CN111434305A - Fetal electrocardiogram extraction system and method based on convolutional coding and decoding neural network - Google Patents
Fetal electrocardiogram extraction system and method based on convolutional coding and decoding neural network Download PDFInfo
- Publication number
- CN111434305A CN111434305A CN201910036810.5A CN201910036810A CN111434305A CN 111434305 A CN111434305 A CN 111434305A CN 201910036810 A CN201910036810 A CN 201910036810A CN 111434305 A CN111434305 A CN 111434305A
- Authority
- CN
- China
- Prior art keywords
- maternal
- neural network
- electrocardio
- fetal
- convolution
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/344—Foetal cardiography
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
Abstract
The invention discloses a fetal electrocardiogram extraction system based on a convolutional encoding and decoding neural network and a method thereof, wherein the system comprises a data acquisition device: the device is used for collecting real abdomen electric signals of the pregnant woman; maternal electrocardiogram component estimation device: the method is used for estimating maternal electrocardio components in the maternal abdominal electric signals by adopting a convolution coding and decoding neural network, taking the simulated abdominal electric signals as input of the neural network during training, taking the maternal electrocardio components in the simulated abdominal electric signals as network tags for training, taking the real maternal abdominal electric signals as input of the neural network during testing, and outputting the neural network as the estimated maternal electrocardio components in the abdominal electric signals; fetus electrocardio composition extraction element: is used for subtracting the obtained maternal electrocardio component from the collected maternal abdomen electric signal so as to extract the fetal electrocardio component in the maternal abdomen electric signal. The method comprises the steps of data preprocessing, and extraction of estimated maternal electrocardio components and fetal electrocardio components. Through the technical scheme, the efficiency and the accuracy of fetal electrocardiogram extraction can be effectively improved.
Description
Technical Field
The invention relates to the technical field of fetal electrocardiogram extraction, in particular to a fetal electrocardiogram extraction system based on a convolutional encoding and decoding neural network, and further provides a method for using the system.
Background
Fetal heart monitoring is a means for monitoring fetus in utero, can reflect the biological and physical activity of fetus in real time, and is widely applied to clinical practice. Plays an important role in reducing perinatal mortality. Perinatal mortality reflects, to some extent, the combined economic development and health care status of a country and region.
The fetal electrocardiosignals can be used for calculating the heart rate of the fetus, more morphological information can be provided, the information records the action condition of the heart of the fetus, various states of physiological activities in the fetus uterus are objectively reflected, and medical personnel can judge the development degree and position of the fetus, and whether the fetus is acidotic or arrhythmia, so that the current health condition of the fetus is obtained. One of the difficulties in extracting the fetal electrocardiosignals is that the abdominal electric signals contain the electrocardio components of the mother, the electrocardio components of the mother usually have larger amplitude than the fetal electrocardio components, and the electrocardio components of the mother and the electrocardio components of the fetus are overlapped in time domain and frequency domain, so that great interference is brought to the extraction of the fetal electrocardio. Therefore, it is a challenging task to extract fetal ecg signals from the mixed signal of the maternal abdomen.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a fetal electrocardiogram extraction system which is based on a convolutional coding and decoding neural network and can improve the extraction efficiency and the extraction accuracy and a method using the system.
In order to solve the first technical problem, the technical scheme adopted by the invention specifically comprises the following steps:
a fetal electrocardiogram extraction system based on a convolutional encoding and decoding neural network comprises the following devices:
a data acquisition device: the device is used for collecting real abdomen electric signals of the pregnant woman;
maternal electrocardiogram component estimation device: the method is used for constructing a convolutional encoding and decoding neural network to be trained, during training, a simulated abdomen electrical signal is used as the input of the neural network, a parent electrocardio component in the simulated abdomen electrical signal is used as a network tag to train the convolutional encoding and decoding neural network to change parameters (namely weight coefficient and deviation) in the neural network, the training is stopped until the error of the neural network is completely converged (the error is completely converged, namely the obtained parent electrocardio component is compared with the known parent electrocardio component in the simulated abdomen electrical signal until no error exists or the error is in a reasonable range), and all the parameters are stored to obtain a trained convolutional encoding and decoding neural network unit. When the real pregnant woman abdominal electric signals are measured, the real single-channel pregnant woman abdominal electric signals collected by the data collecting device are substituted into the trained convolutional encoding and decoding neural network unit as the input of the neural network, and the obtained output of the neural network is the estimated maternal electrocardio component in the abdominal electric signals.
The convolution coding and decoding neural network in the executive program of the maternal electrocardio component estimation device is composed of a plurality of convolution-deconvolution modules and a full-connection module in series, and the convolution kernel size of a convolution layer in each convolution-deconvolution module is 1 × 3, 1 × 4 or 1 × 5.
Preferably, the technical scheme trains the convolutional encoding and decoding neural network by adopting a gradient descent method and a back propagation algorithm. The simulated abdominal electrical signals are obtained according to data in an existing simulation database and can be divided into a training set and a verification set.
