CN106805965A - A kind of electrocardiosignal sorting technique and device - Google Patents
A kind of electrocardiosignal sorting technique and device Download PDFInfo
- Publication number
- CN106805965A CN106805965A CN201611181530.6A CN201611181530A CN106805965A CN 106805965 A CN106805965 A CN 106805965A CN 201611181530 A CN201611181530 A CN 201611181530A CN 106805965 A CN106805965 A CN 106805965A
- Authority
- CN
- China
- Prior art keywords
- electrocardiosignal
- hidden layer
- data
- carried out
- disaggregated model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- 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
-
- 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]
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Animal Behavior & Ethology (AREA)
- Veterinary Medicine (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Surgery (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Pathology (AREA)
- Cardiology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physiology (AREA)
- Psychiatry (AREA)
- Signal Processing (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The present invention relates to ecg analysis technical field, more particularly to a kind of electrocardiosignal sorting technique and device.The electrocardiosignal sorting technique includes：Step a：Electrocardiosignal is split, training set data and test set data are respectively obtained；Step b：Model training is carried out to the training set data by deep learning, prediction disaggregated model is built；Step c：Electrocardiosignal classification is carried out to the test set data by the prediction disaggregated model.The embodiment of the present invention carries out model training by deep learning to training set, obtain predicting disaggregated model, by predicting that disaggregated model carries out electrocardiosignal classification, reduce the incompleteness that artificial design feature is caused, the accuracy rate of electrocardiosignal classification is improve, while can classify to further types of heart rate variability.
Description
Technical field
The present invention relates to ecg analysis technical field, more particularly to a kind of electrocardiosignal sorting technique and device.
Background technology
Arrhythmia cordis is the common phenomenon in crowd, and serious arrhythmia cordis can immediately threaten the life of the mankind, therefore and
When detect arrhythmia cordis, to prevention heart disease and cardiac sudden death be significant.The P ripples in ecg information
(during Electrocardioscopy, the waveform produced during left atrium depolarization), QPS ripples (the maximum wave group of amplitude in normal ECG) and T
Ripple (ripple occurred after the pause of last wave group represents the multipole of ventricle in case the depolarization of ventricle next time) is electrocardiosignal
Key character ripple, their changing features information is the important evidence of heart pathological analysis and diagnosis.Therefore it is accurate to extract QRS wave
The characteristic information of group is the important foundation of ecg analysis.
At present, there are many electrocardiosignal arrhythmia classification algorithms both at home and abroad, for example：
Li Kunyang etc. detects the QRS wave morphological feature of ecg wave form with mathematical morphology and Wavelet Transformation Algorithm, including
The parameter such as phase, QRS wave time limit between RR, the distinguishing rule of electrocardiogram is distinguished with reference to doctor, and the heart is clapped carries out the four classification dominance hearts
Clap, the supraventricular abnormal heart is clapped, the room sexual abnormality heart is clapped and other hearts are clapped, (provided by Massachusetts Institute Technology using MIT-BIH
Study the database of arrhythmia cordis) arrhythmia cordis database inspection-classification rate of accuracy reached is to 94.2%；
The form and a phase based on ecg wave form such as Philip de Chazal are extracted 12 characteristic vectors, and construction is linear
Disaggregated model is trained and tests, and the heart is clapped and is fallen into 5 types, and test result True Positive Rate is 81.9%；
After Guleral carries out four wavelet transforms of yardstick to electrocardiosignal, using the statistic of wavelet coefficient as
Ecg characteristics parameter, the multi-Layer Perceptron Neural Network combined using two-stage realizes the classification to four class electrocardiosignals, correct recognition rata
Reach 96.94%；
Osowski proposes a kind of fuzzy hybrid neural network classifier, will obscure self-organization layer and MLP (Multi-
Layer Perceptron, multilayer perceptron) network cascaded, seven class electrocardiosignals classified, average correct identification
Rate has reached 96%；
Have what document will be based on Higher-Order Statistics Characteristics and SVM (Support Vector Machine, SVMs)
Neural network classifier, is integrated into the neural network classifier based on Herrnite transform characteristics and SVM by weight votes
13 class arrhythmia signals are carried out Classification and Identification by one expert system；
At home, Cao Yuzhen is empty to feature on the basis of wavelet transform acquisition feature space is carried out to electrocardiogram
The optimization combinations of features scanned for obtaining under different dimensions is asked, by studying the divergence value of these optimization combinations of features with dimension
Variation tendency determine the composition of characteristic vector, finally classified with BP neural network, to the classification of four class electrocardiograms just
True rate reaches 93.9%.
Liu Shixiong proposes to be accurately positioned QRS wave using Wavelet Detection algorithm first to electrocardiosignal, then by each QRS wave
Group extracts 26 characteristic feature composition characteristic vectors, and finally combining the fuzzy clustering method based on object function using addition method enters
Row classification；
Luo Dehan is identified with multistage feed forward-fuzzy control to 6 class electrocardiograms, wherein Second-Order Neural Network
Accuracy reaches 90.57%.
In sum, at least there are the following problems for existing electrocardiosignal arrhythmia classification algorithm：Side based on form
Method is relatively simple and directly perceived, but characteristic value is less, and classification type is limited, and electrocardiogram form is very sensitive to noise.And it is each
The definition of kind transform domain and statistical method to types of arrhythmia is more chaotic, and classification results and effect are also different.And,
The method for also carrying out electrocardiosignal classification not over deep learning in the prior art.
The content of the invention
The invention provides a kind of electrocardiosignal sorting technique and device, it is intended to solve existing electrocardiosignal arrhythmia cordis
Sorting technique classification type is limited, and the definition to types of arrhythmia is chaotic, and carries out electrocardiosignal not over deep learning
The technical problem of classification.
In order to solve the above problems, the invention provides following technical scheme：
A kind of electrocardiosignal sorting technique, including：
Step a：Electrocardiosignal is split, training set data and test set data are respectively obtained；
Step b：Model training is carried out to the training set data by deep learning, prediction disaggregated model is built；
Step c：Electrocardiosignal classification is carried out to the test set data by the prediction disaggregated model.
The technical scheme that the embodiment of the present invention is taken also includes：Segmentation step is being carried out to electrocardiosignal in the step a
Also include before：Denoising is carried out to the electrocardiosignal；It is described denoising is carried out to electrocardiosignal to be specially：By small
Wave conversion carries out denoising to electrocardiosignal.
The technical scheme that the embodiment of the present invention is taken also includes：It is described that denoising tool is carried out to electrocardiosignal by wavelet transformation
Body includes：
Step a1：Lifting Wavelet decomposition is carried out to electrocardiosignal, makes contained noise profile on different decomposition subbands；
Step a2：According to electrocardiosignal and its noise behavior on different decomposition subband, each layer of wavelet coefficient of setting
Weighting coefficient threshold value, weight treatment each layer of wavelet coefficient of setting is carried out to the electrocardiosignal on different decomposition subband；
Step a3：Electrocardiosignal after weight treatment is reconstructed, clean electrocardiosignal is obtained.
