CN116250844A - Electrocardiosignal noise reduction optimization method and system based on condition generation countermeasure network - Google Patents
Electrocardiosignal noise reduction optimization method and system based on condition generation countermeasure network Download PDFInfo
- Publication number
- CN116250844A CN116250844A CN202310225141.2A CN202310225141A CN116250844A CN 116250844 A CN116250844 A CN 116250844A CN 202310225141 A CN202310225141 A CN 202310225141A CN 116250844 A CN116250844 A CN 116250844A
- Authority
- CN
- China
- Prior art keywords
- noise reduction
- model
- electrocardiosignal
- noise
- deep neural
- 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.)
- Granted
Links
- 230000009467 reduction Effects 0.000 title claims abstract description 220
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000005457 optimization Methods 0.000 title claims abstract description 43
- 238000013528 artificial neural network Methods 0.000 claims abstract description 62
- 238000004364 calculation method Methods 0.000 claims abstract description 34
- 239000012634 fragment Substances 0.000 claims abstract description 11
- 238000010606 normalization Methods 0.000 claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 37
- 230000015654 memory Effects 0.000 claims description 23
- 210000002569 neuron Anatomy 0.000 claims description 23
- 238000012549 training Methods 0.000 claims description 17
- 230000008569 process Effects 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 13
- 230000000694 effects Effects 0.000 claims description 11
- 238000004422 calculation algorithm Methods 0.000 description 8
- 230000002457 bidirectional effect Effects 0.000 description 6
- 238000005070 sampling Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 239000004973 liquid crystal related substance Substances 0.000 description 4
- 238000007635 classification algorithm Methods 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 208000019622 heart disease Diseases 0.000 description 3
- 238000012821 model calculation Methods 0.000 description 3
- 238000011946 reduction process Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000011218 segmentation Effects 0.000 description 2
- 230000009897 systematic effect Effects 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000003745 diagnosis Methods 0.000 description 1
- 238000002565 electrocardiography Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000003205 muscle Anatomy 0.000 description 1
- 230000037081 physical activity Effects 0.000 description 1
- 238000004064 recycling Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
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]
-
- 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]
- A61B5/346—Analysis of electrocardiograms
-
- 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
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- 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
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Signal Processing (AREA)
- Veterinary Medicine (AREA)
- Heart & Thoracic Surgery (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Pathology (AREA)
- Public Health (AREA)
- Medical Informatics (AREA)
- Artificial Intelligence (AREA)
- Cardiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Theoretical Computer Science (AREA)
- Power Engineering (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Fuzzy Systems (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses an electrocardiosignal noise reduction optimization method and system based on a condition generation countermeasure network, comprising the following steps: acquiring an electrocardiosignal, dividing the electrocardiosignal according to the sample length T, and carrying out maximum and minimum normalization processing on the divided data fragments; inputting the processed data segment into a trained electrocardiosignal noise reduction model to obtain a noise-reduced electrocardiosignal; the electrocardiosignal noise reduction model is constructed based on a deep neural network and a condition generation countermeasure network; and determining the optimal values of modeling parameters such as the sample length, the number of layers of the neural network of the encoder and the decoder and the like according to the fitting function curve of the corresponding modeling parameters and the index on the premise of ensuring the noise reduction performance of the model by calculating the noise reduction performance and the calculation complexity ratio index of the model, so that the noise reduction performance and the calculation cost of the noise reduction model are optimal. The invention provides a noise reduction system with better cost performance by providing a noise reduction performance and calculation complexity ratio index.
Description
Technical Field
The invention relates to the technical field of electrocardiosignal noise reduction, in particular to an electrocardiosignal noise reduction optimization method and system based on a condition generation countermeasure network.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
Electrocardiography is a technology for reflecting the heart activity condition by collecting electrophysiological signals at fixed positions on a body surface, and is widely used for diagnosing heart diseases at present and is one of the gold standards for medical diagnosis of heart diseases. However, the electrocardiographic signal (hereinafter referred to as an electrocardiographic signal) is easily interfered by various noises such as physical activity and circuit noise, and thus the electrocardiographic signal is affected for diagnosing heart diseases.
Traditional electrocardiosignal noise reduction methods include: filters, fourier transforms, wavelet decomposition, etc., are generally directed to a single noise type, and the time-frequency characteristics of the noise should be clearly different from that of the electrocardiograph signal itself. With the development of deep learning, deep learning noise reduction algorithms for various noises are more and more, such as: the method comprises the steps of noise reduction, self-encoder, convolution self-encoder and the like, wherein the method comprises the steps of countermeasure noise reduction based on a fully connected network, generation countermeasure noise reduction based on the convolution self-encoder and the like, and the problems that electrocardiosignals in the algorithm need to be divided according to heart beats before noise reduction, mixed noise cannot be processed and the like are solved to a certain extent.
However, the algorithm also has the defects of insufficient potential mining of model learning data distribution, lack of reasonable model, optimization of a system, high model calculation complexity and difficulty in deployment on remote medical equipment.
Such as:
the prior art discloses a method for reducing electrocardiosignal noise based on generation countermeasure network, wherein a generator network mostly adopts network structures such as full connection, convolution and the like, but modeling is carried out by adopting a Convolution Neural Network (CNN) or a bidirectional long and short time memory network (BiLSTM), and due to the lack of an optimization method for modeling of a high-efficiency noise reduction network model, the structure of a model established in the prior art is complex, the scale is large, the computational complexity is high, and the requirement on deployment environment is harsh.
Disclosure of Invention
In order to solve the problems, the invention provides an electrocardiosignal noise reduction optimization method and system based on a condition generation countermeasure network, an electrocardiosignal noise reduction model framework is constructed based on a deep neural network and the condition generation countermeasure network, and simultaneously, an index which takes noise reduction performance and calculation cost into consideration is provided, and systematic optimization design is carried out on the aspects of the internal structure, the learning framework, data segmentation, model complexity and the like of the noise reduction model, so that the noise reduction model has better noise reduction effect while the cost is lower.