When the convolutional encoding and decoding neural network is trained by adopting a gradient descent method and a back propagation algorithm, in normal operation, firstly, the abdominal electric signals of a simulated pregnant woman in a training set are input into the convolutional encoding and decoding neural network after being preprocessed, the convolutional encoding and decoding neural network is trained, the electrocardio components of a mother body are evaluated through a loss function, then, the gradients of all parameters in the convolutional neural network are calculated through the back propagation algorithm (the gradient descent method), all parameters are updated according to gradient changes (derivatives of all parameters), the training set is substituted into the convolutional encoding and decoding neural network again for training, and all parameters are stored and the model is reserved until the model with the minimum loss function is found through iteration. When real pregnant woman abdominal electric signals need to be tested, calling a convolutional neural network model, preprocessing the real pregnant woman abdominal electric signals acquired by the data acquisition device, inputting the preprocessed real pregnant woman abdominal electric signals into a convolutional encoding and decoding neural network, acquiring estimated maternal electrocardio components, and subtracting the estimated maternal electrocardio components from the real acquired pregnant woman abdominal electric signals, so that fetal electrocardio components in the pregnant woman abdominal electric signals are extracted.
Fetus electrocardio composition extraction element: the method is used for subtracting the obtained estimated maternal electrocardio component from the collected maternal abdomen electric signal so as to extract the fetal electrocardio component in the maternal abdomen electric signal.
It should be noted that, in order to improve the efficiency and accuracy of fetal electrocardiogram extraction in the detection process, the inventor provides an extraction system in the technical scheme, which is convenient for an operator to realize signal processing through the system and is convenient for subsequent detection and diagnosis. The inventor creates a scheme for estimating maternal electrocardio components in the maternal abdomen electric signals by using a convolution coding and decoding neural network innovatively in the technical scheme and subtracting the acquired maternal electrocardio components from the acquired maternal abdomen electric signals so as to extract fetal electrocardio components in the maternal abdomen electric signals.
It should be noted that the data acquisition device also converts the abdomen electrical signals of the pregnant woman into digital signals, which are convenient to be processed on digital equipment such as a computer.
Preferably, the data acquisition device further executes the following program:
and the pregnant woman abdominal electric signals are processed by processing programs such as amplification and/or filtering, and then converted into digital signals by an A/D conversion processing program.
More preferably, the data acquisition device further executes the following program:
preferably, the digital signal is cropped to a size that matches the convolutional codec neural network input, the cropped size of the digital signal is 1 × 1000, where the value in each signal of 1 × 1 is the electrocardiogram value (amplitude) at a point, and 1000 has 1000 points, each point representing the value (amplitude) in the electrocardiogram at a different time.
By tailoring to a matching size, the following training using convolutional codec neural networks can be performed more efficiently.
Preferably, the maternal cardiac component estimation device executes the following procedure during the training process:
the network model is trained using an open full-layer network parameter update approach and using the mean square error as a loss function. That is, the present application adopts a back propagation algorithm (gradient descent method), and the gradient of all parameters in the convolutional neural network needs to be calculated during calculation, and all parameters are updated according to the gradient change (the derivative of each parameter).
Preferably, the loss function is:
wherein M is the number of training samples,
for the estimation of maternal electrocardio components output by the network, X is a network label, namely the maternal electrocardio components in the simulated abdominal electric signals. By adopting the loss function, the characteristics of maternal electrocardio can be completely and quickly learned.
Preferably, when training is performed, the maximum number of iterations is 10000, and the learning rate is 0.0001.
Preferably, the convolution coding and decoding neural network in the executive program of the maternal electrocardiogram component estimation device is formed by connecting a plurality of convolution-deconvolution modules and a full-connection module in series;
the convolution-deconvolution module comprises a convolution layer, a deconvolution layer and a nonlinear layer, wherein the convolution-deconvolution module connected with the full-connection module comprises one or more full-connection layers, the convolution-deconvolution module connected with the full-connection module comprises one convolution layer, one deconvolution layer and two nonlinear layers, the direct connection mode is adopted between every two adjacent convolution-deconvolution modules, the activation function of the nonlinear layer is preferably a tanh function, the convolution kernel size of the convolution layer is preferably 1 × 3, 1 × 4 or 1 × 5, and the convolution kernel is relatively proper, so that local feature extraction is not excessive, and a part of local features cannot be lost due to the large convolution kernel size.
It should be noted that, the convolutional encoding and decoding neural network adopting the above structure can effectively capture the characteristics of maternal electrocardio, accelerate the convergence rate of network training, and simultaneously has a good effect on the problem of gradient dispersion. It should be noted that, by adopting a direct connection mode between two adjacent convolution-deconvolution modules, the shallow features can be transferred to the deep layers, so that more detailed features of the maternal electrocardiogram are retained, and the efficiency and accuracy are further improved.
