CN114129171A - Electrocardiosignal noise reduction method based on improved residual error dense network - Google Patents
Electrocardiosignal noise reduction method based on improved residual error dense network Download PDFInfo
- Publication number
- CN114129171A CN114129171A CN202111460457.7A CN202111460457A CN114129171A CN 114129171 A CN114129171 A CN 114129171A CN 202111460457 A CN202111460457 A CN 202111460457A CN 114129171 A CN114129171 A CN 114129171A
- Authority
- CN
- China
- Prior art keywords
- convolution
- residual
- block
- dense network
- noise
- 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
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
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
- A61B5/7207—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal of noise induced by motion artifacts
-
- 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
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- General Health & Medical Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Artificial Intelligence (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Cardiology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Power Engineering (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
An improved residual error dense network-based electrocardiosignal noise reduction method is provided, the residual error dense network has the capability of characteristic reuse, the electrocardiosignal noise reduction is realized, and meanwhile, the calculation cost is reduced. In the process of applying the residual error dense network, parameters do not need to be set manually according to experience, so that experience errors are avoided, and the generalization capability of the model is improved. In the improved residual error dense network, the input of each improved residual error block is fused with the output of all the previous improved residual error blocks; the electrocardiosignal noise is removed through the network, the output of all the improved residual blocks can be obtained, and the characteristic propagation is enhanced; as the network deepens, the problems of gradient messages, gradient explosion and the like can not occur. And meanwhile, the local characteristics and the global characteristics of the signals are considered, so that the local characteristics of the signals can be captured, useful medical characteristics can be stored, and the global characteristics of the signals can be captured, and the training process is stable.
Description
Technical Field
The invention relates to the technical field of ECG signal noise reduction, in particular to an electrocardiosignal noise reduction method based on an improved residual error dense network.
Background
The electrocardiogram is the main basis for cardiovascular disease diagnosis and is an important auxiliary means for cardiovascular doctors to check. The electrocardiosignals contain important medical information such as physiology and pathology, reflect physiological health of all parts of the heart to a certain extent, and are important biomedical signals. However, because the electrocardiosignals have the characteristics of weakness, randomness and the like, and a large amount of noise is often accompanied in clinical application, the electrocardio waveform is easy to distort, the identification of each wave band of the signals is affected, and even the diagnosis of a doctor is affected, so that the removal of the electrocardio noise signals becomes an important link in the electrocardio signal processing.
At present, a great number of researchers contribute to the development of electrocardiosignal noise reduction technology, and corresponding measures are taken for different interferences, such as the application of an empirical mode decomposition method to remove baseline drift signals. However, the electrocardiographic signal often contains a plurality of kinds of noise, and the kind of the noise cannot be determined, and the effect of a single denoising method is not ideal. In the traditional noise reduction method, because the waveforms of the noise signal and part of the electrocardiosignal are mixed, the waveform of the signal is easy to distort, and a large amount of useful information is lost. Some deep learning noise reduction methods, such as a self-encoder and a method for generating an antagonistic network, have the problems of gradient disappearance or gradient explosion, and the like, so that it is difficult to train a deeper network, and the subsequent research of electrocardiosignals is influenced.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides an improved residual error dense network with local feature fusion, global feature fusion and global residual error learning, and the method for denoising the electrocardiosignals by using the network can effectively remove various noises in the electrocardiosignals, avoid information loss and better keep the waveform characteristics of the electrocardiosignals.
The technical scheme adopted by the invention for overcoming the technical problems is as follows:
an electrocardiosignal noise reduction method based on an improved residual error dense network comprises the following steps:
a) selecting 48 original clean electrocardiosignals from an MIT-BIH arrhythmia database, and selecting a baseline drift signal, an electrode motion artifact signal and an electromyographic interference signal from an MIT-BIH noise pressure database as noise signals;
b) slicing the original clean electrocardiosignals at every 512 sampling points to obtain m divided samples, normalizing the sampling points in each sample, and adding noise signals into the normalized clean electrocardiosignals with the signal-to-noise ratio of 5dB to obtain noisy electrocardiosignals;
c) dividing the clean electrocardiosignals and the noisy electrocardiosignals into a training set and a testing set respectively;
d) constructing an improved residual error dense network model;
e) optimizing an improved residual dense network model by a loss function;
f) and carrying out noise reduction treatment on the electrocardiosignals with noise in the test set by using an optimized and improved residual dense network model.