In some embodiments, the following technical scheme is adopted:
an electrocardiosignal noise reduction optimization method based on a condition generation countermeasure network comprises the following steps:
acquiring an electrocardiosignal, dividing the electrocardiosignal according to a sample length T, and carrying out maximum and minimum normalization processing on the divided data fragments;
inputting the processed data segment into a trained electrocardiosignal noise reduction model to obtain a noise-reduced electrocardiosignal;
the electrocardiosignal noise reduction model is constructed based on a deep neural network and a condition generation countermeasure network and comprises a generator and a discriminator; for each modeling parameter in the electrocardiosignal noise reduction model, respectively establishing a noise reduction model under different values (usually more than or equal to 3 conditions) of the modeling parameter according to a certain step length to obtain a ratio index (The ratio of Signal-to-NoiseRatio to Computational Complexity, SNR-CC) of the corresponding noise reduction performance and the computational complexity, predicting a corresponding optimal value of each modeling parameter by adopting a fitting function, and determining that the noise reduction performance and the computational cost of the model reach the optimal through verification;
the modeling parameters include: sample length T, number of layers N of deep neural network of encoder and decoder in generator, and number of neurons per layer L x And the number of deep neural network layers M of the discriminator and the number of neurons R of each layer x At least one or more of (a) and (b).
As a further scheme, the generator is composed of a noise reduction self-encoder (DAE) composed of depth neural networks, wherein the DAE comprises an encoder composed of N layers of depth neural networks and a decoder composed of N layers of depth neural networks, and the discriminator in the training process is composed of two classes of depth neural networks composed of M layers of depth neural networks;
the input of the generator is an electrocardiosignal data segment with the length T, and the electrocardiosignal data segment is output as a noise-reduced signal; the discriminator is combined with the generator to perform anti-game learning when the electrocardiosignal noise reduction model is trained.
As a further scheme, the loss function of the generator increases the gap l between the noise-reduced electrocardiosignal and the original electrocardiosignal on the basis of the loss function of the condition generation countermeasure network dist And the maximum local error l between the noise-reduced electrocardiosignal and the original electrocardiosignal max 。
As a further scheme, the ratio index of the noise reduction performance and the computational complexity of the electrocardiosignal noise reduction model is specifically as follows: the ratio of the average signal-to-noise ratio of the noise reduced signal during model testing to the time required by the model to process a single data sample.
As a further scheme, the specific procedure for predicting the optimal value of the sample length T using the fitting function is as follows:
dividing the electrocardiosignal according to the sample length T, and carrying out data normalization to obtain a data set corresponding to the sample length T;
setting values of a plurality of different sample lengths T, and respectively obtaining a data set corresponding to each sample length T;
fixing other modeling parameters unchanged, and respectively modeling and training the noise reduction model according to the set T value and the corresponding data set;
for the trained electrocardiosignal noise reduction model, respectively calculating the noise reduction performance and the calculation complexity ratio index of the model; performing curve fitting according to different sample lengths T and corresponding SNR-CC data to obtain a fitting function f (T), deriving the fitted function to obtain f '(T), and calculating to obtain the value T of T when f' (T) =0 0 . By T 0 And reestablishing a database, and training, testing and verifying the noise reduction model. If the average signal-to-noise ratio of the signal after model noise reduction is not less than the set expected value, the current sample length T 0 I.e. the optimal value of T.
Meanwhile, the modeling parameters can be optimized together with other indexes which possibly influence the deployment of the model, such as: the size of the occupied memory of the model parameters, etc. And combining the modeling parameter optimization process, limiting the occupied memory size of the modeling parameter, obtaining a corresponding piecewise fitting function, and analyzing and predicting the piecewise fitting function to determine the optimal modeling parameter value in the piecewise.
The number of deep neural network layers N for encoder and decoder in the generator, and the number of deep neural network neurons per layer L in the generator x And respectively adopting an optimization method which is the same as the sample length T value, fixing modeling parameters which are not to be optimized, and selecting the optimal value of the corresponding modeling parameters to be optimized.
For the layer number M of the deep neural network in the discriminator and the neuron number R of each layer of the deep neural network in the discriminator x Respectively adopting an optimization method which is the same as the sample length T value, and selecting a corresponding optimal value;
after optimization of the optimized parameters of the deep neural network in the discriminator is completed, the noise reduction model is reconstructed by combining the optimized generator parameters, and the noise reduction effect of the optimized parameters is further verified until the noise reduction performance and the calculation complexity ratio index of the model tend to be stable near the maximum value position.
Selecting different kinds of deep neural networks and condition generation countermeasure networks to construct corresponding electrocardiosignal noise reduction models; and respectively optimizing modeling parameters of the corresponding noise reduction model by adopting the optimization method, comparing the noise reduction performance of the optimized corresponding model with the ratio index of the noise reduction performance and the computational complexity, and selecting the noise reduction model with the optimal performance as a final electrocardiosignal noise reduction model.
In other embodiments, the following technical solutions are adopted:
an electrocardiosignal noise reduction optimization system based on a condition generation countermeasure network, comprising:
the data acquisition module is used for acquiring electrocardiosignals, dividing the electrocardiosignals according to the sample length T, and carrying out maximum and minimum normalization processing on the divided data fragments;
the noise reduction module is used for inputting the processed data fragments into a trained electrocardiosignal noise reduction model to obtain a noise-reduced electrocardiosignal;
the electrocardiosignal noise reduction model is constructed based on a deep neural network and a condition generation countermeasure network and comprises a generator and a discriminator; for each modeling parameter in the electrocardiosignal noise reduction model, selecting different values according to a set step length, respectively establishing a noise reduction model of the modeling parameter under the different values, obtaining a corresponding noise reduction performance and calculation complexity ratio index, and predicting an optimal value of each modeling parameter by adopting a fitting function so as to optimize the noise reduction performance and calculation cost of the model;
the modeling parameters include: sample length T, number of layers N of deep neural network of encoder and decoder in generator, and number of neurons per layer L x And the number of deep neural network layers M of the discriminator and the number of neurons R of each layer x At least one of (a)Or a plurality thereof.
In other embodiments, the following technical solutions are adopted:
a terminal device comprising a processor and a memory, the processor for implementing instructions; the memory is for storing a plurality of instructions adapted to be loaded by the processor and to perform the above-described condition-based generation of an electrocardiographic signal noise reduction optimization method against the network.
Compared with the prior art, the invention has the beneficial effects that:
in summary, an index which is lacking at present and can simultaneously give consideration to the computational complexity and the noise reduction performance and an optimization method which is sufficient, reasonable and systematic for modeling a model according to the index are established and proposed, and the problems that modeling time is long, the computational complexity of the noise reduction model is high, and operation cost is high when the existing noise reduction methods are implemented can be solved.
(1) The invention combines the deep neural network with the condition generation countermeasure network to realize the noise reduction optimization of the electrocardiosignal; the deep neural network can learn the distribution of real data in the sample so as to remove noise; the conditional generation countermeasure network is an improved generation countermeasure network, which combines a noisy signal as a condition with an original input signal (a noise-reduced signal or an original signal) as an input of a discriminator, so that a generator and the discriminator can better perform countermeasure game, and the learning direction of the generator is controlled to be closer to the real data distribution in a sample.