Preferably, the system further comprises a display device for splicing the extracted fetal electrocardiogram components into a complete signal and displaying the complete signal.
It should be noted that the signal is cut into a shorter length for learning, and after learning is completed, the network output is spliced into a complete signal, so that more detailed features can be learned by the network, and the effect of fetal electrocardiogram extraction is improved.
Preferably, all parameters of the network are initialized with small random numbers before training is started. Firstly, if the parameter is small, the protection network can not enter a saturation state due to the fact that the weight is too large at the beginning, so that training fails, secondly, the randomized parameter is realized to ensure that the initial value of each parameter is different, and the situation that the same parameter is obtained in each training is avoided. The numerical range of the small random number is 0-1.
In order to solve the second technical problem, the technical scheme adopted by the invention specifically comprises the following steps:
a fetal electrocardiogram extraction method based on a convolutional encoding and decoding neural network comprises the following steps:
data preprocessing: collecting abdominal electric signals of a pregnant woman;
estimation of maternal electrocardiogram components: estimating maternal electrocardio components in the maternal abdominal electric signals by adopting a convolution coding and decoding neural network, namely, taking simulated abdominal electric signals as input of the neural network during training, taking maternal electrocardio components in the simulated abdominal electric signals as network tags for training, taking real single-channel maternal abdominal electric signals as input of the neural network during testing, and outputting the neural network as the estimated maternal electrocardio components in the abdominal electric signals;
elimination of maternal electrocardiographic components: and subtracting the obtained maternal electrocardio component from the acquired pregnant woman abdominal electric signal so as to extract a fetal electrocardio component in the pregnant woman abdominal electric signal.
In order to improve the efficiency and the accuracy of fetal electrocardio extraction, the inventor creatively estimates the maternal electrocardio component in the pregnant woman abdominal electric signal by using a convolution coding and decoding neural network in the technical scheme, and subtracts the obtained maternal electrocardio component from the collected pregnant woman abdominal electric signal so as to extract the fetal electrocardio component in the pregnant woman abdominal electric signal. In the technical scheme, the convolutional encoding and decoding neural network is adopted, the technology has stronger robustness, and the fetal electrocardiosignals can still be extracted under the interference conditions of contact noise, respiratory noise and the like. Moreover, the method performs well when the signal-to-noise ratio between the fetal signal and the maternal signal is low and there is overlap between the QRS wave of the fetus and the QRS wave of the mother. Through the technical means, the relevant noise can be eliminated to a greater extent, so that the extraction is more accurate and the efficiency is higher. The clear fetal electrocardiosignals extracted by the method can provide information such as fetal heart rate and the like, and have important clinical application value and considerable social benefit.
More preferably, the digital signal is obtained by a/D conversion after processing including amplification and/or filtering.
Further, the data preprocessing step further comprises cutting the digital signal into a size matched with the input of the convolutional encoding and decoding neural network.
Preferably, the training method in the estimation of the maternal electrocardiogram components is to use an open full-layer network parameter updating mode and to train a convolutional encoding and decoding neural network model by using a mean square error as a loss function;
preferably, the loss function is:
wherein M is the number of training samples,
the estimated maternal electrocardio component is output by the network, and X is a network label, namely the maternal electrocardio component in the simulated abdominal electric signal.
Preferably, the convolution coding and decoding neural network in the maternal electrocardiogram component estimation device is formed by connecting a plurality of convolution-deconvolution modules and a full-connection module in series;
the convolution-deconvolution module connected with the full-connection module consists of a convolution layer, a deconvolution layer and a nonlinear layer, and all other convolution-deconvolution modules comprise the convolution layer, the deconvolution layer and two nonlinear layers; and a direct connection mode is adopted between two adjacent convolution-deconvolution modules.
Preferably, the technical scheme trains the convolutional encoding and decoding neural network by adopting a gradient descent method and a back propagation algorithm.
Preferably, the method further comprises splicing the extracted fetal electrocardiogram components into a complete signal and displaying the complete signal.
Compared with the prior art, the invention has the beneficial effects that:
1. the fetal electrocardiogram extraction system based on the convolutional encoding and decoding neural network adopts the convolutional encoding and decoding neural network, has stronger robustness, can still extract fetal electrocardiogram signals under the interference conditions of contact noise, breathing noise and the like, can remove relevant noise to a greater extent, and ensures that the extraction is more accurate and the efficiency is higher;
2. according to the fetal electrocardiogram extraction system based on the convolutional encoding and decoding neural network, signals are cut into short lengths for learning, and after learning is completed, network outputs are spliced into complete signals, so that more detailed characteristics can be learned by the network, and the effect of fetal electrocardiogram extraction is improved;
3. the fetal electrocardiogram extraction method based on the convolutional encoding and decoding neural network can realize the extraction purpose by applying the system to execute the corresponding process.
