fully connected layer, z o is the output of the linear transformation, P represents the predicted class probability, y p is the true class label, l represents the class, L p is the total number of classes, where l = [1, 2, ……, L p , k p represents the class index, and e is the base of the natural logarithm.

[0157] In this embodiment, the process of anomaly judgment is as follows:

[0158] According to the output P(y p = l|c o ), the predicted class is expressed as follows:

[0159]

[0160] where argmax represents selecting the class with the highest probability. When , it indicates that the heartbeat signal is normal. When , it indicates that the heartbeat signal is abnormal.

[0161] In this embodiment, the abnormal heartbeat signal z i [n] is normalized as follows:

[0162]

[0163] where c[n] is the abnormal normalized signal, μ is the mean of the heartbeat signal z i [n], and σ is the standard deviation of the heartbeat signal z i [n]. The calculation of the mean and standard deviation of the heartbeat signal is prior art and will not be elaborated here.

[0164] In this embodiment, the process of feature extraction is as follows:

[0165] Extract the abnormal heartbeat signal z iThe time-domain features of [n], where the time-domain features include the RR interval and the QRS complex width;

[0166] The expression for RR interval feature extraction is as follows:

[0167] RR i = r i+1 - r i

[0168] The expression for QRS complex width feature extraction is as follows:

[0169] Q i = r end - r sta

[0170] where r sta represents the start time of the i-th QRS complex, and r end represents the end time of the i-th QRS complex;

[0171] The frequency-domain features of the abnormal normalized signal c[n] are extracted as follows:

[0172]

[0173] where X[v] is the frequency-domain signal representing the complex value of the signal at frequency v, n is the discrete-time index, N is the length of the signal, i.e., n = [0, 1, 2, ……, N - 1], v is the frequency index representing the v-th frequency component in the frequency domain, where v = [0, 1, 2, ……, N - 1], q represents the imaginary unit, satisfying q 2 = -1, and e is the base of the natural logarithm;

[0174] The morphological features of the abnormal normalized signal c[n] are extracted, where the morphological features include the waveform area and the waveform derivative;

[0175] The expression for waveform area feature extraction is as follows:

[0176]

[0177] The expression for waveform derivative feature extraction is as follows:

[0178] D i [n] = c[n] - c[n - 1]

[0179] where A i is the waveform area feature of the i-th heartbeat, D i [n] is the waveform derivative feature of the i-th heartbeat, n is the discrete-time index, and N is the length of the signal, i.e., n = [0, 1, 2, ……, N - 1];

[0180] Combining the above-extracted features to form the joint feature F, then F = [RRi ,Q i ,X[v],A i ,D i [n]]。

[0181] In this embodiment, the detection of abnormal features is based on the comparison between the extracted combined feature F and the features of the normal signal, and the abnormal features in the combined feature F are detected.

[0182] This embodiment provides a method for rapid pre-judgment and analysis of dynamic electrocardiogram, including the following steps:

[0183] S1. Download electrocardiogram data from the cloud platform and import the patient's dynamic electrocardiogram in the electrocardiogram data into the pre-judgment analysis module;

[0184] S2. Analyze the dynamic electrocardiogram through the pre-judgment analysis module. First, preprocess the original dynamic electrocardiogram signal to remove the baseline drift and high-frequency noise in the original dynamic electrocardiogram signal;

[0185] S3. Then segment the denoised signal. First, detect the R wave, record the R wave position, determine the start and end positions of each heartbeat, and then perform heartbeat segmentation. With the R wave position as the center, intercept the signal segment of each heartbeat to obtain the heartbeat signal;

[0186] S4. Input the heartbeat signal into the analysis model to determine whether the heartbeat signal is abnormal and detect the abnormal heartbeat signal;

[0187] S5. Perform normalization processing on the abnormal heartbeat signal to obtain the abnormal normalized signal, and then perform feature extraction;

[0188] S6. Extract time-domain features from the abnormal heartbeat signal, extract frequency-domain features and morphological features from the abnormal normalized signal, and integrate the extracted features to obtain the combined feature;

[0189] S7. According to the extracted combined feature, perform feature comparison to detect the abnormal features in the combined feature, so as to detect the abnormal features of the dynamic electrocardiogram;

[0190] S8. Generate a result report for analyzing the dynamic electrocardiogram, check whether the result report is correct, and then upload the result report to the workstation for viewing or printing through the workstation.

[0191] The above schematically describes the present invention and its implementation manners. This description is not restrictive. What is shown in the drawings is only one of the implementation manners of the present invention, and the actual structure is not limited thereto. Therefore, if those of ordinary skill in the art are inspired by it and design, without creative efforts, structural manners and embodiments similar to the technical solution without departing from the purpose of the present invention, they should all fall within the protection scope of the present invention.

