CN106725420A - VPB recognition methods and VPB identifying system - Google Patents
VPB recognition methods and VPB identifying system Download PDFInfo
- Publication number
- CN106725420A CN106725420A CN201510801501.4A CN201510801501A CN106725420A CN 106725420 A CN106725420 A CN 106725420A CN 201510801501 A CN201510801501 A CN 201510801501A CN 106725420 A CN106725420 A CN 106725420A
- Authority
- CN
- China
- Prior art keywords
- ecg signal
- vpb
- pretreated
- qrs
- ripples
- 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
Abstract
The invention provides a kind of VPB recognition methods and VPB identifying system, including：Receive ECG signal；The ECG signal is pre-processed；The pretreated ECG signal is classified using several disaggregated models, obtains several original probability values；Fusion decision-making is carried out to described several original probability values using predetermined fusion decision rule, obtain fusion decision-making after combined chance value, recognized according to combined chance value the pretreated ECG signal be the ECG signal of VPB or it is to be determined be the ECG signal of VPB；If the pretreated ECG signal be it is to be determined be the ECG signal of VPB, extract simultaneously again identified that according to the characteristic parameter of the pretreated ECG signal.VPB recognition methods proposed by the present invention and VPB identifying system, enable to PVC to recognize that sensitivity is more preferable, performance is more excellent, accuracy rate is higher.
Description
Technical field
The present invention relates to medical electronics technical field, specifically, be related to a kind of VPB recognition methods and
VPB identifying system.
Background technology
VPB (Premature Ventricular Contraction, PVC) is anticipatory to originate from the heart
The abnormal heartbeats of room, are also one of clinical common arrhythmia cordis.Therefore, it is possible to correct detection, automatic knowledge
Other PVC has very important clinical meaning.
At present, classify automatically on PVC or the method for PVC classification is a lot, such as SVMs,
The methods such as neutral net, wavelet transformation, template matches.In the studies above method, traditional neural network is calculated
Method operand is big, the training time is more long, it is difficult to realize real-time detection.And for small wave converting method, how
One appropriate wavelet basis of selection is the problem that researcher needs to solve.Additionally, being directed to template matches side
Method, because electrocardiogram has difference between different patients, even for same patient in not its electrocardio in the same time
Figure is also difference, therefore in using template matching method identification PVC, how to find one and compare
Complete ATL is the problem that researcher needs to consider emphatically.
Therefore, prior art has yet to be improved and developed.
The content of the invention
In order to solve above-mentioned problems of the prior art, the present invention proposes a kind of PVC recognition methods, makes
Obtain PVC and recognize that sensitivity is more preferable, performance is more excellent, accuracy rate is higher.
The present invention the proposed technical scheme that solves the above problems is as follows：
It is an object of the present invention to provide a kind of VPB recognition methods, including：
Receive electrocardiogram (Electrocardiogram, ECG) signal；
ECG signal to receiving is pre-processed, to obtain pretreated ECG signal；
The pretreated ECG signal is classified using several disaggregated models, obtains several
Original probability value, wherein, the disaggregated model is lead nerve convolutional network (Lead Convolutional
Neural Network, LCNN) disaggregated model；
It is room property morning to recognize the pretreated ECG signal according to several original probability values for obtaining
The ECG signal fought or it is to be determined be the ECG signal of VPB.
Further, the bandpass filtering pretreatment of 0.5Hz-40Hz is carried out to the ECG signal for receiving, to obtain
Obtain the pretreated ECG signal.
Further, the pretreated electrocardiogram letter is recognized according to several original probability values for obtaining
Number it is the specific method bag of the ECG signal of the ECG signal or VPB to be determined of VPB
Include：
Fusion decision-making is carried out to described several original probability values using predetermined fusion decision rule, to obtain
Combined chance value；
Judge the combined chance value whether more than a predetermined threshold；Wherein, if the combined chance value is more than
The predetermined threshold, then the pretreated ECG signal is the ECG signal of VPB；If institute
State combined chance value and be not more than the predetermined threshold, then the pretreated ECG signal is to be determined
The ECG signal of VPB.