Further, step b) comprises:
b-1) by the formulaCalculating to obtain a Normalized value Normalized (x) of the ith sampling point in the samplei) In the formula xiIs the amplitude, x, of the ith sample point in the sampleminIs the minimum of the amplitude in the sample, xmaxIs the maximum value of the amplitude in the sample;
b-2) sequentially selecting the normalized samples of the front z slices from each signal for 48 original clean electrocardiosignals to form D samples, selecting a baseline drift signal, an electrode motion artifact signal and a myoelectric interference signal from an MIT-BIH noise pressure database to perform slicing processing according to the ratio of 1:1:1, performing normalization processing on the sliced noise samples to form noise samples with sampling points of 512, and randomly extracting D samples from the normalized noise samples to serve as noise signals F;
b-3) calculating to obtain the electrocardiosignal Y with noise through a formula Y ═ X + F, wherein X ═ X1,x2,…,xn)TX is a clean electrocardiosignal, XiIs the amplitude of the ith sampling point in the sample, i is more than or equal to 1 and less than or equal to n, n is the length of the sample, and Y is (Y)1,y2,…,yn)T,yiThe amplitude of the ith sampling point of the electrocardiosignal containing noise is more than or equal to 1 and less than or equal to n, and n is 512.
Further, in the step c), the clean electrocardiosignals and the electrocardiosignals with noise are divided into a training set and a testing set according to the proportion of 80 percent and 20 percent respectively.
Further, step d) comprises the following steps:
d-1) setting a residual dense network model, wherein the residual dense network model is sequentially composed of 2 convolution blocks, 6 MRSB residual blocks, a global-fusion convolution layer and a final convolution layer;
d-2) inputting the noisy electrocardiosignal Y in the training set into a first convolution block through a formula H-1=δ(w-1*Y+b-1) Calculating to obtain the extracted shallow layer characteristic H-1Where δ is the nonlinear activation function, w-1As a convolution filter, b-1For bias, shallow feature H-1Input into a second volume block by formula H0=δ(w0*H-1+b0) Calculating to obtain the extracted shallow layer characteristic H0,w0As a convolution filter, b0Is a deviation;
d-3) shallow feature H0Inputting the result into the 1 st MRSB residual block to obtain the output result H1Shallow feature H0And H1Inputting the spliced result into the 2 nd MRSB residual block to obtain an output result H2Shallow feature H0、H1And H2Inputting the spliced result into the 3 rd MRSB residual block to obtain an output result H3Shallow feature H0、H1、H2And H3After splicing operation, deliveringPutting the data into a 4 th MRSB residual block to obtain an output result H4Shallow feature H0、H1、H2、H3And H4Inputting the spliced result into the 5 th MRSB residual block to obtain an output result H5Shallow feature H0、H1、H2、H3、H4And H5Inputting the spliced result into the 6 th MRSB residual block to obtain an output result H6The MRSB residual block is composed of 10 convolution blocks in sequence, each convolution block is composed of a convolution kernel, a batch normalization process and a nonlinear activation function in sequence, the 2 nd convolution block and the 3 rd convolution block form a first residual block, the 4 th convolution block and the 5 th convolution block form a second residual block, the 6 th convolution block and the 7 th convolution block form a third residual block, the 8 th convolution block and the 9 th convolution block form a fourth residual block, the output result of the 1 st convolution block is input into the first residual block, the output result of the first residual block is input into the second residual block, the output result of the second residual block is input into the third residual block, the output result of the third residual block is input into the fourth residual block, the output result of the first residual block, the output result of the second residual block, the output result of the third residual block and the output result of the fourth residual block are operated, and then the output result of the 10 th 1 x 1 residual block is processed Rolling up blocks to obtain local fusion characteristics;
d-4) reacting H1、H2、H3、H4、H5And H6Inputting the global-fused convolutional layer after splicing operation for global feature fusion to obtain a global feature HGb;
d-5) combining the global features HGbAnd shallow feature H-1Splicing operation to obtain a global residual HGR;
d-6) combining the global residuals HGRInputting a final convolution layer, wherein the final convolution layer is sequentially provided with a convolution layer and a Leaky ReLU activation function structure to obtain a final output electrocardiosignal X after noise reduction*。
Further, in the step d-2), both the convolution blocks are composed of a convolution layer and a leakage ReLU activation function layer, the convolution kernel sizes of both the convolution blocks are set to be 3, the step size is 1, padding is set to be 1, and the dimensions of both the convolution blocks are set to be 32.
Further, the convolution kernel sizes of the 1 st convolution block and the 10 th convolution block in the MRSB residual block in step d-3) are 1 × 1, the step size is 1, the dimension of the convolution kernel is set to 32, the convolution kernel sizes of the 2 nd convolution block to the 9 th convolution block are 1 × 3, the step size is 1, padding is set to 1, and the dimension of the convolution kernel is set to 32.
Further, the convolution layer in step d-4) has a convolution kernel size of 1 × 1.