The technical scheme that the embodiment of the present invention is taken also includes：It is described that electrocardiosignal is split in the step a
Partitioning scheme be：The crest centered on R crests, the phase between each electrocardiosignal RR of choosing has adopting for the sampled point of data characteristics
Sample frequency and time span are split to electrocardiosignal.
The technical scheme that the embodiment of the present invention is taken also includes：It is described to build prediction disaggregated model tool in the step b
Body is the optimum structure for building the prediction disaggregated model based on depth belief network：When the prediction classification based on depth belief network
When the hidden layer number of model is 1, in input layer, the number of input node is set to 10 different values, by hidden layer
Node number is set to five different values；In to the classification results of the test set data, when searching out accuracy rate highest
Corresponding input layer point number and hidden layer node number, are further added by new hidden layer afterwards, judge node in new hidden layer
Influence of the change of number to classification results, it is determined that optimal hidden layer node number, while determine the hidden layer number of plies, it is final to determine
The optimum structure of the prediction disaggregated model based on depth belief network.
Another technical scheme that the embodiment of the present invention is taken is：A kind of electrocardiosignal sorter, including：
Data segmentation module：For splitting to electrocardiosignal, training set data and test set data are respectively obtained；
Model training module：For carrying out model training to the training set data by deep learning, prediction point is built
Class model；
Modulation recognition module：For the test set data to be carried out with electrocardiosignal point by the prediction disaggregated model
Class.
The technical scheme that the embodiment of the present invention is taken also includes data preprocessing module, and the data preprocessing module is used for
Denoising is carried out to the electrocardiosignal；It is described denoising is carried out to electrocardiosignal to be specially：By wavelet transformation to the heart
Electric signal carries out denoising.
The technical scheme that the embodiment of the present invention is taken also includes：It is described that denoising tool is carried out to electrocardiosignal by wavelet transformation
Body includes：Lifting Wavelet decomposition is carried out to electrocardiosignal, makes contained noise profile on different decomposition subbands；According to difference point
Electrocardiosignal and its noise behavior on solution subband, set each layer of weighting coefficient threshold value of wavelet coefficient, to different decomposition
The electrocardiosignal for taking carries out weight treatment；Electrocardiosignal after weight treatment is reconstructed, clean electrocardiosignal is obtained.
The technical scheme that the embodiment of the present invention is taken also includes：The data preprocessing module is split to electrocardiosignal
Specially：The crest centered on R crests, the phase between each electrocardiosignal RR of choosing has the sample frequency of the sampled point of data characteristics
Electrocardiosignal is split with time span.
The technical scheme that the embodiment of the present invention is taken also includes：The model training module is additionally operable to build based on depth letter
Read the optimum structure of the prediction disaggregated model of network：When the hidden layer number of the prediction disaggregated model based on depth belief network is
When 1, in input layer, the number of input node is set to 10 different values, the node number of hidden layer is set to five
Different values；In to the classification results of the test set data, corresponding input layer point number during accuracy rate highest is searched out
With hidden layer node number, new hidden layer is further added by afterwards, judge that the change of node number in new hidden layer is tied to classification
The influence of fruit, it is determined that optimal hidden layer node number, while determining the hidden layer number of plies, final determination is described to be based on depth conviction net
The optimum structure of the prediction disaggregated model of network.
Relative to prior art, the beneficial effect that the embodiment of the present invention is produced is：The electrocardiosignal of the embodiment of the present invention
After sorting technique to electrocardiosignal by splitting, training set and test set are obtained, training set is carried out by deep learning
Model training, obtains predicting disaggregated model, by predicting that disaggregated model carries out electrocardiosignal classification；The present invention passes through deep learning
Prediction disaggregated model is built, the incompleteness that artificial design feature is caused is reduced, the accuracy rate of electrocardiosignal classification is improve,
Further types of heart rate variability can be classified simultaneously.
Brief description of the drawings
Fig. 1 is the flow chart of the electrocardiosignal sorting technique of the embodiment of the present invention；
Fig. 2 is the structural representation of the electrocardiosignal categorizing system of the embodiment of the present invention；
Fig. 3 is DBN model and BP neural network model accuracy rate comparison diagram.
Specific embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the present invention, not
For limiting the present invention.
Fig. 1 is referred to, is the flow chart of the electrocardiosignal sorting technique of the embodiment of the present invention.The electrocardio of the embodiment of the present invention
Modulation recognition method is comprised the following steps：
Step 100：By sample database select electrocardiosignal data sample, and to select data sample go
Make an uproar treatment；
In step 100, the embodiment of the present invention selects the heart using MIT-BIH arrhythmia cordis database as sample database
The data sample of electric signal, MIT-BIH arrhythmia cordis databases are closed by Massachusetts Institute Technology and Beth Israel hospitals
What work was set up, the database is had altogether comprising 48 electrocardiosignals, there are about 109500 bats, wherein it is the normal heart that about 70% heart is clapped
Clap, remaining is that the abnormal heart is clapped, have 15 kinds of abnormal hearts and clap, it is by least two electrocardiogram expert that each heart is clapped
It is independent by hand to mark.48 electrocardiosignals take from 47 individualities, including 25 male individuals of the age from 32 years old to 89 years old and
22 female individuals of the age from 23 years old to 89 years old, wherein data record number 201 and 202 come from same individuality, data record number
23 data between 100 to 124 are randomly selected from the above-mentioned data sets of Holter, are have the various changes for representing meaning
The waveform and artefact recording mechanism of change；25 data between 200 to 234 are not comprising common but have extremely important clinical picture
Data, including some complicated room property, knot property, SA and conduction abnormalities.Each electrocardiosignal in database
Record includes three files respectively, is respectively the entitled .hea of expansion of header files, the entitled .dat of data file extension, comment file expansion
Open up entitled .air.Header file is used for the name of data file and the attribute for illustrating associate with it, and storage mode is ASCII character word
Symbol, wherein the relevant information of the form including signal, sample frequency, length and this record patient is saved, for example locality,
Conditions of patients, medicining condition etc..Data file is the signal initial data stored with " 212 " form, and " 212 " form is pin
To two data-base recordings of signal, the data of the two signals are alternately stored, and every three bytes store two data, in head text
It is explained in part.Comment file is the result for recording cardiac diagnosis expert to ECG Signal Analysis, mainly including the heart
The information such as jump, the rhythm and pace of moving things and signal quality, are stored with binary system.