The foregoing description is only an overview of the technical solutions of the present invention, and in order to make the technical means of the present invention more clearly understood, the present invention may be implemented in accordance with the content of the description, and in order to make the above and other objects, features, and advantages of the present invention more clearly understood, the following preferred embodiments are described in detail with reference to the accompanying drawings.
Drawings
FIG. 1 is a schematic view of a flow framework of a preferred embodiment of a fetal electrocardiogram extraction method based on a convolutional encoding and decoding neural network according to the present invention;
FIG. 2 is a schematic diagram of a simulated abdominal signal and corresponding maternal electrocardiac components during training in one embodiment of the method of the present invention;
FIG. 3 is a schematic diagram of the fetal ECG signal obtained according to the method, the corresponding real abdominal signal and the maternal ECG component output by the network;
fig. 4 is a schematic structural diagram of a convolutional codec neural network.
Detailed Description
To further illustrate the technical means and effects of the present invention adopted to achieve the predetermined objects, the following detailed description of the embodiments, structures, features and effects according to the present invention with reference to the accompanying drawings and preferred embodiments is as follows:
embodiment 1 (fetal electrocardiogram extraction system based on convolutional encoding and decoding neural network)
A fetal electrocardiogram extraction system based on a convolutional coding and decoding neural network comprises the following devices:
a data acquisition device: is used for collecting a path of abdomen electric signals of the pregnant woman;
maternal electrocardiogram component estimation device: the method is used for estimating maternal electrocardio components in the maternal abdominal electric signals by adopting a convolutional encoding and decoding neural network, namely, the simulated abdominal electric signals are used as the input of the neural network during training, and the maternal electrocardio components in the simulated abdominal electric signals are used as network labels for training to obtain the trained convolutional encoding and decoding neural network; when the real fetal electrocardio components are measured, the real single-channel pregnant woman abdominal electric signals are input into the trained convolutional encoding and decoding neural network, and the estimated maternal electrocardio components in the real abdominal electric signals are obtained through the trained convolutional encoding and decoding neural network. (ii) a
Fetus electrocardio composition extraction element: is used for subtracting the obtained maternal electrocardio component from the collected maternal abdomen electric signal so as to extract the fetal electrocardio component in the maternal abdomen electric signal.
The above is the basic embodiment of the present technical solution. In order to improve the efficiency and the accuracy of fetal electrocardiogram extraction in the detection process, the inventor provides an extraction system in the technical scheme, so that an operator can conveniently realize signal processing through the system, and subsequent detection and diagnosis and treatment are facilitated. The inventor creates a scheme for estimating maternal electrocardio components in the maternal abdomen electric signals by using a convolution coding and decoding neural network innovatively in the technical scheme and subtracting the acquired maternal electrocardio components from the acquired maternal abdomen electric signals so as to extract fetal electrocardio components in the maternal abdomen electric signals. The data acquisition device can also convert the abdomen electric signals of the pregnant woman into digital signals, and the digital signals can be conveniently processed on digital equipment such as a computer.
In some embodiments, the data acquisition device further performs the following procedure:
and after the pregnant woman abdominal electric signals are processed by an amplifying and/or filtering processing program, the electric signals are converted into digital signals by an A/D conversion program.
In some embodiments, the data acquisition device further performs the following procedure: and cutting the digital signal into a size matched with the input of the convolutional encoding and decoding neural network. By tailoring to a matching size, the following training using convolutional codec neural networks can be performed more efficiently.
In some embodiments, the maternal cardiac component estimation device, when taken for the training process, performs the following procedure: the network model is trained using an open full-layer network parameter update approach and using the mean square error as a loss function.
In some embodiments, the loss function is:
wherein M is the number of training samples,
for the estimation of maternal electrocardio components output by the network, X is a network label, namely the maternal electrocardio components in the simulated abdominal electric signals. By adopting the loss function, the characteristics of maternal electrocardio can be completely and quickly learned.
In some embodiments, the convolution coding and decoding neural network in the program executed by the maternal electrocardiogram component estimation device is composed of a plurality of convolution-deconvolution modules and a full-connection module which are connected in series; the convolution-deconvolution module connected with the full-connection module consists of a convolution layer, a deconvolution layer and a nonlinear layer, and all other convolution-deconvolution modules comprise the convolution layer, the deconvolution layer and two nonlinear layers; and a direct connection mode is adopted between two adjacent convolution-deconvolution modules. The convolutional encoding and decoding neural network adopts the structure, can effectively capture the characteristics of maternal electrocardio, accelerates the convergence speed of network training, and has good effect on the problem of gradient dispersion. It should be noted that, by adopting a direct connection mode between two adjacent convolution-deconvolution modules, the shallow features can be transferred to the deep layers, so that more detailed features of the maternal electrocardiogram are retained, and the efficiency and accuracy are further improved.
