CN106725426A - A kind of method and system of electrocardiosignal classification - Google Patents
A kind of method and system of electrocardiosignal classification Download PDFInfo
- Publication number
- CN106725426A CN106725426A CN201611150154.4A CN201611150154A CN106725426A CN 106725426 A CN106725426 A CN 106725426A CN 201611150154 A CN201611150154 A CN 201611150154A CN 106725426 A CN106725426 A CN 106725426A
- Authority
- CN
- China
- Prior art keywords
- oscillogram
- data
- model
- test data
- electrocardiogram
- 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
- A61B5/24—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
-
- 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)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Veterinary Medicine (AREA)
- Public Health (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Heart & Thoracic Surgery (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Signal Processing (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Fuzzy Systems (AREA)
- Cardiology (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The present invention relates to a kind of method and system of electrocardiosignal classification.Method includes：Obtain electrocardio test data, electrocardio test data is processed to obtain the oscillogram of preliminary dimension, with reference to the oscillogram and the heart disease disaggregated model set up of preliminary dimension, determine the classification results corresponding to the electrocardio test data, wherein, heart disease disaggregated model is the model that can reflect original waveform feature and classification results.Using method of the invention, it is possible to improve the accuracy rate of classification results, while can realize classifying further types of heart rate variability.
Description
Technical field
The present invention relates to technical field of data processing, more particularly to a kind of method and system of electrocardiosignal classification.
Background technology
Electrocardiogram (Electrocardiograph, ECG) is a kind of data as measured by electronic equipment out, for retouching
Paint the whole process of cardiomotility, including the recovery process of excited generation, excited conduction and excitement potential change feelings
The curve of condition, mainly includes the multipole of depolarization, the multipole in atrium, the depolarization of ventricle and the ventricle in the atrium of heart.
Holter, also known as Holter.It is the present age most important noninvasive cardioelectric monitor technology.It is recordable lower 24 hours
Ecg data, display monitored in 24 hours during total heart beats, HRmax, HR min, average heart rate and per small
When heart rate；Display is whole or supraventricular, VPB and the dystopy tachycardia that calculate by the hour, the situation of cardiac arrest；Ⅰ、
IIth, change of III Aminophyline and ST-T etc..The Two-dimensional morphology of Holter is the heart represented with image format
Electrograph, is also to be engaged in the electrocardiogram form that the professional person in the fields such as medical treatment, biological study is analyzed.
The automatic detection of current electrocardiogram and the research of identification are essentially all to be present in theoretical field and exploration field, and
Effect is unsatisfactory in the actual diagnosis of angiocardiopathy.This be primarily due to previous researcher research electrocardiogram from
When dynamic detection and identification, doctor is have ignored when cardiovascular disease diagnosis are carried out to the concern of electrocardiogram morphological feature.
Therefore, in the urgent need to it is a kind of can accurately automatic detection and identification heart disease classification method, to improve study
Classification capacity and adaptive stress, the preferably diagnosis of auxiliary related cardiac conditions.
The content of the invention
Based on this, it is necessary to regarding to the issue above, there is provided it is a kind of can electrocardiosignal classification method and system, can be direct
On the basis of by original electro-cardiologic signals, the classification of heart disease is accurately realized, it is possible to which further types of heart rate variability is entered
Row classification.
A kind of method of electrocardiosignal classification, the described method comprises the following steps：
Obtain electrocardio test data；
The electrocardio test data is processed to obtain the oscillogram of preliminary dimension；
With reference to the standard-sized oscillogram and the heart disease disaggregated model set up, the electrocardio test is determined
Classification results corresponding to data, wherein, the heart disease disaggregated model is can to reflect that original waveform feature is tied with classification
The model of fruit.
A kind of system of electrocardiosignal classification, the system includes
Data acquisition module, data processing module and result determining module, wherein：
The data acquisition module is used to obtain electrocardio test data；
The data processing module is used to that the electrocardio test data to be processed to obtain the oscillogram of preliminary dimension；
The result determining module is used for the heart disease classification for combining the standard-sized oscillogram and having set up
Model, determines the classification results corresponding to the electrocardio test data, wherein, the heart disease disaggregated model is to reflect
The model of original waveform feature and classification results.
The method and system of above-mentioned electrocardiosignal classification, obtain electrocardio test data, and electrocardio test data is processed
To obtain the oscillogram of preliminary dimension, with reference to the oscillogram and the heart disease disaggregated model set up of preliminary dimension, it is determined that
Classification results corresponding to the electrocardio test data, wherein, heart disease disaggregated model is can to reflect original waveform feature
With the model of classification results.Heart disease disaggregated model of the invention is to carry out learning training by electrocardio test sample data
Obtained from, can farthest reflect original electrocardiographicdigital wave character and heart disease classification results.Therefore, based on the present invention
The heart disease disaggregated model of structure, it is possible to increase the accuracy rate of classification results, while can realize to further types of
Heart rate variability is classified.
Brief description of the drawings
Fig. 1 is the flow chart of the method for an embodiment center telecommunications number classification；
Fig. 2 is a kind of flow chart of the method for building heart disease disaggregated model in an embodiment；
Fig. 3 is the flow chart for sample data process the implementation method for obtaining electrocardiogram training set in an embodiment；
Fig. 4 is single heartbeat image coordinate system schematic diagram in an embodiment；
Fig. 5 is convolutional neural networks model framework schematic diagram in an embodiment；
Fig. 6 is the structural representation of the system of an embodiment center telecommunications number classification；
Fig. 7 is the structural representation of model construction module in an embodiment.
Specific embodiment
In one embodiment, a kind of method of electrocardiosignal classification, as shown in figure 1, comprising the following steps：
Step S101：Obtain electrocardio test data.
Here electrocardio test data, refers to the primary data obtained by device measuring.
Step S102：Electrocardio test data is processed to obtain the oscillogram of preliminary dimension.
Wherein, in a kind of possible implementation of the embodiment of the present invention, the oscillogram of preliminary dimension for height and width all
It is the oscillogram of 28 pixels, certainly, in other possible implementations, is also not limited to the size, only mays other
Can there is certain difference in the result that possible size is likely to be obtained.In present invention below embodiment, it is with preliminary dimension mainly
Height and width are all to illustrate as a example by 28 pixels.
In one of which implementation of the invention, the R ripples position of heartbeat each time is determined by wavelet transformation,
To split to electrocardiosignal, it is all the oscillogram of 28 pixels to obtain height and width.