Further, if the pretreated ECG signal is the electrocardiogram letter of VPB to be determined
Number, then the VPB recognition methods also includes：
Extract the characteristic parameter of the pretreated ECG signal；
Recognize whether the pretreated ECG signal is VPB according to the characteristic parameter for extracting
ECG signal.
Further, the characteristic parameter of the extraction includes：Phase, T ripples direction between width, the RR of QRS wave
And the main ripple directions of QRS.
Further, according to extract characteristic parameter come recognize the pretreated ECG signal whether be
The specific method of the ECG signal of VPB includes：
Judge the width of the QRS wave whether not less than 0.1 second；If the width of the QRS wave is less than 0.1
Second, then the pretreated ECG signal is the ECG signal of non-VPB；
If the width of the QRS wave is not less than 0.1 second, whether the phase is not less than normal between judging the RR
Twice during RR；If the phase is described pretreated less than the twice during normal RR between the RR
ECG signal is the ECG signal of non-VPB；
If the phase, not less than the twice during normal RR, judges the pretreated electrocardiogram between the RR
Whether signal there are QRS-T ripples in advance；If the pretreated ECG signal does not occur QRS-T in advance
Ripple, then the pretreated ECG signal is the ECG signal of non-VPB；
If there are QRS-T ripples in advance in the pretreated ECG signal, judge the T ripples direction with
Whether the main ripple directions of QRS are opposite；If the T ripples direction and the main ripple directions of the QRS not conversely,
Then the pretreated ECG signal is the ECG signal of non-VPB；
If the T ripples direction is in opposite direction with the main ripples of the QRS, the pretreated ECG signal
It is the ECG signal of VPB.
Present invention also offers a kind of VPB identifying system, including
Receiver module, is configured to receive ECG signal；
Pretreatment module, is configured to pre-process the ECG signal, pretreated to obtain
ECG signal；
Sort module, is configured to enter the pretreated ECG signal using several disaggregated models
Row classification, to obtain several original probability values, wherein, the disaggregated model is lead nerve convolutional network
Disaggregated model；
Identification module, is configured to described pretreated to recognize according to several original probability values for obtaining
ECG signal be the ECG signal of VPB or it is to be determined be the ECG signal of VPB.
Further, the identification module is further configured to using predetermined fusion decision rule to described
Several original probability values carry out fusion decision-making, to obtain combined chance value, and judge the combined chance value
Whether a predetermined threshold is more than；
Wherein, it is described if the identification module is judged as the combined chance value more than the predetermined threshold
Pretreated ECG signal is the ECG signal of VPB；If the identification module is judged as described
Combined chance value is not more than the predetermined threshold, then the pretreated ECG signal is room to be determined
The ECG signal of property premature beat.
Further, the VPB identifying system also includes：Identification module again, if being configured to described
Identification module recognizes that the pretreated ECG signal is the ECG signal of VPB to be determined,
Then extract the characteristic parameter of the pretreated ECG signal, and recognized according to the characteristic parameter for extracting
The pretreated ECG signal whether be VPB ECG signal；
Wherein, the characteristic parameter of the extraction includes：Phase between width, the RR of QRS wave, T ripples direction and
The main ripple directions of QRS.