Further, step d) comprises the following steps:
d-1) by the formulaCalculating to obtain a loss function L, X*The electrocardiosignals after noise reduction are processed by the method,is the amplitude of the ith sampling point of the electrocardiosignal after noise reduction, | · | | survival1Is L1 norm, lambda is parameter, lambda is more than or equal to 0.2 and less than or equal to 0.7;
d-2) setting the initial learning rate to be 0.001, the batch size to be 64, the optimization function to be Adam, the iteration times epoch to be 2000, training the residual error dense network model by using a loss function L, stopping training if the minimum training loss is not updated within continuous 30 epochs, and storing the model and parameters to obtain the optimized and improved residual error dense network model.
Inputting the electrocardiosignals Y with noise concentrated in the test into an optimized and improved residual dense network model to obtain the electrocardiosignals X with noise reduced in the step f)*。
The invention has the beneficial effects that: the residual error dense network has the capability of characteristic reuse, and the calculation cost is reduced while the electrocardiosignal noise reduction is realized. In the process of applying the residual error dense network, parameters do not need to be set manually according to experience, so that experience errors are avoided, and the generalization capability of the model is improved. In the improved residual error dense network, the input of each improved residual error block is fused with the output of all the previous improved residual error blocks; the electrocardiosignal noise is removed through the network, the output of all the improved residual blocks can be obtained, and the characteristic propagation is enhanced; as the network deepens, the problems of gradient messages, gradient explosion and the like can not occur. And meanwhile, the local characteristics and the global characteristics of the signals are considered, so that the local characteristics of the signals can be captured, useful medical characteristics can be stored, and the global characteristics of the signals can be captured, and the training process is stable. The improved residual error dense network is applied to noise reduction of the electrocardiosignals, and noise signals such as baseline drift, electromyographic interference, electrode motion artifacts and the like in the electrocardiosignals can be effectively removed. In the automatic diagnosis of clinical medicine, the network is applied to electrocardiosignal noise reduction, so that the accuracy and the working efficiency of doctors for disease diagnosis can be improved.
Drawings
FIG. 1 is a block diagram of an improved residual block (MRSB) of the present invention;
fig. 2 is a structural diagram of a residual error dense network (RDN) of the present invention.
Detailed Description
The invention will be further explained with reference to fig. 1 and 2.
An electrocardiosignal noise reduction method based on an improved residual error dense network comprises the following steps:
a) selecting 48 original clean electrocardiosignals from an MIT-BIH arrhythmia database, and selecting a baseline drift signal, an electrode motion artifact signal and an electromyographic interference signal from an MIT-BIH noise pressure database as noise signals;
b) slicing the original clean electrocardiosignals at every 512 sampling points to obtain m divided samples, normalizing the sampling points in each sample, and adding noise signals into the normalized clean electrocardiosignals with the signal-to-noise ratio of 5dB to obtain noisy electrocardiosignals;
c) dividing the clean electrocardiosignals and the noisy electrocardiosignals into a training set and a testing set respectively;
d) constructing an improved residual error dense network model;
e) optimizing an improved residual dense network model by a loss function;
f) and carrying out noise reduction treatment on the electrocardiosignals with noise in the test set by using an optimized and improved residual dense network model.
The residual error dense network has the capability of characteristic reuse, and the calculation cost is reduced while the electrocardiosignal noise reduction is realized. In the process of applying the residual error dense network, parameters do not need to be set manually according to experience, so that experience errors are avoided, and the generalization capability of the model is improved. In the improved residual error dense network, the input of each improved residual error block is fused with the output of all the previous improved residual error blocks; the electrocardiosignal noise is removed through the network, the output of all the improved residual blocks can be obtained, and the characteristic propagation is enhanced; as the network deepens, the problems of gradient messages, gradient explosion and the like can not occur. And meanwhile, the local characteristics and the global characteristics of the signals are considered, so that the local characteristics of the signals can be captured, useful medical characteristics can be stored, and the global characteristics of the signals can be captured, and the training process is stable. The improved residual error dense network is applied to noise reduction of the electrocardiosignals, and noise signals such as baseline drift, electromyographic interference, electrode motion artifacts and the like in the electrocardiosignals can be effectively removed. In the automatic diagnosis of clinical medicine, the network is applied to electrocardiosignal noise reduction, so that the accuracy and the working efficiency of doctors for disease diagnosis can be improved.