Wavelet transformation (wavelet transform, WT) is a kind of new transform analysis method, and it inherits and has developed short
When Fourier transform localization thought, while the shortcomings of overcoming window size again and do not change with frequency, using the teaching of the invention it is possible to provide one
It is the ideal tools for carrying out signal time frequency analysis and treatment with " T/F " window of frequency shift.Being mainly for it passes through
Conversion can abundant feature of the outstanding problem in terms of some, the localization of time (space) frequency can be analyzed, put down by flexible
Movement is calculated and progressively carries out multi-scale refinement to signal (function), is finally reached high frequency treatment time subdivision, frequency subdivision, energy at low frequency
The automatic requirement for adapting to time frequency signal analysis, so as to any details of signal can be focused on.
Wherein, all it is 28 pixels the treatment of electrocardio test data is obtained into height and width during implementing
During reference waveform, possible implementation can be：
Mode one：1 is set to by the pixel value of the point of amplitude present in the electrocardiosignal array by electrocardio test data,
Remaining is set to 0, is all the oscillogram of 28 pixels so as to obtain height and width.
Mode two：By electrocardio test data conversion into oscillogram, oscillogram is carried out directly by Matlab drawing instruments
It is all the oscillogram of 28 pixels to preserve to obtain height and width.
Mode three：By electrocardio test data conversion into oscillogram, by sectional drawing instrument, oscillogram is carried out sectional drawing preserve with
It is all the oscillogram of 28 pixels to obtain height and width.
It is several possible implementation cited by the present invention with upper type, is not limited thereto, other is possible
As long as implementation can reach the above-mentioned effect of the present invention, the present invention is can be applied to, the present invention is not described in detail one by one.
Step S103：With reference to the oscillogram and the heart disease disaggregated model set up of preliminary dimension, determine that electrocardio is surveyed
Classification results corresponding to examination data.
Wherein, heart disease disaggregated model is the model that can reflect original waveform feature and classification results.
Wherein, in embodiments of the invention, before step S103 is performed, also including heart disease disaggregated model training
Journey.
Heart disease disaggregated model in the embodiment of the present invention is the mould for reflecting electrocardio original waveform feature and classification results
Type.That is, be the continuous learning training by numerous electrocardio test data samples, obtain different electrocardio original waveforms and its
The model of corresponding classification results.
To further describe the training process of heart disease disaggregated model, the embodiment of the present invention provides a kind of possible
Implementation, refers to Fig. 2, and Fig. 2 is a kind of method for building heart disease disaggregated model provided in an embodiment of the present invention, its bag
Include following steps：
Step S201：Sample data is collected, sample data includes multiple dynamic electrocardiogram record datas.
Wherein, in a kind of implementation of the invention, the sample number is collected by from MIT-BIH arrhythmia cordis database
According to.
MIT-BIH arrhythmia cordis databases, are set up cooperatively by Massachusetts Institute Technology and Beth Israel hospitals
, between its data comes from 1975 to 1979, the multiple dynamic heart of Beth Israel hospital's arrhythmia cordis laboratory collection
Electrographic recording data.The database is had altogether comprising 48 electrocardiosignals, takes from 47 individualities, and wherein recording mechanism 201 and 202 is from same
One is individual.22 female individuals of 25 male individuals and age including the age from 32 years old to 89 years old from 23 years old to 89 years old, its
In about 69% data come from inpatient.23 data wherein between data record number 100 to 124 are from Holter
Randomly selected in above-mentioned data set, had between the waveform and artefact recording mechanism 200 to 234 of the various change for representing meaning
25 data include data that are uncommon but having extremely important clinical picture, including some complicated room property, knot property, on room
Property arrhythmia cordis and conduction abnormalities.Whole database has 48 electrocardiogram (ECG) datas, there are about 109500 bats, wherein about 70% heart is clapped being
The normal heart is clapped, and remaining is that the abnormal heart is clapped, and is had 15 kinds of abnormal hearts and is clapped, and it is by least two electrocardio that each heart is clapped
What figure expert independently marked by hand.Each record of database includes three files, is respectively the entitled .hea of expansion of header files, number
Entitled .air is extended according to the entitled .dat of file extent, comment file.Header file is used for the name of data file for illustrating to be associated with it
Word and attribute, storage mode are ASCII character character, wherein the form including signal, sample frequency, length are saved, and this
Record patient relevant information, such as locality, conditions of patients, medicining condition detail information data file be with " 212 " form
The signal initial data for being stored, 212 forms are directed to two data-base recordings of signal, and the data of the two signals replace
Storage, every three bytes store two data, and it is that record cardiac diagnosis are special that comment file is explained in header file
Family to the result of signal analysis, mainly including heartbeat, the rhythm and pace of moving things and information etc. signal quality, with binary storage.
Step S202：Sample data process and obtains electrocardiogram training set.
Wherein, in embodiments of the present invention, there is provided it is a kind of it is possible sample data process obtain electrocardiogram training
The method of collection, further referring to Fig. 3, Fig. 3 is provided in an embodiment of the present invention sample data process to obtain electrocardiogram
The possibility implementation method of training set, comprises the following steps：
S2021：Each sample data process and obtains cardiac pulse waveform figure.
The treatment of each sample data is obtained into electrocardio pulse waveform figure, for the electrocardiosignal wave band by treatment adjustment,
Search for its data point maximum ymax, minimum value ymin, unit is millivolt (mV)., it is necessary to generate cardiac pulse waveform figure during training,
Every image heartbeat of correspondence, is highly (ymax-ymin)/0.001 pixel, width is 540 pixels, each pixel
Point initial value is 0.
To every picture construction coordinate system, as shown in Figure 4.
Because the R ripples position of each training data has been indicated, therefore centered on R ripples position, totally 540 are taken before and after it
Individual data point.
In order to obtain the input of oscillogram sample, first have to set (x, y) position to each data point for the waveform signal for obtaining
Put a little.Leftmost side data point x values are 0, are then assigned according to data point sequential and are incremented by x values, then low order end data point x values are 539；Often
The y values of individual data point are set to ((the point data value-ymin)/0.001)。
After to all 540 data points set (x, y) location point, to each data point, image (x, y) position is set
Pixel value be 1.So as to obtain oscillogram according to this mode.
S2022：Each cardiac pulse waveform figure is compressed, and using bilinear interpolation method process obtain height
All it is 28 pictures of pixel with width, as electrocardiogram training set.
If when data of having an electro-cardiogram are converted the electrocardiogram that changes into as when, image is too small to lose script electrocardio number
According to feature, the electrocardiogram picture too conference for changing into causes the speed of service to decline.Therefore, the spy of embodiment of the present invention complex electrocardio figure
Point, while having carried out the analysis of test of many times data, is ensureing that the data characteristics of electrocardiogram ensures algorithm again as far as possible
On the premise of operational efficiency, so it is all 28 images of pixel (28*28) that electrocardiogram (ECG) data is changed into height and width by selection.