(2) The invention provides a noise reduction performance and calculation complexity ratio index which is used for evaluating the noise reduction performance and calculation cost of a model, curve fitting is carried out by utilizing the provided index and corresponding modeling parameter experimental data, a fitting function is obtained and derived, the optimal value of the modeling parameter is further predicted, and the optimal value of each modeling parameter to be optimized is verified through experiments. The optimization method can utilize a small amount of experiments to realize the combination optimization of model parameters, so that the model has the lowest possible model calculation cost on the premise of achieving the same noise reduction performance.
(3) The loss function of the noise reduction model generator is based on the loss function of a conventional CGAN network, and the distance between the noise reduction signal and the pure electrocardiosignal and the maximum local error are also added. The distance between the noise reduction signal and the pure electrocardiosignal reflects the integral difference between the noise reduction signal and the pure electrocardiosignal, and can control the big direction of the data generated by the generator; the maximum local error between the noise reduction signal and the pure electrocardiosignal represents the local difference to be improved between the noise reduction signal and the pure signal, namely the signal detail which needs to be improved when the generator generates data. When the loss function of the generator is added with the two parts, the data generation process of the generator can be controlled in the whole direction and the local detail, and further, a better noise reduction effect is obtained.
Additional features and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
FIG. 1 is a schematic diagram of an electrocardiosignal noise reduction model optimization process in an embodiment of the invention;
FIG. 2 is a schematic diagram of the overall structure of an electrocardiosignal noise reduction model in an embodiment of the invention;
FIG. 3 is a graph showing a fitted function of modeling parameter T and SNR-CC in an embodiment of the present invention;
fig. 4 is a schematic diagram of an electrocardiosignal noise reduction process based on a condition generation countermeasure network in an embodiment of the invention.
Detailed Description
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
Example 1
In one or more embodiments, an electrocardiosignal noise reduction optimization method based on a condition generation countermeasure network is disclosed, which specifically comprises the following steps:
(1) Acquiring an electrocardiosignal, dividing the electrocardiosignal according to the sample length T, and carrying out maximum and minimum normalization processing on the divided data fragments;
(2) Inputting the processed data segment into a trained electrocardiosignal noise reduction model to obtain a noise-reduced electrocardiosignal;
the electrocardiosignal noise reduction model is constructed based on a deep neural network and a condition generation countermeasure network; and (3) calculating a ratio index of the noise reduction performance and the computational complexity of the electrocardiosignal noise reduction model, predicting an optimal modeling parameter value through curve fitting, and determining that the noise reduction performance and the computational cost of the model reach the optimal value through verification. Wherein the modeling parameters can be the sample length T, the number of layers N of the deep neural network of the encoder and the decoder and the number of neurons L of each layer x The number of deep neural network layers M of the discriminator and the number of neurons R of each layer x One or more of the following.
In the embodiment, with reference to fig. 2, the electrocardiosignal noise reduction model is constructed based on a deep neural network (such as a two-way long-short time memory network, a convolutional neural network, a cyclic neural network and the like) and a condition generation countermeasure network combination; the electrocardiosignal noise reduction model consists of a generator and a discriminator, wherein the generator can be formed by a noise reduction self-encoder (DAE) formed by a depth neural network, the DAE comprises an encoder formed by N layers of depth neural networks and a decoder formed by the N layers of depth neural networks, and the discriminator consists of two types of depth neural networks formed by M layers of depth neural networks.
Noisy signal with length T input to the generator
The output is the noise reduced signal->
The input of the discriminator is the combination of the noise-carrying signal and the noise-reducing signal or the combination of the noise-carrying signal and the original signal, and the output is true or false.
The training and optimizing process of the electrocardiosignal noise reduction model is described below with reference to fig. 1.
a) And (5) adding noise to the signals.
A cleaner electrocardiographic signal is selected from a public or specific electrocardiographic database as the original signal (x n ) Selecting a plurality of common electrocardiographic noises from an electrocardiographic noise database as noise signals (e n ) The sampling frequency of the original signal and the noise signal is f. According to different input signal-to-noise ratios (Signal to Noise Ratio, SNR), respectively superposing multiple single noises and multiple mixed noises with the original signal to obtain a noisy signal
b) And setting a modeling parameter set to be optimized of the model.
The modeling parameter set C to be optimized of the electrocardiosignal noise reduction model comprises the following components: modeling parameters (sample length T, encoder and decoder deep neural network layer number N, neuron number L of each layer are generated x ) The modeling parameters of the arbiter (the number of layers M of the deep neural network of the arbiter, the number of neurons R of each layer x )。
Firstly, the modeling parameters to be optimized of the noise reduction model are preset according to the following rules.
The sample length T determines both the size of the amount of information that the model inputs and the number of start-layer neurons of the generator and the arbiter (i.e., the model start size). Typically, the heart rate is in the range of 50-150 beats/min, i.e. the period of a single beat is 0.4-1.2s, and at a sampling rate f, to ensure that each sample segment contains at least one beat period, it is preferred that T should be selected from a plurality of values in the range of ≡1.2 f.
In order to enable the generator and the arbiter of the noise reduction model to better resist games during training, a better noise reduction effect is achieved. In general, the neural network architecture used by the generator and the arbiter should be similar, and where M should be N or greater. That is, an appropriate N should be selected according to the characteristics of the deep neural network constituting the generator, and then the number of neurons of the corresponding network layer should be set according to the type of the deep neural network selected by selecting M to be a value equal to or greater than N (L x 、R x )。
c) Data segmentation and normalization: for noisy signals according to the sample length (T)
And the original signal (x n ) Dividing to obtain electrocardiosignal data segments with corresponding lengths, normalizing the maximum and minimum values respectively, and forming corresponding data set (D T ). Data set (D) T ) Of these, data representing C% of the total data are used for training and 1-C% of the data are used for testing.
The calculation formula for maximum and minimum normalization of the data is as follows:
wherein x is n For the value of the nth sampling point of the signal to be normalized, x min Is the minimum value, x, in the sampling points of the signal to be normalized max Is the maximum of the signal sample points to be normalized.
d) And (5) model training.