The embodiment of the present invention carries out denoising by wavelet transformation to the data sample of electrocardiosignal, and wavelet transformation is Fu
In leaf transformation a new development, retain Fourier transformation analysis characteristic while also compensate for Fourier transformation in non-stationary
Some defects in signal, make to have more flexible, effective characteristic in the analysis of signal.In embodiments of the present invention, small echo
Conversion denoising mode specifically includes following steps：
Step 101：Lifting Wavelet decomposition is carried out to ecg signal data sample, makes contained noise profile in different decomposition
On subband；
In a step 101, electrocardiosignal belongs to body surface bioelectrical signals, and information is mainly distributed between 0.7Hz~45Hz,
It is faint changeable, easily by noise jamming；Noise is derived mainly from：Hz noise, baseline drift and myoelectricity interference.And correctly select small
Ripple basic function, the yardstick of wavelet decomposition, threshold value and its processing mode play pass in the practical application of wavelet threshold denoising algorithm
Key is acted on.The embodiment of the present invention is entered on the basis of analysis electrocardiosignal and noise band distribution to ecg signal data sample
The lifting factorization of 8 yardsticks of the row based on db4 small echos, makes contained noise profile on different decomposition subbands；Wherein, electrocardiosignal
It is mainly distributed in the 3rd~8 layer of wavelet coefficient, myoelectricity interference is mainly distributed in the 1st~5 layer of wavelet coefficient, Hz noise master
It is present in the 2nd layer of high frequency coefficient, baseline drift is primarily present in the 8th layer of low frequency coefficient.
Step 102：According to the characteristics of electrocardiosignal and its noise on different decomposition subband, each layer of wavelet coefficient is set
Weighting coefficient threshold value, weight treatment is carried out to the electrocardiosignal on different decomposition subband；
In a step 102, it is contemplated that P, T ripple information are primarily present in the 5th layer of high frequency coefficient, due to P, T wave amplitude compared with
It is small, the individual features information easy to lose during electrocardiosignal denoising, therefore, by the 5th layer of weighting system in the embodiment of the present invention
Number is set to 0.5, and by the 1st, 2 layers of high frequency coefficient and the 8th layer of low frequency coefficient zero setting so that eliminate Hz noise, baseline drift and
The myoelectricity interference of HFS.
Step 103：Electrocardiosignal after weight treatment is reconstructed, clean ecg signal data sample is obtained.
In step 100, electrocardiosignal denoising can also be the denoising methods such as bandpass filtering, by ecg signal data
Sample carries out denoising, realizes data sample compression to a certain extent, improves the operational efficiency of program.
Step 200：(it, for calculating ventricular rate, is R in two QRS waves that the phase is between RR the phase between choosing each electrocardiosignal RR
Time between ripple) data window of a number of sampled point with data characteristics divided ecg signal data sample
Cut, respectively obtain training set and test set data；
In step 200, the phase of normal person is between the 0.20s of 0.12s mono-, and the width of QRS wave typically exists
Between 0.06s-0.10s, should, 0.11s.Therefore, in ECG's data compression, it is desirable to which choosing can most show the QRS that the heart claps feature
Ripple without by P ripples and ST wave interference, therefore in the embodiment of the present invention centered on R crests crest, choose each electrocardiosignal number
The sample frequency and time span according to the phase between sample RR with 300 sampled points of data characteristics are split to electrocardiosignal,
The compression of data is furthermore achieved that, the operational efficiency of program is improve.Wherein, 100 sampled points are in ECG (electrocardiogram) peak value
In the past, 200 sampled points are after ECG peak values R for R.In MIT-BIH arrhythmia cordis databases in each annotation of electrocardiosignal
All have recorded electrocardiosignal essential information and cardiac diagnosis expert to the result of signal analysis, such as signal duration, lead, the heart
Rate, signal quality, there is position, number and arrhythmia cordis type that the arrhythmia cordis heart is clapped etc., according in MIT-BIH databases this
A little hearts bats for having marked and types of arrhythmia information, the heart that choosing needs is clapped, and is assembled respectively according to heart bat type
Training set and test set, constitute arrhythmia classification sample；Because the electrocardiosignal in MIT-BIH arrhythmia cordis databases is all
Measured using two leads (MLII leads and V5 leads), the electrocardiogram that MLII leads therein are chosen in the embodiment of the present invention is constituted
Arrhythmia classification sample；Sample type, numbering and number are as shown in table 1 below:
The arrhythmia classification sample of table 1
Step 300：Model training is carried out to training set by deep learning, prediction disaggregated model is built；
In step 300, the purpose of deep learning is exactly to be connected by building the neuron in the imitative human brain of pattern die
Structure, when the signals such as picture, audio and word are processed, by entering to above-mentioned data for the different levels of neutral net
Row feature is described, and finally provides the explanation of data.In embodiments of the present invention, the prediction disaggregated model for being built by deep learning
It is the prediction disaggregated model based on DBN (depth belief network).The learning training core of DBN model includes limited Boltzmann machine
Unsupervised autonomous training and BP (backpropagation) algorithm Training.When DBN model is trained, if entered simultaneously
The training that all layers of row whole network, can cause time complexity too high, and if using greedy successively unsupervised learning algorithm
This problem can just be solved.The greedy successively basic thought of unsupervised learning algorithm is to be divided a DBN network model
Layer study, each layer is all unsupervised study, after all-network layer study is finished, then is had to whole DBN network models
The fine setting of supervised learning.
Comprising multiple hidden layers in DBN model, it is the neural network model of generation type.It is more complicated in data relationship
In model construction process, because DBN model includes multiple hidden layers, so it can show more powerful ability.Thus,
In the method for model performance optimization, it becomes possible to carried out using the node number aspect for changing hidden layer number and each hidden layer
Optimization, this is just for model construction provides direction.Then can just be constructed according to different data sets, different application fields
The DBN model of different hidden layer numbers.However, so far, also without DBN model can be determined using unified method
Optimum structure (for example determines node number and Internet number etc.).
Adopt experimentally to determine the optimum structure of DBN model in the embodiment of the present invention.Specifically determination mode is：When
When the hidden layer number of DBN model is 1, in input layer, the number of input node is set to 10 of the change between 1-10
Individual different value；The node number of hidden layer is set to 4,8,12,16 and 20 5 different values.It has been investigated that, it is relative with
The number change of input node, the change of the prediction effect of network to hidden layer node number is more sensitive；To test set number
According to classification results in, search out corresponding input layer point number and hidden layer node number during accuracy rate highest, increase again afterwards
Plus new hidden layer, influence of the change of node number in new hidden layer to prediction effect is judged, so that it is determined that Best knots
Number, while also determining the number of plies of hidden layer.The final optimum structure for determining DBN model.The embodiment of the present invention is by deep
Degree study builds prediction disaggregated model, reduces the incompleteness that artificial design feature is caused, and improves electrocardiosignal classification
Accuracy rate, while can classify to further types of heart rate variability.In other embodiments of the present invention, deep learning builds
Prediction disaggregated model can also be and accumulate the disaggregated models such as network, convolutional neural networks or recirculating network.