Preferably, the digital signal is cut to a size of 1 × 1000, where the value in each signal of 1 × 1 is the electrocardiogram value (amplitude) at a certain point, and 1000 has 1000 points, each point representing the value (amplitude) in the electrocardiogram at a different time, and the convolution kernel size of the convolutional layer is 1 × 3, 1 × 4 or 1 × 5, and the number of convolution kernels in each convolutional layer is 32, 64 or 128, preferably 64.
In one embodiment, as shown in fig. 4, the system includes 3 convolution-deconvolution modules, that is, a first convolution-deconvolution module, a second convolution-deconvolution module, and a third convolution-deconvolution module, in the first convolution-deconvolution module, the clipped digital signal is convolved with a convolution kernel, a matrix obtained by convolution is further subjected to calculation of an activation function by a nonlinear layer, and after deconvolution is performed, calculation of the activation function is further performed by the nonlinear layer, an output of the first convolution-deconvolution module may be used as an input of the second convolution-deconvolution module or the third convolution-deconvolution module, and an output of the second convolution-deconvolution module may be used as an input of the third convolution-deconvolution module. And finally, the output of the third convolution-deconvolution module is used as the input of a fully connected layer (dense), and the finally output image is obtained through the processing of the fully connected layer. In this embodiment, the number of neurons in the fully-connected layer is 1000.
In some embodiments, the system further comprises a display device for splicing the extracted fetal electrocardiogram components into a complete signal and displaying the complete signal. The method of cutting the signal into a shorter length for learning and splicing the network output into a complete signal after learning is completed is adopted, so that more detailed characteristics can be learned by the network, and the effect of fetal electrocardiogram extraction is improved.
When the simulation pregnant woman abdominal electrical signals are normally operated, firstly, simulation pregnant woman abdominal electrical signals in a training set and/or a verification set are input into a convolution encoding and decoding neural network after being preprocessed, the convolution encoding and decoding neural network is trained, maternal electrocardio components are evaluated through a loss function, gradients of all parameters in the convolution neural network are calculated through a back propagation algorithm (a gradient descent method), all parameters are updated according to gradient changes (derivatives of all parameters), the training set is substituted into the convolution encoding and decoding neural network again for training, and all parameters are stored and the model is reserved until iteration searching is completed and the model with the minimum loss function is found. When real pregnant woman abdominal electric signals need to be tested, calling a convolutional neural network model, preprocessing the real pregnant woman abdominal electric signals, inputting the preprocessed real pregnant woman abdominal electric signals into a convolutional encoding and decoding neural network, obtaining estimated maternal electrocardio components, and subtracting the obtained estimated maternal electrocardio components from the real collected pregnant woman abdominal electric signals, so that fetal electrocardio components in the pregnant woman abdominal electric signals are extracted.
Embodiment 2 (fetal electrocardiogram extraction method based on convolutional coding and decoding neural network)
As shown in fig. 1, the present invention provides a fetal electrocardiogram extraction method based on a convolutional encoding and decoding neural network, which comprises the following steps:
data preprocessing: collecting abdominal electric signals of a pregnant woman;
estimation of maternal electrocardiogram components: estimating maternal electrocardio components in the maternal abdominal electric signals by adopting a convolution coding and decoding neural network, namely, taking simulated abdominal electric signals as input of the neural network during training, taking maternal electrocardio components in the simulated abdominal electric signals as network tags for training, taking real single-channel maternal abdominal electric signals as input of the neural network during testing, and outputting the neural network as the estimated maternal electrocardio components in the abdominal electric signals;
extracting fetal electrocardio components: and subtracting the obtained estimated maternal electrocardio component from the acquired maternal abdomen electric signal, thereby extracting the fetal electrocardio component from the maternal abdomen electric signal.
The above is the basic embodiment of the present invention. In the technical scheme, in order to improve the efficiency and the accuracy of fetal electrocardio extraction, the inventor creatively estimates the maternal electrocardio component in the pregnant woman abdominal electric signal by using a convolution coding and decoding neural network in the technical scheme, and subtracts the estimated maternal electrocardio component from the collected pregnant woman abdominal electric signal so as to extract the fetal electrocardio component in the pregnant woman abdominal electric signal. The technology has strong robustness by adopting the convolutional encoding and decoding neural network, and fetal electrocardiosignals can still be extracted under the interference conditions of contact noise, breathing noise and the like. Moreover, the method also performs well when the signal-to-noise ratio between the fetal and maternal signals is low and there is an overlap of the QRS wave of the fetus with the QRS wave of the mother. Through the technical means, the relevant noise can be eliminated to a greater extent, so that the extraction is more accurate and the efficiency is higher. The clear fetal electrocardiosignals extracted by the method can provide information such as fetal heart rate and the like, and have important clinical application value and considerable social benefit.
Embodiment 3 (fetal electrocardiogram extraction method based on convolutional coding and decoding neural network)
The implementation of the basic embodiment is described below with reference to a specific embodiment. However, this embodiment is only for illustrating the present invention and does not represent a limitation to the scope of the present invention.
