CN109846476A - A kind of ventricular fibrillation recognition methods based on machine learning techniques - Google Patents
A kind of ventricular fibrillation recognition methods based on machine learning techniques Download PDFInfo
- Publication number
- CN109846476A CN109846476A CN201910208962.9A CN201910208962A CN109846476A CN 109846476 A CN109846476 A CN 109846476A CN 201910208962 A CN201910208962 A CN 201910208962A CN 109846476 A CN109846476 A CN 109846476A
- Authority
- CN
- China
- Prior art keywords
- ventricular fibrillation
- electrocardiogram
- feature
- ecg
- probability
- 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
Landscapes
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The ventricular fibrillation recognition methods based on machine learning techniques that the invention discloses a kind of, comprising the following steps: original electro-cardiologic signals are handled, to extract ventricular fibrillation feature；Using marked ecg database training for judge each ECG signal whether be ventricular fibrillation Logic Regression Models；The probability that each ECG signal is ventricular fibrillation is acquired using trained model.The present invention chooses most effective combination from the various features that the prior art proposes, effectively increases the performance of ventricular fibrillation detection algorithm.Meanwhile heart rate feature is applied in ventricular fibrillation detection by the present invention.When in order to solve the problems, such as that ventricular fibrillation occurs QRS detection difficult, heart rate may serious inaccuracy, the electrocardiogram that the present invention using the quality information of signal enables model accurate judgement under what circumstances and should ignore heart rate information, which kind of by heart rate can be excluded obvious non-ventricular fibrillation in the case of.
Description
Technical field
The present invention relates to field of medical technology more particularly to a kind of ventricular fibrillation recognition methods based on machine learning techniques.
Background technique
In the prior art, relative to rule-based ventricular fibrillation detection method, the method based on machine learning techniques is proved to
Performance is more preferable, robustness is more excellent.One step of key of ventricular fibrillation detection algorithm based on machine learning techniques is the selection of characteristic parameter.
To extract feature, the prior art applies the various analysis methods based on frequency domain or time domain, such as analysis of complexity, ventricular fibrillation filtering point
Analysis, spectrum analysis, time delay algorithm, bandpass filtering analysis, the analysis of covariance, Kurtosis etc..
Due to QRS detection difficult when ventricular fibrillation occurs, the prior art often eliminates heart rate feature when detecting ventricular fibrillation, this leads
Cause the prior art that can will obviously judge that the electrocardiogram of non-ventricular fibrillation is judged as ventricular fibrillation by heart rate sometimes.
Summary of the invention
Technical problems based on background technology, the ventricular fibrillation identification based on machine learning techniques that the invention proposes a kind of
Method.
The technical solution adopted by the present invention is that:
A kind of ventricular fibrillation recognition methods based on machine learning techniques, which comprises the following steps:
(1) ECG signal is pre-processed, to filter out the noises such as baseline drift, Hz noise, will be believed except the electrocardio after making an uproar
Number it is resampled to a certain fixed sample rate；
(2) to except make an uproar and resampling after ECG signal carry out bandpass filtering analysis, calculate feature 1, method particularly includes:
Assuming that the output of bandpass filtering is FS, to each second window, the maximum value and average value of FS absolute value is calculated, is denoted as MAX_ respectively
ABS_FS and MEAN_ABS_FS；The 1st feature for detecting ventricular fibrillation is defined as filtering output absolute value in electrocardiogram and is in
The number of sample between MEAN_ABS_FS and MAX_ABS_FS accounts for the ratio of all numbers of samples of electrocardiogram；
(3) to except make an uproar and resampling after ECG signal carry out ventricular fibrillation Filtering Analysis, calculation question 2, method particularly includes:
The meaning of ventricular fibrillation filtering is explained as follows: since ventricular fibrillation electrocardiogram is similar to sine curve, energy can be distributed mainly on average frequency
Near, thus can be than non-ventricular fibrillation electrocardio for surplus after the bandstop filter of ECG signal average frequency by a centre frequency
Small more of the surplus of figure；For the average frequency of calculating ECG signal, estimate T average period of signal be it is as follows, wherein Vi is
Signal value, m are ecg samples numbers:
It can be calculated by following equation by the surplus of the ECG signal after the bandstop filter, this is as the 2nd feature:
(4) ECG-QRS Wave characteristic point, specific method are extracted using based on the method for wavelet transformation and logistic regression algorithm
Are as follows: comprehensive multiple waveforms feature, using machine learning algorithm, study obtains each feature weight from mass data automatically, differentiates and doubts
It is whether true like QRS wave；Meanwhile to each doubtful QRS wave provide one be true QRS wave probability, when signal quality compared with
When poor, the probability value is generally lower, therefore these values can be used for measuring the quality of electrocardiosignal；
(5) calculating ECG average heart rate, as the 3rd feature；
(6) each QRS wave of calculating ECG is the average value of the probability of true QRS wave, as the 4th feature；
(7) ecg database marked through electrocardio expert is chosen, each electrocardiographic recording is divided into multiple electrocardiogram segments,
Each electrocardiogram segment 10 seconds；
(8) each 10 seconds electrocardiogram segments are handled by step (1)-step (6), obtains the spy of a differentiation ventricular fibrillation
Parameter matrix is levied, the line number of the matrix is equal to the number of all 10 seconds electrocardiogram segments, and columns is equal to 4, and columns is step (1)-
The number of the ventricular fibrillation characteristic parameter obtained in step (6)；To each electrocardiogram segment, it is compared with expert's label, judges that it is
No is ventricular fibrillation；
(9) by step (8) obtain characteristic parameter matrix and whether be ventricular fibrillation one Logic Regression Models of information input
In, using gradient descent method as optimization method, setting learning rate is 0.02, and iteration is until convergence；In training, using staying 1 intersection to test
Training data is randomly divided into K parts by the method for card, take and wherein do training for K-1 parts, and remaining 1 part is verified, so repeatedly n times；
It is all averaged after the completion of training to the N number of model of gained, obtains final mask；Finally, using F1 score as selecting index probability threshold
Value, below threshold value, it is believed that non-ventricular fibrillation, more than threshold value, it is believed that be ventricular fibrillation；
(10) to any new to be analyzed 10 second electrocardio segment, at the method using step (1) to step (6)
Reason obtains the ventricular fibrillation characteristic parameter combination of the electrocardiogram；Classified using the model that training obtains in step (9), obtains it
The probability of ventricular fibrillation recycles, and the probability threshold value determined in step (9) finally determines whether the electrocardiogram is ventricular fibrillation.
A kind of ventricular fibrillation recognition methods based on machine learning techniques, which is characterized in that the band in the step (2)
The centre frequency of bandpass filter is 14.6Hz, is realized using integral digital filter.
The principle of the present invention is:
1. the present invention is selected from ten various features by systematic test using the training method for staying a verifying first
Two kinds of characteristic parameters are taken, both features are obtained by bandpass filtering analysis and ventricular fibrillation Filtering Analysis respectively, these two types of features
Calculation see step (1) to step (4).
2. present invention uses electrocardiogram average heart rates as the 3rd feature as described in step (5) and step (6), use
Each QRS wave of electrocardiogram is that the average value of the probability of true QRS wave is the 4th feature, thus effectivelying prevent will can by heart rate
Obviously judge that the electrocardiogram of non-ventricular fibrillation is judged as ventricular fibrillation.
The invention has the advantages that
The present invention chooses most effective combination from the various features that the prior art proposes, effectively increases ventricular fibrillation detection and calculates
The performance of method.Meanwhile heart rate feature is applied in ventricular fibrillation detection by the present invention.In order to solve QRS detection difficult when ventricular fibrillation occurs,
The problem of the possible serious inaccuracy of heart rate, the present invention enable model accurate judgement under what circumstances using the quality information of signal
The electrocardiogram that heart rate information should be ignored, which kind of by heart rate can be excluded obvious non-ventricular fibrillation in the case of.
Detailed description of the invention
Fig. 1 is to handle original electro-cardiologic signals, to extract the flow chart of ventricular fibrillation feature.
Fig. 2 be using marked ecg database training for judge each ECG signal whether be ventricular fibrillation logistic regression
The flow chart of model.
Fig. 3 is the flow chart that the probability that each ECG signal is ventricular fibrillation is acquired using trained model.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.
Embodiment.