Step 400：Electrocardiosignal classification will be carried out in test set data input prediction disaggregated model in step 200, and
Output electrocardiosignal classification type；
In step 400, the electrocardiosignal classification type of output can be compared with the classification type of setting, judges defeated
Whether the electrocardiosignal classification type for going out coincide with the classification type of setting, so as to calculate the accuracy rate of DBN model.
Fig. 2 is referred to, is the structural representation of the electrocardiosignal sorter of the embodiment of the present invention.The embodiment of the present invention
Electrocardiosignal sorter includes data selecting module, data preprocessing module, data segmentation module, model training module and letter
Number sort module.
Data selecting module is used to be selected from sample database the data sample of electrocardiosignal；Wherein, the present invention is implemented
Example selects the data sample of electrocardiosignal, the MIT-BIH rhythms of the heart to lose using MIT-BIH arrhythmia cordis database as sample database
Regular data storehouse is set up cooperatively by Massachusetts Institute Technology and Beth Israel hospitals, and the database is had altogether comprising 48
Electrocardiosignal, there are about 109500 bats, wherein about 70% heart is clapped as the normal heart is clapped, remaining is abnormal heart bat, have 15 kinds it is different
The normal heart is clapped, and it is by the manual independent mark of at least two electrocardiogram expert that each heart is clapped.
The data sample that data preprocessing module is used for the electrocardiosignal to selecting carries out denoising；Wherein, the present invention
Embodiment carries out denoising by wavelet transformation to the data sample of electrocardiosignal；Noise Elimination from Wavelet Transform mode is specially：
1st, Lifting Wavelet decomposition is carried out to ecg signal data sample, makes contained noise profile in different decomposition subbands
On；Wherein, electrocardiosignal belongs to body surface bioelectrical signals, and information is mainly distributed between 0.7Hz~45Hz, faint changeable, easily
By noise jamming；Noise is derived mainly from：Hz noise, baseline drift and myoelectricity interference.And correctly select wavelet basis function, small
The yardstick of Wave Decomposition, threshold value and its processing mode play key effect in the practical application of wavelet threshold denoising algorithm.This hair
Bright embodiment is carried out small based on db4 on the basis of analysis electrocardiosignal and noise band distribution to ecg signal data sample
The lifting factorization of 8 yardsticks of ripple, makes contained noise profile on different decomposition subbands；Wherein, electrocardiosignal is mainly distributed on
In 3rd~8 layer of wavelet coefficient, myoelectricity interference is mainly distributed in the 1st~5 layer of wavelet coefficient, and Hz noise is primarily present in the 2nd
In layer high frequency coefficient, baseline drift is primarily present in the 8th layer of low frequency coefficient.
2nd, according to the characteristics of the electrocardiosignal and its noise on different decomposition subband, each layer of weighting of wavelet coefficient is set
Coefficient threshold, weight treatment is carried out to the electrocardiosignal on different decomposition subband；Wherein, it is contemplated that P, T ripple information are primarily present
In the 5th layer of high frequency coefficient, because P, T wave amplitude are smaller, the individual features information easy to lose during electrocardiosignal denoising, because
This, 0.5 is set in the embodiment of the present invention by the 5th layer of weight coefficient, and by the 1st, 2 layers of high frequency coefficient and the 8th layer of low frequency coefficient
Zero setting, so as to eliminate the myoelectricity interference of Hz noise, baseline drift and HFS.
3rd, the electrocardiosignal after weight treatment is reconstructed, obtains clean ecg signal data sample；Electrocardiosignal
Denoising method can also be the denoising methods such as bandpass filtering, and denoising is carried out by ecg signal data sample, realize
Data sample compression to a certain extent, improves the operational efficiency of program.
Data segmentation module is used to choosing the phase between each electrocardiosignal RR has a number of sampled point of data characteristics
Data window ecg signal data sample is split, respectively obtain training set and test set data；Wherein, normal person
One phase is between the 0.20s of 0.12s mono-, and the width of QRS wave is general between 0.06s-0.10s, should, 0.11s.Therefore, exist
In ECG's data compression, it is desirable to choose the QRS wave that can most show heart bat feature without by P ripples and ST wave interference, therefore the present invention
In embodiment centered on R crests crest, 300 with data characteristics of phase between each ecg signal data sample RR of choosing adopt
The sample frequency and time span of sampling point are split to electrocardiosignal, furthermore achieved that the compression of data, improve program
Operational efficiency.Wherein, before ECG (electrocardiogram) peak values R, 200 sampled points are after ECG peak values R for 100 sampled points.
The essential information and electrocardio of electrocardiosignal are all have recorded in MIT-BIH arrhythmia cordis databases in each annotation of electrocardiosignal
For example signal duration, lead, heart rate, signal quality, there is the position that the arrhythmia cordis heart is clapped to the result of signal analysis in diagnostician
Put, number and arrhythmia cordis type etc., clapped according to the heart that these have been marked in MIT-BIH databases and types of arrhythmia
Information, the heart that choosing needs is clapped, and assembles training set and test set respectively according to heart bat type, constitutes arrhythmia classification sample
This；All it is to be measured using two leads (MLII leads and V5 leads) due to the electrocardiosignal in MIT-BIH arrhythmia cordis databases,
The electrocardiogram that MLII leads therein are chosen in the embodiment of the present invention constitutes arrhythmia classification sample；Sample type, numbering and
Number is as shown in table 1 below:
The arrhythmia classification sample of table 1
Model training module is used to carry out training set model training by deep learning, builds prediction disaggregated model；Its
In, the prediction disaggregated model built by deep learning is the prediction disaggregated model based on DBN (depth belief network).DBN moulds
What the learning training core of type included the unsupervised autonomous training of limited Boltzmann machine and BP (backpropagation) algorithm has supervision
Training.When DBN model is trained, if carrying out the training of all layers of whole network simultaneously, time complexity can be caused too
Height, and can solve this problem if using greedy successively unsupervised learning algorithm.Greedy successively unsupervised learning algorithm
Basic thought be that a DBN network model is carried out into Layered Learning, each layer is all unsupervised study, all-network layer
After study is finished, then the fine setting that supervised learning is carried out to whole DBN network models.
Comprising multiple hidden layers in DBN model, it is the neural network model of generation type.It is more complicated in data relationship
In model construction process, because DBN model includes multiple hidden layers, so it can show more powerful ability.Thus,
In the method for model performance optimization, it becomes possible to carried out using the node number aspect for changing hidden layer number and each hidden layer
Optimization, this is just for model construction provides direction.Then can just be constructed according to different data sets, different application fields
The DBN model of different hidden layer numbers.However, so far, also without DBN model can be determined using unified method
Optimum structure (for example determines node number and Internet number etc.).