Example 1:
the step b) comprises the following steps:
b-1) by the formulaCalculating to obtain a Normalized value Normalized (x) of the ith sampling point in the samplei) In the formula xiIs the amplitude, x, of the ith sample point in the sampleminIs the minimum of the amplitude in the sample, xmaxIs the maximum value of the amplitude in the sample;
b-2) sequentially selecting the normalized samples of the front z slices from each signal for 48 original clean electrocardiosignals to form D samples, selecting a baseline drift signal, an electrode motion artifact signal and a myoelectric interference signal from an MIT-BIH noise pressure database to perform slicing processing according to the ratio of 1:1:1, performing normalization processing on the sliced noise samples to form noise samples with sampling points of 512, and randomly extracting D samples from the normalized noise samples to serve as noise signals F;
b-3) calculating to obtain the electrocardiosignal Y with noise through a formula Y ═ X + F, wherein X ═ X1,x2,…,xn)TX is a clean electrocardiosignal, XiIs the amplitude of the ith sampling point in the sample, i is more than or equal to 1 and less than or equal to n, n is the length of the sample, and Y is (Y)1,y2,…,yn)T,yiThe amplitude of the ith sampling point of the electrocardiosignal containing noise is more than or equal to 1 and less than or equal to n, and n is 512.
Example 2:
in the step c), the clean electrocardiosignals and the electrocardiosignals with noises are divided into a training set and a testing set according to the proportion of 80 percent and 20 percent respectively.
Example 3:
the step d) comprises the following steps:
d-1) setting a residual dense network model, wherein the residual dense network model is sequentially composed of 2 convolution blocks, 6 MRSB residual blocks, a global-fusion convolution layer and a final convolution layer;
d-2) inputting the noisy electrocardiosignal Y in the training set into a first convolution block through a formula H-1=δ(w-1*Y+b-1) Calculating to obtain the extracted shallow layer characteristic H-1Where δ is the nonlinear activation function, w-1As a convolution filter, b-1For bias, shallow feature H-1Input into a second volume block by formula H0=δ(w0*H-1+b0) Calculating to obtain the extracted shallow layer characteristic H0,w0As a convolution filter, b0Is a deviation;
d-3) shallow feature H0Inputting the result into the 1 st MRSB residual block to obtain the output result H1Shallow feature H0And H1Inputting the spliced result into the 2 nd MRSB residual block to obtain an output result H2Shallow feature H0、H1And H2Inputting the spliced result into the 3 rd MRSB residual block to obtain an output result H3Shallow feature H0、H1、H2And H3Inputting the spliced result into the 4 th MRSB residual block to obtain an output result H4Shallow feature H0、H1、H2、H3And H4Inputting the spliced result into the 5 th MRSB residual block to obtain an output result H5Shallow feature H0、H1、H2、H3、H4And H5Inputting the spliced result into the 6 th MRSB residual block to obtain an output result H6The MRSB residual block is composed of 10 convolution blocks in sequence, each convolution block is composed of a convolution kernel, a batch normalization process and a nonlinear activation function in sequence, the 2 nd convolution block and the 3 rd convolution block form a first residual block, the 4 th convolution block and the 5 th convolution block form a second residual block, the 6 th convolution block and the 7 th convolution block form a third residual block, the 8 th convolution block and the 9 th convolution block form a fourth residual block, the output result of the 1 st convolution block is input into the first residual block, the output result of the first residual block is input into the second residual block, the output result of the second residual block is input into the third residual block, the output result of the third residual block is input into the fourth residual block, the output result of the first residual block, the output result of the second residual block, the output result of the third residual block and the output result of the fourth residual block are operated, and then the output result of the 10 th 1 x 1 residual block is processed Rolling up blocks to obtain local fusion characteristics;
d-4) reacting H1、H2、H3、H4、H5And H6Inputting the global-fused convolutional layer after splicing operation for global feature fusion to obtain a global feature HGb;
d-5) combining the global features HGbAnd shallow feature H-1Splicing operation to obtain a global residual HGR;
d-6) combining the global residuals HGRInputting the final convolution layer, said finalThe convolution layer is sequentially provided with a convolution layer and a Leaky ReLU activation function structure to obtain the finally output electrocardiosignal X after noise reduction*。
Example 4:
preferably, in step d-2), both convolution blocks are composed of one convolution layer and one leakage ReLU activation function layer, the convolution kernel sizes of both convolution blocks are set to 3, the step size is 1, padding is set to 1, and the dimensions of both convolution blocks are set to 32.
Example 5:
preferably, the convolution kernel sizes of the 1 st convolution block and the 10 th convolution block in the MRSB residual block in step d-3) are 1 × 1, the step size is 1, the dimension of the convolution kernel is set to 32, the convolution kernel sizes of the 2 nd convolution block to the 9 th convolution block are 1 × 3, the step size is 1, padding is set to 1, and the dimension of the convolution kernel is set to 32.