A kind of ventricular fibrillation recognition methods based on machine learning techniques, which comprises the following steps:
(1) ECG signal is pre-processed, to filter out the noises such as baseline drift, Hz noise, will be believed except the electrocardio after making an uproar
Number it is resampled to a certain fixed sample rate；
(2) to except make an uproar and resampling after ECG signal carry out bandpass filtering analysis, calculate feature 1, bandpass filter
Centre frequency be 14.6Hz, realized using integral digital filter；Method particularly includes: assuming that the output of bandpass filtering is
FS calculated the maximum value and average value of FS absolute value, is denoted as MAX_ABS_FS and MEAN_ABS_FS respectively to each second window；
The 1st feature for detecting ventricular fibrillation is defined as filtering output absolute value in electrocardiogram and is in MEAN_ABS_FS and MAX_
The number of sample between ABS_FS accounts for the ratio of all numbers of samples of electrocardiogram；
(3) to except make an uproar and resampling after ECG signal carry out ventricular fibrillation Filtering Analysis, calculate feature 2, method particularly includes:
The meaning of ventricular fibrillation filtering is explained as follows: since ventricular fibrillation electrocardiogram is similar to sine curve, energy can be distributed mainly on average frequency
Near, thus can be than non-ventricular fibrillation electrocardio for surplus after the bandstop filter of ECG signal average frequency by a centre frequency
Small more of the surplus of figure；For the average frequency of calculating ECG signal, estimate T average period of signal be it is as follows, wherein Vi is
Signal value, m are ecg samples numbers:
It can be calculated by following equation by the surplus of the ECG signal after the bandstop filter, this is as the 2nd feature:
(4) ECG-QRS Wave characteristic point, specific method are extracted using based on the method for wavelet transformation and logistic regression algorithm
Are as follows: comprehensive multiple waveforms feature, using machine learning algorithm, study obtains each feature weight from mass data automatically, differentiates and doubts
It is whether true like QRS wave；Meanwhile to each doubtful QRS wave provide one be true QRS wave probability, when signal quality compared with
When poor, the probability value is generally lower, therefore these values can be used for measuring the quality of electrocardiosignal；
(5) calculating ECG average heart rate, as the 3rd feature；
(6) each QRS wave of calculating ECG is the average value of the probability of true QRS wave, as the 4th feature；
(7) ecg database marked through electrocardio expert is chosen, each electrocardiographic recording is divided into multiple electrocardiogram segments,
Each electrocardiogram segment 10 seconds；
(8) each 10 seconds electrocardiogram segments are handled by step (1)-step (6), obtains the spy of a differentiation ventricular fibrillation
Parameter matrix is levied, the line number of the matrix is equal to the number of all 10 seconds electrocardiogram segments, and columns is equal to 4, and columns is step (1)-
The number of the ventricular fibrillation characteristic parameter obtained in step (6)；To each electrocardiogram segment, it is compared with expert's label, judges that it is
No is ventricular fibrillation；
(9) by step (8) obtain characteristic parameter matrix and whether be ventricular fibrillation one Logic Regression Models of information input
In, using gradient descent method as optimization method, setting learning rate is 0.02, and iteration is until convergence；In training, using staying 1 intersection to test
Training data is randomly divided into K parts by the method for card, take and wherein do training for K-1 parts, and remaining 1 part is verified, so repeatedly n times；
It is all averaged after the completion of training to the N number of model of gained, obtains final mask；Finally, using F1 score as selecting index probability threshold
Value, below threshold value, it is believed that non-ventricular fibrillation, more than threshold value, it is believed that be ventricular fibrillation；
(10) to any new to be analyzed 10 second electrocardio segment, at the method using step (1) to step (6)
Reason obtains the ventricular fibrillation characteristic parameter combination of the electrocardiogram；Classified using the model that training obtains in step (9), obtains it
The probability of ventricular fibrillation recycles, and the probability threshold value determined in step (9) finally determines whether the electrocardiogram is ventricular fibrillation.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
Anyone skilled in the art in the technical scope disclosed by the present invention, according to the technique and scheme of the present invention and its
Inventive concept is subject to equivalent substitution or change, should be covered by the protection scope of the present invention.