With the prepared dataset (D T ) Through the countermeasure game learning, the constructed electrocardiosignal noise reduction model is trained. When the arbiter cannot distinguish whether the input is the original signal or the noise reduction signal, the countermeasure learning is considered to reach Nash equilibrium, and the model converges at this time.
e) Noise reduction model test: and using a trained generator to input a data set test noisy signal to obtain a noise-reduced signal, and calculating noise reduction performance indexes such as a signal-to-noise ratio (SNR), a root mean square error (Root mean square error, RMSE) and the like of the noise-reduced signal for noise reduction effect evaluation. Then, the parameters such as the parameter quantity, the memory occupied by the parameters, the calculated quantity and the like of the noise reduction model are used for evaluating the calculation consumption (cost) of the model. And the model noise reduction performance indexes and the model calculation cost under different modeling parameter values to be optimized are comprehensively compared, and the modeling parameter values to be optimized are optimized.
In order to facilitate comprehensive comparison of the above noise reduction performance and the calculation complexity index, the present embodiment proposes an index of a ratio of noise reduction performance to calculation complexity (The ratio of Signal-to-Noise Ratio to Computational Complexity, SNR-CC), and the calculation method thereof is as follows:
where O is the value of SNR-CC, representing the ratio between the noise reduction performance of the model and the computational complexity, the larger the index, the smaller the computational time (computational overhead) consumed when the noise reduction signal reaches the same signal-to-noise ratio. SNR of Average Is the average signal-to-noise ratio of the noise reduced signal during the model test. t is t s The method is the time required by the trained noise reduction model (using the generator part) to calculate a single sample, and the comprehensive influence of indexes such as the parameter quantity, the memory occupied by the parameter, the calculated quantity and the like of the noise reduction model is reflected to a certain extent. When SNR is Average Greater than or equal to the desired value S 0 When the SNR-CC of the noise reduction model is larger, the better. Wherein S is 0 The noise reduction performance, i.e., the expected value of the average SNR, for the optimized model can be obtained by referencing other advanced noise reduction algorithm performance indicators.
When the algorithm model architecture is optimized, curve fitting is firstly carried out according to modeling parameters p and SNR-CC corresponding data to obtain a fitting function f (p), and a fitting function diagram of different values of the sample length T and corresponding SNR-CC values is given in FIG. 3. Deriving the fitting function to obtain f' (p). Then, when f' (p) =0, the value p of p is obtained through calculation 0 . A common third order polynomial fitting function and its derivatives are shown below:
O＝f(p)＝ap 3 +bp 2 +cp+d(3)
O′＝f′(p)＝3ap 2 +2bp+c(4)
subsequently, according to the modeling parameter p 0 Modeling again, training and testing, calculating SNR-CC value O corresponding to the SNR value and the parameter 0 And verifying the noise reduction performance and the calculation loss of the model at the moment. That is, when f' (p) =0 and SNR Average ≥S 0 And the model parameter is considered to be the preferred value of the modeling parameter under the current condition of the model.
At this time, if the memory size occupied by the noise reduction model parameters is required, the fitting function f (p) is transformed into a piecewise function according to the specific limitation.
O＝f(p)＝ap 3 +bp 2 +cp+d(SNR Average ≥S 0 MC is less than or equal to MC 0 )(5)
Wherein MC is the memory size occupied by model parameters, MC 0 The maximum value of the memory is occupied for the desired model parameters. When SNR is Average ≥S 0 MC is less than or equal to MC 0 At the time, p is the minimum value that f' (p) can take min Is considered to be the preferred value of the modeling parameter in the current situation of the model.
In addition, a noise reduction model constructed when the modeling parameters to be optimized take the preferred values is adopted, noise reduction processing is carried out on the test data, and then classification performance evaluation is carried out on the original electrocardiosignals before noise reduction, after noise reduction and by utilizing an electrocardiosignal classification algorithm based on the electrocardiosignal characteristics, so that the capability of the noise reduction model for retaining medical value information is evaluated.
f) And optimizing a generator and a discriminator model.
Only the generator part of the trained noise reduction model is used in the noise reduction process, and therefore T, N is two important parameters in the model optimization process. But the value of M must also be optimized because M can affect the performance of the arbiter in response to game play throughout the training process.
Analyzed, T, N, M takes value, and each deep neural network layer is generatedNeuron number (L) x ) And the number of neurons per deep neural network layer (R x ) The noise reduction effect of the model can be greatly affected. Therefore, the model modeling parameters need to be optimized.
The general optimization process of the model modeling parameters is as follows: the current non-optimized modeling parameters are fixed, noise reduction models under different values (usually more than or equal to 3 conditions) of the modeling parameters to be optimized are respectively built according to a certain step length, corresponding noise reduction performance and calculation complexity ratio indexes are obtained according to the steps b) -e), the optimal value of the modeling parameters to be optimized is predicted by using a fitting function, a training noise reduction model is built according to the optimal value, and the comprehensive performance of the noise reduction model is tested and verified. The remaining modeling parameter values are then optimized sequentially in order.
Since the model finally implements the noise reduction process by the generator, the generator model should be optimized first. The method specifically adopts the following scheme: firstly, the modeling parameter T is optimized, and after the optimization is completed, a corresponding data set (D T ) By recycling the above preferred steps, according to N, L x Sequentially, the generator model is sequentially preferred over a range (i.e., N, L x Is a value of (c).
Then, the modeling parameters to be optimized in the discriminators are further optimized according to the same method (M, R x ) Is optimized. The N, M value affects the performance of the generator and the arbiter in the opponent game, so that a certain mutual influence relationship exists between the generator and the arbiter. Therefore, firstly, the optimal value of N is determined in generator optimization, then, after optimization of the optimized parameters of the deep neural network in the discriminator is completed, a noise reduction model is reconstructed together with the optimized generator parameters, and the noise reduction effect of the optimized parameters is further verified until the noise reduction performance and the calculation complexity ratio index of the model tend to be stable at a high position (near the maximum position).
Finally, the preferred T, N, M, L is obtained x 、R x Is used for optimizing a model by a trained noise reduction algorithm.
After training and optimizing the electrocardiosignal noise reduction model, when noise reduction is implemented, the noise reduction is implemented through a generator part in the optimized model; the specific process is as follows:
and (3) dividing the actually acquired electrocardiosignal data by a fixed length according to a preferable T value, and then normalizing the maximum and minimum values of the divided data fragments to obtain the noisy data to be processed. And then inputting the noise signals into a generator part of the trained noise reduction algorithm optimization model for calculation, and further obtaining noise-reduced signals.
Example two
The electrocardiograph noise reduction model of the embodiment is constructed by a bidirectional long short time memory network (bidirectional long short-term memory, biLSTM) and a condition generation countermeasure network CGAN, the model adopts a learning framework of the CGAN, and the model is composed of a generator and a discriminator which are composed of the bidirectional long short time memory network.