A fetal electrocardiogram extraction method based on a convolutional encoding and decoding neural network comprises the following steps:
1) data preprocessing, as shown in fig. 3, firstly, collecting a path of signal on the abdomen of the mother body by using an electrode, wherein the sampling frequency is 250Hz, filtering the signal by using a band-pass filter with a pass band of 0.5-100Hz and a notch filter with a pass band of 50Hz, secondly, amplifying the signal by using an amplifying circuit, converting the electric signal into a digital signal by using an analog-to-digital conversion circuit, and finally, cutting the filtered data into a fixed size (1 × 1000).
In a more specific preferred embodiment, the signal samples are uniformly cropped to a fixed size that matches the size of the convolutional codec neural network input used in the method (or the signal is acquired directly at the time of acquiring the signal, e.g., 2 or 4 seconds in length).
2) Estimation of maternal electrocardiogram components: training a convolution coding and decoding neural network to predict maternal electrocardio components in the maternal abdomen electric signals. In some preferred embodiments, a single-channel input data processing framework is used to extract the electrocardiogram sample data, the order of the input samples is randomly disturbed, and the sample data is input with a batch _ size of 64. During training, the simulated abdomen electric signal is used as the input of the neural network, the maternal electrocardio component in the simulated abdomen electric signal is used as the network label for training, during testing, the real single-channel pregnant woman abdomen electric signal is used as the input of the neural network, and the output of the neural network is the estimated maternal electrocardio component in the abdomen electric signal, as shown in fig. 3.
In some preferred embodiments, the convolutional encoding and decoding neural network is used as a deep neural network model to estimate the maternal electrocardio-component in the abdominal electric signal of the pregnant woman. The adopted convolutional encoding and decoding neural network is formed by connecting 5 convolutional-deconvolution modules and a full-connection module in series; the convolution-deconvolution module connected with the full-connection module consists of a convolution layer, a deconvolution layer and a nonlinear layer, and all other convolution-deconvolution modules comprise the convolution layer, the deconvolution layer and the two nonlinear layers; meanwhile, a direct connection mode is adopted between two adjacent convolution-deconvolution modules, so that the characteristics of a shallow layer can be transferred to a deep layer.
In some preferred embodiments, the network model is trained using an open full-layer network parameter update approach and using the mean square error as a loss function.
In more specific preferred embodiments, the convolutional codec neural network is trained by using an open full-layer network parameter updating method, and using a mean square error as a loss function, where the function is expressed as:
wherein M is the number of training samples,
is the estimated maternal electrocardio component output by the network, X is a network label, namely the mother in the simulated abdominal electric signalBody electrocardio component.
3) Extracting fetal electrocardio components: and subtracting the estimated maternal electrocardio component in the abdominal electric signal from the acquired abdominal electric signal of the pregnant woman, thereby extracting the fetal electrocardio component in the abdominal electric signal of the pregnant woman.
More specifically, the estimated maternal electrocardio component output by the convolutional encoding and decoding neural network is subtracted from the acquired abdominal electric signal, and the rest part is the fetal electrocardio component, so that the fetal electrocardio component in the maternal abdominal electrocardio is extracted.
4) Displaying maternal and fetal electrocardiograms: splicing the extracted fetal electrocardio components into a complete signal, and displaying the maternal electrocardio and the fetal electrocardio as shown in figure 3.
To verify the effect of the present invention, the applicant substitutes the test set into the trained convolutional encoding and decoding neural network for many times (at this time, the convolutional encoding and decoding neural network is as shown in fig. 4, the size of the convolutional kernel is 1 × 3, and the number of the convolutional kernels is 64), so as to obtain the following data:
| mean SE(％) | mean PPV(％) | mean F1(％) | |
| trained neural network | 93.08 | 91.21 | 92.13 |
Wherein the content of the first and second substances,
TP is the number of finding the R peak (R peak of QRS peak), FN is the number of missing R peak, FP is the number of finding the wrong R peak, and the finding the position of the R peak and the real position are all calculated within 50 ms.
The above embodiments are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited thereby, and any insubstantial changes and substitutions made by those skilled in the art based on the present invention are within the protection scope of the present invention.
Claims (10)
1. A fetal electrocardiogram extraction system based on a convolutional coding and decoding neural network is characterized by comprising the following devices:
a data acquisition device: is used for collecting the abdomen electric signals of the pregnant woman;
maternal electrocardiogram component estimation device: the training device is used for constructing a convolutional coding and decoding neural network to be trained and training the convolutional coding and decoding to be trained according to the simulated pregnant woman abdominal electric signals; when real fetal electrocardio components are measured, single-channel pregnant woman abdominal electric signals collected by a data collecting device are input into a trained convolutional encoding and decoding neural network, and estimated maternal electrocardio components of the real abdominal electric signals are obtained through the trained convolutional encoding and decoding neural network;
fetus electrocardio composition extraction element: the pregnant woman abdominal electric signal acquisition module is used for subtracting the estimated maternal electrocardio component from the acquired pregnant woman abdominal electric signal so as to extract the fetal electrocardio component in the pregnant woman abdominal electric signal;
the convolution coding and decoding neural network in the executive program of the maternal electrocardio component estimation device is composed of a plurality of convolution-deconvolution modules and a full-connection module in series, and the convolution kernel size of a convolution layer in each convolution-deconvolution module is 1 × 3, 1 × 4 or 1 × 5.