[0192] In addition, it should be understood that although this specification is described in terms of embodiments, not every embodiment only contains an independent technical solution. This narrative manner of the specification is only for clarity. Those skilled in the art should regard the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A dynamic electrocardiogram rapid pre-judgment and analysis system, characterized in that Including: A cloud platform for storing electrocardiogram data of patients, where the electrocardiogram data includes ambulatory electrocardiogram; A pre-judgment analysis module for pre-judging and analyzing the ambulatory electrocardiogram stored in the cloud platform, determining whether the ambulatory electrocardiogram is abnormal, and analyzing the abnormal characteristics of the detected ambulatory electrocardiogram to output a result report. The pre-judgment analysis module is installed in the terminal device, and the ambulatory electrocardiogram data in the cloud platform is imported into the pre-judgment analysis module of the terminal device for analysis; A workstation for viewing or printing the result report of ambulatory electrocardiogram detection. After the pre-judgment analysis module outputs the result report, it is uploaded to the workstation for viewing or printing.

2. The dynamic electrocardiogram rapid pre-judgment and analysis system according to claim 1, wherein, The ambulatory electrocardiogram analysis process of the pre-judgment analysis module includes data preprocessing, input into the analysis model, abnormality judgment, feature extraction, and detection of abnormal features.

3. The dynamic electrocardiogram rapid pre-judgment and analysis system according to claim 2, wherein The process of the data preprocessing is as follows: Let the original ambulatory electrocardiogram signal be x(t). The process of removing baseline drift and high-frequency noise from the original ambulatory electrocardiogram signal is as follows: High-pass filtering to remove baseline drift: x[n] = x(t = nT s ) y f [n] = x[n] - x[n - 1] + β·y f [n - 1] Among them, \(x(t)\) represents the original continuous-time dynamic electrocardiogram signal, and \(x[n]\) represents the discrete-time signal obtained by sampling the continuous-time signal \(x(t)\), where \(n\) is the discrete-time index, \(T\) s is the sampling interval in seconds, and \(T\) s represents the time interval between two consecutive sampling points, and \(nT\) s represents the time corresponding to the \(n\)th sampling point. \(x[n]\) is the sampling value of \(x(t)\) at the discrete time point \(t = nT\) s and \(\beta\) is the filtering coefficient that controls the attenuation characteristic of the filter. \(y\) f [n] is the signal after removing the filtering; Discrete wavelet transform to remove high-frequency noise: Among them, ψ j,k [n] is a discrete wavelet basis function, N is the length of the signal, that is, n = [0, 1, 2, ……, N - 1], j is the scale parameter, k is the translation parameter, W[j, k] is the wavelet coefficient. Threshold processing is performed on the detail coefficients in the wavelet coefficient W[j, k] to remove the high-frequency noise in the filtered signal y f [n], and the signal y d [n] after noise removal is obtained; For the noise-removed signal y d [n], signal segmentation is performed, and the signal segmentation includes R-wave detection and heartbeat segmentation; The process of R wave detection is as follows: Calculate the derivative signal d[n]: Calculate the squared signal s[n]: s[n] = (d[n]) 2 Calculate the sliding window integral signal I[n]: Where M is the window size in the sliding window integral and m is the summation variable in the sliding window integral; Detect the R-wave peak and record the R-wave position r i , r i represents the time position of the i-th R-wave. Determine the start and end positions of each heartbeat through r i ; The process of cardiac beat segmentation is as follows: With the R-wave position r i as the center, intercept the signal segment of each heartbeat to obtain the heartbeat signal z i [n] is as follows: z i [n] = y d [r i -Δt:r i +Δt] where z i [n] is the signal segment of the i-th heartbeat, Δt is the time offset, and the length of each heartbeat signal segment is L, where 4. The dynamic electrocardiogram rapid prediction and analysis system according to claim 3, wherein, The process of the input analysis model is as follows: Construct an analysis model, which includes a dynamic depth convolutional layer, an adaptive dilated convolutional layer, and a fully connected layer; The expression of the dynamic depth convolutional layer is as follows: Among them, V c [p c ,q c is the convolution output of the dynamic depth convolution layer, K c is the size of the convolution kernel in the dynamic depth convolution, p c represents the height index of the output feature map, q c represents the width index of the output feature map, j c represents the width index of the convolution kernel, H(·) is the dynamic convolution kernel generation function, and the input signal of the dynamic depth convolution layer is the heartbeat signal z i [n]. The one-dimensional signal z i [n] is converted into a two-dimensional feature map Channel attention mechanism: Weighted features: V a [p a ,q a =a c ·V c [p c ,q c ​ Pointwise convolution: Among them, a c is the channel attention weight, σ c is the activation function, P c is the dimension of the intermediate layer, W c is the attention weight matrix, V a [p a ,q a is the weighted feature map, g o [q o ,r o is the convolution kernel, q o is the input channel index, r o is the output channel index, C o is the number of input channels for pointwise convolution, w o [q o ,r o is the output feature; The expression of the adaptive dilated convolutional layer is as follows: Among them, y t [t o is the convolution output of the adaptive dilation convolutional layer, where t o is the time index, K t is the size of the convolutional kernel in the adaptive dilation convolution, k t is the index of the product kernel, w t is the convolutional kernel weight of the adaptive dilation convolution, d t is the dilation factor, and the input signal of the adaptive dilation convolutional layer is the heartbeat signal z i [n], where z i [t o -d t ·k t represents the value of the input heartbeat signal z i [n] at time t o -d t ·k t ; Temporal attention mechanism: Weighted features: y a [t a = a t ·y t [t o ​ Lightweight residual connection: y r [t r = z i [t o + y a [t a ​ Among them, W t is the attention weight, σ t is the activation function, a t is the temporal attention weight, y a [t a is the weighted feature, z i [t o represents the value of the input heartbeat signal z i [n] at time t o ; y r [t r is the residual connection output, and the context feature vector c o is obtained by summarizing the residual connection output; The expression of the fully connected layer is as follows: Linear transformation: z o = W o · c o + b o Probability mapping: Among them, W o is the weight matrix of the fully connected layer, b o is the bias vector of the fully connected layer, z o is the output of the linear transformation, P represents the predicted class probability, y p is the true class label, l represents the class, L p is the total number of classes, k p represents the class index.