For the oscillogram of each Zhang Shengcheng, a newly-built height and width are all 28 pictures of pixel, are then used
The method of bilinear interpolation, obtains new oscillogram, as electrocardiogram training set by the oscillogram generated in previous step.
Step S203：Model framework is built with reference to ecg characteristics.
Wherein, as possible implementation of the invention, ecg characteristics can be combined and builds convolutional neural networks model
Framework, or build and product network model framework, or build recirculating network framework.In these implementations, difference master
It is the method for electrocardiogram (ECG) data feature extraction and the network structure of use.Recognition with Recurrent Neural Network is one kind in feedforward neural network
The middle neutral net for increasing feedback link, can produce the memory state to past data, for the treatment to sequence data, and
The dependence set up between different periods data.It is complicated multivariate probability by the decomposition of many levels with product network
Distribution and expression for single argument probability distribution and with product, its node is made up of variable.Convolutional neural networks are a kind of feedforward god
Through network, its artificial neuron can respond the surrounding cells in a part of coverage, have for large-scale image procossing
Color table shows.
Used as specific embodiment of the invention, the present invention builds convolutional neural networks model framework to combine ecg characteristics
As a example by be described in detail.
Fig. 5 is referred to, Fig. 5 is the convolutional neural networks model framework schematic diagram that one of embodiment of the invention builds.
As illustrated, the convolutional neural networks model constructed by the embodiment of the present invention has 1 input layer, 2 hidden layers, 1 full connection
Layer and 1 output layer.
Step S204：Learning training is carried out to electrocardiogram training set based on model framework to obtain heart disease classification mould
Type.
With reference to Fig. 5, the embodiment of the present invention is described in further detail data learning training process.
According to the graphic characteristicses of the electrocardiogram the training set more than present invention being converted into, the present invention builds level 2 volume lamination knot
The convolutional neural networks model of structure.The size 28*28 of image due to by 12*12 can be reduced to after 1 convolutional layer, by 2
It is changed into 4*4 after convolutional layer, so as to can be reached on the premise of accuracy rate is ensured improve computing using the structural model framework
The effect of speed.
The image of the 28*28 obtained after pretreatment is directly inputted from data input layer (Input).C1 layers of convolutional layer output 20
Individual Feature Mapping, that is to say, that it is 20 that C1 convolutional layers are needed by the convolution nuclear volume for training study, i.e., to the every of input picture
20 kinds of different features of individual extracted region, the size of convolution kernel is 5 × 5.C1 convolutional layers need the parameter of training：Each convolution kernel
There are 5 × 5=25 parameter and an offset parameter, because C1 convolutional layers there are 20 convolution kernels, the parameter that study is needed altogether is
(5 × 5+1) × 20=520.
Because input picture size is 28*28, convolution kernel convolution operation sliding step is 1 pixel, and feature is reflected after convolution
It is 24 × 24 to penetrate size, then the connecting line number of C1 convolutional layers is 520 × (24 × 24)=299520.By contrasting C1 convolutional layers
The parameter and connecting line number of study is needed to understand, convolutional layer can greatly reduce convolutional Neural net by the way of weights are shared
Network needs the number of parameters of study, and wherein weights are shared means that each convolution kernel is deconvoluted figure using same weights
Picture, different convolution kernels use different weights.The output characteristic mapping amount of S2 layers of down-sampling layer S2 down-sampling layers and above
The Feature Mapping quantity of C1 convolutional layers output is identical, is 20.The sample area 2 × 2 of S2 down-samplings layer, non-overlapping sampling,
Therefore the size of S2 each Feature Mapping be C1 convolutional layers output characteristic mapping size a quarter (row and column each two/
One).C3 layers of convolutional layer exports 50 Feature Mappings, therefore it is 50 that C3 convolutional layers need the convolution nuclear volume of training, i.e., input is schemed
50 kinds of different features of each extracted region of picture, the size of convolution kernel is 5 × 5.
C3 convolutional layers need the training parameter to be：5 × 5=25 parameter of each convolution kernel adds an offset parameter, due to
There are 50 convolution kernels, altogether (5 × 5+1) × 50=1300 parameter.S2 layers of Feature Mapping resolution ratio is 12 × 12, convolution operation
Sliding step is that the Feature Mapping exported after 1, C3 layers of convolution is 8 × 8, thus the connecting line number of C3 convolutional layers come to 1300 ×
8 × 8=83200.Each convolution kernel of C3 convolutional layers carries out convolution, i.e., the side of full connection to S2 layers of all of Feature Mapping
Formula.S4 down-samplings layer S4 down-samplings layer is operated using with S2 identicals mode to the mapping of C3 convolutional layers output characteristic.F5 is complete
512 neurodes of articulamentum, each neurode and the connection of S4 layers of all neurodes.The full articulamentums of F5 need to learn
The parameter of habit is common according to network final output to the quantity of the Feature Mapping of S4 down-samplings layer and the resolution ratio of Feature Mapping
Determine.Output layer is actually many classification softmax graders, and output node number is determined according to classification task.Such as one
Plant in implementation, could be arranged to 16 (15 abnormal signals, 1 normal signal).
The test sample data of the known electrocardiosignal abnormal class that will be obtained, the heart of 28*28 is changed into by pretreatment
Electrical image, is input to the convolutional neural networks model that training is obtained, and ratio is done according to the classification results for finally giving and concrete class
To calculating accuracy.The time complexity of the algorithm and traditional BP neural network algorithm can be compared simultaneously.
Convolutional neural networks model and its training belong to supervised learning, can allow meter by the data set by mark
, be dissolved into electrocardiogram (ECG) data feature in model using its adaptive learning ability by the method that calculation machine autonomous learning goes out data characteristics,
The incompleteness that artificial selected characteristic is caused is reduced, the accuracy rate of ECG detecting and the categorical measure of classification is improved.And be based on
Learning process that the behavioral characteristics learning simulation of the Two-dimensional morphology of the convolutional neural networks mankind differentiate for Characteristics of electrocardiogram and
Determination methods, farthest enhancing differentiates for the profound level of complex situations, so as to improve accuracy.
Because the above-mentioned model training method of the embodiment of the present invention is that, to normal beats, left bundle branch block, right bundle branch is passed
Retardance is led, atrial premature beats, VPB, ventricle fusion heartbeat joins premature beat, and supraventricular premature beat, room property ease Baudot kind data are entered
Row training, the final network model for obtaining can be to various heart classification of type.Used when data training managing is done simultaneously
Electrocardiogram (ECG) data is converted into the method for EGC pattern, do so has more obvious data characteristics and shows in data training
, final classification result is more accurate.
The reference waveform figure of the 28*28 that the treatment of electrocardio test data is obtained is input to the heart disease classification having been built up
In model, you can to obtain the classification results corresponding to electrocardio test data, classification results here refer to electrocardio test data
Corresponding heart disease classification results.