The initial model structure is shown in fig. 2, wherein the generator is composed of a 2-layer (N) BiLSTM encoder, a 2-layer (N) BiLSTM decoder and a one-layer fully-connected-layer noise-reducing self-encoder, and the discriminator is composed of a 3-layer (M) BiLSTM, one-layer fully-connected-layer and a two-class deep learning network composed of an activation function Sigmoid. The input sample length of the generator and the discriminator is T, and the number of neurons of the encoder and the decoder in the generator is L x = {400, 200}, the number of neurons in the arbiter is R x = {400, 200, 100}. The formula for calculating the activation function Sigmoid is as follows:
the input of the generator is a noise signal, and the output is a noise-reduced signal; the input of the discriminator is the combination of the noise-contained signal and the noise-reduced signal or the combination of the noise-contained signal and the original signal, and the output is true or false.
In this embodiment, the loss function of the electrocardiographic noise reduction model generator is based on the loss function of the conventional CGAN network, and the difference l between the noise-reduced signal and the pure electrocardiographic signal is also added dist Maximum local error l max 。
The loss function of the generator is specifically:
wherein, the liquid crystal display device comprises a liquid crystal display device,
is a noisy signal, obeys the noisy signal distribution +.>
Is the generator input as noisy signal +.>
Output of time->
Is the arbiter input +.>
And->
Output of time lambda 1 And lambda (lambda) 2 Are respectively l dist And l max Weight coefficients of 0.7 and 0.2, respectively.
Difference l between noise-reduced signal and pure electrocardiosignal dist The calculation formula of (2) is as follows:
wherein, the liquid crystal display device comprises a liquid crystal display device,
for the noise reduced signal, x n Is the original signal, T is the sampleThe length is the same.
Maximum local error l between noise-reduced signal and pure electrocardiosignal max The calculation formula of (2) is as follows:
the overall loss function of the model at this time is:
wherein x is the original signal, obeys the original signal distribution p data (x)；
Is a noisy signal, obeys the noisy signal distribution
The model is realized by Pytorch programming, and parameter updating is performed by adopting an RMSProp (Root Mean Square Prop root mean square transfer) optimizer, and the learning rate is set to be 0.0001. Model training and testing was run on a Dell T640 server (NVIDIA GTX 3090 24GB). The model training and optimizing process is the same as that in the first embodiment, and the specific implementation process is as follows:
a) Electrocardiographic signals denoted by reference numerals 103, 105, 111, 116, 122, 205, 213, 219, 223, 230 are selected as raw signals (x n ). Three common electrocardiosignals including muscle artifact, electrode motion artifact and baseline drift are selected from an electrocardiosignal database to be used as noise signals (e n ). According to the input SNR being respectively set as 0dB, 1dB, 2dB, 3dB, 4dB and 5dB, three single noise, three mixed noise overlapped by two and three mixed noise overlapped by three noise are respectively overlapped with the original signal to obtain noisy signals with different input SNRs and different types of noise
The sampling rate of the electrocardiosignal is 360Hz, so the sample length T is more than or equal to 432 sampling points.
b) According to the optimization strategy, a modeling parameter set C to be optimized is preset. Tentative t=512, n=2, m=3, L of generator according to data feature extraction capability of the BiLSTM network x = {400, 200}, arbiter R x ＝{400，200，100}。
c) Dividing the noisy signal and the original signal with different input SNR and noise according to corresponding T value (512 at this time) to obtain electrocardiosignal data segments with different lengths, normalizing the maximum and minimum values respectively, and forming corresponding data set (D 512 ). Data sets 80% of the data were used for training and 20% of the data were used for testing.
d) The electrocardiograph noise reduction model of the embodiment is trained, and different data sets are adopted to train the model through the countermeasure game learning process, so that a generator and a discriminator are continuously optimized. When the arbiter cannot distinguish whether the input is the original signal or the noise reduction signal or the SNR of the generated noise reduction signal is high enough or the RMSE is low enough, the model converges, and the training is finished.
e) At this time, the noise reduction model constructed with the modeling parameter (t=512) is statistically and calculated as follows: snr=28.71 dB, SNR-cc= 97.32dB/ms. Setting expected noise reduction performance index S of model 0 =28 dB, at which time the model noise reduction performance SNR > S 0 。
f) Generator model optimization: first, taking the preferred example of the value of T, the procedure is as follows: setting a plurality of sample length (T) values (e.g. 512, 1024, 2048, 4096, etc.), obtaining a data set (D) T ) Other modeling parameters are fixed to construct a noise reduction model, and a data set (D T ) The noise reduction model is trained. And then the trained generator is used to input the corresponding data set (D T ) The method is used for testing the noisy signal to obtain a noise-reduced signal, calculating the signal-to-noise ratio (SNR), the SNR-CC noise reduction performance and the calculated overhead index, performing curve fitting according to the T value and the SNR-CC value, and the fitting result is shown in figure 3When f' (T) =0, the result T is fitted 0 For ease of calculation, T is taken here to be =2127 0 =2048. By T 0 After re-modeling, comprehensively comparing various indexes, and enabling SNR (noise reduction performance) of the model to be more than S 0 A better level was reached (snr= 36.21 dB), SNR-cc= 109.74dB/ms, which proved to be less computationally expensive. And then, adopting an electrocardiosignal characteristic-based electrocardiosignal classification algorithm to compare classification performances of the original electrocardiosignals before noise reduction, after noise reduction and evaluate the capability of the noise reduction model for retaining medical value information.
Repeating the above steps, and predicting and optimizing other modeling parameters (N and L) in the generator by comprehensively comparing the noise reduction performance index, the capability of reserving medical information and the calculation cost of the model, namely calculating and analyzing the SNR-CC, modeling parameters and SNR-CC fitting function and derivatives thereof x ). Sequentially determining N and L x Is used for optimizing the value of (a).
Then repeating the steps, and sequentially carrying out M and R by fixing the value of the modeling parameter to be optimized of the generator x The optimization is performed, i.e. the arbiter model is optimized. After determining the preferred value of M, repeating the above steps to perform re-optimization of the N value, so as to determine the validity of N.