Adopt experimentally to determine the optimum structure of DBN model in the embodiment of the present invention.Specifically determination mode is：When
When the hidden layer number of DBN model is 1, in input layer, the number of input node is set to 10 of the change between 1-10
Individual different value；The node number of hidden layer is set to 4,8,12,16 and 20 5 different values.It has been investigated that, it is relative with
The number change of input node, the change of the prediction effect of network to hidden layer node number is more sensitive；To test set number
According to classification results in, search out corresponding input layer point number and hidden layer node number during accuracy rate highest, increase again afterwards
Plus new hidden layer, influence of the change of node number in new hidden layer to prediction effect is judged, so that it is determined that Best knots
Number, while also determining the number of plies of hidden layer.The final optimum structure for determining DBN model.The embodiment of the present invention is by deep
Degree study builds prediction disaggregated model, reduces the incompleteness that artificial design feature is caused, and improves electrocardiosignal classification
Accuracy rate, while can classify to further types of heart rate variability.In other embodiments of the present invention, deep learning builds
Prediction disaggregated model can also be and accumulate the disaggregated models such as network, convolutional neural networks or recirculating network.
Modulation recognition module is used to carry out electrocardiosignal classification during test set data input is predicted into disaggregated model, and exports
Electrocardiosignal classification type；Wherein, the electrocardiosignal classification type that Modulation recognition module can be exported and the classification type for setting
It is compared, judges whether the electrocardiosignal classification type of output coincide with the classification type of setting, so as to calculates DBN model
Accuracy rate.
In order to check the electrocardiosignal classifying quality of the DBN model constructed by the embodiment of the present invention, by the present invention in that with
Same group of test data is compared to DBN model with the structure and prediction effect of BP neural network model.Two models
Prediction effect comparing result is as shown in figure 3, be DBN model and BP neural network model accuracy rate comparison diagram.Can from Fig. 3
Go out, same group of test data is higher than the accuracy rate that the electrocardiosignal that BP model is obtained is classified using DBN model.
The electrocardiosignal sorting technique of the embodiment of the present invention carries out denoising to original electro-cardiologic signals first, then chooses
The sampled point that the phase has data characteristics between each electrocardiosignal RR is split to electrocardiosignal, obtains training set and test set,
Model training is carried out to training set by deep learning, obtains predicting disaggregated model, by predicting that disaggregated model carries out electrocardio letter
Number classification；The present invention builds prediction disaggregated model by deep learning, reduces the incompleteness that artificial design feature is caused, and carries
The accuracy rate of electrocardiosignal classification high, while can classify to further types of heart rate variability；By to electrocardiosignal
Carry out denoising, and the sampled point chosen the phase between each electrocardiosignal RR and have data characteristics is divided electrocardiosignal
Cut, the data sample compression for realizing improves the operational efficiency of program.
The foregoing description of the disclosed embodiments, enables professional and technical personnel in the field to realize or uses the present invention.
Various modifications to these embodiments will be apparent for those skilled in the art, as defined herein
General Principle can be realized in other embodiments without departing from the spirit or scope of the present invention.Therefore, the present invention
The embodiments shown herein is not intended to be limited to, and is to fit to and principles disclosed herein and features of novelty phase one
The scope most wide for causing.
Claims (10)
1. a kind of electrocardiosignal sorting technique, it is characterised in that including：
Step a：Electrocardiosignal is split, training set data and test set data are respectively obtained；
Step b：Model training is carried out to the training set data by deep learning, prediction disaggregated model is built；
Step c：Electrocardiosignal classification is carried out to the test set data by the prediction disaggregated model.
2. electrocardiosignal sorting technique according to claim 1, it is characterised in that to electrocardiosignal in the step a
Also include before segmentation step：Denoising is carried out to the electrocardiosignal；It is described that denoising is carried out to electrocardiosignal
Specially：Denoising is carried out to electrocardiosignal by wavelet transformation.
3. electrocardiosignal sorting technique according to claim 2, it is characterised in that described to be believed electrocardio by wavelet transformation
Number carrying out denoising specifically includes：
Step a1：Lifting Wavelet decomposition is carried out to electrocardiosignal, makes contained noise profile on different decomposition subbands；
Step a2：According to electrocardiosignal and its noise behavior on different decomposition subband, each layer of weighting of wavelet coefficient is set
Coefficient threshold, weight treatment is carried out to the electrocardiosignal on different decomposition subband；
Step a3：Electrocardiosignal after weight treatment is reconstructed, clean electrocardiosignal is obtained.
4. electrocardiosignal sorting technique according to claim 3, it is characterised in that described to electrocardio in the step a
The partitioning scheme that signal is split is：The crest centered on R crests, the phase has data characteristics between choosing each electrocardiosignal RR
Sampled point sample frequency and time span electrocardiosignal is split.
5. electrocardiosignal sorting technique according to claim 1, it is characterised in that in the step b, it is described build it is pre-
Survey disaggregated model and be specially the optimum structure for building the prediction disaggregated model based on depth belief network：When based on depth conviction net
When the hidden layer number of the prediction disaggregated model of network is 1, in input layer, the number of input node is set to 10 differences
Value, five different values are set to by the node number of hidden layer；In to the classification results of the test set data, search out
Corresponding input layer point number and hidden layer node number, are further added by new hidden layer afterwards during accuracy rate highest, judge new
Influence of the change of node number to classification results in hidden layer, it is determined that optimal hidden layer node number, while determining hidden layer
The number of plies, the final optimum structure for determining the prediction disaggregated model based on depth belief network.
6. a kind of electrocardiosignal sorter, it is characterised in that including：
Data segmentation module：For splitting to electrocardiosignal, training set data and test set data are respectively obtained；
Model training module：For carrying out model training to the training set data by deep learning, prediction classification mould is built
Type；
Modulation recognition module：For carrying out electrocardiosignal classification to the test set data by the prediction disaggregated model.
7. electrocardiosignal sorter according to claim 6, it is characterised in that also including data preprocessing module, institute
Data preprocessing module is stated for carrying out denoising to the electrocardiosignal；It is described that to carry out denoising to electrocardiosignal specific
For：Denoising is carried out to electrocardiosignal by wavelet transformation.
8. electrocardiosignal sorter according to claim 7, it is characterised in that described to be believed electrocardio by wavelet transformation
Number carrying out denoising specifically includes：Lifting Wavelet decomposition is carried out to electrocardiosignal, makes contained noise profile in different decomposition subbands
On；According to electrocardiosignal and its noise behavior on different decomposition subband, each layer of weighting coefficient threshold value of wavelet coefficient is set,
Weight treatment is carried out to the electrocardiosignal on different decomposition subband；Electrocardiosignal after weight treatment is reconstructed, is done
Net electrocardiosignal.
9. electrocardiosignal sorter according to claim 8, it is characterised in that the data preprocessing module is to electrocardio
Signal is split specially：The crest centered on R crests, the phase between each electrocardiosignal RR of choosing has the sampling of data characteristics
The sample frequency and time span of point are split to electrocardiosignal.