2. A fetal electrocardiogram extraction system as claimed in claim 1 wherein the data acquisition means further executes the following procedures:
and after the pregnant woman abdominal electric signal is subjected to an amplification and/or filtering processing program, the electric signal is converted into a digital signal through an AD conversion program.
3. A fetal electrocardiogram extraction system as claimed in claim 2 wherein the data acquisition means further executes the following procedures:
and cutting the digital signal into a size matched with the input of the convolutional encoding and decoding neural network.
4. A fetal electrocardiogram extraction system as claimed in claim 1 wherein the simulated abdominal electrical signal used by the maternal electrocardiogram component estimation device in training is comprised of a superposition of simulated maternal and fetal electrocardiogram components and some noise.
5. A fetal electrocardiogram extraction system as claimed in claim 1 wherein the maternal electrocardiogram component estimation means performs the following procedures for the training process:
training a convolution decoding neural network by using an open full-layer network parameter updating mode and adopting a mean square error as a loss function;
preferably, the convolutional codec neural network is trained using a gradient descent method and a back propagation algorithm.
6. A fetal electrocardiogram extraction system as claimed in claim 5 wherein the loss function is:
wherein M is the number of training samples,
the estimated maternal electrocardio component is output by the network, and X is a network label, namely the maternal electrocardio component in the simulated abdominal electric signal.
7. A fetal ecg extraction system as claimed in claim 1 wherein the fully connected module is also directly connected to the output layer and the fully connected module is comprised of one or more fully connected layers, the convolution-deconvolution module connected to the fully connected module is comprised of a convolution layer, a deconvolution layer and a non-linear layer, all other convolution-deconvolution modules including a convolution layer, a deconvolution layer and two non-linear layers; and a direct connection mode is adopted between two adjacent convolution-deconvolution modules.
8. A fetal electrocardiogram extraction system as claimed in claim 1 wherein the system further comprises a display means for stitching and displaying the extracted fetal electrocardiogram components into a complete signal.
9. The fetal electrocardiogram extraction method based on the convolutional encoding and decoding neural network is characterized by comprising the following steps of:
data acquisition: collecting a path of abdomen electric signals of the pregnant woman;
estimation of maternal electrocardiogram components: constructing a convolutional encoding and decoding neural network to be trained, and training the convolutional encoding and decoding neural network to be trained by adopting an artificial abdomen electric signal; when the real fetal electrocardio components are measured, the real single-channel pregnant woman abdominal electric signals are input into the trained convolutional encoding and decoding neural network, and the estimated maternal electrocardio components in the real abdominal electric signals are obtained through the trained convolutional encoding and decoding neural network.
Extracting fetal electrocardio components: subtracting the estimated maternal electrocardio component from the acquired pregnant woman abdominal electric signal, thereby extracting a fetal electrocardio component from the pregnant woman abdominal electric signal;
preferably, the process of converting the maternal abdominal electrical signals into digital signals comprises an amplification and/or filtering processing procedure and an a/D conversion procedure;
more preferably, the data preprocessing step further comprises cutting the digital signal into a size matched with the input of the convolutional encoding and decoding neural network;
preferably, the training method in the estimation of the maternal electrocardiogram components is to use an open full-layer network parameter updating mode and to train a network model by using a mean square error as a loss function;
preferably, the convolutional encoding and decoding neural network in the maternal electrocardiogram component estimation is formed by connecting a plurality of convolutional-deconvolution modules and a full-connection module in series;
the convolution-deconvolution module connected with the full-connection module consists of a convolution layer, a deconvolution layer and a nonlinear layer, and all other convolution-deconvolution modules comprise the convolution layer, the deconvolution layer and two nonlinear layers; and a direct connection mode is adopted between two adjacent convolution-deconvolution modules. Preferably, the activation function of the non-linear layer is a tanh function.
Preferably, the method further comprises splicing the extracted fetal electrocardiogram components into a complete signal and displaying;
preferably, the technical scheme trains the convolutional encoding and decoding neural network by adopting a gradient descent method and a back propagation algorithm.