The results show that when the range of N is {1,2,3,4} and the range of M is {3,4,5,6}, L is obtained when N=3, M=4, the generator x = {800, 400, 200}, arbiter R x When = {800, 400, 200, 100}, the average SNR of the electrocardiosignal after noise reduction is=45.12 dB, and the SNR-CC of the model is= 131.54dB/ms, at the moment, the calculation consumption of the model is lower than that of other algorithms similar to the noise reduction effect, so that the better balance of the calculation consumption and the noise reduction effect is achieved, namely, the SNR-CC value of the model can be high-order stable at the moment. In addition, an electrocardiosignal characteristic-based electrocardiosignal classification algorithm (support vector machine) is adopted to evaluate four classification performances of abnormal electrocardiosignals before noise reduction, after noise reduction and the original electrocardiosignals, and the classification accuracy of the signals after noise reduction by adopting the model is close to the classification accuracy (more than or equal to 95%) of the original signals, so that a great amount of medical value information is reserved for the signals after noise reduction.
Referring to fig. 4, for the trained electrocardiograph noise reduction model, electrocardiograph signal data is acquired, the electrocardiograph signal data is segmented according to a fixed length of t=2048, and then maximum and minimum normalization is performed on the segmented data segments to obtain noisy electrocardiograph signal segments to be processed. And then inputting the signals into a generator in the electrocardio noise reduction model with optimized training for noise reduction treatment, and obtaining noise-reduced signals.
Example III
In this embodiment, two noise reduction models are respectively constructed:
species (1): generating a noise reduction model constructed by the countermeasure network based on the convolutional neural network and the conditions;
species (2): and generating a noise reduction model constructed by the countermeasure network based on the bidirectional long-short-term memory network and the conditions.
Firstly, based on the same hardware computing platform or system, the two noise reduction models are respectively optimized by adopting the method provided in the first embodiment, and the optimized noise reduction models composed of different depth neural networks are respectively obtained. And then, comparing the noise reduction performance, SNR-CC and other indexes of the two optimized noise reduction models, and determining the better noise reduction model in the two noise reduction models.
The noise reduction model (1) is a model constructed under the optimal modeling parameters after optimization, and the SNR=41.03 dB and the SNR-CC=2.28 dB/ms are tested.
And (2) the noise reduction model is subjected to optimization and then is constructed under the optimal modeling parameters, and the SNR of the model is 45.12dB, and the SNR of the model is-CC= 131.54dB/ms.
Therefore, the noise reduction model constructed based on the bidirectional long-short-time memory network and the condition generation countermeasure network has better noise reduction performance and noise reduction cost performance than the noise reduction model constructed based on the convolutional neural network and the condition generation countermeasure network.
Example IV
In one or more embodiments, an electrocardiosignal noise reduction optimization system based on a condition generation countermeasure network is disclosed, comprising:
the data acquisition module is used for acquiring electrocardiosignals, dividing the electrocardiosignals according to the length T, and carrying out maximum and minimum normalization processing on the divided data fragments;
the noise reduction module is used for inputting the processed data fragments into a trained electrocardiosignal noise reduction model to obtain a noise-reduced electrocardiosignal;
the electrocardiosignal noise reduction model is constructed based on a deep neural network and a condition generation countermeasure network and comprises a generator and a discriminator; for each modeling parameter in the electrocardiosignal noise reduction model, selecting different values according to a set step length, respectively establishing a noise reduction model of the modeling parameter under the different values, obtaining a corresponding noise reduction performance and calculation complexity ratio index, and predicting an optimal value of each modeling parameter by adopting a fitting function so as to optimize the noise reduction performance and calculation cost of the model;
the modeling parameters include: sample length T, number of layers N of deep neural network of encoder and decoder in generator, and number of neurons per layer L x And the number of deep neural network layers M of the discriminator and the number of neurons R of each layer x At least one or more of (a) and (b).
The specific implementation of each module and the first embodiment are described in detail, and will not be described in detail here.
Example five
In one or more embodiments, a terminal device is disclosed, comprising a server comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the condition-based generation of an electrocardiosignal noise reduction optimization method of embodiment one, when the program is executed. For brevity, the description is omitted here.
It should be understood that in this embodiment, the processor may be a central processing unit CPU, and the processor may also be other general purpose processors, digital signal processors DSP, application specific integrated circuits ASIC, off-the-shelf programmable gate array FPGA or other programmable logic device, discrete gate or transistor logic devices, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The memory may include read only memory and random access memory and provide instructions and data to the processor, and a portion of the memory may also include non-volatile random access memory. For example, the memory may also store information of the device type.
In implementation, the steps of the above method may be performed by integrated logic circuits of hardware in a processor or by instructions in the form of software.
While the foregoing description of the embodiments of the present invention has been presented in conjunction with the drawings, it should be understood that it is not intended to limit the scope of the invention, but rather, it is intended to cover all modifications or variations within the scope of the invention as defined by the claims of the present invention.
Claims (10)
1. An electrocardiosignal noise reduction optimization method based on a condition generation countermeasure network is characterized by comprising the following steps:
acquiring an electrocardiosignal, dividing the electrocardiosignal according to a sample length T, and carrying out maximum and minimum normalization processing on the divided data fragments;
inputting the processed data segment into a trained electrocardiosignal noise reduction model to obtain a noise-reduced electrocardiosignal;
the electrocardiosignal noise reduction model is constructed based on a deep neural network and a condition generation countermeasure network and comprises a generator and a discriminator; for each modeling parameter in the electrocardiosignal noise reduction model, selecting different values according to a set step length, respectively establishing a noise reduction model of the modeling parameter under the different values, obtaining a corresponding noise reduction performance and calculation complexity ratio index, and predicting an optimal value of each modeling parameter by adopting a fitting function so as to optimize the noise reduction performance and calculation cost of the model;
the modeling parameters include: sample length T, number of layers N of deep neural network of encoder and decoder in generator, and number of neurons per layer L x Distinguishing deviceThe number of deep neural network layers M and the number of neurons per layer R x At least one or more of (a) and (b).
2. The method for optimizing the noise reduction of an electrocardiograph signal based on a condition generating countermeasure network according to claim 1, wherein the generator is composed of a noise reduction self-encoder composed of a deep neural network, and the noise reduction self-encoder comprises an encoder composed of an N-layer deep neural network and a decoder composed of an N-layer deep neural network; the discriminator consists of two classified deep neural networks consisting of M layers of deep neural networks;
the input of the generator is an electrocardiosignal data segment with the sample length T, and the electrocardiosignal data segment is output as a noise-reduced signal; the discriminator is combined with the generator to perform the countermeasure game learning when the electrocardiosignal noise reduction model is trained.