Further, whether not the identification module again be further configured to judge the width of the QRS wave
Less than 0.1 second；If the identification module again judged the width of the QRS wave less than 0.1 second, described pre-
ECG signal after treatment is the ECG signal of non-VPB；
If the identification module again judged the width of the QRS wave not less than 0.1 second, described to recognize mould again
Block be further configured to judge the RR between the phase whether not less than the twice during normal RR；If it is described again
Identification module is judged as between the RR phase less than the twice during normal RR, then the pretreated electrocardio
Figure signal is the ECG signal of non-VPB；
If it is described not less than the twice during normal RR that the identification module again is judged as between the RR phase
Identification module is further configured to judge whether the pretreated ECG signal occurs in advance again
QRS-T ripples；If the identification module again is judged as that the pretreated ECG signal does not occur in advance
QRS-T ripples, then the pretreated ECG signal is the ECG signal of non-VPB；
If the identification module again is judged as that QRS-T ripples occurs in advance in the pretreated ECG signal,
Then the identification module again is further configured to judge whether are the T ripples direction and the main ripple directions of the QRS
Conversely；If the identification module again be judged as the T ripples direction and the main ripple directions of the QRS not conversely, if
The pretreated ECG signal is the ECG signal of non-VPB；
If the identification module again is judged as that the T ripples direction is in opposite direction with the main ripples of the QRS, described
Pretreated ECG signal is the ECG signal of VPB.
VPB recognition methods proposed by the present invention and VPB identifying system, using multiple classification mould
Type is classified, and whole knowledge is lifted using the otherness of classification results between each disaggregated model in integrated study
The classification performance of other method；While base disaggregated models of the LCNN as integrated study so that PVC identification spirits
Sensitivity is more preferable, performance is more excellent, accuracy rate is higher.
Brief description of the drawings
By the following description carried out with reference to accompanying drawing, above and other aspect of embodiments of the invention, feature
Be will become clearer with advantage, in accompanying drawing：
Fig. 1 is the flow chart of the VPB recognition methods of first embodiment of the invention；
Fig. 2 is the flow chart of VPB recognition methods according to the second embodiment of the present invention；
Fig. 3 is the module map of VPB identifying system according to the third embodiment of the invention.
Specific embodiment
Hereinafter, with reference to the accompanying drawings to describing embodiments of the invention in detail.However, it is possible to many different
Form implements the present invention, and the present invention should not be construed as limited to the specific embodiment that illustrates here.
Conversely, there is provided these embodiments are in order to explain principle of the invention and its practical application, so that this area
Others skilled in the art it will be appreciated that various embodiments of the present invention and being suitable for the various of specific intended application and repairing
Change.
<First embodiment>
Fig. 1 is the flow chart of the VPB recognition methods of first embodiment of the invention.
Reference picture 1, in step sl, receives ECG signal.
In the present embodiment, the ECG signal can be gathered from Chinese angiocardiopathy database (Chinese
Cardiovascular Disease Database, abbreviation CCDD), but the present invention is not restricted to this.
In step s 2, the ECG signal is pre-processed, obtains pretreated ECG signal.
Typically, can be comprising noises such as power frequency, myoelectricity, baseline drifts in the ECG signal of reception.Power frequency is made an uproar
Sound can produce influence to tiny turnover in ECG signal, so that the feature of the ECG signal occurs
Change and influence the ECG signal for the diagnosis of the state of an illness, its frequency is fixed as 50Hz.Baseline drift one
As it is caused by human body respiration and electrode movement, the datum line of ECG signal can be caused to be presented what is drifted about up and down
Situation, its frequency is less than 10Hz.Myoelectricity interference is mainly caused by human muscle trembles, and its frequency typically exists
Between 5Hz~2kHz.Wherein, influence of the baseline drift to ECG signal is maximum.The present invention makes an uproar to reduce
Acoustic jamming, the ECG signal to receiving carries out advance filtering process, for example, to the electrocardiogram letter for receiving
Number carry out the bandpass filtering pretreatment of 0.5Hz-40Hz.
In step s3, the pretreated ECG signal is classified using several disaggregated models,
Several original probability values are obtained, wherein, in view of LCNN training parameters are few, self adaptation is strong, simple structure
Deng good characteristic, several disaggregated models are using LCNN as base disaggregated model.
In step s 4, the pretreated electrocardiogram is recognized according to several original probability values for obtaining
Signal be the ECG signal of VPB or it is to be determined be the ECG signal of VPB.