Example 6:
preferably, the convolution kernel size of the convolution layer in step d-4) is 1 × 1.
Example 7:
the step d) comprises the following steps:
d-1) by the formulaCalculating to obtain a loss function L, X*The electrocardiosignals after noise reduction are processed by the method,is the amplitude of the ith sampling point of the electrocardiosignal after noise reduction, | · | | survival1Is L1 norm, lambda is parameter, lambda is more than or equal to 0.2 and less than or equal to 0.7;
d-2) setting the initial learning rate to be 0.001, the batch size to be 64, the optimization function to be Adam, the iteration times epoch to be 2000, and training a residual error dense network model by using a loss function L. Parameters of the model are updated through an optimization function Adam, and after multiple times of training, the training model and the parameters are stored. Each epoch outputs a training signal-to-noise ratio and training loss, as well as the minimum training loss of all epochs present. And (3) according to the signal-to-noise ratio and the loss function data, properly adjusting and learning the model, thereby gradually improving the learning capability of the model, stopping training if the minimum training loss is not updated within continuous 30 epochs, and storing the model and parameters to obtain the optimized and improved residual error dense network model.
Example 8:
inputting the electrocardiosignals Y with noise concentrated in the test into an optimized and improved residual dense network model to obtain the electrocardiosignals X with noise reduced in the step f)*。
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (9)
1. An electrocardiosignal noise reduction method based on an improved residual error dense network is characterized by comprising the following steps:
a) selecting 48 original clean electrocardiosignals from an MIT-BIH arrhythmia database, and selecting a baseline drift signal, an electrode motion artifact signal and an electromyographic interference signal from an MIT-BIH noise pressure database as noise signals;
b) slicing the original clean electrocardiosignals at every 512 sampling points to obtain m divided samples, normalizing the sampling points in each sample, and adding noise signals into the normalized clean electrocardiosignals with the signal-to-noise ratio of 5dB to obtain noisy electrocardiosignals;
c) dividing the clean electrocardiosignals and the noisy electrocardiosignals into a training set and a testing set respectively;
d) constructing an improved residual error dense network model;
e) optimizing an improved residual dense network model by a loss function;
f) and carrying out noise reduction treatment on the electrocardiosignals with noise in the test set by using an optimized and improved residual dense network model.
2. The improved residual error dense network-based electrocardiosignal noise reduction method according to claim 1, wherein the step b) comprises the following steps:
b-1) by the formulaCalculating to obtain a Normalized value Normalized (x) of the ith sampling point in the samplei) In the formula xiIs the amplitude, x, of the ith sample point in the sampleminIs the minimum of the amplitude in the sample, xmaxIs the maximum value of the amplitude in the sample;
b-2) sequentially selecting the normalized samples of the front z slices from each signal for 48 original clean electrocardiosignals to form D samples, selecting a baseline drift signal, an electrode motion artifact signal and a myoelectric interference signal from an MIT-BIH noise pressure database to perform slicing processing according to the ratio of 1:1:1, performing normalization processing on the sliced noise samples to form noise samples with sampling points of 512, and randomly extracting D samples from the normalized noise samples to serve as noise signals F;
b-3) calculating to obtain the electrocardiosignal Y with noise through a formula Y ═ X + F, wherein X ═ X1,x2,…,xn)TX is a clean electrocardiosignal, XiIs the amplitude of the ith sampling point in the sample, i is more than or equal to 1 and less than or equal to n, n is the length of the sample, and Y is (Y)1,y2,…,yn)T,yiThe amplitude of the ith sampling point of the electrocardiosignal containing noise is more than or equal to 1 and less than or equal to n, and n is 512.
3. The improved residual error dense network-based electrocardiosignal noise reduction method according to claim 1, wherein the method comprises the following steps: in the step c), the clean electrocardiosignals and the electrocardiosignals with noises are divided into a training set and a testing set according to the proportion of 80 percent and 20 percent respectively.