Claims (2)
1. a kind of ventricular fibrillation recognition methods based on machine learning techniques, which comprises the following steps:
(1) ECG signal is pre-processed, carries out that a certain fixed sample rate will be resampled to except the electrocardiosignal after making an uproar except making an uproar；
(2) to except make an uproar and resampling after ECG signal carry out bandpass filtering analysis, calculate feature 1, method particularly includes: assuming that
The output of bandpass filtering is FS, to each second window, calculates the maximum value and average value of FS absolute value, is denoted as MAX_ABS_ respectively
FS and MEAN_ABS_FS；The 1st feature for detecting ventricular fibrillation is defined as filtering output absolute value in electrocardiogram and is in
The number of sample between MEAN_ABS_FS and MAX_ABS_FS accounts for the ratio of all numbers of samples of electrocardiogram；
(3) to except make an uproar and resampling after ECG signal carry out ventricular fibrillation Filtering Analysis, calculate feature 2, method particularly includes: ventricular fibrillation
The meaning of filtering is explained as follows: since ventricular fibrillation electrocardiogram is similar to sine curve, energy can be distributed mainly near average frequency,
It thus can be than non-ventricular fibrillation electrocardiogram for surplus after the bandstop filter of ECG signal average frequency by a centre frequency
Small more of surplus；For the average frequency of calculating ECG signal, estimate T average period of signal be it is as follows, wherein Vi is signal
Value, m is ecg samples number:
It can be calculated by following equation by the surplus of the ECG signal after the bandstop filter, this is as the 2nd feature:
(4) ECG-QRS Wave characteristic point is extracted using based on the method for wavelet transformation and logistic regression algorithm, method particularly includes:
Comprehensive multiple waveforms feature, using machine learning algorithm, study obtains each feature weight from mass data automatically, differentiates doubtful
Whether QRS wave is true；Meanwhile to each doubtful QRS wave provide one be true QRS wave probability, when signal quality is poor
When, the probability value is generally lower, therefore these values can be used for measuring the quality of electrocardiosignal；
(5) calculating ECG average heart rate, as the 3rd feature；
(6) each QRS wave of calculating ECG is the average value of the probability of true QRS wave, as the 4th feature；
(7) ecg database marked through electrocardio expert is chosen, each electrocardiographic recording is divided into multiple electrocardiogram segments, each
Electrocardiogram segment 10 seconds；
(8) each 10 seconds electrocardiogram segments are handled by step (1)-step (6), obtains the feature ginseng of a differentiation ventricular fibrillation
Matrix number, the line number of the matrix are equal to the number of all 10 seconds electrocardiogram segments, and columns is equal to 4, and columns is step (1)-step
(6) number of the ventricular fibrillation characteristic parameter obtained in；To each electrocardiogram segment, be compared with expert's label, judge its whether be
Ventricular fibrillation；
(9) by characteristic parameter matrix that step (8) obtain and whether be in one Logic Regression Models of information input of ventricular fibrillation, with
Gradient descent method is optimization method, and setting learning rate is 0.02, and iteration is until convergence；In training, using the side for staying 1 cross validation
Training data is randomly divided into K parts by method, take and wherein do training for K-1 parts, and remaining 1 part is verified, so repeatedly n times；All instructions
It is averaged after the completion of white silk to the N number of model of gained, obtains final mask；Finally, using F1 score as selecting index probability threshold value, in threshold
Value is following, it is believed that non-ventricular fibrillation, more than threshold value, it is believed that be ventricular fibrillation；
(10) it to any new to be analyzed 10 second electrocardio segment, is handled, is obtained using the method for step (1) to step (6)
Obtain the ventricular fibrillation characteristic parameter combination of the electrocardiogram；Classified using the model that training obtains in step (9), obtains its ventricular fibrillation
Probability recycles, and the probability threshold value determined in step (9) finally determines whether the electrocardiogram is ventricular fibrillation.