Specifically, fusion decision-making is carried out to described several original probability values using predetermined fusion decision rule,
To obtain combined chance value；Judge the combined chance value whether more than a predetermined threshold；Wherein, if described
Combined chance value is more than the predetermined threshold, then the pretreated ECG signal is the heart of VPB
Electrical picture signal；If the combined chance value is not more than the predetermined threshold, the pretreated electrocardiogram
Signal is the ECG signal of VPB to be determined.
In the present embodiment, step S3, S4 uses the method based on integrated study to carry out PVC knowledges
Not, wherein, integrated learning approach validity is determined from suitable disaggregated model, typically in order to lift collection
Into effect, each disaggregated model in some disaggregated models of selection must be fulfilled for it is following some：1st, each point
The error probability of class model is not above 0.5；2nd, the performance between each disaggregated model will have larger difference.
The technical scheme of the present embodiment is described below by a specific example.
First, the ECG signal is obtained from the record of CCDD, takes from some electrocardios of the database
Figure signal carries out denoising by the bandpass filtering of 0.5~40Hz.
Using pretreated some ECG signals as training sample, specifically, as an example, choosing
35840 pretreated ECG signals are taken as training sample (including 3112 PVC)；
141046 ECG signals are chosen in 35840 pretreated ECG signals (including 2148
PVC) as test sample.In view of the excellent spy such as few, strong, simple structure of self adaptation of LCNN training parameters
Property, using LCNN as the base disaggregated model of integrated study, 6 classification with LCNN as base are chosen here
Model is classified, wherein, this 6 LCNN models are chosen from some LCNN models for having trained
The error probability elected 6 models more larger than otherness between relatively low and each model.This 6 LCNN
3 layers of sampling core of model are 1*3, and 3 convolution kernels are respectively 1*18,1*10,1*5.
In order to by 6 LCNN Construction of A Model into different disaggregated models, in 6 LCNN models
There are 3 structures of LCNN models for [15,20,15,50,1], wherein, 15,20,15 represent respectively
3 characteristic face numbers of convolution unit, 50 and 1 represents two neuron numbers of full articulamentum respectively.By
PVC is differed with non-VPB (NPVC) sampling bar number in training sample, in order to solve sampling bar
The disequilibrium that number brings, sets different in this 3 loss functions of LCNN models to each classification
Mistake point cost, wherein, the mistake point cost ratio for having 2 is 3:1, i.e. NPVC divide cost with the mistake of PVC
The ratio between be 3:1；Although this 2 LCNN model structures and wrong point of cost are than identical, its training initial value
It is different, so this 2 LCNN models are also different；And the mistake of another LCNN model point cost ratio
It is 1:4.
In addition, in other 3 LCNN models, wherein, 1 LCNN model structure for [8,
14,20,50,2], its mistake point cost ratio is 4:1, the structure of another 2 LCNN models for [10,15,20,
50,2], their mistake point cost ratio respectively 4:1 and 3:1.
After having chosen disaggregated model, training sample is input in 6 LCNN models carries out independent parallel
Training, by after study, then carries out test sample independent test and obtains by this 6 LCNN models
6 original probability values.
In the present embodiment, the number of the disaggregated model of selection and its building method are merely possible to example and show,
Without limiting the invention, in other embodiments, it is also possible to select different disaggregated model numbers with
And the building method of different disaggregated models.
Specifically, as an example, using tmjRepresent that m-th disaggregated model obtains belonging to the original of jth class
Probable value (j=0,1, wherein 0 represents NPVC classes, 1 represents PVC classes), is so classified by each
Model can obtain an original probability value, if i.e. tm1More than 0.5, then m-th disaggregated model is by institute
State test sample and be judged to PVC classes.But, in order to obtain a more reliable recognition result, here by
6 original probability values that 6 LCNN models are exported are carried out with fusion decision-making using certain fusion rule to obtain
One more reliable combined chance value.