4. The improved residual error dense network-based electrocardiosignal noise reduction method according to claim 2, wherein the step d) comprises the following steps:
d-1) setting a residual dense network model, wherein the residual dense network model is sequentially composed of 2 convolution blocks, 6 MRSB residual blocks, a global-fusion convolution layer and a final convolution layer;
d-2) inputting the noisy electrocardiosignal Y in the training set into a first convolution block through a formula H-1=δ(w-1*Y+b-1) Calculating to obtain the extracted shallow layer characteristic H-1Where δ is the nonlinear activation function, w-1As a convolution filter, b-1For bias, shallow feature H-1Input into a second volume block by formula H0=δ(w0*H-1+b0) Calculating to obtain the extracted shallow layer characteristic H0,w0As a convolution filter, b0Is a deviation;
d-3) shallow feature H0Inputting the result into the 1 st MRSB residual block to obtain the output result H1Shallow feature H0And H1Inputting the spliced result into the 2 nd MRSB residual block to obtain an output result H2Shallow feature H0、H1And H2Inputting the spliced result into the 3 rd MRSB residual block to obtain an output result H3Shallow feature H0、H1、H2And H3Inputting the spliced result into the 4 th MRSB residual block to obtain an output result H4Shallow feature H0、H1、H2、H3And H4Inputting the spliced result into the 5 th MRSB residual block to obtain an output result H5Shallow feature H0、H1、H2、H3、H4And H5Inputting the spliced result into the 6 th MRSB residual block to obtain an output result H6The MRSB residual block is composed of 10 convolution blocks in turn, each convolution block is composed of a convolution kernel, a batch normalization process and a nonlinear activation function in turn, and the 2 nd convolution block and the 3 rd convolution block form a first convolution blockA residual module, wherein a second residual module is formed by a 4 th convolution block and a 5 th convolution block, a third residual module is formed by a 6 th convolution block and a 7 th convolution block, a fourth residual module is formed by an 8 th convolution block and a 9 th convolution block, an output result of the 1 st convolution block is input into the first residual module, an output result of the first residual module is input into the second residual module, an output result of the second residual module is input into the third residual module, an output result of the third residual module is input into the fourth residual module, and a result output by the first residual module, an output result of the second residual module, an output result of the third residual module and an output result of the fourth residual module are spliced and then pass through the 10 th 1 × 1 convolution block to obtain local fusion characteristics;
d-4) reacting H1、H2、H3、H4、H5And H6Inputting the global-fused convolutional layer after splicing operation for global feature fusion to obtain a global feature HGb;
d-5) combining the global features HGbAnd shallow feature H-1Splicing operation to obtain a global residual HGR;
d-6) combining the global residuals HGRInputting a final convolution layer, wherein the final convolution layer is sequentially provided with a convolution layer and a LeakyReLU activation function structure to obtain a final output electrocardiosignal X after noise reduction*。
5. The improved residual error dense network-based electrocardiosignal noise reduction method according to claim 4, wherein the method comprises the following steps: in the step d-2), the two convolution blocks are both composed of a convolution layer and a leakage ReLU activation function layer, the convolution kernel sizes of the two convolution blocks are both set to be 3, the step length is 1, padding is set to be 1, and the dimensionalities of the two convolution blocks are both set to be 32.
6. The improved residual error dense network-based electrocardiosignal noise reduction method according to claim 4, wherein the method comprises the following steps: the convolution kernel sizes of the 1 st convolution block and the 10 th convolution block in the MRSB residual block in the step d-3) are 1 × 1, the step size is 1, the dimension of the convolution kernel is set to 32, the convolution kernel sizes of the 2 nd convolution block to the 9 th convolution block are 1 × 3, the step size is 1, padding is set to 1, and the dimension of the convolution kernel is set to 32.
7. The improved residual error dense network-based electrocardiosignal noise reduction method according to claim 4, wherein the method comprises the following steps: the convolution kernel size of the convolution layer in step d-4) is 1 × 1.
8. The improved residual error dense network-based electrocardiosignal noise reduction method according to claim 4, wherein the step d) comprises the following steps:
d-1) by the formulaCalculating to obtain a loss function L, X*The electrocardiosignals after noise reduction are processed by the method,is the amplitude of the ith sampling point of the electrocardiosignal after noise reduction, | · | | survival1Is L1 norm, lambda is parameter, lambda is more than or equal to 0.2 and less than or equal to 0.7;
d-2) setting the initial learning rate to be 0.001, the batch size to be 64, the optimization function to be Adam, the iteration times epoch to be 2000, training the residual error dense network model by using a loss function L, stopping training if the minimum training loss is not updated within continuous 30 epochs, and storing the model and parameters to obtain the optimized and improved residual error dense network model.