10. electrocardiosignal sorter according to claim 6, it is characterised in that the model training module is additionally operable to
Build the optimum structure of the prediction disaggregated model based on depth belief network：When the prediction disaggregated model based on depth belief network
Hidden layer number be 1 when, in input layer, the number of input node is set to 10 different values, by the node of hidden layer
Number is set to five different values；In to the classification results of the test set data, correspondence when searching out accuracy rate highest
Input layer point number and hidden layer node number, new hidden layer is further added by afterwards, judge node number in new hidden layer
Influence of the change to classification results, it is determined that optimal hidden layer node number, while determine the hidden layer number of plies, it is final determine it is described
The optimum structure of the prediction disaggregated model based on depth belief network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611181530.6A CN106805965A (en) | 2016-12-19 | 2016-12-19 | A kind of electrocardiosignal sorting technique and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611181530.6A CN106805965A (en) | 2016-12-19 | 2016-12-19 | A kind of electrocardiosignal sorting technique and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106805965A true CN106805965A (en) | 2017-06-09 |
Family
ID=59109402
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611181530.6A Pending CN106805965A (en) | 2016-12-19 | 2016-12-19 | A kind of electrocardiosignal sorting technique and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106805965A (en) |
Cited By (39)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107495959A (en) * | 2017-07-27 | 2017-12-22 | 大连大学 | A kind of electrocardiosignal sorting technique based on one-dimensional convolutional neural networks |
| CN107510452A (en) * | 2017-09-30 | 2017-12-26 | 扬美慧普（北京）科技有限公司 | A kind of ECG detecting method based on multiple dimensioned deep learning neutral net |
| CN107890348A (en) * | 2017-11-21 | 2018-04-10 | 郑州大学 | One kind is based on deep approach of learning electrocardio tempo characteristic automation extraction and sorting technique |
| CN108039203A (en) * | 2017-12-04 | 2018-05-15 | 北京医拍智能科技有限公司 | The detecting system of arrhythmia cordis based on deep neural network |
| CN108597336A (en) * | 2018-02-28 | 2018-09-28 | 天津天堰科技股份有限公司 | Ecg wave form emulation mode |
| CN108647565A (en) * | 2018-03-28 | 2018-10-12 | 浙江工业大学 | A kind of data preprocessing method classified to electrocardiosignal based on deep learning model |
| CN108836307A (en) * | 2018-05-14 | 2018-11-20 | 广东工业大学 | A kind of intelligent ECG detection device, equipment and mobile terminal |
| CN108968941A (en) * | 2018-05-25 | 2018-12-11 | 深圳市太空科技南方研究院 | A kind of arrhythmia detection method, apparatus and terminal |
| CN109033990A (en) * | 2018-07-02 | 2018-12-18 | 河南大学 | CNN model beat classification method based between class distance in class |
| CN109077715A (en) * | 2018-09-03 | 2018-12-25 | 北京工业大学 | A kind of electrocardiosignal automatic classification method based on single lead |
| CN109102848A (en) * | 2018-07-11 | 2018-12-28 | 上海夏先机电科技发展有限公司 | Electrocardiogram (ECG) data management system based on web |
| CN109303561A (en) * | 2018-11-01 | 2019-02-05 | 杭州质子科技有限公司 | It is a kind of to clap the recognition methods clapped with the abnormal heart based on the artifact heart of misclassification and supervised learning |
| CN109363659A (en) * | 2018-09-30 | 2019-02-22 | 平安科技（深圳）有限公司 | Heart rate monitoring method, apparatus and storage medium based on deep learning |
| CN109480825A (en) * | 2018-12-13 | 2019-03-19 | 武汉中旗生物医疗电子有限公司 | The processing method and processing device of electrocardiogram (ECG) data |
| CN109497986A (en) * | 2018-11-22 | 2019-03-22 | 杭州脉流科技有限公司 | Electrocardiogram intelligent analysis method, device, computer equipment and system based on deep neural network |
| CN109567789A (en) * | 2018-12-03 | 2019-04-05 | 东软集团股份有限公司 | Division processing method, device and the readable storage medium storing program for executing of ECG data |
| CN109620152A (en) * | 2018-12-16 | 2019-04-16 | 北京工业大学 | A kind of electrocardiosignal classification method based on MutiFacolLoss-Densenet |
| CN109700450A (en) * | 2018-12-28 | 2019-05-03 | 联想(北京)有限公司 | A kind of heart rate detection method and electronic equipment |
| CN109770861A (en) * | 2019-03-29 | 2019-05-21 | 广州视源电子科技股份有限公司 | Method, device, equipment and storage medium for training and detecting cardioelectric rhythm model |
| CN109864714A (en) * | 2019-04-04 | 2019-06-11 | 北京邮电大学 | A kind of ECG Signal Analysis method based on deep learning |
| CN110141214A (en) * | 2019-04-23 | 2019-08-20 | 首都师范大学 | A kind of mask method of electrocardiogram identification and its application |
| CN110226917A (en) * | 2019-06-26 | 2019-09-13 | 广州视源电子科技股份有限公司 | Electrocardiosignal type detection method and device, computer equipment and storage medium |
| CN111091092A (en) * | 2019-12-16 | 2020-05-01 | 桂林电子科技大学 | Data enhancement method based on optimized convolutional neural network classification |
| WO2020031105A3 (en) * | 2018-08-07 | 2020-06-11 | Goldtech Sino Ltd | Noninvasive systems and methods for continuous hemodynamic monitoring |
| CN111476282A (en) * | 2020-03-27 | 2020-07-31 | 东软集团股份有限公司 | Data classification method and device, storage medium and electronic equipment |
| CN111488793A (en) * | 2020-02-21 | 2020-08-04 | 广州视源电子科技股份有限公司 | Electrocardiosignal classification method and device, electronic equipment and storage medium |
| CN111557660A (en) * | 2020-06-08 | 2020-08-21 | 东北大学 | Arrhythmia identification method under sub-population deep learning framework |
| CN112617855A (en) * | 2020-12-31 | 2021-04-09 | 平安科技（深圳）有限公司 | Electrocardiogram analysis method and device based on federal learning and related equipment |
| WO2021098461A1 (en) * | 2019-11-22 | 2021-05-27 | 华为技术有限公司 | Heart state evaluation method and apparatus, terminal device and storage medium |
| WO2021108950A1 (en) * | 2019-12-02 | 2021-06-10 | 深圳迈瑞生物医疗电子股份有限公司 | Monitoring method, monitoring apparatus, monitoring device, and computer readable storage medium |