5. A dynamic electrocardiogram rapid pre-judgment and analysis system according to claim 4, characterized in that The process of the abnormality judgment is as follows: According to the output P(y p = l|c o ), the predicted class is expressed as follows: Among them, argmax represents selecting the category with the highest probability. When it indicates that the heartbeat signal is normal. When it indicates that the heartbeat signal is abnormal.

6. The rapid pre-judgment analysis system for dynamic electrocardiogram according to claim 5, characterized in that, The abnormal heartbeat signal z detected by the abnormal judgment i [n] The expression for the normalization process is as follows: Among them, c[n] is the abnormal normalized signal, and μ is the mean value of the heartbeat signal z i [n], and σ is the standard deviation of the heartbeat signal z i [n].

7. The dynamic electrocardiogram rapid prediction and analysis system according to claim 6, wherein The process of the feature extraction is as follows: Extract the abnormal heartbeat signal z i The time-domain features of [n], where the time-domain features include the RR interval and the QRS wave width; The expression of RR interval feature extraction is as follows: RR i = r i+1 - r i The expression of QRS complex width feature extraction is as follows: Q i ＝r end -r sta where, r sta represents the start time of the i-th QRS complex, and r end represents the end time of the i-th QRS complex; Extract the frequency domain features of the abnormal normalized signal c[n] as follows: Among them, X[v] is the complex value of the frequency-domain signal representing the signal at frequency v, v is the frequency index representing the v-th frequency component in the frequency domain, where v = [0, 1, 2, ……, N-1], q represents the imaginary unit, satisfying q 2 = -1, e is the base of the natural logarithm; Extract the morphological features of the abnormal normalized signal c[n], where the morphological features include waveform area and waveform derivative; The expression of the waveform area feature extraction is as follows: The expression of the waveform derivative feature extraction is as follows: D i [n] = c[n] - c[n - 1] Among them, A i is the waveform area feature of the i-th heartbeat, and D i [n] is the waveform derivative feature of the i-th heartbeat; Combine the above-extracted features to form a combined feature F, then F = [RR i , Q i , X[v], A i , D i [n]].

8. A dynamic electrocardiogram rapid pre-judgment and analysis system according to claim 7, characterized in that, The detection of abnormal features is to compare the extracted joint features F with the features of the normal signal to detect the abnormal features in the joint features F.

9. A method for rapid pre-judgment and analysis of ambulatory electrocardiogram, characterized in that, Including the following steps: S1. Download the electrocardiogram data from the cloud platform and import the patient's ambulatory electrocardiogram in the electrocardiogram data into the pre-judgment analysis module; S2. Analyze the ambulatory electrocardiogram through the pre-judgment analysis module. First, preprocess the original ambulatory electrocardiogram signal to remove baseline drift and high-frequency noise from the original ambulatory electrocardiogram signal; S3. Then segment the signal after removing noise. First, perform R wave detection, record the R wave position, determine the start and end positions of each cardiac beat, and then perform cardiac beat segmentation. Centering on the R wave position, intercept the signal segment of each cardiac beat to obtain the cardiac beat signal; S4. Input the cardiac beat signal into the analysis model to determine whether the cardiac beat signal is abnormal and detect the abnormal cardiac beat signal. S5. Standardize the abnormal heartbeat signal to obtain an abnormal standardized signal, and then perform feature extraction; S6. Extract time-domain features from the abnormal heartbeat signal, extract frequency-domain features and morphological features from the abnormal standardized signal, and integrate the extracted features to obtain combined features; S7. According to the extracted combined features, perform feature comparison to detect the abnormal features in the combined features, thereby detecting the abnormal features of the dynamic electrocardiogram; S8. Generate a result report for analyzing the dynamic electrocardiogram, check whether the result report is correct, and then upload the result report to the workstation for viewing or printing through the workstation.

Images