9. The method for denoising cardiac signals based on an improved residual error dense network according to claim 8, wherein: inputting the electrocardiosignals Y with noise concentrated in the test into an optimized and improved residual dense network model to obtain the electrocardiosignals X with noise reduced in the step f)*。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111460457.7A CN114129171B (en) | 2021-12-01 | 2021-12-01 | Electrocardiosignal noise reduction method based on improved residual error dense network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111460457.7A CN114129171B (en) | 2021-12-01 | 2021-12-01 | Electrocardiosignal noise reduction method based on improved residual error dense network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114129171A true CN114129171A (en) | 2022-03-04 |
CN114129171B CN114129171B (en) | 2022-06-03 |
Family
ID=80387151
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111460457.7A Active CN114129171B (en) | 2021-12-01 | 2021-12-01 | Electrocardiosignal noise reduction method based on improved residual error dense network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114129171B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115905996A (en) * | 2022-09-19 | 2023-04-04 | 山东省人工智能研究院 | Electrocardiosignal noise reduction method of multi-scale expansion convolution residual error dense network |
CN116250844A (en) * | 2023-03-03 | 2023-06-13 | 山东大学 | Electrocardiosignal noise reduction optimization method and system based on condition generation countermeasure network |
Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180286037A1 (en) * | 2017-03-31 | 2018-10-04 | Greg Zaharchuk | Quality of Medical Images Using Multi-Contrast and Deep Learning |
CN109784242A (en) * | 2018-12-31 | 2019-05-21 | 陕西师范大学 | EEG Noise Cancellation based on one-dimensional residual error convolutional neural networks |
CN110974217A (en) * | 2020-01-03 | 2020-04-10 | 苏州大学 | Dual-stage electrocardiosignal noise reduction method based on convolution self-encoder |
WO2020077232A1 (en) * | 2018-10-12 | 2020-04-16 | Cambridge Cancer Genomics Limited | Methods and systems for nucleic acid variant detection and analysis |
CN111184508A (en) * | 2020-01-19 | 2020-05-22 | 武汉大学 | Electrocardiosignal detection device and analysis method based on joint neural network |
US20200311914A1 (en) * | 2017-04-25 | 2020-10-01 | The Board Of Trustees Of Leland Stanford University | Dose reduction for medical imaging using deep convolutional neural networks |
CN111814656A (en) * | 2020-07-02 | 2020-10-23 | 山东省人工智能研究院 | Electrocardiosignal noise reduction method based on countermeasure generation 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 |
US20210174149A1 (en) * | 2018-11-20 | 2021-06-10 | Xidian University | Feature fusion and dense connection-based method for infrared plane object detection |
CN112991199A (en) * | 2021-02-08 | 2021-06-18 | 西安理工大学 | Image high-low frequency decomposition noise removing method based on residual error dense network |
CN113274031A (en) * | 2021-04-30 | 2021-08-20 | 西安理工大学 | Arrhythmia classification method based on deep convolution residual error network |
CN113436089A (en) * | 2021-06-15 | 2021-09-24 | 山东省人工智能研究院 | ECG noise reduction method based on combination of BilSTM and generation countermeasure network |
CN113440149A (en) * | 2021-07-12 | 2021-09-28 | 齐鲁工业大学 | ECG signal classification method based on twelve-lead electrocardiogram data two-dimensional multi-input residual error neural network |
-
2021
- 2021-12-01 CN CN202111460457.7A patent/CN114129171B/en active Active
Patent Citations (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20180286037A1 (en) * | 2017-03-31 | 2018-10-04 | Greg Zaharchuk | Quality of Medical Images Using Multi-Contrast and Deep Learning |
US20200311914A1 (en) * | 2017-04-25 | 2020-10-01 | The Board Of Trustees Of Leland Stanford University | Dose reduction for medical imaging using deep convolutional neural networks |
WO2020077232A1 (en) * | 2018-10-12 | 2020-04-16 | Cambridge Cancer Genomics Limited | Methods and systems for nucleic acid variant detection and analysis |
US20210174149A1 (en) * | 2018-11-20 | 2021-06-10 | Xidian University | Feature fusion and dense connection-based method for infrared plane object detection |
CN109784242A (en) * | 2018-12-31 | 2019-05-21 | 陕西师范大学 | EEG Noise Cancellation based on one-dimensional residual error convolutional neural networks |
CN110974217A (en) * | 2020-01-03 | 2020-04-10 | 苏州大学 | Dual-stage electrocardiosignal noise reduction method based on convolution self-encoder |
CN111184508A (en) * | 2020-01-19 | 2020-05-22 | 武汉大学 | Electrocardiosignal detection device and analysis method based on joint neural network |
CN111814656A (en) * | 2020-07-02 | 2020-10-23 | 山东省人工智能研究院 | Electrocardiosignal noise reduction method based on countermeasure generation 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 |