3. The method for optimizing the noise reduction of an electrocardiograph signal based on a condition generating countermeasure network according to claim 2, wherein the loss function of the generator increases the difference L between the electrocardiograph signal after noise reduction and the original electrocardiograph signal based on the loss function of the condition generating countermeasure network dist And the maximum local error l between the noise-reduced electrocardiosignal and the original electrocardiosignal max 。
4. The method for optimizing the noise reduction of the electrocardiosignal based on the condition generation countermeasure network as claimed in claim 1, wherein the ratio index of the noise reduction performance to the computational complexity of the electrocardiosignal noise reduction model is specifically: the ratio of the average signal-to-noise ratio of the noise reduced signal during model testing to the time required by the model to process a single data sample.
5. The method for optimizing the noise reduction of an electrocardiographic signal based on a condition generating countermeasure network according to claim 1, wherein the specific process of predicting the optimal value of the sample length T by using a fitting function is as follows:
setting values of a plurality of different sample lengths T, and respectively obtaining a data set corresponding to each sample length T;
fixing other modeling parameters unchanged, and respectively modeling and training the noise reduction model according to the set T value and the corresponding data set;
for the trained electrocardiosignal noise reduction model, respectively calculating the noise reduction performance and the calculation complexity ratio index of the model; constructing a fitting function of the sample length T and the ratio index of the noise reduction performance and the computational complexity;
deriving the fitting function, and calculating the value of the corresponding sample length T when the derivative of the fitting function is zero; and performing model noise reduction test based on the value of the sample length T, and if the average signal-to-noise ratio of the noise reduced signal is not less than the set expected value, obtaining the value of the sample length T as an optimal value.
6. The method for optimizing the noise reduction of an electrocardiographic signal based on a condition-generating countermeasure network according to claim 5, wherein the number of layers N of the deep neural network for the encoder and the decoder in the generator, and the number L of neurons per layer of the deep neural network in the generator x And respectively adopting an optimization method which is the same as the sample length T value, and selecting a corresponding optimal value.
7. The method for optimizing the noise reduction of an electrocardiographic signal based on a condition generating countermeasure network according to claim 6, wherein for the number of layers M of the deep neural network in the arbiter, and the number of neurons R of the deep neural network per layer in the arbiter x Respectively adopting an optimization method which is the same as the sample length T value, and selecting a corresponding optimal value;
after optimization of the optimized parameters of the deep neural network in the discriminator is completed, the noise reduction model is reconstructed by combining the optimized generator parameters, and the noise reduction effect of the optimized parameters is further verified until the noise reduction performance and the calculation complexity ratio index of the model tend to be stable near the maximum value position.
8. The method for optimizing the noise reduction of an electrocardiograph signal based on a condition-generating countermeasure network of claim 1, further comprising:
selecting different kinds of deep neural networks and condition generation countermeasure networks to construct corresponding electrocardiosignal noise reduction models; the optimization method of claim 1 is adopted to optimize modeling parameters of the corresponding noise reduction model, noise reduction performance and a ratio index of the noise reduction performance to the computational complexity of the optimized corresponding model are compared, and the noise reduction model with the optimal performance is selected to be used as a final electrocardiosignal noise reduction model.
9. An electrocardiosignal noise reduction optimization system based on a condition generation countermeasure network, which is characterized by comprising:
the data acquisition module is used for acquiring electrocardiosignals, dividing the electrocardiosignals according to the sample length T, and carrying out maximum and minimum normalization processing on the divided data fragments;
the noise reduction module is used for inputting the processed data fragments into a trained electrocardiosignal noise reduction model to obtain a noise-reduced electrocardiosignal;
the electrocardiosignal noise reduction model is constructed based on a deep neural network and a condition generation countermeasure network and comprises a generator and a discriminator; for each modeling parameter in the electrocardiosignal noise reduction model, selecting different values according to a set step length, respectively establishing a noise reduction model of the modeling parameter under the different values, obtaining a corresponding noise reduction performance and calculation complexity ratio index, and predicting an optimal value of each modeling parameter by adopting a fitting function so as to optimize the noise reduction performance and calculation cost of the model;
the setting of the modeling parameters includes: sample length T, number of layers N of deep neural network of encoder and decoder in generator, and number of neurons per layer L x And the number of deep neural network layers M of the discriminator and the number of neurons R of each layer x At least one or more of (a) and (b).
10. A terminal device comprising a processor and a memory, the processor for implementing instructions; the memory for storing a plurality of instructions adapted to be loaded by the processor and to perform the condition-based generation of an electrocardiographic signal noise reduction optimization method of any one of claims 1-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310225141.2A CN116250844B (en) | 2023-03-03 | 2023-03-03 | Electrocardiosignal noise reduction optimization method and system based on condition generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310225141.2A CN116250844B (en) | 2023-03-03 | 2023-03-03 | Electrocardiosignal noise reduction optimization method and system based on condition generation countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116250844A true CN116250844A (en) | 2023-06-13 |
| CN116250844B CN116250844B (en) | 2024-04-26 |
Family
ID=86682353
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310225141.2A Active CN116250844B (en) | 2023-03-03 | 2023-03-03 | Electrocardiosignal noise reduction optimization method and system based on condition generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116250844B (en) |
Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190274108A1 (en) * | 2018-03-02 | 2019-09-05 | DeepSig Inc. | Learning communication systems using channel approximation |