10. The fetal electrocardiogram extraction method of claim 9 wherein the loss function is:
wherein M is the number of training samples,
the estimated maternal electrocardio component is output by the network, and X is a network label, namely the maternal electrocardio component in the simulated abdominal electric signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910036810.5A CN111434305A (en) | 2019-01-15 | 2019-01-15 | Fetal electrocardiogram extraction system and method based on convolutional coding and decoding neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910036810.5A CN111434305A (en) | 2019-01-15 | 2019-01-15 | Fetal electrocardiogram extraction system and method based on convolutional coding and decoding neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111434305A true CN111434305A (en) | 2020-07-21 |
Family
ID=71580830
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910036810.5A Pending CN111434305A (en) | 2019-01-15 | 2019-01-15 | Fetal electrocardiogram extraction system and method based on convolutional coding and decoding neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111434305A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023064574A1 (en) * | 2021-10-15 | 2023-04-20 | Children's National Medical Center | Development of a beat-to-beat fetal electrocardiogram |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005117692A2 (en) * | 2004-05-28 | 2005-12-15 | University Of Florida Research Foundation, Inc. | Maternal-fetal monitoring system |
| JP2009160410A (en) * | 2008-01-08 | 2009-07-23 | General Electric Co <Ge> | Method of fetal and maternal ecg identification across multiple epochs |
| CN107693004A (en) * | 2017-09-05 | 2018-02-16 | 广东工业大学 | Fetal ECG extraction and fetal heart frequency recognition methods based on hilbert conversion |
| CN108814585A (en) * | 2018-05-03 | 2018-11-16 | 深圳竹信科技有限公司 | ECG's data compression method, apparatus and computer readable storage medium |
-
2019
- 2019-01-15 CN CN201910036810.5A patent/CN111434305A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005117692A2 (en) * | 2004-05-28 | 2005-12-15 | University Of Florida Research Foundation, Inc. | Maternal-fetal monitoring system |
| JP2009160410A (en) * | 2008-01-08 | 2009-07-23 | General Electric Co <Ge> | Method of fetal and maternal ecg identification across multiple epochs |
| CN107693004A (en) * | 2017-09-05 | 2018-02-16 | 广东工业大学 | Fetal ECG extraction and fetal heart frequency recognition methods based on hilbert conversion |
| CN108814585A (en) * | 2018-05-03 | 2018-11-16 | 深圳竹信科技有限公司 | ECG's data compression method, apparatus and computer readable storage medium |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023064574A1 (en) * | 2021-10-15 | 2023-04-20 | Children's National Medical Center | Development of a beat-to-beat fetal electrocardiogram |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110840402B (en) | Atrial fibrillation signal identification method and system based on machine learning | |
| US11529103B2 (en) | Artificial intelligence self-learning-based automatic electrocardiography analysis method and apparatus | |
| US11324455B2 (en) | Artificial intelligence-based interference recognition method for electrocardiogram | |
| US11617528B2 (en) | Systems and methods for reduced lead electrocardiogram diagnosis using deep neural networks and rule-based systems | |
| CN110619322A (en) | Multi-lead electrocardio abnormal signal identification method and system based on multi-flow convolution cyclic neural network | |
| Belforte et al. | A contribution to the automatic processing of electrocardiograms using syntactic methods | |
| CN102626310A (en) | Electrocardiogram signal feature detection algorithm based on wavelet transformation lifting and approximate envelope improving | |
| CN109674464B (en) | Multi-lead electrocardiosignal composite feature extraction method and corresponding monitoring system | |
| Fotiadou et al. | End-to-end trained encoder–decoder convolutional neural network for fetal electrocardiogram signal denoising | |
| CN109893118A (en) | A kind of electrocardiosignal classification diagnosis method based on deep learning | |
| CN103750835A (en) | Electrocardiosignal characteristic detection algorithm | |
| CN109567792A (en) | A kind of single channel abdomen record fetal electrocardiogram extracting method | |
| Fotiadou et al. | Deep convolutional long short-term memory network for fetal heart rate extraction | |
| CN114901145A (en) | System and method for electrocardiographic diagnosis using deep neural networks and rule-based systems | |
| CN111667921A (en) | Artificial intelligence ECG atrial fibrillation and arrhythmia detection system | |
| CN111434305A (en) | Fetal electrocardiogram extraction system and method based on convolutional coding and decoding neural network | |
| CN113384277A (en) | Electrocardiogram data classification method and classification system | |
| Zairi et al. | Intelligent system for detecting cardiac arrhythmia on FPGA | |
| CN110141245A (en) | A kind of Characteristics of electrocardiogram vector extracting method | |
| Warbhe et al. | A single channel phonocardiograph processing using EMD, SVD, and EFICA | |
| Petrolis et al. | Multi stage principal component analysis based method for detection of fetal heart beats in abdominal ECGs | |
| CN110537907B (en) | Electrocardiosignal compression and identification method based on singular value decomposition | |
| CN115607167A (en) | Lightweight model training method, atrial fibrillation detection method, device and system | |
| Sivaranjini et al. | Computationally-efficient pulse rate estimation from compressed ppg measurements for continuous vital signs monitoring | |
| CN109567743A (en) | A kind of signal reconfiguring method based on EMD, device, terminal device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200721 |
|
| RJ01 | Rejection of invention patent application after publication |