| US11089989B2 (en) | 2018-09-14 | 2021-08-17 | Avive Solutions, Inc. | Shockable heart rhythm classifier for defibrillators |
| CN113855042A (en) * | 2021-08-29 | 2021-12-31 | 北京工业大学 | Sequence labeling-based multi-lead electrocardiosignal classification method fusing depth and medical characteristics |
| CN114271836A (en) * | 2022-01-25 | 2022-04-05 | 合肥学院 | Intelligent myoelectricity detection processing method and device based on wavelet transformation |
| CN114521899A (en) * | 2020-11-23 | 2022-05-24 | 武汉心络科技有限公司 | Electrocardiosignal filtering method and device, computer equipment and storage medium |
| WO2022121107A1 (en) * | 2020-12-10 | 2022-06-16 | 青岛大爱慈康智能医疗科技有限公司 | Electrocardiographic signal detection method and device |
| CN115293300A (en) * | 2022-10-09 | 2022-11-04 | 广东技术师范大学 | Arrhythmia classification method and system based on TSK fuzzy semantics |
| CN115470832A (en) * | 2022-11-14 | 2022-12-13 | 南京邮电大学 | Electrocardiosignal data processing method based on block chain |
| CN116369936A (en) * | 2023-05-31 | 2023-07-04 | 深圳市奋达智能技术有限公司 | Electrocardiosignal processing method, system, device and storage medium |
| WO2024061275A1 (en) * | 2022-09-20 | 2024-03-28 | 杭州辉图康科技有限公司 | Surface electrophysiological signal processing method for eliminating signal phase shift and system |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102247143A (en) * | 2011-06-03 | 2011-11-23 | 吉林大学珠海学院 | Integratable fast algorithm for denoising electrocardiosignal and identifying QRS waves |
-
2016
- 2016-12-19 CN CN201611181530.6A patent/CN106805965A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102247143A (en) * | 2011-06-03 | 2011-11-23 | 吉林大学珠海学院 | Integratable fast algorithm for denoising electrocardiosignal and identifying QRS waves |
Non-Patent Citations (1)
| Title |
|---|
| 樊承柱: "基于深度神经网络的心电图自动分类方法研究与实现", 《山东大学硕士学位论文》 * |
Cited By (53)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107495959A (en) * | 2017-07-27 | 2017-12-22 | 大连大学 | A kind of electrocardiosignal sorting technique based on one-dimensional convolutional neural networks |
| CN107510452A (en) * | 2017-09-30 | 2017-12-26 | 扬美慧普（北京）科技有限公司 | A kind of ECG detecting method based on multiple dimensioned deep learning neutral net |
| CN107510452B (en) * | 2017-09-30 | 2019-10-08 | 扬美慧普（北京）科技有限公司 | A kind of ECG detecting method based on multiple dimensioned deep learning neural network |
| CN107890348A (en) * | 2017-11-21 | 2018-04-10 | 郑州大学 | One kind is based on deep approach of learning electrocardio tempo characteristic automation extraction and sorting technique |
| CN108039203A (en) * | 2017-12-04 | 2018-05-15 | 北京医拍智能科技有限公司 | The detecting system of arrhythmia cordis based on deep neural network |
| CN108597336A (en) * | 2018-02-28 | 2018-09-28 | 天津天堰科技股份有限公司 | Ecg wave form emulation mode |
| CN108597336B (en) * | 2018-02-28 | 2021-11-05 | 天津天堰科技股份有限公司 | Electrocardiogram waveform simulation method |
| CN108647565B (en) * | 2018-03-28 | 2021-08-03 | 浙江工业大学 | Data preprocessing method for classifying electrocardiosignals based on deep learning model |
| CN108647565A (en) * | 2018-03-28 | 2018-10-12 | 浙江工业大学 | A kind of data preprocessing method classified to electrocardiosignal based on deep learning model |
| CN108836307A (en) * | 2018-05-14 | 2018-11-20 | 广东工业大学 | A kind of intelligent ECG detection device, equipment and mobile terminal |
| CN108968941A (en) * | 2018-05-25 | 2018-12-11 | 深圳市太空科技南方研究院 | A kind of arrhythmia detection method, apparatus and terminal |
| CN109033990A (en) * | 2018-07-02 | 2018-12-18 | 河南大学 | CNN model beat classification method based between class distance in class |
| CN109033990B (en) * | 2018-07-02 | 2021-08-17 | 河南大学 | CNN model heartbeat classification method based on intra-class and inter-class distances |
| CN109102848A (en) * | 2018-07-11 | 2018-12-28 | 上海夏先机电科技发展有限公司 | Electrocardiogram (ECG) data management system based on web |
| WO2020031105A3 (en) * | 2018-08-07 | 2020-06-11 | Goldtech Sino Ltd | Noninvasive systems and methods for continuous hemodynamic monitoring |
| CN109077715A (en) * | 2018-09-03 | 2018-12-25 | 北京工业大学 | A kind of electrocardiosignal automatic classification method based on single lead |
| CN109077715B (en) * | 2018-09-03 | 2021-09-17 | 北京工业大学 | Electrocardiosignal automatic classification method based on single lead |
| US11089989B2 (en) | 2018-09-14 | 2021-08-17 | Avive Solutions, Inc. | Shockable heart rhythm classifier for defibrillators |
| US11903741B2 (en) | 2018-09-14 | 2024-02-20 | Avive Solutions, Inc. | Shockable heart rhythm classifier for defibrillators |
| CN109363659A (en) * | 2018-09-30 | 2019-02-22 | 平安科技（深圳）有限公司 | Heart rate monitoring method, apparatus and storage medium based on deep learning |
| CN109303561A (en) * | 2018-11-01 | 2019-02-05 | 杭州质子科技有限公司 | It is a kind of to clap the recognition methods clapped with the abnormal heart based on the artifact heart of misclassification and supervised learning |
| CN109497986A (en) * | 2018-11-22 | 2019-03-22 | 杭州脉流科技有限公司 | Electrocardiogram intelligent analysis method, device, computer equipment and system based on deep neural network |
| CN109567789A (en) * | 2018-12-03 | 2019-04-05 | 东软集团股份有限公司 | Division processing method, device and the readable storage medium storing program for executing of ECG data |
| CN109480825A (en) * | 2018-12-13 | 2019-03-19 | 武汉中旗生物医疗电子有限公司 | The processing method and processing device of electrocardiogram (ECG) data |
| CN109480825B (en) * | 2018-12-13 | 2021-08-06 | 武汉中旗生物医疗电子有限公司 | Electrocardio data processing method and device |
| CN109620152A (en) * | 2018-12-16 | 2019-04-16 | 北京工业大学 | A kind of electrocardiosignal classification method based on MutiFacolLoss-Densenet |
| CN109620152B (en) * | 2018-12-16 | 2021-09-14 | 北京工业大学 | MutifacolLoss-densenert-based electrocardiosignal classification method |
| CN109700450A (en) * | 2018-12-28 | 2019-05-03 | 联想(北京)有限公司 | A kind of heart rate detection method and electronic equipment |
| CN109770861A (en) * | 2019-03-29 | 2019-05-21 | 广州视源电子科技股份有限公司 | Method, device, equipment and storage medium for training and detecting cardioelectric rhythm model |
| CN109864714A (en) * | 2019-04-04 | 2019-06-11 | 北京邮电大学 | A kind of ECG Signal Analysis method based on deep learning |
| CN110141214A (en) * | 2019-04-23 | 2019-08-20 | 首都师范大学 | A kind of mask method of electrocardiogram identification and its application |