CN112991199A (en) * | 2021-02-08 | 2021-06-18 | 西安理工大学 | Image high-low frequency decomposition noise removing method based on residual error dense network |
CN113274031A (en) * | 2021-04-30 | 2021-08-20 | 西安理工大学 | Arrhythmia classification method based on deep convolution residual error network |
CN113436089A (en) * | 2021-06-15 | 2021-09-24 | 山东省人工智能研究院 | ECG noise reduction method based on combination of BilSTM and generation countermeasure network |
CN113440149A (en) * | 2021-07-12 | 2021-09-28 | 齐鲁工业大学 | ECG signal classification method based on twelve-lead electrocardiogram data two-dimensional multi-input residual error neural network |
Non-Patent Citations (8)
Title |
---|
CHEN, CHANGFANG: "ECG Signal Denoising and Reconstruction Based on Basis Pursuit", 《APPLIED SCIENCES-BASEL》 * |
CHEN, CHANGFANG: "ECG Signal Denoising and Reconstruction Based on Basis Pursuit", 《APPLIED SCIENCES-BASEL》, 28 February 2021 (2021-02-28) * |
XU, XUE: "Interpretation of Electrocardiogram (ECG) Rhythm by Combined CNN and BiLSTM", 《IEEE ACCESS》 * |
XU, XUE: "Interpretation of Electrocardiogram (ECG) Rhythm by Combined CNN and BiLSTM", 《IEEE ACCESS》, 31 December 2020 (2020-12-31) * |
李传栋: "基于改进残差密集网络的心律失常自动分类", 《计算机与现代化》 * |
李传栋: "基于改进残差密集网络的心律失常自动分类", 《计算机与现代化》, 15 November 2021 (2021-11-15), pages 106 - 111 * |
李端: "面向智慧医疗的生物电信号分类识别算法研究", 《中国优秀硕士学位论文全文数据库》 * |
李端: "面向智慧医疗的生物电信号分类识别算法研究", 《中国优秀硕士学位论文全文数据库》, 15 October 2018 (2018-10-15) * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115905996A (en) * | 2022-09-19 | 2023-04-04 | 山东省人工智能研究院 | Electrocardiosignal noise reduction method of multi-scale expansion convolution residual error dense network |
CN115905996B (en) * | 2022-09-19 | 2023-08-08 | 山东省人工智能研究院 | Electrocardiosignal noise reduction method for multi-scale expansion convolution residual error dense network |
CN116250844A (en) * | 2023-03-03 | 2023-06-13 | 山东大学 | Electrocardiosignal noise reduction optimization method and system based on condition generation countermeasure network |
CN116250844B (en) * | 2023-03-03 | 2024-04-26 | 山东大学 | Electrocardiosignal noise reduction optimization method and system based on condition generation countermeasure network |
Also Published As
Publication number | Publication date |
---|---|
CN114129171B (en) | 2022-06-03 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111814656B (en) | Electrocardiosignal noise reduction method based on countermeasure generation network | |
CN114129171B (en) | Electrocardiosignal noise reduction method based on improved residual error dense network | |
CN111046824A (en) | Time series signal efficient denoising and high-precision reconstruction modeling method and system | |
CN109949235A (en) | A kind of chest x-ray piece denoising method based on depth convolutional neural networks | |
CN107495959A (en) | A kind of electrocardiosignal sorting technique based on one-dimensional convolutional neural networks | |
CN111481192B (en) | Electrocardiosignal R wave detection method based on improved U-Net | |
CN111067507B (en) | Electrocardiosignal denoising method based on generation of countermeasure network and strategy gradient | |
CN111370120B (en) | Heart diastole dysfunction detection method based on heart sound signals | |
CN114648048B (en) | Electrocardiosignal noise reduction method based on variational self-coding and PixelCNN model | |
CN112806977A (en) | Physiological parameter measuring method based on multi-scale fusion network | |
CN110811591A (en) | Heart failure grading method based on heart rate variability | |
CN112336369B (en) | Coronary heart disease risk index evaluation system of multichannel heart sound signals | |
CN110327034B (en) | Tachycardia electrocardiogram screening method based on depth feature fusion network | |
CN113033358A (en) | Cuff-free blood pressure modeling method based on deep neural network | |
CN116720056A (en) | Enhancement decoding-based AE-GAN ECG signal reconstruction method | |
CN116712083A (en) | QRS complex wave detection method based on U-Net network | |
He et al. | Dual attention convolutional neural network based on adaptive parametric ReLU for denoising ECG signals with strong noise | |
CN111803060B (en) | Electrocardio artifact signal removing method and device | |
CN115363594A (en) | Real-time heart disease screening method based on recurrent neural network | |
CN114757236A (en) | Electroencephalogram signal denoising optimization method and system based on TQWT and SVMD | |
Sheu et al. | Lightweight denoising autoencoder design for noise removal in electrocardiography | |
Srivastava et al. | ECG Pattern Analysis using Artificial Neural Network | |
Tripathi et al. | Statistical analysis of HRV parameters for the detection of Arrhythmia | |
NSVN et al. | Optimal threshold estimation using cultural algorithm for EMD-DWT based ECG denoising | |
EP4111953A1 (en) | Method and device for translating between stochastic signals |
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 |