The method of the electrocardiosignal classification that the above embodiment of the present invention is provided, obtains electrocardio test data, and electrocardio is surveyed
Examination data are processed to obtain the oscillogram of preliminary dimension, with reference to the oscillogram and the heart disease set up of preliminary dimension
Disaggregated model, determines the classification results corresponding to the electrocardio test data, wherein, heart disease disaggregated model is to reflect
The model of original waveform feature and classification results.Heart disease disaggregated model of the invention combines morphological analysis, digitized map
As obtained from processing method and deep learning method carry out learning training to electrocardio test sample data, at utmost retaining
While original waveform feature, classification learning ability and adaptivity are improve.Therefore, the heart disease for being built based on the present invention
Disaggregated model, it is possible to increase the accuracy rate of classification results, while can realize dividing further types of heart rate variability
Class.
Fig. 6 is referred to, Fig. 6 is a kind of structural representation of the system of electrocardiosignal classification provided in an embodiment of the present invention,
The method that the system of the electrocardiosignal classification that the present embodiment is provided is used to perform the electrocardiosignal classification described in above-described embodiment.Such as
Shown in figure, the system 100 of the electrocardiosignal classification of the present embodiment includes data acquisition module 11, data processing module 12 and result
Determining module 13, wherein：
Data acquisition module 11 is used to obtain electrocardio test data.
Here electrocardio test data, refers to the primary data obtained by device measuring.
Data processing module 12 is used to that electrocardio test data to be processed to obtain the oscillogram of preliminary dimension.
In one of which implementation of the invention, data processing module 12 determines heart each time by wavelet transformation
The R ripples position of beating, to split to electrocardiosignal, obtains the oscillogram of preliminary dimension.
Wherein, in a kind of possible implementation of the embodiment of the present invention, the oscillogram of preliminary dimension for height and width all
It is the oscillogram of 28 pixels, certainly, in other possible implementations, is also not limited to the size, only mays other
Can there is certain difference in the result that possible size is likely to be obtained.In present invention below embodiment, it is with preliminary dimension mainly
Height and width are all to illustrate as a example by 28 pixels.
Wavelet transformation (wavelet transform, WT) is a kind of new transform analysis method, and it inherits and has developed short
When Fourier transform localization thought, while the shortcomings of overcoming window size again and do not change with frequency, using the teaching of the invention it is possible to provide one
It is the ideal tools for carrying out signal time frequency analysis and treatment with " T/F " window of frequency shift.Being mainly for it passes through
Conversion can abundant feature of the outstanding problem in terms of some, the localization of time (space) frequency can be analyzed, put down by flexible
Movement is calculated and progressively carries out multi-scale refinement to signal (function), is finally reached high frequency treatment time subdivision, frequency subdivision, energy at low frequency
The automatic requirement for adapting to time frequency signal analysis, so as to any details of signal can be focused on.
Wherein, during implementing, the treatment of electrocardio test data is being obtained height and width by data processing module 12
When degree is all for the waveform of 28 pixels, possible implementation can be：
Mode one：1 is set to by the pixel value of the point of amplitude present in the electrocardiosignal array by electrocardio test data,
Remaining is set to 0, is all the oscillogram of 28 pixels so as to obtain height and width.
Mode two：By electrocardio test data conversion into oscillogram, oscillogram is carried out directly by Matlab drawing instruments
It is all the oscillogram of 28 pixels to preserve to obtain height and width.
Mode three：By electrocardio test data conversion into oscillogram, by sectional drawing instrument, oscillogram is carried out sectional drawing preserve with
It is all the oscillogram of 28 pixels to obtain height and width.
It is several possible implementation cited by the present invention with upper type, is not limited thereto, other is possible
As long as implementation can reach the above-mentioned effect of the present invention, the present invention is can be applied to, the present invention is not described in detail one by one.
As a result determining module 13 is used for the heart disease disaggregated model for combining the oscillogram of preliminary dimension and having set up, really
Centering electrical test data corresponding to classification results, wherein, heart disease disaggregated model be can reflect original waveform feature with
The model of classification results.
Wherein in one embodiment, the system of electrocardiosignal classification of the invention can further include model construction
Module 14, model construction module 14 is for being processed electrocardiogram (ECG) data sample, learning training is so as to obtain heart disease classification
Model.Heart disease disaggregated model in the embodiment of the present invention is the model for reflecting electrocardio original waveform feature and classification results.
That is, being that, by numerous electrocardio test data samples, continuous learning training obtains different electrocardio original waveforms corresponding
Classification results model.
For model construction module 14 is discussed in detail, please further combined with Fig. 7, Fig. 7 is model provided in an embodiment of the present invention
The structural representation of module 14 is built, as illustrated, the model construction module 14 of the embodiment of the present invention includes sample collection unit
141st, processing unit 142, framework establishment unit 143 and model training unit 144, wherein：
Sample collection unit 141 is used to collect sample data, and sample data includes multiple dynamic electrocardiogram record datas.
Wherein, in a kind of implementation of the invention, sample collection unit collects sample from MIT-BIH arrhythmia cordis database
Data.
MIT-BIH arrhythmia cordis databases, are set up cooperatively by Massachusetts Institute Technology and Beth Israel hospitals
, between its data comes from 1975 to 1979, the multiple dynamic heart of Beth Israel hospital's arrhythmia cordis laboratory collection
Electrographic recording data.The database is had altogether comprising 48 electrocardiosignals, takes from 47 individualities, and wherein recording mechanism 201 and 202 is from same
One is individual.22 female individuals of 25 male individuals and age including the age from 32 years old to 89 years old from 23 years old to 89 years old, its
In about 69% data come from inpatient.23 data wherein between data record number 100 to 124 are from Holter
Randomly selected in above-mentioned data set, had between the waveform and artefact recording mechanism 200 to 234 of the various change for representing meaning
25 data include data that are uncommon but having extremely important clinical picture, including some complicated room property, knot property, on room
Property arrhythmia cordis and conduction abnormalities.Whole database has 48 electrocardiogram (ECG) datas, there are about 109500 bats, wherein about 70% heart is clapped being
The normal heart is clapped, and remaining is that the abnormal heart is clapped, and is had 15 kinds of abnormal hearts and is clapped, and it is by least two electrocardio that each heart is clapped
What figure expert independently marked by hand.Each record of database includes three files, is respectively the entitled .hea of expansion of header files, number
Entitled .air is extended according to the entitled .dat of file extent, comment file.Header file is used for the name of data file for illustrating to be associated with it
Word and attribute, storage mode are ASCII character character, wherein the form including signal, sample frequency, length are saved, and this
Record patient relevant information, such as locality, conditions of patients, medicining condition detail information data file be with " 212 " form
The signal initial data for being stored, 212 forms are directed to two data-base recordings of signal, and the data of the two signals replace
Storage, every three bytes store two data, and it is that record cardiac diagnosis are special that comment file is explained in header file
Family to the result of signal analysis, mainly including heartbeat, the rhythm and pace of moving things and information etc. signal quality, with binary storage.
Processing unit 142 is used to that sample data process to obtain electrocardiogram training set.
Processing unit 142 is used to that each sample data process to obtain cardiac pulse waveform figure, and each heart is fought
Dynamic oscillogram is compressed, and method using bilinear interpolation processes that to obtain height and width be all the pictures of 28 pixels,
As electrocardiogram training set.
The treatment of each sample data is obtained into electrocardio pulse waveform figure, for the electrocardiosignal wave band by treatment adjustment,
Search for its data point maximum ymax, minimum value ymin, unit is millivolt (mV)., it is necessary to generate cardiac pulse waveform figure during training,
Every image heartbeat of correspondence, is highly (ymax-ymin)/0.001 pixel, width is 540 pixels, each pixel
Point initial value is 0.
To every picture construction coordinate system, because the R ripples position of each training data has been indicated, therefore with R ripples position
Center is set to, totally 540 data points are taken before and after it.
In order to obtain the input of oscillogram sample, first have to set (x, y) position to each data point for the waveform signal for obtaining
Put a little.Leftmost side data point x values are 0, are then assigned according to data point sequential and are incremented by x values, then low order end data point x values are 539；Often
The y values of individual data point are set to ((the point data value-ymin)/0.001)。
After to all 540 data points set (x, y) location point, to each data point, image (x, y) position is set
Pixel value be 1.So as to obtain oscillogram according to this mode.
If when data of having an electro-cardiogram are converted the electrocardiogram that changes into as when, image is too small to lose script electrocardio number
According to feature, the electrocardiogram picture too conference for changing into causes the speed of service to decline.Therefore, the spy of embodiment of the present invention complex electrocardio figure
Point, while having carried out the analysis of test of many times data, is ensureing that the data characteristics of electrocardiogram ensures algorithm again as far as possible
On the premise of operational efficiency, so it is all 28 images of pixel (28*28) that electrocardiogram (ECG) data is changed into height and width by selection.
For the oscillogram of each Zhang Shengcheng, a newly-built height and width are all 28 pictures of pixel, are then used
The method of bilinear interpolation, obtains new oscillogram, as electrocardiogram training set by the oscillogram generated in previous step.
Model construction unit 143 is used to combine ecg characteristics structure model framework.
Model construction unit 143 is used to combine ecg characteristics structure convolutional neural networks model framework；Or combine electrocardio
Figure feature builds and product network model framework；Or combine ecg characteristics structure recirculating network framework.
In these implementations, difference essentially consists in the method for electrocardiogram (ECG) data feature extraction and the network knot of use
Structure.Recognition with Recurrent Neural Network is a kind of neutral net for increasing feedback link in feedforward neural network, can be produced to past number
According to memory state, for the treatment to sequence data, and the dependence set up between different periods data.It is with product network
By the decomposition of many levels complicated multivariate probability distribution and expression for single argument probability distribution and with product, its node is
It is made up of variable.Convolutional neural networks are a kind of feedforward neural networks, and its artificial neuron can respond part covering model
Interior surrounding cells are enclosed, has outstanding performance for large-scale image procossing.
Used as specific embodiment of the invention, the present invention is to combine electrocardiogram structure convolutional neural networks model framework
Example.The convolutional neural networks model framework built in the embodiment of the present invention is the convolutional neural networks model of level 2 volume lamination structure
(as shown in Figure 5).
The size 28*28 of image due to by 12*12 can be reduced to after 1 convolutional layer, by being changed into 4* after 2 convolutional layers
4, so as to the effect of the speed for improving computing can be reached on the premise of accuracy rate is ensured using the structural model framework.
Model training unit 144 is used to carry out learning training to electrocardiogram training set to obtain heart disease based on model framework
Sick disaggregated model.
The image of the 28*28 obtained after pretreatment is directly inputted from data input layer (Input).C1 layers of convolutional layer output 20
Individual Feature Mapping, that is to say, that it is 20 that C1 convolutional layers are needed by the convolution nuclear volume for training study, i.e., to the every of input picture
20 kinds of different features of individual extracted region, the size of convolution kernel is 5 × 5.C1 convolutional layers need the parameter of training：Each convolution kernel
There are 5 × 5=25 parameter and an offset parameter, because C1 convolutional layers there are 20 convolution kernels, the parameter that study is needed altogether is
(5 × 5+1) × 20=520.
Because input picture size is 28*28, convolution kernel convolution operation sliding step is 1 pixel, and feature is reflected after convolution
It is 24 × 24 to penetrate size, then the connecting line number of C1 convolutional layers is 520 × (24 × 24)=299520.By contrasting C1 convolutional layers
The parameter and connecting line number of study is needed to understand, convolutional layer can greatly reduce convolutional Neural net by the way of weights are shared
Network needs the number of parameters of study, and wherein weights are shared means that each convolution kernel is deconvoluted figure using same weights
Picture, different convolution kernels use different weights.The output characteristic mapping amount of S2 layers of down-sampling layer S2 down-sampling layers and above
The Feature Mapping quantity of C1 convolutional layers output is identical, is 20.The sample area 2 × 2 of S2 down-samplings layer, non-overlapping sampling,
Therefore the size of S2 each Feature Mapping be C1 convolutional layers output characteristic mapping size a quarter (row and column each two/
One).C3 layers of convolutional layer exports 50 Feature Mappings, therefore it is 50 that C3 convolutional layers need the convolution nuclear volume of training, i.e., input is schemed
50 kinds of different features of each extracted region of picture, the size of convolution kernel is 5 × 5.
C3 convolutional layers need the training parameter to be：5 × 5=25 parameter of each convolution kernel adds an offset parameter, due to
There are 50 convolution kernels, altogether (5 × 5+1) × 50=1300 parameter.S2 layers of Feature Mapping resolution ratio is 12 × 12, convolution operation
Sliding step is that the Feature Mapping exported after 1, C3 layers of convolution is 8 × 8, thus the connecting line number of C3 convolutional layers come to 1300 ×
8 × 8=83200.Each convolution kernel of C3 convolutional layers carries out convolution, i.e., the side of full connection to S2 layers of all of Feature Mapping
Formula.S4 down-samplings layer S4 down-samplings layer is operated using with S2 identicals mode to the mapping of C3 convolutional layers output characteristic.F5 is complete
512 neurodes of articulamentum, each neurode and the connection of S4 layers of all neurodes.The full articulamentums of F5 need to learn
The parameter of habit is common according to network final output to the quantity of the Feature Mapping of S4 down-samplings layer and the resolution ratio of Feature Mapping
Determine.Output layer is actually many classification softmax graders, and output node number is determined according to classification task.Such as one
Plant in implementation, could be arranged to 16 (15 abnormal signals, 1 normal signal).
The test sample data of the known electrocardiosignal abnormal class that will be obtained, the heart of 28*28 is changed into by pretreatment
Electrical image, is input to the convolutional neural networks model that training is obtained, and ratio is done according to the classification results for finally giving and concrete class
To calculating accuracy.The time complexity of the algorithm and traditional BP neural network algorithm can be compared simultaneously.
Convolutional neural networks model and its training belong to supervised learning, can allow meter by the data set by mark
, be dissolved into electrocardiogram (ECG) data feature in model using its adaptive learning ability by the method that calculation machine autonomous learning goes out data characteristics,
The incompleteness that artificial selected characteristic is caused is reduced, the accuracy rate of ECG detecting and the categorical measure of classification is improved.And be based on
Learning process that the behavioral characteristics learning simulation of the Two-dimensional morphology of the convolutional neural networks mankind differentiate for Characteristics of electrocardiogram and
Determination methods, farthest enhancing differentiates for the profound level of complex situations, so as to improve accuracy.
Because the above-mentioned model training method of the embodiment of the present invention is that, to normal beats, left bundle branch block, right bundle branch is passed
Retardance is led, atrial premature beats, VPB, ventricle fusion heartbeat joins premature beat, and supraventricular premature beat, room property ease Baudot kind data are entered
Row training, the final network model for obtaining can be to various heart classification of type.Used when data training managing is done simultaneously
Electrocardiogram (ECG) data is converted into the method for EGC pattern, do so has more obvious data characteristics and shows in data training
, final classification result is more accurate.
The oscillogram of the 28*28 that the treatment of electrocardio test data is obtained is input to the heart disease disaggregated model having been built up
In, you can to obtain the classification results corresponding to electrocardio test data, classification results here refer to that electrocardio test data institute is right
The heart disease classification results answered.
Above is the method and system of the electrocardiosignal classification that the embodiment of the present invention is provided, obtain electrocardio test data,
Electrocardio test data is processed to obtain the oscillogram of preliminary dimension, with reference to preliminary dimension oscillogram and set up
Heart disease disaggregated model, determines the classification results corresponding to the electrocardio test data, wherein, heart disease disaggregated model is
The model of original waveform feature and classification results can be reflected.Heart disease disaggregated model of the invention combines form credit
Analysis, digital image processing method and deep learning method are carried out obtained from learning training to electrocardio test sample data, most
While big degree retains original waveform feature, classification learning ability and adaptivity are improve.Therefore, built based on the present invention
Heart disease disaggregated model, it is possible to increase the accuracy rate of classification results, while can realize to further types of heart rate
Variation is classified.
In several embodiments provided herein, it should be understood that disclosed system, apparatus and method can be with
Realize by another way.For example, device embodiment described above is only schematical, for example, the module or
The division of unit, only a kind of division of logic function, can there is other dividing mode when actually realizing, such as multiple units
Or component can be combined or be desirably integrated into another system, or some features can be ignored, or not perform.It is another, institute
Display or the coupling each other for discussing or direct-coupling or communication connection can be by some interfaces, device or unit
INDIRECT COUPLING or communication connection, can be electrical, mechanical or other forms.
The unit that is illustrated as separating component can be or may not be it is physically separate, it is aobvious as unit
The part for showing can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On NE.Some or all of unit therein can be according to the actual needs selected to realize the mesh of this embodiment scheme
's.
In addition, during each functional unit in the application each embodiment can be integrated in a processing unit, it is also possible to
It is that unit is individually physically present, it is also possible to which two or more units are integrated in a unit.Above-mentioned integrated list
Unit can both be realized in the form of hardware, it would however also be possible to employ the form of SFU software functional unit is realized.
If the integrated unit is to realize in the form of SFU software functional unit and as independent production marketing or use
When, can store in a computer read/write memory medium.Based on such understanding, the technical scheme of the application is substantially
The part for being contributed to prior art in other words or all or part of the technical scheme can be in the form of software products
Embody, the computer software product is stored in a storage medium, including some instructions are used to so that a computer
Equipment (can be personal computer, server, or network equipment etc.) or processor (processor) perform the application each
The all or part of step of implementation method methods described.And foregoing storage medium includes：USB flash disk, mobile hard disk, read-only storage
(ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disc or CD
Etc. it is various can be with the medium of store program codes.
Embodiments herein is the foregoing is only, the scope of the claims of the application is not thereby limited, it is every to utilize this Shen
Please the equivalent structure made of specification and accompanying drawing content or equivalent flow conversion, or be directly or indirectly used in other related skills
Art field, is similarly included in the scope of patent protection of the application.
Claims (10)
1. a kind of method that electrocardiosignal is classified, it is characterised in that the described method comprises the following steps：
Obtain electrocardio test data；
The electrocardio test data is processed to obtain the oscillogram of preliminary dimension；
With reference to the oscillogram and the heart disease disaggregated model set up of the preliminary dimension, the electrocardio test data is determined
Corresponding classification results, wherein, the heart disease disaggregated model can reflect original waveform feature and classification results
Model.
2. the method that electrocardiosignal according to claim 1 is classified, it is characterised in that preliminary dimension described in the combination
Oscillogram and the heart disease disaggregated model set up, before determining the classification results corresponding to the electrocardio test data,
Also include：
Sample data is collected, the sample data includes multiple dynamic electrocardiogram record datas；
The sample data process and obtains electrocardiogram training set；
Model framework is built with reference to ecg characteristics；
The electrocardiogram training set is carried out learning training to obtain the heart disease disaggregated model based on the model framework.
3. the method that electrocardiosignal according to claim 2 is classified, it is characterised in that the collection sample data includes：
The sample data is collected from MIT-BIH arrhythmia cordis database；
It is described the sample data process obtain electrocardiogram training set and comprise the following steps：
Sample data each described process and obtains cardiac pulse waveform figure；
Cardiac pulse waveform figure each described is compressed, and using bilinear interpolation method process obtain height and width
All it is 28 pictures of pixel, as the electrocardiogram training set；
The combination ecg characteristics build model framework to be included：
Convolutional neural networks model framework is built with reference to ecg characteristics；Or
Built with reference to ecg characteristics and product network model framework；Or
Recirculating network framework is built with reference to ecg characteristics.
4. the method that electrocardiosignal according to claim 3 is classified, it is characterised in that the convolutional neural networks model frame
Frame is the convolutional neural networks model framework of level 2 volume lamination structure.
5. the method that electrocardiosignal according to claim 1 is classified, it is characterised in that described to the electrocardio test data
Processed is included with obtaining the oscillogram of preliminary dimension：
1 is set to by by the pixel value of amplitude point present in the electrocardiosignal array of the electrocardio test data, remaining sets
0 is set to, is all the oscillogram of 28 pixels so as to obtain height and width；Or
By the electrocardio test data conversion into oscillogram, the oscillogram is directly preserved by Matlab drawing instruments
All it is the oscillogram of 28 pixels to obtain height and width；Or
By the electrocardio test data conversion into oscillogram, by sectional drawing instrument, sectional drawing is carried out to the oscillogram and preserves to obtain
All it is the oscillogram of 28 pixels to height and width.
6. the system that a kind of electrocardiosignal is classified, it is characterised in that the system includes data acquisition module, data processing module
With result determining module, wherein：
The data acquisition module is used to obtain electrocardio test data；
The data processing module is used to that the electrocardio test data to be processed to obtain the oscillogram of preliminary dimension；
The result determining module is used for the heart disease disaggregated model for combining the oscillogram of the preliminary dimension and having set up,
Determine the classification results corresponding to the electrocardio test data, wherein, the heart disease disaggregated model is can to reflect original
The model of wave character and classification results.
7. the system that electrocardiosignal according to claim 6 is classified, it is characterised in that the system also includes model construction
Module, wherein, the model construction module further includes sample collection unit, processing unit, framework establishment unit and mould
Type training unit, wherein：
The sample collection unit is used to collect sample data, and the sample data includes multiple dynamic electrocardiogram record datas；
The processing unit is used to that the sample data process to obtain electrocardiogram training set；
The model construction unit is used to combine ecg characteristics structure model framework；
The model training unit is used to that the electrocardiogram training set to be carried out learning training to obtain based on the model framework
The heart disease disaggregated model.
8. the system of electrocardiosignal according to claim 7 classification, it is characterised in that the sample collection unit is from described
MIT-BIH arrhythmia cordis database collects the sample data；
The processing unit is used to that sample data each described process to obtain cardiac pulse waveform figure, to the heart each described
Dirty pulse waveform figure is compressed, and method using bilinear interpolation processes that to obtain height and width be all the figures of 28 pixels
Piece, as the electrocardiogram training set；
The model construction unit is used to combine ecg characteristics structure convolutional neural networks model framework；Or combine electrocardiogram spy
Point builds and product network model framework；Or combine ecg characteristics structure recirculating network framework.
9. the system that electrocardiosignal according to claim 8 is classified, it is characterised in that the convolutional neural networks model frame
Frame is the convolutional neural networks model framework of level 2 volume lamination structure.
10. the system that electrocardiosignal according to claim 6 is classified, it is characterised in that the oscillogram of the preliminary dimension
It is height and the oscillogram of width all pixels of position 28, the data processing module is believed by by the electrocardio of the electrocardio test data
The pixel value of the point of amplitude present in number group is set to 1, and remaining is set to 0, is all 28 pixels so as to obtain height and width
Oscillogram；Or
The data processing module by the electrocardio test data conversion into oscillogram, by Matlab drawing instruments to the ripple
It is all the oscillogram of 28 pixels that shape figure directly preserve to obtain height and width；Or
The data processing module by the electrocardio test data conversion into oscillogram, by sectional drawing instrument, to the oscillogram
Carry out sectional drawing to preserve to obtain height and width is all the oscillogram of 28 pixels.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611150154.4A CN106725426A (en) | 2016-12-14 | 2016-12-14 | A kind of method and system of electrocardiosignal classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611150154.4A CN106725426A (en) | 2016-12-14 | 2016-12-14 | A kind of method and system of electrocardiosignal classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106725426A true CN106725426A (en) | 2017-05-31 |
Family
ID=58880989
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611150154.4A Pending CN106725426A (en) | 2016-12-14 | 2016-12-14 | A kind of method and system of electrocardiosignal classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106725426A (en) |
Cited By (19)
| 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 |
| CN107887010A (en) * | 2017-11-29 | 2018-04-06 | 天津中科爱乐芙医疗科技有限公司 | A kind of angiocardiopathy data acquisition examines platform with dividing |
| CN108030488A (en) * | 2017-11-30 | 2018-05-15 | 北京医拍智能科技有限公司 | The detecting system of arrhythmia cordis based on convolutional neural networks |
| CN108403107A (en) * | 2018-02-06 | 2018-08-17 | 北京大学深圳研究生院 | A kind of arrhythmia cordis method of discrimination and system |
| CN108416277A (en) * | 2018-02-11 | 2018-08-17 | 广州市碳码科技有限责任公司 | A kind of cardioelectric monitor method, apparatus, terminal and computer readable storage medium |
| CN108764457A (en) * | 2018-05-12 | 2018-11-06 | 鲁东大学 | The automatic arrhythmia analysis method of deep neural network is merged based on sampling channel |
| CN108784680A (en) * | 2018-03-19 | 2018-11-13 | 武汉海星通技术股份有限公司 | Electrocardiogram intelligent analysis method based on scatter plot and system |
| 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 |
| CN108926341A (en) * | 2018-04-20 | 2018-12-04 | 平安科技（深圳）有限公司 | Detection method, device, computer equipment and the storage medium of ECG signal |
| CN109009047A (en) * | 2018-05-24 | 2018-12-18 | 东软集团股份有限公司 | A kind of arrhythmia detection method, device and equipment |
| CN109009074A (en) * | 2018-07-19 | 2018-12-18 | 上海工程技术大学 | A kind of sudden cardiac death auxiliary prior-warning device based on deep learning |
| CN109157211A (en) * | 2018-08-14 | 2019-01-08 | 江苏师范大学 | A kind of portable cardiac on-line intelligence monitoring diagnosis system design method |
| 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 |
| CN109620152A (en) * | 2018-12-16 | 2019-04-16 | 北京工业大学 | A kind of electrocardiosignal classification method based on MutiFacolLoss-Densenet |
| CN110495872A (en) * | 2019-08-27 | 2019-11-26 | 中科麦迪人工智能研究院(苏州)有限公司 | Ecg analysis method, apparatus, equipment and medium based on picture and heart-beat messages |
| CN110811608A (en) * | 2019-11-19 | 2020-02-21 | 中电健康云科技有限公司 | Atrial fibrillation monitoring method based on ECG (ECG) signals |
| WO2020047750A1 (en) * | 2018-09-04 | 2020-03-12 | 深圳先进技术研究院 | Arrhythmia detection method and apparatus, electronic device, and computer storage medium |
| CN111261289A (en) * | 2018-11-30 | 2020-06-09 | 上海图灵医疗科技有限公司 | Heart disease detection method based on artificial intelligence model |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104523266A (en) * | 2015-01-07 | 2015-04-22 | 河北大学 | Automatic classification method for electrocardiogram signals |
| CN104970789A (en) * | 2014-04-04 | 2015-10-14 | 中国科学院苏州纳米技术与纳米仿生研究所 | Electrocardiogram classification method and system |
-
2016
- 2016-12-14 CN CN201611150154.4A patent/CN106725426A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104970789A (en) * | 2014-04-04 | 2015-10-14 | 中国科学院苏州纳米技术与纳米仿生研究所 | Electrocardiogram classification method and system |
| CN104523266A (en) * | 2015-01-07 | 2015-04-22 | 河北大学 | Automatic classification method for electrocardiogram signals |
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 (24)
| 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 |
| 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 |
| 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 |
| CN107887010A (en) * | 2017-11-29 | 2018-04-06 | 天津中科爱乐芙医疗科技有限公司 | A kind of angiocardiopathy data acquisition examines platform with dividing |
| CN108030488A (en) * | 2017-11-30 | 2018-05-15 | 北京医拍智能科技有限公司 | The detecting system of arrhythmia cordis based on convolutional neural networks |
| CN108403107A (en) * | 2018-02-06 | 2018-08-17 | 北京大学深圳研究生院 | A kind of arrhythmia cordis method of discrimination and system |
| CN108403107B (en) * | 2018-02-06 | 2019-12-31 | 北京大学深圳研究生院 | Arrhythmia discrimination method and system |
| CN108416277A (en) * | 2018-02-11 | 2018-08-17 | 广州市碳码科技有限责任公司 | A kind of cardioelectric monitor method, apparatus, terminal and computer readable storage medium |
| CN108784680A (en) * | 2018-03-19 | 2018-11-13 | 武汉海星通技术股份有限公司 | Electrocardiogram intelligent analysis method based on scatter plot and system |
| CN108926341A (en) * | 2018-04-20 | 2018-12-04 | 平安科技（深圳）有限公司 | Detection method, device, computer equipment and the storage medium of ECG signal |
| WO2019200746A1 (en) * | 2018-04-20 | 2019-10-24 | 平安科技（深圳）有限公司 | Ecg signal detection method, device, computer apparatus, and storage medium |
| CN108764457A (en) * | 2018-05-12 | 2018-11-06 | 鲁东大学 | The automatic arrhythmia analysis method of deep neural network is merged based on sampling channel |
| 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 |
| CN109009047A (en) * | 2018-05-24 | 2018-12-18 | 东软集团股份有限公司 | A kind of arrhythmia detection method, device and equipment |
| CN109009047B (en) * | 2018-05-24 | 2021-10-26 | 东软集团股份有限公司 | Arrhythmia detection method, device and equipment |
| CN109009074A (en) * | 2018-07-19 | 2018-12-18 | 上海工程技术大学 | A kind of sudden cardiac death auxiliary prior-warning device based on deep learning |
| CN109157211A (en) * | 2018-08-14 | 2019-01-08 | 江苏师范大学 | A kind of portable cardiac on-line intelligence monitoring diagnosis system design method |
| WO2020047750A1 (en) * | 2018-09-04 | 2020-03-12 | 深圳先进技术研究院 | Arrhythmia detection method and apparatus, electronic device, and computer storage medium |
| 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 |
| CN111261289A (en) * | 2018-11-30 | 2020-06-09 | 上海图灵医疗科技有限公司 | Heart disease detection method based on artificial intelligence model |
| 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 |
| CN110495872A (en) * | 2019-08-27 | 2019-11-26 | 中科麦迪人工智能研究院(苏州)有限公司 | Ecg analysis method, apparatus, equipment and medium based on picture and heart-beat messages |
| CN110811608A (en) * | 2019-11-19 | 2020-02-21 | 中电健康云科技有限公司 | Atrial fibrillation monitoring method based on ECG (ECG) signals |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106725426A (en) | A kind of method and system of electrocardiosignal classification | |
| Oh et al. | Automated diagnosis of arrhythmia using combination of CNN and LSTM techniques with variable length heart beats | |
| Kachuee et al. | Ecg heartbeat classification: A deep transferable representation | |
| JP7091451B2 (en) | Automatic ECG analysis method based on artificial intelligence self-learning, equipment used to execute the analysis method, computer program products and computer-readable storage media | |
| CN110890155B (en) | Multi-class arrhythmia detection method based on lead attention mechanism | |
| Deng et al. | Extracting cardiac dynamics within ECG signal for human identification and cardiovascular diseases classification | |
| US20230143594A1 (en) | Systems and methods for reduced lead electrocardiogram diagnosis using deep neural networks and rule-based systems | |
| CN111990989A (en) | Electrocardiosignal identification method based on generation countermeasure and convolution cyclic network | |
| CN107714023A (en) | Static ecg analysis method and apparatus based on artificial intelligence self study | |
| CN110974214A (en) | Automatic electrocardiogram classification method, system and equipment based on deep learning | |
| US20080103403A1 (en) | Method and System for Diagnosis of Cardiac Diseases Utilizing Neural Networks | |
| WO2020011033A1 (en) | Determining device and mapping system for origin of arrhythmia | |
| Bergquist et al. | Body surface potential mapping: contemporary applications and future perspectives | |
| CN112906748A (en) | 12-lead ECG arrhythmia detection classification model construction method based on residual error network | |
| CN111329469A (en) | Arrhythmia prediction method | |
| Feng et al. | Unsupervised semantic-aware adaptive feature fusion network for arrhythmia detection | |
| CN113901893B (en) | Electrocardiosignal identification and classification method based on multi-cascade deep neural network | |
| Zeng et al. | Ensemble learning on heartbeat type classification | |
| JP2023104885A (en) | Electrocardiographic heart rate multi-type prediction method based on graph convolution | |
| WO2021071646A1 (en) | Systems and methods for electrocardiogram diagnosis using deep neural networks and rule-based systems | |
| Deng et al. | ST-Net: Synthetic ECG tracings for diagnosing various cardiovascular diseases | |
| Daydulo et al. | Cardiac arrhythmia detection using deep learning approach and time frequency representation of ECG signals | |
| CN116172573A (en) | Arrhythmia image classification method based on improved acceptance-ResNet-v 2 | |
| Ali et al. | Comparative evaluation for two and five classes ECG signal classification: applied deep learning | |
| Zhang et al. | Multi-scale and multi-channel information fusion for exercise electrocardiogram feature extraction and classification |
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: 20170531 |
|
| RJ01 | Rejection of invention patent application after publication |