| CN110226917A (en) * | 2019-06-26 | 2019-09-13 | 广州视源电子科技股份有限公司 | Electrocardiosignal type detection method and device, computer equipment and storage medium |
| WO2021098461A1 (en) * | 2019-11-22 | 2021-05-27 | 华为技术有限公司 | Heart state evaluation method and apparatus, terminal device and storage medium |
| WO2021108950A1 (en) * | 2019-12-02 | 2021-06-10 | 深圳迈瑞生物医疗电子股份有限公司 | Monitoring method, monitoring apparatus, monitoring device, and computer readable storage medium |
| CN111091092A (en) * | 2019-12-16 | 2020-05-01 | 桂林电子科技大学 | Data enhancement method based on optimized convolutional neural network classification |
| CN111488793B (en) * | 2020-02-21 | 2024-04-30 | 广州视源电子科技股份有限公司 | Electrocardiosignal classification method, electrocardiosignal classification device, electronic equipment and storage medium |
| CN111488793A (en) * | 2020-02-21 | 2020-08-04 | 广州视源电子科技股份有限公司 | Electrocardiosignal classification method and device, electronic equipment and storage medium |
| CN111476282A (en) * | 2020-03-27 | 2020-07-31 | 东软集团股份有限公司 | Data classification method and device, storage medium and electronic equipment |
| CN111557660A (en) * | 2020-06-08 | 2020-08-21 | 东北大学 | Arrhythmia identification method under sub-population deep learning framework |
| CN114521899B (en) * | 2020-11-23 | 2024-02-13 | 武汉心络科技有限公司 | Electrocardiosignal filtering method, electrocardiosignal filtering device, computer equipment and storage medium |
| CN114521899A (en) * | 2020-11-23 | 2022-05-24 | 武汉心络科技有限公司 | Electrocardiosignal filtering method and device, computer equipment and storage medium |
| WO2022121107A1 (en) * | 2020-12-10 | 2022-06-16 | 青岛大爱慈康智能医疗科技有限公司 | Electrocardiographic signal detection method and device |
| CN112617855A (en) * | 2020-12-31 | 2021-04-09 | 平安科技（深圳）有限公司 | Electrocardiogram analysis method and device based on federal learning and related equipment |
| CN113855042B (en) * | 2021-08-29 | 2023-10-03 | 北京工业大学 | Sequence labeling-based multi-lead electrocardiosignal classification method integrating depth and medical features |
| CN113855042A (en) * | 2021-08-29 | 2021-12-31 | 北京工业大学 | Sequence labeling-based multi-lead electrocardiosignal classification method fusing depth and medical characteristics |
| CN114271836B (en) * | 2022-01-25 | 2023-08-29 | 合肥学院 | Intelligent myoelectricity detection processing method and device based on wavelet transformation |
| CN114271836A (en) * | 2022-01-25 | 2022-04-05 | 合肥学院 | Intelligent myoelectricity detection processing method and device based on wavelet transformation |
| WO2024061275A1 (en) * | 2022-09-20 | 2024-03-28 | 杭州辉图康科技有限公司 | Surface electrophysiological signal processing method for eliminating signal phase shift and system |
| CN115293300A (en) * | 2022-10-09 | 2022-11-04 | 广东技术师范大学 | Arrhythmia classification method and system based on TSK fuzzy semantics |
| CN115470832B (en) * | 2022-11-14 | 2023-05-02 | 南京邮电大学 | Electrocardiosignal data processing method based on block chain |
| CN115470832A (en) * | 2022-11-14 | 2022-12-13 | 南京邮电大学 | Electrocardiosignal data processing method based on block chain |
| CN116369936A (en) * | 2023-05-31 | 2023-07-04 | 深圳市奋达智能技术有限公司 | Electrocardiosignal processing method, system, device and storage medium |
| CN116369936B (en) * | 2023-05-31 | 2023-10-03 | 深圳市奋达智能技术有限公司 | Electrocardiosignal processing method, system, device and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106805965A (en) | A kind of electrocardiosignal sorting technique and device | |
| CN107822622B (en) | Electrocardiogram diagnosis method and system based on deep convolutional neural network | |
| Çınar et al. | Classification of normal sinus rhythm, abnormal arrhythmia and congestive heart failure ECG signals using LSTM and hybrid CNN-SVM deep neural networks | |
| Berkaya et al. | A survey on ECG analysis | |
| CN111449645B (en) | Intelligent classification and identification method for electrocardiogram and heartbeat | |
| CN106725428A (en) | A kind of electrocardiosignal sorting technique and device | |
| Vijayavanan et al. | Automatic classification of ECG signal for heart disease diagnosis using morphological features | |
| CN111053549A (en) | Intelligent biological signal abnormality detection method and system | |
| Zhang et al. | ECG signal classification with deep learning for heart disease identification | |
| Daqrouq et al. | Neural network and wavelet average framing percentage energy for atrial fibrillation classification | |
| Emanet | ECG beat classification by using discrete wavelet transform and Random Forest algorithm | |
| CN107239684A (en) | A kind of feature learning method and system for ECG identifications | |
| Diker et al. | A diagnostic model for identification of myocardial infarction from electrocardiography signals | |
| CN115530788A (en) | Arrhythmia classification method based on self-attention mechanism | |
| Jambukia et al. | ECG beat classification using machine learning techniques | |
| Al Rahhal et al. | Automatic premature ventricular contractions detection for multi-lead electrocardiogram signal | |
| CN110974213A (en) | Electrocardiosignal identification method based on deep stack network | |
| Oleiwi et al. | Efficient ECG Beats Classification Techniques for The Cardiac Arrhythmia Detection Based on Wavelet Transformation. | |
| CN113855037A (en) | Atrial fibrillation identification method and device based on Transformer | |
| Bashar et al. | Heart abnormality classification using phonocardiogram (PCG) signals | |
| Rai et al. | Classification of ECG waveforms for abnormalities detection using DWT and back propagation algorithm | |
| Ebrahimpoor et al. | Premature Ventricular Contractions Detection by Multi-Domain Feature Extraction and Auto-Encoder-based Feature Reduction | |
| Mohseni et al. | Heart arrhythmias classification via a sequential classifier using neural network, principal component analysis and heart rate variation | |
| Wen et al. | ECG beat classification using GreyART network | |
| Shafaatfar et al. | Improving the Efficiency of Automatic Cardiac Arrhythmias Classification by a Novel Patient-Specific Feature Space Mapping |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170609 |
|
| RJ01 | Rejection of invention patent application after publication |