2. a kind of ventricular fibrillation recognition methods based on machine learning techniques according to claim 1, which is characterized in that the step
Suddenly the centre frequency of the bandpass filter in (2) is 14.6Hz, is realized using integral digital filter.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910208962.9A CN109846476A (en) | 2019-03-19 | 2019-03-19 | A kind of ventricular fibrillation recognition methods based on machine learning techniques |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910208962.9A CN109846476A (en) | 2019-03-19 | 2019-03-19 | A kind of ventricular fibrillation recognition methods based on machine learning techniques |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109846476A true CN109846476A (en) | 2019-06-07 |
Family
ID=66901216
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910208962.9A Pending CN109846476A (en) | 2019-03-19 | 2019-03-19 | A kind of ventricular fibrillation recognition methods based on machine learning techniques |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109846476A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6865329B1 (en) * | 2019-07-29 | 2021-04-28 | 株式会社カルディオインテリジェンス | Display device, display method and program |
| CN113100779A (en) * | 2020-01-10 | 2021-07-13 | 深圳市理邦精密仪器股份有限公司 | Ventricular fibrillation detection method and device and monitoring equipment |
| CN113425272A (en) * | 2021-08-02 | 2021-09-24 | 北京雪扬科技有限公司 | Method for analyzing atrial fibrillation and ventricular fibrillation by collecting data through wearable equipment |
| JP7032747B1 (en) | 2021-03-24 | 2022-03-09 | アステラス製薬株式会社 | ECG analysis support device, program, ECG analysis support method, and ECG analysis support system |
-
2019
- 2019-03-19 CN CN201910208962.9A patent/CN109846476A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6865329B1 (en) * | 2019-07-29 | 2021-04-28 | 株式会社カルディオインテリジェンス | Display device, display method and program |
| CN113100779A (en) * | 2020-01-10 | 2021-07-13 | 深圳市理邦精密仪器股份有限公司 | Ventricular fibrillation detection method and device and monitoring equipment |
| JP7032747B1 (en) | 2021-03-24 | 2022-03-09 | アステラス製薬株式会社 | ECG analysis support device, program, ECG analysis support method, and ECG analysis support system |
| WO2022202942A1 (en) * | 2021-03-24 | 2022-09-29 | アステラス製薬株式会社 | Electrocardiogram analysis assistance device, program, electrocardiogram analysis assistance method, electrocardiogram analysis assistance system, peak estimation model generation method, and segment estimation model generation method |
| JP2022148631A (en) * | 2021-03-24 | 2022-10-06 | アステラス製薬株式会社 | Electrocardiogram analysis support device, program, electrocardiogram analysis support method and electrocardiogram analysis support system |
| CN113425272A (en) * | 2021-08-02 | 2021-09-24 | 北京雪扬科技有限公司 | Method for analyzing atrial fibrillation and ventricular fibrillation by collecting data through wearable equipment |
| CN113425272B (en) * | 2021-08-02 | 2024-02-20 | 北京雪扬科技有限公司 | Method for analyzing atrial fibrillation through data acquired by wearable equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109846476A (en) | A kind of ventricular fibrillation recognition methods based on machine learning techniques | |
| Hwang et al. | Deep ECGNet: An optimal deep learning framework for monitoring mental stress using ultra short-term ECG signals | |
| US20200367800A1 (en) | Method for identifying driving fatigue based on cnn-lstm deep learning model | |
| CN109961017A (en) | A kind of cardiechema signals classification method based on convolution loop neural network | |
| CN107837082A (en) | Electrocardiogram automatic analysis method and device based on artificial intelligence self study | |
| CN111759345B (en) | Heart valve abnormality analysis method, system and device based on convolutional neural network | |
| KR100750662B1 (en) | A biometics system and method using electrocardiogram | |
| CN108186011A (en) | Atrial fibrillation detection method, device and readable storage medium storing program for executing | |
| CN109171708B (en) | Defibrillation heart rhythm recognition device | |
| CN104274191B (en) | A kind of Psychological Evaluation method and system thereof | |
| CN110600053A (en) | Cerebral stroke dysarthria risk prediction method based on ResNet and LSTM network | |
| CN111481192B (en) | Electrocardiosignal R wave detection method based on improved U-Net | |
| CN113128585B (en) | Deep neural network based multi-size convolution kernel method for realizing electrocardiographic abnormality detection and classification | |
| CN110491506A (en) | Auricular fibrillation prediction model and its forecasting system | |
| CN109864739A (en) | A method of identifying second degree A-V block from 10 seconds electrocardiograms | |
| CN113349793A (en) | Arrhythmia classification method based on layered attention model | |
| CN114093501A (en) | Intelligent auxiliary analysis method for children's motor epilepsy based on synchronous video and electroencephalogram | |
| CN116269355A (en) | Safety monitoring system based on figure gesture recognition | |
| CN101897578B (en) | Method for segmenting arterial pressure signal by beats | |
| CN109846473A (en) | A method of detecting 10 seconds electrocardiogram noise jamming degree of single lead | |
| CN115836847A (en) | Blood pressure prediction device and equipment | |
| CN111345815B (en) | Method, device, equipment and storage medium for detecting QRS wave in electrocardiosignal | |
| CN101879058B (en) | Method for segmenting intracranial pressure signal beat by beat | |
| CN113171102A (en) | ECG data classification method based on continuous deep learning | |
| Nikolaev et al. | Structural architectural solutions for an intelligence system of cardiological screening of diabetes patients |
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: 20190607 |
|
| RJ01 | Rejection of invention patent application after publication |