| CN110879254A (en) * | 2018-09-05 | 2020-03-13 | 哈尔滨工业大学 | Steel rail crack acoustic emission signal detection method based on improved least square generation type countermeasure network |
| CN111067507A (en) * | 2019-12-26 | 2020-04-28 | 常熟理工学院 | Electrocardiosignal denoising method based on generation of countermeasure network and strategy gradient |
| CN111564160A (en) * | 2020-04-21 | 2020-08-21 | 重庆邮电大学 | Voice noise reduction method based on AEWGAN |
| WO2020248492A1 (en) * | 2019-06-14 | 2020-12-17 | 平安科技（深圳）有限公司 | Method and device for denoising oct image based on cyclic generative adversarial network |
| CN112487914A (en) * | 2020-11-25 | 2021-03-12 | 山东省人工智能研究院 | ECG noise reduction method based on deep convolution generation countermeasure network |
| CN113963213A (en) * | 2021-10-27 | 2022-01-21 | 上海交通大学 | Method and system for removing antagonistic noise aiming at antagonistic sample of deep neural network |
| CN114129171A (en) * | 2021-12-01 | 2022-03-04 | 山东省人工智能研究院 | Electrocardiosignal noise reduction method based on improved residual error dense network |
| CN114418886A (en) * | 2022-01-19 | 2022-04-29 | 电子科技大学 | Robustness denoising method based on deep convolution self-encoder |
| CN114428234A (en) * | 2021-12-23 | 2022-05-03 | 西安电子科技大学 | Radar high-resolution range profile noise reduction identification method based on GAN and self-attention |
| CN114966860A (en) * | 2022-05-13 | 2022-08-30 | 重庆科技学院 | Seismic data denoising method based on convolutional neural network |
| CN115442191A (en) * | 2022-11-08 | 2022-12-06 | 中国人民解放军战略支援部队航天工程大学 | Communication signal noise reduction method and system based on relative average generation countermeasure network |
| CN115429285A (en) * | 2022-09-26 | 2022-12-06 | 哈尔滨理工大学 | Electrocardiosignal denoising method based on generation countermeasure network |
| WO2023279366A1 (en) * | 2021-07-09 | 2023-01-12 | Oppo广东移动通信有限公司 | Noise reduction method based on transfer learning, terminal device, network device and storage medium |
-
2023
- 2023-03-03 CN CN202310225141.2A patent/CN116250844B/en active Active
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190274108A1 (en) * | 2018-03-02 | 2019-09-05 | DeepSig Inc. | Learning communication systems using channel approximation |
| CN110879254A (en) * | 2018-09-05 | 2020-03-13 | 哈尔滨工业大学 | Steel rail crack acoustic emission signal detection method based on improved least square generation type countermeasure network |
| WO2020248492A1 (en) * | 2019-06-14 | 2020-12-17 | 平安科技（深圳）有限公司 | Method and device for denoising oct image based on cyclic generative adversarial network |
| CN111067507A (en) * | 2019-12-26 | 2020-04-28 | 常熟理工学院 | Electrocardiosignal denoising method based on generation of countermeasure network and strategy gradient |
| CN111564160A (en) * | 2020-04-21 | 2020-08-21 | 重庆邮电大学 | Voice noise reduction method based on AEWGAN |
| CN112487914A (en) * | 2020-11-25 | 2021-03-12 | 山东省人工智能研究院 | ECG noise reduction method based on deep convolution generation countermeasure network |
| WO2023279366A1 (en) * | 2021-07-09 | 2023-01-12 | Oppo广东移动通信有限公司 | Noise reduction method based on transfer learning, terminal device, network device and storage medium |
| CN113963213A (en) * | 2021-10-27 | 2022-01-21 | 上海交通大学 | Method and system for removing antagonistic noise aiming at antagonistic sample of deep neural network |
| CN114129171A (en) * | 2021-12-01 | 2022-03-04 | 山东省人工智能研究院 | Electrocardiosignal noise reduction method based on improved residual error dense network |
| CN114428234A (en) * | 2021-12-23 | 2022-05-03 | 西安电子科技大学 | Radar high-resolution range profile noise reduction identification method based on GAN and self-attention |
| CN114418886A (en) * | 2022-01-19 | 2022-04-29 | 电子科技大学 | Robustness denoising method based on deep convolution self-encoder |
| CN114966860A (en) * | 2022-05-13 | 2022-08-30 | 重庆科技学院 | Seismic data denoising method based on convolutional neural network |
| CN115429285A (en) * | 2022-09-26 | 2022-12-06 | 哈尔滨理工大学 | Electrocardiosignal denoising method based on generation countermeasure network |
| CN115442191A (en) * | 2022-11-08 | 2022-12-06 | 中国人民解放军战略支援部队航天工程大学 | Communication signal noise reduction method and system based on relative average generation countermeasure network |
Non-Patent Citations (2)
| Title |
|---|
| XIAOYU WANG，BINGCHU CHEN，等: "An ECG Signal Denoising Method Using Conditional Generative Adversarial Net", 《IEEE JOURNAL OF BIOMEDICAL AND HEALTH INFORMATICS》, vol. 26, no. 7, 31 July 2022 (2022-07-31) * |
| 王晓宇: "基于生成对抗网络的心电信号降噪算法及性能分析", 《中国优秀硕士学位论文全文数据库（医药卫生科技辑）》, 15 February 2023 (2023-02-15) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116250844B (en) | 2024-04-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109948647B (en) | Electrocardiogram classification method and system based on depth residual error network | |
| US11562222B2 (en) | Systems and methods of identity analysis of electrocardiograms | |
| CN111461176A (en) | Multi-mode fusion method, device, medium and equipment based on normalized mutual information | |
| KR20210037614A (en) | Arrhythmia detection method, device, electronic device and computer storage medium | |
| CN111067507B (en) | Electrocardiosignal denoising method based on generation of countermeasure network and strategy gradient | |
| CN109243604B (en) | Neural network algorithm-based Kawasaki disease risk assessment model construction method and system | |
| CN115470832A (en) | Electrocardiosignal data processing method based on block chain | |
| CN114357237B (en) | Electrocardiosignal and music signal matching method, system, device and medium | |
| CN115363586A (en) | Psychological stress grade assessment system and method based on pulse wave signals | |
| CN111297327A (en) | Sleep analysis method, system, electronic equipment and storage medium | |
| CN113920123B (en) | Addictive brain network analysis method and device | |
| CN116250844B (en) | Electrocardiosignal noise reduction optimization method and system based on condition generation countermeasure network | |
| CN112288718B (en) | Image processing method and apparatus, electronic device, and computer-readable storage medium | |
| WO2021179198A1 (en) | Image feature visualization method, image feature visualization apparatus, and electronic device | |
| CN115064257A (en) | Method, device, medium and equipment for establishing detrusor overactivity diagnosis model | |
| CN115171896A (en) | System and method for predicting long-term death risk of critically ill patient | |
| Shi et al. | Fuzzy neural networks to detect cardiovascular diseases hierarchically | |
| Lu et al. | An end-to-end model for ECG signals classification based on residual attention network | |
| CN112466461A (en) | Medical image intelligent diagnosis method based on multi-network integration | |
| Van Gorp et al. | Aleatoric Uncertainty Estimation of Overnight Sleep Statistics Through Posterior Sampling Using Conditional Normalizing Flows | |
| CN117290709B (en) | Method, system, device and storage medium for continuous dynamic intent decoding | |
| CN114596621B (en) | Tongue picture data processing method and system based on machine vision | |
| CN118016303A (en) | Method and device for predicting occurrence probability of hypotension in hemodialysis | |
| CN112668556B (en) | Breathing sound identification method and system | |
| CN113538451B (en) | Method and device for segmenting magnetic resonance image of deep vein thrombosis, electronic equipment 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |