CN109117730A - Electrocardiogram auricular fibrillation real-time judge method, apparatus, system and storage medium - Google Patents
Electrocardiogram auricular fibrillation real-time judge method, apparatus, system and storage medium Download PDFInfo
- Publication number
- CN109117730A CN109117730A CN201810760168.0A CN201810760168A CN109117730A CN 109117730 A CN109117730 A CN 109117730A CN 201810760168 A CN201810760168 A CN 201810760168A CN 109117730 A CN109117730 A CN 109117730A
- Authority
- CN
- China
- Prior art keywords
- electrocardiogram
- interphase
- model
- wave
- result
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/318—Heart-related electrical modalities, e.g. electrocardiography [ECG]
- A61B5/346—Analysis of electrocardiograms
- A61B5/349—Detecting specific parameters of the electrocardiograph cycle
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/50—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for simulation or modelling of medical disorders
Landscapes
- Health & Medical Sciences (AREA)
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Public Health (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- Medical Informatics (AREA)
- Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Heart & Thoracic Surgery (AREA)
- Surgery (AREA)
- Animal Behavior & Ethology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Veterinary Medicine (AREA)
- Cardiology (AREA)
- Psychiatry (AREA)
- Physiology (AREA)
- Theoretical Computer Science (AREA)
- Power Engineering (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Epidemiology (AREA)
- Primary Health Care (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
A kind of electrocardiogram auricular fibrillation real-time judge method, apparatus, system and its computer storage medium can be based on wearable device, analyze electrocardiogram in real time.This method extracts the features such as RR interphase, root mean square, wavelet coefficient, makes full use of the hiding information in waveform, binding rule judgement and machine learning model, judges auricular fibrillation.Finally, the judging result for merging multiple models provides the marking to atrial fibrillation, that is, suffers from the Suspected Degree of atrial fibrillation.The present invention only needs to acquire the electrocardiogram of 20s, can in real time, accurately analyze and determine auricular fibrillation.
Description
Technical field
The present invention relates to electrocardiogram monitoring fields, and in particular to a kind of electrocardiogram auricular fibrillation real-time judge method, apparatus,
System and its computer storage medium.
Background technique
In recent years, cardiovascular disease worldwide becomes more and more popular, because the number of cardiovascular disease death is also more next
It is more.Electrocardiogram is a kind of effective method observed cardiac potential variation and judge cardiovascular disease.However, due to electrocardiogram
The variation of waveforms amplitude and frequency content be it is small, for doctor, by electrocardiogram judge cardiovascular disease it is more difficult and
It is time-consuming.In addition, prolonged range estimation may also lead to mistaken diagnosis.Therefore, area of computer aided electrocardiogram judgement is increasingly taken seriously,
It has obtained a large amount of research and has gradually started to be applied.Sentencing for heart disease can be markedly improved in it under acceptable price
It is disconnected.
Auricular fibrillation is a kind of common important supraventricular arrhythmia cordis, very common especially in old group.Room
Quivering is to dominate reentrant cycle by atrium to cause rule disorder in room caused by many micro reentry rings, and atrium is in unordered excitement and nothing when generation
Effect is shunk.It has biggish harm to patient.Since ventricular beat is extremely irregular neat, patient often shows nervous, out of strength.And
Atrial fibrillation is found in all organic heart disease patients, and the disease incidence high duration is long, it is also possible to deteriorate heart function, cause tight
The complication of weight causes patient disabled or case fatality rate increases such as heart failure and arterial embolism.It is therefore desirable to have one kind can
The intelligent determination method for judging atrial fibrillation promptly and accurately.
Instantly, in addition to the conventional method judged to hospital's detection electrocardiogram by doctor, for Paroxysmal Atrial Fibrillation, frequently with
Holter carrys out continuous collecting electrocardiogram within 24 hours, and 2-3 days data are transmitted to hospital and are judged by doctor.Existing area of computer aided
Judgment method is analyzed and determined after being usually also required to the signal for acquiring suitable duration by computer.Have already appeared based on portable
The method of formula hardware, it is also desirable to acquire a few minutes a waveforms up to a hundred and be judged.And much also need the inspection based on implanted
Survey device.In existing atrial fibrillation judgment method, to the judgement of atrial fibrillation generally according to the information of RR interphase, such as the kind of RR interphase
Class, size variation, turning point etc..It is analyzed using one or more Rule of judgment, whether finally provide has sentencing for atrial fibrillation
It is disconnected.
1. existing atrial fibrillation judges that scheme is usually only absolutely uneven for according to the relevant criterion of formulation with RR interphase.This is resulted in
Other for including in waveform to atrial fibrillation judge that useful information is not utilized.It is inadequate to the utilization of hiding disease information
Sufficiently influence the accuracy of final judging result.
2. existing method is seen to doctor after either acquiring a certain amount of data, or analyzed using computer, all need
Analysis result can be just obtained after wanting some time.The existing method based on portable hard, it is also desirable to acquire the long period.Cause
This is unable to quick real-time judge.
3. existing atrial fibrillation judgment method is judging to provide the judgement for whether suffering from atrial fibrillation when completing.It cannot be guaranteed
Absolute judgement is provided in the case where absolutely accurate, will affect judgement of the patient to disease.Existing method can not provide trouble
There is a possibility that atrial fibrillation.
Summary of the invention
Present invention seek to address that electrocardiographic wave information, judgement cannot be made full use of to have delay or take a long time, judge
As a result excessively absolute etc. technical problems provide a kind of electrocardiogram auricular fibrillation real-time judge method, apparatus, system and its calculating
Machine storage medium can fast implement real-time auricular fibrillation monitoring and judgement by the electrocardiogram of 20s.
The present invention realizes that the first aspect of the present invention provides a kind of electrocardiogram auricular fibrillation by following technical solution
Real-time judge method, includes the following steps:
Step 1, electrocardiogram pretreatment, available electrocardiogram for obtaining being labeled with wave character and to available electrocardiogram into
The electrocardiogram wave band of row segmentation；
Step 2, the available electrocardiogram input rule model that will be labeled with wave character are based on room escape through rule model
Rule model result is exported after the judgement of medical judgment rule；The electrocardiogram wave band input of segmentation is trained by hospital data
Machine learning model obtains machine learning model result after machine learning model judges；
Rule model result and machine learning model result are obtained atrial fibrillation Suspected Degree most by fusion calculation by step 3
Terminate fruit.
In the particular embodiment, the electrocardiogram pretreatment of the step 1 includes the following steps:
Step 11 is filtered the electrocardiogram of acquired original, for removing signal noise, obtains available electrocardiogram；
Step 12 can carry out waveform recognition with electrocardiogram to described, identify and mark P wave and QRS wave in electrocardiosignal
Group；
Step 13 carries out waveform segments to the available electrocardiogram of mark, and waveform, which is divided into, certain can be used for machine learning
Wave band, obtain electrocardiogram wave band.
In the particular embodiment, the filtering carries out de-noising using wavelet thresholding methods, using db4 small echo by signal 8
Layer decomposes, and the wavelet coefficient decomposed is handled by Soft thresholding, obtains new wavelet coefficient, carries out letter by new wavelet coefficient
Number reconstruct, obtains filtered electrocardiosignal, can use electrocardiogram to be described；
And/or the waveform recognition is based on B- batten biorthogonal wavelet and detects QRS complex, determines the position of Q, R, S point；
P wave is detected based on first-order difference, determines P point position；
And/or the waveform segments are using the position of 0.3s before R wave wave crest as starting point, the position of 0.3s after R wave wave crest
It as end point, is clapped as a heart, 6 heart bats are divided into a wave band as an electrocardiogram wave band sample and input institute
State machine learning model.
In the particular embodiment, the rule model judgement of the step 2 includes the following steps:
Step A21, RR interphase, P wave height and R wave height are calculated according to the available electrocardiogram of input；
Step A22, exceptional value interim between detection RR and removal；
Step A23, by the maximum minimum to RR interphase than judged rule model one, to RR interphase and
Rule model two that value deviation is judged carries out the RR interphase group number for meeting complete compensatory gap and approximate premature beat type
It is the rule model three of judgement, more above-mentioned than the rule model four that the waveform accounting in threshold range is judged to P wave R wave height
The judgement of four rule models respectively obtains one result S1 of model, two result S2 of model, three result S3 of model and four result of model
S4；
Step A24, judge whether S1 is equal to 0: if so, S=0；If it is not, then S=S1+S2-S3-S4；
Step A25, S is exported.
In the particular embodiment, the rejecting outliers in the step A22 are interim with the presence or absence of different between RR for determining
Constant value calculates the mean value of all RR interphases first；Judge whether the value of each RR interphase is greater than 0.5 times of mean value and less than 1.6 times
Mean value, if it is not, exceeding threshold range, it is determined that the RR interphase is exceptional value and removes.
In the particular embodiment, the rule model one in the step A23 includes:
(1) input can use the maximum of all RR interphases and the ratio q of minimum in electrocardiogram, and initiation parameter d, make
D=0；
(2) q value is judged:
If q≤1.1, parameter d=5；
If 1.1 q≤1.9 <, calculating parameter d=-4.745 × q+10.2195；
If 1.9 q≤2.1 <, parameter d=1.204；
If 2.1 q≤3 <, calculating parameter d=-0.5677 × q+2.3962；
If q > 3, then parameter d=0.6931；
(3) the score S1=100exp (- d) of rule model one is calculated；
Rule model two in the step A23 includes:
(1) mean value and standard deviation of all RR interphases in electrocardiogram can be used by calculating；Then each RR interphase and mean value are calculated
Deviation, judges whether the deviation is greater than standard deviation；Calculate the RR interphase accounting r that it is poor that deviation is above standard；
(2) the RR interphase accounting r, and initiation parameter d are inputted, d=0 is made；
(3) r value is judged:
If r≤0.25, parameter d=5；
If 0.25 r≤0.35 <, calculating parameter d=-26.974 × r+11.7435；
If 0.35 r≤0.45 <, calculating parameter d=-10.896 × r+6.1477；
If r > 0.45, then parameter d=1.204；
(4) the score S2=100exp (- d) of rule model two is calculated；
Rule model three in the step A23 includes:
(1) number for the RR interphase group for meeting complete compensatory pause and approximate premature beat type in electrocardiogram can be used described in judgement
N: enabling four continuous RR interphases is a RR interphase group, by the sum of the 2nd RR interphase and the 3rd RR interphase and average RR interphase phase
Compare, if average RR interphase of the sum of the 2nd RR interphase and the 3rd RR interphase less than 2.2 times and greater than between 1.1 times of average RR
Phase, it is determined that the 2nd RR interphase and the 3rd RR interphase meet complete compensatory pause；If the first RR interphase is greater than the 2nd RR interphase
And the 3rd RR interphase be greater than the 4th RR interphase, it is determined that it is approximate premature beat type；Meet between the RR of above-mentioned two condition simultaneously
Phase group is then to meet the RR interphase group of complete compensatory pause and approximate premature beat type；
(2) meet the number n of complete compensatory pause and the RR interphase group of approximate premature beat type described in input, and initialize ginseng
Number d, makes d=0；
(3) n value is judged:
If n≤1, parameter d=5；
If n=2, parameter d=2.9957；
If n=3, parameter d=1.8971；
If n=4, then parameter d=1.204；
If n >=5, parameter d=0.6971；
(4) the score S3=100exp (- d) of rule model three is calculated；
Rule model four in the step A23 includes:
(1) each waveform P wave R wave height ratio in electrocardiogram can be used described in input；
(2) if P wave R wave height ratio is within the scope of 0.1-0.2, it is determined that it is normal level ratio, is included in；Calculate P wave R
Wave height is than waveform within the above range in the accounting p with all waveforms in electrocardiogram；And initiation parameter d, make d
=0；
(3) p value is judged:
If p≤0.4, parameter d=5；
If 0.4 p≤0.6 <, calculating parameter d=-10.0215 × p+9.0086；
If 0.6 p≤0.8 <, calculating parameter d=-8.9585 × p+8.3708；
If 0.8 p≤0.9 <, calculating parameter d=-5.109 × p+5.2912；
If p > 0.9, then parameter d=0.6931；
(4) the score S4=100exp (- d) of rule model four is calculated.
In the particular embodiment, the machine learning model judgement in the step 2 includes the following steps:
Step B21, extracting RR interphase, average absolute amplitude, root mean square, wavelet coefficient to the electrocardiogram wave band of input is spy
Sign decomposes to 4 layers using heart bat of the db4 small echo to input, takes the wavelet coefficient of a4 frequency range as feature；
Step B22, z-score standardization is carried out to the feature, it is identical for having the feature of different dimensions
Scale；
Step B23, dimensionality reduction: for reducing the dimension of feature, retain the principal component that weight is more than 98%；
Step B24, the final feature obtained after dimensionality reduction is input to five machine learning submodels, is linearly to sentence respectively
Not Fen Xi, in gauss hybrid models, least square method supporting vector machine, reverse transmittance nerve network, probabilistic neural network model into
Row judgement, respectively obtains judging result；Result is divided into 2 classes, atrial fibrillation and non-atrial fibrillation by the linear discriminant analysis；The Gauss is mixed
Molding type is divided into atrial fibrillation and non-atrial fibrillation aiming at the problem that two classification, by sample, and parameter Estimation uses EM algorithm and maximum likelihood
Estimation；The least square method supporting vector machine uses radial basis function as kernel function to classify；The backpropagation mind
It include 1 layer of input layer, 1 layer of hidden layer and 1 layer of output layer through network；Wherein neuronal quantity, that is, sample feature drop of input layer
Dimension after dimension；Hidden layer is set as 6 neurons；Output layer includes 2 neurons, i.e. atrial fibrillation and non-atrial fibrillation；The probability
Neural network includes 1 layer of radial base and one layer of competition layer；Radial base includes two neurons of atrial fibrillation and non-atrial fibrillation；Competition layer
Determine the class of maximum probability and to its assignment 1；
Step B25, each judging result is merged to obtain final result.
In the particular embodiment, each machine learning submodel is respectively trained, and obtains model ginseng with data training in advance
It counts, only has parameter in the portable device that the method is based on, directly the waveform of input is judged.
In the particular embodiment, the result of each submodel uses the side voted and given a mark in proportion in the step B25
Formula merges, i.e., ratio shared by atrial fibrillation is the judgement final result of machine learning model in the result of each submodel.
In the particular embodiment, the fusion of the rule model and machine learning model judging result are as follows: determine two
As a result it whether there is larger difference, if the difference of the two score is greater than 0.6, this result is cancelled；Otherwise the two is averaged
As final atrial fibrillation Suspected Degree.
The second aspect of the present invention provides a kind of electrocardiogram auricular fibrillation real-time judge device, comprising:
Electrocardiogram preprocessing module, available electrocardiogram for obtaining being labeled with wave character and carries out available electrocardiogram
The electrocardiogram wave band of segmentation；The electrocardiogram preprocessing module includes: filter unit, is filtered to the electrocardiogram of acquired original
Wave obtains available electrocardiogram for removing signal noise；Waveform recognition unit can carry out waveform recognition with electrocardiogram to described,
It identifies and marks P wave and QRS complex in electrocardiosignal；
Waveform segments unit carries out waveform segments to the available electrocardiogram of mark, and waveform, which is divided into, certain can be used for machine
The wave band of device study, obtains electrocardiogram wave band；
Rule model will be labeled with the available electrocardiogram input rule model of wave character, defeated after rule model judges
Rule model result out；The rule model includes: parameter calculation unit, calculates RR interphase, P according to the available electrocardiogram of input
Wave height and R wave height；RR interphase rejecting outliers unit detects interim exceptional value and removal between RR；To the pole of RR interphase
Big value minimum than the rule model one judged, the rule model two judged with mean bias to RR interphase, to symbol
Close rule model three that the RR interphase group number of compensatory gap completely and approximate premature beat type judged, to P wave R wave height ratio
The rule model four that waveform accounting in threshold range is judged, aforementioned four rule model respectively obtain mould by judgement
One result S1 of type, four result S4 of two result S2 of model, three result S3 of model and model；Judging unit, judges whether S1 is equal to 0:
If so, S=0；If it is not, then S=S1+S2-S3-S4；Output unit exports result S；
The electrocardiogram wave band of segmentation is inputted machine learning model, after machine learning model judges by machine learning model
Obtain machine learning model result；The machine learning model judgement includes: feature extraction unit, to the electrocardiogram wave band of input
RR interphase, average absolute amplitude, root mean square, wavelet coefficient is extracted to be characterized；Standardisation Cell carries out z- to the feature
Score standardization, for making the feature scale having the same of different dimensions；Dimensionality reduction unit: for reducing the dimension of feature
Degree；Machine learning submodel is linear discriminant analysis, gauss hybrid models, least square method supporting vector machine, backpropagation respectively
Neural network, probabilistic neural network model, by the feature obtained after dimensionality reduction be input in above-mentioned five machine learning submodels into
Row judgement, respectively obtains judging result；Machine learning model computing unit, each judging result carry out fusion calculation and are most terminated
Fruit.
Rule model result and machine learning model result are obtained atrial fibrillation Suspected Degree by fusion calculation by computing unit
Final result.
In the particular embodiment, the rule model one determines maximum and minimum interim between all RR, determines
RR interphase maximum minimum ratio, determines range of the RR interphase maximum minimum than place, and in response to the maximum pole
Than the range at place the value of coefficient is calculated, and thus the score of model one is calculated in coefficient in small value；
The rule model two determines the mean value and standard deviation of all RR interphases, determines each RR interphase and mean bias
Whether difference, the RR interphase number accounting that it is poor that determination deviation is above standard if being above standard, the RR interphase that it is poor that determination deviation is above standard
Range where number accounting, and in response to the number accounting range, the value of coefficient is calculated, and thus coefficient calculates
To the score of model two；
The rule model three handles continuous 4 RR interphases as a RR interphase group, determines each RR interphase
Whether group meets complete compensatory pause, determines the whether approximate premature beat type of each RR interphase group, is determined for compliance with complete compensatory pause
And the RR interphase group number of approximate premature beat type, it is determined for compliance with the RR interphase group of complete compensatory pause and approximate premature beat type
Range where number, and in response to the number location, the value of coefficient is calculated, and thus mould is calculated in coefficient
The score of type three；
The rule model four determines all waveform P wave R wave height ratios, whether determines each waveform P wave R wave height ratio
In threshold range, determine that P wave R wave height than the waveform accounting in threshold range, determines P wave R wave height ratio in threshold value model
The range where interior waveform accounting is enclosed, and in response to the accounting location, the value of coefficient is calculated, and thus coefficient
The score of model four is calculated.The third aspect of the present invention provides a kind of electrocardiogram auricular fibrillation real-time judge system, should
Judgement system includes:
Memory and one or more processors；
Wherein, the memory is connect with one or more of processor communications, and being stored in the memory can quilt
The instruction that one or more of processors execute, described instruction is executed by one or more of processors, so that described one
A or multiple processors are for executing method above-mentioned.
The fourth aspect of the present invention provides a kind of computer readable storage medium, is stored thereon with the executable finger of computer
It enables, when the computer executable instructions are executed by a computing apparatus, is operable to execute method above-mentioned.
In conclusion the present invention provides a kind of electrocardiogram auricular fibrillation real-time judge method, apparatus, system and its calculating
Machine storage medium, method provided by the invention can be based on wearable device, analyze electrocardiogram in real time.This method is extracted between RR
The features such as phase, root mean square, wavelet coefficient make full use of the hiding information in waveform, binding rule judgement and machine learning model,
Auricular fibrillation is judged.Finally, the judging result for merging multiple models provides the marking to atrial fibrillation, i.e., with atrial fibrillation
Suspected Degree.
Above-mentioned technical proposal of the invention has following beneficial technical effect:
1. usually only using the correlated characteristic of RR interphase compared with the prior art, this method extracts RR interphase, and root mean square is small
The features such as wave system number, make full use of the hiding information in waveform, binding rule judgement and machine learning model, to auricular fibrillation into
Row judgement.
2. judgement has delay or takes a long time compared with the prior art, this method only needs to acquire the electrocardiogram of 20s, can
To analyze and determine auricular fibrillation in real time.
3. final result can provide the absolute judgement for whether suffering from atrial fibrillation compared with the prior art, this method provides result and is
A possibility that with atrial fibrillation.The more conducively decision of user.
Detailed description of the invention
Fig. 1 is the method general flow chart that the embodiment of the present invention is used to detect auricular fibrillation；
Fig. 2 is the flow chart that the embodiment of the present invention is used to detect rule model in the method for auricular fibrillation；
Fig. 3 is the algorithm flow chart for the method rule model one that the embodiment of the present invention is used to detect auricular fibrillation；
Fig. 4 is the algorithm flow chart for the method rule model two that the embodiment of the present invention is used to detect auricular fibrillation；
Fig. 5 is the algorithm flow chart for the method rule model three that the embodiment of the present invention is used to detect auricular fibrillation；
Fig. 6 is the algorithm flow chart for the method rule model four that the embodiment of the present invention is used to detect auricular fibrillation；
Fig. 7 is the algorithm flow chart that the embodiment of the present invention is used to detect machine learning model in the method for auricular fibrillation.
Specific embodiment
In order to make the objectives, technical solutions and advantages of the present invention clearer, With reference to embodiment and join
According to attached drawing, the present invention is described in more detail.It should be understood that these descriptions are merely illustrative, and it is not intended to limit this hair
Bright range.In addition, in the following description, descriptions of well-known structures and technologies are omitted, to avoid this is unnecessarily obscured
The concept of invention.
A kind of electrocardiogram auricular fibrillation real-time intelligent judgment method provided according to the present invention, including filtering, waveform recognition,
Segmentation, rule model, machine learning model.As shown in Figure 1.
Fig. 1 is the method general flow chart that embodiment according to the present invention is used to detect auricular fibrillation.It is adopted by portable hard
The electrocardiogram of the original 20s duration of collection obtains the available electrocardiogram of removal interference after filtering.Next, to available electrocardio
Figure carries out waveform recognition detection, identifies P wave and QRS complex.By the position of P wave and QRS complex, available electrocardiogram is divided
Section.Then by the waveform of marker characteristic point and electrocardiogram wave band difference input rule model and machine learning model.Finally, by two
The result of large-sized model output is merged, and final result, i.e. atrial fibrillation Suspected Degree are obtained.
The filtering, using wavelet thresholding methods de-noising.Using db6 small echo, 8 layers of signal are decomposed.Decomposition obtains small
Wave system number, is handled by Soft thresholding, obtains new wavelet coefficient.Signal reconstruction is carried out by new wavelet coefficient again, is filtered
Electrocardiosignal afterwards.
The waveform recognition detects QRS complex based on B- batten biorthogonal wavelet, determines the position of Q, R, S point.In addition,
P wave is detected based on first-order difference, determines P point position.
The segmentation, using the position of 0.3s before R wave wave crest as starting point, the position of 0.3s is as terminating after R wave wave crest
Point is clapped as a heart.6 heart bats are divided into a wave band, a sample as the machine learning model.
Described rule model, including parameter calculating, rejecting outliers, four rule models, result fusion etc., such as Fig. 2 institute
Show.
Fig. 2 is the flow chart that embodiment according to the present invention is used to detect rule model in the method for auricular fibrillation.For root
It is judged that rule carries out auricular fibrillation judgement to 20s electrocardiogram, parameter calculating is carried out by the position of P wave, QRS wave, obtains a system
Arrange RR interphase, P wave height and R wave height.Then exceptional value interim between detection RR and removal.Next, according to RR interphase
Maximum minimum is than, deviation, the number and P wave P wave of the RR interphase group for meeting complete compensatory pause and approximate premature beat type
The features such as height ratio, are judged respectively in four rule models.Finally, the result fusion of rule model one, two, three, four
After obtain final result.
The test of outlier, it is interim between determining RR to whether there is exceptional value, calculate the mean value of all RR interphases.It is right
Whether each RR interphase makes it less than 1.6 times of mean values and greater than the judgement of 0.5 times of mean value.If exceeding threshold range, it is determined that
The RR interphase is exceptional value and removes.
The rule model one needs to input RR interphase maximum minimum ratio, determines interim maximum between all RR
And minimum, it determines RR interphase maximum minimum ratio, determines range of the RR interphase maximum minimum than place, and in response to
The value of coefficient is calculated than the range at place in the maximum minimum, and thus the score of model one is calculated in coefficient.
The rule model one includes that 4 layers of judgement and score calculate, as shown in Figure 3.
Fig. 3 is the algorithm flow that embodiment according to the present invention is used to detect rule model one in the method for auricular fibrillation
Figure.By the maximum minimum ratio q input model of all RR interphases in 20s, the judgement of ratio q range is made.If q≤1.1,
Then parameter d=5；If 1.1 q≤1.9 <, calculating parameter d=-4.745 × q+10.2195；If 1.9 q≤2.1 <, parameter d
=1.204；If 2.1 q≤3 <, calculating parameter d=-0.5677 × q+2.3962；If q > 3, then parameter d=0.6931.Then
The score S1=100exp (- d) of computation model one.
The rule model two needs the RR interphase accounting that it is poor that input deviation is above standard, including 3 layers of judgement and score meter
It calculates, determines the mean value and standard deviation of all RR interphases, determine each RR interphase and whether mean bias is above standard difference, determine inclined
The RR interphase number accounting that it is poor that difference is above standard, the range where the RR interphase number accounting that it is poor that determination deviation is above standard, and
In response to the number accounting range, the value of coefficient is calculated, and thus the score of model two, such as Fig. 4 is calculated in coefficient
It is shown.
Fig. 4 is the algorithm flow that embodiment according to the present invention is used to detect rule model two in the method for auricular fibrillation
Figure.The mean value and standard deviation that all RR interphases in 20s are first calculated before starting, then calculate the inclined of each RR interphase and mean value
Difference, makes and whether the deviation of mean value is greater than the judgement of standard deviation.By the RR interphase accounting r input model that it is poor that deviation is above standard
Two.Make the judgement of accounting r range.If r≤0.25, parameter d=5；If 0.25 r≤0.35 <, calculating parameter d=-
26.974×r+11.7435；If 0.35 r≤0.45 <, calculating parameter d=-10.896 × r+6.1477；If r > 0.45, then
Parameter d=1.204.Then the score S2=100exp (- d) of computation model two.
The rule model three needs to input for meeting complete compensatory pause and the RR interphase group of approximate premature beat type
Number, including 4 layers of judgement and score calculate, and are handled as a RR interphase group continuous 4 RR interphases, determine between each RR
Whether phase group meets complete compensatory pause, determines the whether approximate premature beat type of each RR interphase group, is determined for compliance with compensatory completely
It has a rest and the RR interphase group number of approximate premature beat type, is determined for compliance with the RR interphase of complete compensatory pause and approximate premature beat type
Range where group number, and in response to the number location, the value of coefficient is calculated, and thus coefficient is calculated
The score of model three, as shown in Figure 5.
Fig. 5 is the algorithm flow that embodiment according to the present invention is used to detect rule model three in the method for auricular fibrillation
Figure.Four continuous RR interphases are by as a RR interphase group.To determine whether a RR interphase group meets complete compensatory pause,
The sum of 2nd RR interphase and the 3rd RR interphase is compared with average RR interphase.If the 2nd RR interphase with it is the 3rd RR interphase and small
In 2.2 times of average RR interphases and it is greater than 1.1 times of average RR interphase, it is determined that the 2nd RR interphase has met with the 3rd RR interphase
Full compensatory pause.Next, this four RR interphases are made comparisons to determine if to meet approximate premature beat type.If first
RR interphase is greater than the 2nd RR interphase and the 3rd RR interphase is greater than the 4th RR interphase, it is determined that it is approximate premature beat type.It will meet
The number n input model three of the RR interphase group of complete compensatory pause and approximate premature beat type.Make the judgement of number n range.
If n≤1, parameter d=5；If n=2, parameter d=2.9957；If n=3, parameter d=1.8971；If n=4, then parameter
D=1.204；If n >=5, parameter d=0.6971.Then the score S3=100exp (- d) of computation model three.
The rule model four needs to input each waveform P wave R wave height ratio, including calculates PR height ratio in threshold range
Number, accounting, 4 layers of judgement and the score of interior waveform calculate, and determine all waveform P wave R wave height ratios, determine each waveform P
Wave R wave height ratio determines that P wave R wave height than the waveform accounting in threshold range, determines P wave R wave whether in threshold range
Coefficient is calculated than the range where the waveform accounting in threshold range, and in response to the accounting location in height
Value, and thus the score of model four is calculated in coefficient.As shown in Figure 6.
Fig. 6 is the algorithm flow that embodiment according to the present invention is used to detect rule model four in the method for auricular fibrillation
Figure.By each waveform P wave R wave height than input model four.It is determining P wave R wave height than whether being normal P wave R wave height ratio,
Make P wave R wave height than whether the judgement in threshold range.If P wave R wave height ratio is within the scope of 0.1-0.2, it is determined that
It is normal level ratio, is included in.Then calculating P wave R wave height, all waveforms in 20s account for than the waveform in threshold range
Than.Make the judgement of waveform accounting p range in the normal range.If p≤0.4, parameter d=5；If 0.4 p≤0.6 <,
Calculating parameter d=-10.0215 × p+9.0086；If 0.6 p≤0.8 <, calculating parameter d=-8.9585 × p+8.3708；If
0.8 p≤0.9 <, then calculating parameter d=-5.109 × p+5.2912；If p > 0.9, then parameter d=0.6931.Then mould is calculated
The score S4=100exp (- d) of type four.
Result fusion, make first one result S1 of model value whether the judgement for being 0.If S1=0, final score
S=0；If S1 ≠ 0, final score S=S1+S2-S3-S4 is calculated.This is the final score of rule model.
Machine learning model described in Fig. 1, including feature extraction, standardization, dimensionality reduction, 5 machine learning models, moulds
Type fusion, as shown in Figure 7.
Fig. 7 is the algorithm flow that embodiment according to the present invention is used to detect machine learning model in the method for auricular fibrillation
Figure.To the electrocardiogram section as unit of 6 hearts bats completed is divided, progress feature extraction first uses feature: between RR here
Phase, average absolute amplitude, root mean square, wavelet coefficient.After feature normalization, principal component analysis dimensionality reduction is used.Next by sample
Originally linear discriminant analysis, gauss hybrid models, least square method supporting vector machine, reverse transmittance nerve network, probability mind are input to
Through respectively obtaining classification results in network model.Finally merge each model result to obtain final result.
Sample is carried out 4 layers of decomposition using db4 small echo, takes the wavelet coefficient conduct of a4 frequency range by the Wavelet Coefficients Characteristic
Feature.
The principal component analysis retains preceding several principal components that its weight is more than 98%, completes dimensionality reduction.
Result is divided into 2 classes, atrial fibrillation and non-atrial fibrillation by the linear discriminant analysis.
Sample is equally divided into atrial fibrillation and non-atrial fibrillation aiming at the problem that two classification by the gauss hybrid models.Its parameter is estimated
Meter uses EM algorithm Sum Maximum Likelihood Estimate.
The least square method supporting vector machine uses radial basis function as kernel function to classify.
The reverse transmittance nerve network includes 1 layer of input layer, 1 layer of hidden layer and 1 layer of output layer.Wherein input layer
Dimension after neuronal quantity, that is, sample Feature Dimension Reduction；Hidden layer is set as 6 neurons；Output layer includes 2 neurons,
That is atrial fibrillation and non-atrial fibrillation.
The probabilistic neural network includes 1 layer of radial base and one layer of competition layer.Radial base includes atrial fibrillation and non-atrial fibrillation
Two neurons.Competition layer determines the class of maximum probability and to its assignment 1.
Above-mentioned 5 kinds of machine learning models fulfil training ahead of schedule using the data that hospital doctor has marked, and obtain model ginseng
Number.Directly input waveform is judged when analysis.
The Model Fusion, after a sample, which inputs each model, obtains result, all models are voted together, are provided
Respective result.Wherein ratio, that is, machine learning model atrial fibrillation shared by atrial fibrillation judges score.
Finally, the result of rule model and machine learning model is integrated.To determine that the result of two large-sized models whether there is
The two is compared by larger difference.If the difference of the two score is greater than 0.6, this result is cancelled.Otherwise the two is made even
It is final atrial fibrillation Suspected Degree.
Another aspect provides a kind of electrocardiogram auricular fibrillation real-time judge devices, comprising:
Electrocardiogram preprocessing module, available electrocardiogram for obtaining being labeled with wave character and carries out available electrocardiogram
The electrocardiogram wave band of segmentation；
Rule model will be labeled with the available electrocardiogram input rule model of wave character, defeated after rule model judges
Rule model result out；
The electrocardiogram wave band of segmentation is inputted machine learning model, after machine learning model judges by machine learning model
Obtain machine learning model result；
Rule model result and machine learning model result are obtained atrial fibrillation Suspected Degree by fusion calculation by computing unit
Final result.
Further, the electrocardiogram preprocessing module includes:
Filter unit is filtered the electrocardiogram of acquired original, for removing signal noise, obtains available electrocardiogram；
Waveform recognition unit can carry out waveform recognition with electrocardiogram to described, identify and mark the P wave in electrocardiosignal and
QRS complex；
Waveform segments unit carries out waveform segments to the available electrocardiogram of mark, and waveform, which is divided into, certain can be used for machine
The wave band of device study, obtains electrocardiogram wave band.
Further, the rule model includes:
Parameter calculation unit calculates RR interphase, P wave height and R wave height according to the available electrocardiogram of input；
RR interphase rejecting outliers unit detects interim exceptional value and removal between RR；
To the maximum minimum of RR interphase than judged rule model one, RR interphase and mean bias are sentenced
Disconnected rule model two, the regular mould that the RR interphase group number for meeting complete compensatory gap and approximate premature beat type is judged
Type three, to P wave R wave height than the rule model four that the waveform accounting in threshold range is judged, aforementioned four rule mould
Type respectively obtains one result S1 of model, four result S4 of two result S2 of model, three result S3 of model and model by judgement；
Judging unit, judges whether S1 is equal to 0: if so, S=0；If it is not, then S=S1+S2-S3-S4；
Output unit exports result S.
Further, machine learning model judgement includes:
Feature extraction unit extracts RR interphase, average absolute amplitude, root mean square, wavelet systems to the electrocardiogram wave band of input
Number is characterized；
Standardisation Cell is standardized the feature, for making the feature ruler having the same of different dimensions
Degree.Here it is standardized using z-score: the standard deviation of y=(average value of x-X)/X；X is the set of all values of this feature, x
It is single value.
Dimensionality reduction unit: for reducing the dimension of feature；
Machine learning submodel is linear discriminant analysis respectively, gauss hybrid models, least square method supporting vector machine, anti-
To Propagation Neural Network, probabilistic neural network model, the final feature obtained after dimensionality reduction is input to above-mentioned five engineerings
It practises and being judged in submodel, respectively obtain judging result；
Computing unit, each judging result carry out fusion calculation and obtain final result.
It is yet another aspect of the present invention to provide a kind of electrocardiogram auricular fibrillation real-time judge system, which includes:
Memory and one or more processors；
Wherein, the memory is connect with one or more of processor communications, and being stored in the memory can quilt
The instruction that one or more of processors execute, described instruction is executed by one or more of processors, so that described one
A or multiple processors are for executing method above-mentioned.
Last aspect of the invention provides a kind of computer readable storage medium, and it is executable to be stored thereon with computer
Instruction, when the computer executable instructions are executed by a computing apparatus, is operable to execute method above-mentioned.
In conclusion the present invention provides a kind of electrocardiogram auricular fibrillation real-time judge method, apparatus, system and its calculating
Machine storage medium, method provided by the invention can be based on wearable device, analyze electrocardiogram in real time.This method is extracted between RR
The features such as phase, root mean square, wavelet coefficient make full use of the hiding information in waveform, binding rule judgement and machine learning model,
Auricular fibrillation is judged.Finally, the judging result for merging multiple models provides the marking to atrial fibrillation, i.e., with atrial fibrillation
Suspected Degree.
It should be understood that above-mentioned specific embodiment of the invention is used only for exemplary illustration or explains of the invention
Principle, but not to limit the present invention.Therefore, that is done without departing from the spirit and scope of the present invention is any
Modification, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.In addition, appended claims purport of the present invention
Covering the whole variations fallen into attached claim scope and boundary or this range and the equivalent form on boundary and is repairing
Change example.
Claims (14)
1. a kind of electrocardiogram auricular fibrillation real-time judge method, which comprises the steps of:
Step 1, electrocardiogram pretreatment, available electrocardiogram for obtaining being labeled with wave character and divide available electrocardiogram
The electrocardiogram wave band of section；
Step 2, the available electrocardiogram input rule model that will be labeled with wave character are based on room escape medicine through rule model
Rule model result is exported after judgment rule judgement；The electrocardiogram wave band input of segmentation is passed through into the trained machine of hospital data
Learning model obtains machine learning model result after machine learning model judges；
Rule model result and machine learning model result are obtained the most termination of atrial fibrillation Suspected Degree by fusion calculation by step 3
Fruit.
2. electrocardiogram auricular fibrillation real-time judge method as described in claim 1, which is characterized in that the electrocardio of the step 1
Figure pretreatment includes the following steps:
Step 11 is filtered the electrocardiogram of acquired original, for removing signal noise, obtains available electrocardiogram；
Step 12 can carry out waveform recognition with electrocardiogram to described, identify and mark P wave and QRS complex in electrocardiosignal；
Step 13 carries out waveform segments to the available electrocardiogram of mark, and waveform is divided into certain wave that can be used for machine learning
Section, obtains electrocardiogram wave band.
3. electrocardiogram auricular fibrillation real-time judge method as claimed in claim 2, which is characterized in that the filtering uses small echo
Threshold method carries out de-noising, is decomposed 8 layers of signal using db6 small echo, and the wavelet coefficient decomposed is handled by Soft thresholding,
New wavelet coefficient is obtained, signal reconstruction is carried out by new wavelet coefficient, obtains filtered electrocardiosignal, electrocardio can be used to be described
Figure；
And/or the waveform recognition is based on B- batten biorthogonal wavelet and detects QRS complex, determines the position of Q, R, S point；It is based on
First-order difference detects P wave, determines P point position；
And/or the waveform segments are using the position of 0.3s before R wave wave crest as starting point, the position conduct of 0.3s after R wave wave crest
End point is clapped as a heart, and 6 heart bats are divided into a wave band as an electrocardiogram wave band sample and input the machine
Device learning model.
4. electrocardiogram auricular fibrillation real-time judge method as described in claim 1, which is characterized in that the rule of the step 2
Model judgement includes the following steps:
Step A21, RR interphase, P wave height and R wave height are calculated according to the available electrocardiogram of input；
Step A22, exceptional value interim between detection RR and removal；
Step A23, by the maximum minimum to RR interphase than the rule model one, inclined to RR interphase and mean value that is judged
Rule model two that difference is judged judges the RR interphase group number for meeting complete compensatory gap and approximate premature beat type
Rule model three, more aforementioned four than the rule model four that the waveform accounting in threshold range is judged to P wave R wave height
The judgement of rule model respectively obtains one result S1 of model, four result S4 of two result S2 of model, three result S3 of model and model；
Step A24, judge whether S1 is equal to 0: if so, S=0；If it is not, then S=S1+S2-S3-S4；
Step A25, S is exported.
5. auricular fibrillation real-time intelligent judgment method as claimed in claim 4, which is characterized in that different in the step A22
Constant value detection is interim with the presence or absence of exceptional value between RR for determining, calculates the mean value of all RR interphases first；Judge between each RR
Whether the value of phase is greater than 0.5 times of mean value and less than 1.6 times mean values, if it is not, exceeding threshold range, it is determined that the RR interphase is different
Constant value simultaneously removes.
6. auricular fibrillation real-time intelligent judgment method as claimed in claim 4, which is characterized in that
Rule model one in the step A23 includes:
(1) input can use the maximum of all RR interphases and the ratio q of minimum in electrocardiogram, and initiation parameter d, make d=
0；
(2) q value is judged:
If q≤1.1, parameter d=5；
If 1.1 q≤1.9 <, calculating parameter d=-4.745 × q+10.2195；
If 1.9 q≤2.1 <, parameter d=1.204；
If 2.1 q≤3 <, calculating parameter d=-0.5677 × q+2.3962；
If q > 3, then parameter d=0.6931；
(3) the score S1=100exp (- d) of rule model one is calculated；
Rule model two in the step A23 includes:
(1) mean value and standard deviation of all RR interphases in electrocardiogram can be used by calculating；Then the deviation of each RR interphase and mean value is calculated,
Judge whether the deviation is greater than standard deviation；Calculate the RR interphase accounting r that it is poor that deviation is above standard；
(2) the RR interphase accounting r, and initiation parameter d are inputted, d=0 is made；
(3) r value is judged:
If r≤0.25, parameter d=5；
If 0.25 r≤0.35 <, calculating parameter d=-26.974 × r+11.7435；
If 0.35 r≤0.45 <, calculating parameter d=-10.896 × r+6.1477；
If r > 0.45, then parameter d=1.204；
(4) the score S2=100exp (- d) of rule model two is calculated；
Rule model three in the step A23 includes:
It (1) can be with the number n for the RR interphase group for meeting complete compensatory pause and approximate premature beat type in electrocardiogram described in judgement: enabling
Four continuous RR interphases are a RR interphase group, by the sum of the 2nd RR interphase and the 3rd RR interphase compared with average RR interphase
Compared with, if average RR interphase of the sum less than 2.2 times of the 2nd RR interphase and the 3rd RR interphase and the average RR interphase greater than 1.1 times,
Then determine that the 2nd RR interphase and the 3rd RR interphase meet complete compensatory pause；If the first RR interphase is greater than the 2nd RR interphase and the
Three RR interphases are greater than the 4th RR interphase, it is determined that it is approximate premature beat type；Meet the RR interphase group of above-mentioned two condition simultaneously
The then RR interphase group to meet complete compensatory pause and approximate premature beat type；
(2) meet the number n of complete compensatory pause and the RR interphase group of approximate premature beat type, and initiation parameter d described in input,
Make d=0；
(3) n value is judged:
If n≤1, parameter d=5；
If n=2, parameter d=2.9957；
If n=3, parameter d=1.8971；
If n=4, then parameter d=1.204；
If n >=5, parameter d=0.6971；
(4) the score S3=100exp (- d) of rule model three is calculated；
Rule model four in the step A23 includes:
(1) each waveform P wave R wave height ratio in electrocardiogram can be used described in input；
(2) if P wave R wave height ratio is within the scope of 0.1-0.2, it is determined that it is normal level ratio, is included in；Calculate P wave R wave height
It spends than waveform within the above range in the accounting p with all waveforms in electrocardiogram；And initiation parameter d, make d=0；
(3) p value is judged:
If p≤0.4, parameter d=5；
If 0.4 p≤0.6 <, calculating parameter d=-10.0215 × p+9.0086；
If 0.6 p≤0.8 <, calculating parameter d=-8.9585 × p+8.3708；
If 0.8 p≤0.9 <, calculating parameter d=-5.109 × p+5.2912；
If p > 0.9, then parameter d=0.6931；
(4) the score S4=100exp (- d) of rule model four is calculated.
7. electrocardiogram auricular fibrillation real-time judge method as described in claim 1, which is characterized in that the machine in the step 2
The judgement of device learning model includes the following steps:
Step B21, RR interphase, average absolute amplitude, root mean square, wavelet coefficient is extracted to the electrocardiogram wave band of input to be characterized,
4 layers are decomposed to using heart bat of the db4 small echo to input, takes the wavelet coefficient of a4 frequency range as feature；
Step B22, z-score standardization is carried out to the feature, for making the feature ruler having the same of different dimensions
Degree；
Step B23, dimensionality reduction: for reducing the dimension of feature, retain the principal component that weight is more than 98%；
Step B24, the final feature obtained after dimensionality reduction is input to five machine learning submodels, is linear discriminant point respectively
Analysis, least square method supporting vector machine, reverse transmittance nerve network, is sentenced in probabilistic neural network model gauss hybrid models
It is disconnected, respectively obtain judging result；Result is divided into 2 classes, atrial fibrillation and non-atrial fibrillation by the linear discriminant analysis；The Gaussian Mixture mould
Type is divided into atrial fibrillation and non-atrial fibrillation aiming at the problem that two classification, by sample, and parameter Estimation uses EM algorithm Sum Maximum Likelihood Estimate；
The least square method supporting vector machine uses radial basis function as kernel function to classify；The reverse transmittance nerve network
Include 1 layer of input layer, 1 layer of hidden layer and 1 layer of output layer；Wherein after neuronal quantity, that is, sample Feature Dimension Reduction of input layer
Dimension；Hidden layer is set as 6 neurons；Output layer includes 2 neurons, i.e. atrial fibrillation and non-atrial fibrillation；The probabilistic neural net
Network includes 1 layer of radial base and one layer of competition layer；Radial base includes two neurons of atrial fibrillation and non-atrial fibrillation；Competition layer determines general
The maximum class of rate simultaneously gives its assignment 1；
Step B25, each judging result is merged to obtain final result.
8. electrocardiogram auricular fibrillation real-time judge method as claimed in claim 7, which is characterized in that each machine learning submodule
Type is respectively trained, and obtains model parameter with data training in advance, only has parameter in the portable device that the method is based on,
Directly the waveform of input is judged.
9. electrocardiogram auricular fibrillation real-time judge method as claimed in claim 7, which is characterized in that each in the step B25
The result of submodel using ballot and in proportion give a mark by the way of merge, i.e., ratio shared by atrial fibrillation is in the result of each submodel
The judgement final result of machine learning model.
10. such as the described in any item auricular fibrillation real-time judge methods of claim 1-7, which is characterized in that the rule model
With the fusion of machine learning model judging result are as follows: determine two results with the presence or absence of larger difference, if the difference of the two score
Value is greater than 0.6, then this result is cancelled；Otherwise the two is averaged as final atrial fibrillation Suspected Degree.
11. a kind of electrocardiogram auricular fibrillation real-time judge device characterized by comprising
Electrocardiogram preprocessing module, available electrocardiogram for obtaining being labeled with wave character and is segmented available electrocardiogram
Electrocardiogram wave band；The electrocardiogram preprocessing module includes: filter unit, is filtered to the electrocardiogram of acquired original, is used
In removal signal noise, available electrocardiogram is obtained；Waveform recognition unit can carry out waveform recognition with electrocardiogram to described, identification
And mark P wave and QRS complex in electrocardiosignal；
Waveform segments unit carries out waveform segments to the available electrocardiogram of mark, and waveform, which is divided into, certain can be used for engineering
The wave band of habit obtains electrocardiogram wave band；
Rule model will be labeled with the available electrocardiogram input rule model of wave character, and rule are exported after rule model judges
Then model result；The rule model includes: parameter calculation unit, calculates RR interphase, P wave height according to the available electrocardiogram of input
Degree and R wave height；RR interphase rejecting outliers unit detects interim exceptional value and removal between RR；To the maximum of RR interphase
Minimum than the rule model one judged, the rule model two judged with mean bias to RR interphase, to having met
Rule model three that the RR interphase group number of complete compensatory gap and approximate premature beat type is judged, to P wave R wave height ratio in threshold
The rule model four that waveform accounting within the scope of value is judged, aforementioned four rule model respectively obtain model one by judgement
As a result S1, four result S4 of two result S2 of model, three result S3 of model and model；Judging unit, judges whether S1 is equal to 0: if it is not,
Then S=S1+S2-S3-S4；If so, S=0；Output unit exports result S；
The electrocardiogram wave band of segmentation is inputted machine learning model, obtained after machine learning model judges by machine learning model
Machine learning model result；The machine learning model judgement includes: feature extraction unit, is extracted to the electrocardiogram wave band of input
RR interphase, average absolute amplitude, root mean square, wavelet coefficient are characterized；Standardisation Cell carries out z-score mark to the feature
Quasi-ization processing, for making the feature scale having the same of different dimensions；Dimensionality reduction unit: for reducing the dimension of feature；Machine
Learn submodel, is linear discriminant analysis, gauss hybrid models, least square method supporting vector machine, backpropagation neural network respectively
The feature obtained after dimensionality reduction is input in above-mentioned five machine learning submodels and judges by network, probabilistic neural network model,
Respectively obtain judging result；Machine learning model computing unit, each judging result carry out fusion calculation and obtain final result.
Rule model result and machine learning model result are obtained the final of atrial fibrillation Suspected Degree by fusion calculation by computing unit
As a result.
12. auricular fibrillation real-time judge device as claimed in claim 11, which is characterized in that
The rule model one determines maximum and minimum interim between all RR, determines RR interphase maximum minimum ratio,
It determines range of the RR interphase maximum minimum than place, and the range in response to the maximum minimum than place, calculates
The value of coefficient is obtained, and thus the score of model one is calculated in coefficient；
The rule model two determines the mean value and standard deviation of all RR interphases, determines whether are each RR interphase and mean bias
Be above standard difference, the RR interphase number accounting that it is poor that determination deviation is above standard, the RR interphase number that it is poor that determination deviation is above standard
Range where accounting, and in response to the number accounting range, the value of coefficient is calculated, and thus mould is calculated in coefficient
The score of type two；
The rule model three handles continuous 4 RR interphases as a RR interphase group, determines that each RR interphase group is
It is no to meet complete compensatory pause, determine the whether approximate premature beat type of each RR interphase group, be determined for compliance with complete compensatory pause and
The RR interphase group number of approximate premature beat type is determined for compliance with the RR interphase group number of complete compensatory pause and approximate premature beat type
The range at place, and in response to the number location, the value of coefficient is calculated, and thus model three is calculated in coefficient
Score；
Whether the rule model four determines all waveform P wave R wave height ratios, determine each waveform P wave R wave height ratio in threshold
It is worth in range, determines that P wave R wave height than the waveform accounting in threshold range, determines P wave R wave height ratio in threshold range
Waveform accounting where range the value of coefficient is calculated and in response to the accounting location, and thus coefficient calculates
Obtain the score of model four.
13. a kind of electrocardiogram auricular fibrillation real-time judge system, which is characterized in that the judgement system includes:
Memory and one or more processors；
Wherein, the memory is connect with one or more of processor communications, and being stored in the memory can be described
The instruction that one or more processors execute, described instruction executed by one or more of processors so that it is one or
Multiple processors require the described in any item methods of 1-10 for perform claim.
14. a kind of computer readable storage medium, is stored thereon with computer executable instructions, refer to when the computer is executable
When order is executed by a computing apparatus, it is operable to perform claim and requires the described in any item methods of 1-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810760168.0A CN109117730B (en) | 2018-07-11 | 2018-07-11 | Real-time electrocardiogram atrial fibrillation judgment method, device and system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810760168.0A CN109117730B (en) | 2018-07-11 | 2018-07-11 | Real-time electrocardiogram atrial fibrillation judgment method, device and system and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109117730A true CN109117730A (en) | 2019-01-01 |
| CN109117730B CN109117730B (en) | 2021-05-11 |
Family
ID=64862095
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810760168.0A Active CN109117730B (en) | 2018-07-11 | 2018-07-11 | Real-time electrocardiogram atrial fibrillation judgment method, device and system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109117730B (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109948647A (en) * | 2019-01-24 | 2019-06-28 | 西安交通大学 | A kind of electrocardiogram classification method and system based on depth residual error network |
| CN110151164A (en) * | 2019-03-14 | 2019-08-23 | 南京赢创智金科技成果转化有限公司 | A kind of perimeter mapping method for auricular fibrillation |
| CN110807443A (en) * | 2019-11-27 | 2020-02-18 | 南京信息工程大学 | Somatosensory network electrocardiosignal segmentation method based on parameter self-adaptive HMM model |
| CN110826631A (en) * | 2019-11-08 | 2020-02-21 | 广州视源电子科技股份有限公司 | Arrhythmia disease classification method and device, storage medium and processor |
| CN111000553A (en) * | 2019-12-30 | 2020-04-14 | 山东省计算中心（国家超级计算济南中心） | Intelligent classification method for electrocardiogram data based on voting ensemble learning |
| CN111419220A (en) * | 2020-03-27 | 2020-07-17 | 山东省人工智能研究院 | Resonance sparse decomposition-based twelve-lead electrocardiosignal atrial fibrillation detection method |
| CN111839505A (en) * | 2020-07-02 | 2020-10-30 | 复旦大学 | Atrial fibrillation classification method based on cardiovascular system electrical-mechanical activity information |
| CN112244860A (en) * | 2020-09-27 | 2021-01-22 | 山东平伟医疗技术有限公司 | Dynamic electrocardiogram data analysis method |
| CN112826514A (en) * | 2019-11-22 | 2021-05-25 | 华为技术有限公司 | Atrial fibrillation signal classification method, device, terminal and storage medium |
| CN113543710A (en) * | 2019-03-08 | 2021-10-22 | 国立大学法人东京医科齿科大学 | Atrial fibrillation analysis device, atrial fibrillation analysis method, and program |
| CN113940682A (en) * | 2021-11-10 | 2022-01-18 | 浙江好络维医疗技术有限公司 | Atrial fibrillation identification method based on statistical characteristics |
| CN113995419A (en) * | 2021-12-10 | 2022-02-01 | 华中科技大学 | Atrial fibrillation risk prediction system based on heartbeat rhythm signal and application thereof |
| WO2022073220A1 (en) * | 2020-10-10 | 2022-04-14 | 上海市第一人民医院 | Atrial fibrillation detection device and method, and system and storage medium |
| US20220125376A1 (en) * | 2019-02-13 | 2022-04-28 | Kyoto University | Sleep apnea syndrome determination apparatus, sleep apnea syndrome determination method, and sleep apnea syndrome determination program |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080167567A1 (en) * | 2006-09-19 | 2008-07-10 | The Cleveland Clinic Foundation | Prediction and prevention of postoperative atrial fibrillation in cardiac surgery patients |
| CN106073755A (en) * | 2016-05-27 | 2016-11-09 | 成都信汇聚源科技有限公司 | The implementation method that in a kind of miniature holter devices, atrial fibrillation identifies automatically |
| 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 |
| CN107951485A (en) * | 2017-11-27 | 2018-04-24 | 乐普（北京）医疗器械股份有限公司 | Ambulatory ECG analysis method and apparatus based on artificial intelligence self study |
-
2018
- 2018-07-11 CN CN201810760168.0A patent/CN109117730B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080167567A1 (en) * | 2006-09-19 | 2008-07-10 | The Cleveland Clinic Foundation | Prediction and prevention of postoperative atrial fibrillation in cardiac surgery patients |
| CN106073755A (en) * | 2016-05-27 | 2016-11-09 | 成都信汇聚源科技有限公司 | The implementation method that in a kind of miniature holter devices, atrial fibrillation identifies automatically |
| 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 |
| CN107951485A (en) * | 2017-11-27 | 2018-04-24 | 乐普（北京）医疗器械股份有限公司 | Ambulatory ECG analysis method and apparatus based on artificial intelligence self study |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109948647A (en) * | 2019-01-24 | 2019-06-28 | 西安交通大学 | A kind of electrocardiogram classification method and system based on depth residual error network |
| US20220125376A1 (en) * | 2019-02-13 | 2022-04-28 | Kyoto University | Sleep apnea syndrome determination apparatus, sleep apnea syndrome determination method, and sleep apnea syndrome determination program |
| CN113543710A (en) * | 2019-03-08 | 2021-10-22 | 国立大学法人东京医科齿科大学 | Atrial fibrillation analysis device, atrial fibrillation analysis method, and program |
| CN110151164B (en) * | 2019-03-14 | 2024-03-19 | 无锡市人民医院 | Zhou Changbiao assay for atrial fibrillation |
| CN110151164A (en) * | 2019-03-14 | 2019-08-23 | 南京赢创智金科技成果转化有限公司 | A kind of perimeter mapping method for auricular fibrillation |
| CN110826631A (en) * | 2019-11-08 | 2020-02-21 | 广州视源电子科技股份有限公司 | Arrhythmia disease classification method and device, storage medium and processor |
| CN110826631B (en) * | 2019-11-08 | 2022-11-08 | 广州视源电子科技股份有限公司 | Arrhythmia disease classification method and device, storage medium and processor |
| CN112826514A (en) * | 2019-11-22 | 2021-05-25 | 华为技术有限公司 | Atrial fibrillation signal classification method, device, terminal and storage medium |
| CN112826514B (en) * | 2019-11-22 | 2022-07-22 | 华为技术有限公司 | Atrial fibrillation signal classification method, device, terminal and storage medium |
| CN110807443A (en) * | 2019-11-27 | 2020-02-18 | 南京信息工程大学 | Somatosensory network electrocardiosignal segmentation method based on parameter self-adaptive HMM model |
| CN110807443B (en) * | 2019-11-27 | 2023-04-25 | 南京信息工程大学 | Somatosensory network electrocardiosignal segmentation method based on parameter self-adaptive HMM model |
| CN111000553A (en) * | 2019-12-30 | 2020-04-14 | 山东省计算中心（国家超级计算济南中心） | Intelligent classification method for electrocardiogram data based on voting ensemble learning |
| CN111000553B (en) * | 2019-12-30 | 2022-09-27 | 山东省计算中心（国家超级计算济南中心） | Intelligent classification method for electrocardiogram data based on voting ensemble learning |
| CN111419220A (en) * | 2020-03-27 | 2020-07-17 | 山东省人工智能研究院 | Resonance sparse decomposition-based twelve-lead electrocardiosignal atrial fibrillation detection method |
| CN111839505A (en) * | 2020-07-02 | 2020-10-30 | 复旦大学 | Atrial fibrillation classification method based on cardiovascular system electrical-mechanical activity information |
| CN112244860A (en) * | 2020-09-27 | 2021-01-22 | 山东平伟医疗技术有限公司 | Dynamic electrocardiogram data analysis method |
| CN112244860B (en) * | 2020-09-27 | 2023-06-20 | 山东平伟医疗技术有限公司 | Dynamic electrocardiogram data analysis method |
| WO2022073220A1 (en) * | 2020-10-10 | 2022-04-14 | 上海市第一人民医院 | Atrial fibrillation detection device and method, and system and storage medium |
| CN113940682A (en) * | 2021-11-10 | 2022-01-18 | 浙江好络维医疗技术有限公司 | Atrial fibrillation identification method based on statistical characteristics |
| CN113940682B (en) * | 2021-11-10 | 2024-08-06 | 浙江好络维医疗技术有限公司 | Atrial fibrillation recognition method based on statistical characteristics |
| CN113995419A (en) * | 2021-12-10 | 2022-02-01 | 华中科技大学 | Atrial fibrillation risk prediction system based on heartbeat rhythm signal and application thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109117730B (en) | 2021-05-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109117730A (en) | Electrocardiogram auricular fibrillation real-time judge method, apparatus, system and storage medium | |
| Kropf et al. | ECG classification based on time and frequency domain features using random forests | |
| CN107951485B (en) | Ambulatory ECG analysis method and apparatus based on artificial intelligence self study | |
| CN106108889B (en) | Electrocardiogram classification method based on deep learning algorithm | |
| CN110811609B (en) | Epileptic spike intelligent detection device based on self-adaptive template matching and machine learning algorithm fusion | |
| CN108511055B (en) | Ventricular premature beat recognition system and method based on classifier fusion and diagnosis rules | |
| CN104473629B (en) | Automatic electrocardioelectrode placement error detection method based on kernel function classification algorithm | |
| US20050222508A1 (en) | Cardiac diagnostic system and method | |
| CN109276242A (en) | The method and apparatus of electrocardiosignal type identification | |
| CN109948396B (en) | Heart beat classification method, heart beat classification device and electronic equipment | |
| EP2872039B1 (en) | Device for evaluation of myocardial damages based on the current density variations | |
| CN106725420A (en) | VPB recognition methods and VPB identifying system | |
| Malek et al. | Automated detection of premature ventricular contraction in ECG signals using enhanced template matching algorithm | |
| Rohmantri et al. | Arrhythmia classification using 2D convolutional neural network | |
| CN109044340B (en) | Electrocardiogram data classification method and device, electronic equipment and storage medium | |
| CN109124620A (en) | A kind of atrial fibrillation detection method, device and equipment | |
| CN109960968A (en) | Generation method, atrial fibrillation detection device and the storage medium of classifier | |
| CN105611872A (en) | An apparatus and method for evaluating multichannel ECG signals | |
| CN109117729A (en) | Electrocardiogram room escape real-time judge method, apparatus, system and storage medium | |
| Yang et al. | Automated intra-patient and inter-patient coronary artery disease and congestive heart failure detection using EFAP-Net | |
| Qu et al. | ECG signal classification based on BPNN | |
| Papadogiorgaki et al. | Heart rate classification using ECG signal processing and machine learning methods | |
| CN111053552A (en) | QRS wave detection method based on deep learning | |
| WO2012173583A1 (en) | Method and device for evaluation of myocardial ischemia based on current density maps | |
| CN107837083B (en) | J wave automatic testing method based on least square method supporting vector machine |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210714 Address after: 518100 4th floor, manufacturing center, SEG Navigation Science Park, 28 cuibao Road, Longgang street, Longgang District, Shenzhen City, Guangdong Province Patentee after: Shenzhen Hart Intelligent Technology Co.,Ltd. Address before: 201499 room 1331, No.13, Lane 65, Huandong 1st Road, Fengjing town, Jinshan District, Shanghai Patentee before: SHANGHAI XIAXIAN ELECTROMECHANICAL SCIENCE AND TECHNOLOGY DEVELOPMENT Co.,Ltd. |
|
| TR01 | Transfer of patent right |