Specifically, as an example, using average fusion rule, also can be using other fusion rules such as
Multiplication rule, weighted average, maximum etc. is taken, fusion rule technology known to those skilled in the art,
Here repeat no more, the original probability value of 6 LCNN models merge obtain combined chance value Pj,
Its formula is as follows：
Pj=(1/6) * ∑s6 M=1tmj, j=0,1
PjRepresent that 6 LCNN models finally merge the probability that probability, i.e. sample belong to jth class.If PjGreatly
In 0.5, then sample belongs to 1 class i.e. PVC classes, and otherwise sample belongs to 0 class i.e. NPVC classes.To test specimens
Originally the result after being judged is as shown in the table：
Table 1
In the present embodiment, it is that sensitivity, Acc are that accuracy rate, γ are comprehensive for specificity, Se to use Sp
Index youden index, Gm measure the quality of classifying quality for these parameters of geometric average, their own
Shown in being defined as follows：
Acc=(TP+TN)/(TP+TN+FP+FN)
Se=TP/ (TP+FN)
Sp=TN/ (TN+FP)
γ=Se+Sp-1
Wherein, TP is positive sample by the quantity of algorithm test positive；TN is examined for negative sample by algorithm
It is negative quantity to survey；FN is mistakenly detected as the quantity of feminine gender for positive sample；FP is tested for negative sample
It is positive quantity to survey.As can be seen from the table, wherein, be correctly identified as PVC in 2148 PVC
Be 1380, the recognition methods provided using the present embodiment, its recognition accuracy can reach 98.47%.
In sum, first embodiment of the invention is carried out using the method based on integrated study to ECG signal
PVC identifications, cause that PVC recognition accuracies are higher using LCNN in integrated study as base disaggregated model.
<Second embodiment>
Fig. 2 is the flow chart of the VPB recognition methods of first embodiment of the invention.
Reference picture 2, in step sl, receives ECG signal.
In the present embodiment, the ECG signal can be gathered from Chinese angiocardiopathy database (Chinese
Cardiovascular Disease Database, abbreviation CCDD), but the present invention is not restricted to this.
In step s 2, the ECG signal is pre-processed, obtains pretreated ECG signal.
Typically, can be comprising noises such as power frequency, base electricity, baseline drifts in the ECG signal of reception.Power frequency is made an uproar
Sound can produce influence to tiny turnover in ECG signal, so that the feature of the ECG signal occurs
Change and influence the ECG signal for the diagnosis of the state of an illness, its frequency is fixed as 50Hz.Baseline drift one
As it is caused by human body respiration and electrode movement, the datum line of ECG signal can be caused to be presented what is drifted about up and down
Situation, its frequency is less than 10Hz.Myoelectricity interference is mainly caused by human muscle trembles, and its frequency typically exists
Between 5Hz~2kHz.Wherein, influence of the baseline drift to ECG signal is maximum.The present invention makes an uproar to reduce
Acoustic jamming, the ECG signal to receiving carries out advance filtering process, for example, to the electrocardiogram letter for receiving
Number carry out the bandpass filtering pretreatment of 0.5Hz-40Hz.
In step s3, the pretreated ECG signal is classified using several disaggregated models,
Obtain several original probability values, wherein several disaggregated models are using LCNN as base disaggregated model.
In step s 4, the pretreated electrocardiogram is recognized according to several original probability values for obtaining
Signal be the ECG signal of VPB or it is to be determined be the ECG signal of VPB.
Specifically, fusion decision-making is carried out to described several original probability values using predetermined fusion decision rule,
To obtain combined chance value；Judge the combined chance value whether more than a predetermined threshold；Wherein, if described
Combined chance value is more than the predetermined threshold, then the pretreated ECG signal is the heart of VPB
Electrical picture signal；If the combined chance value is not more than the predetermined threshold, the pretreated electrocardiogram
Signal is the ECG signal of VPB to be determined.
In the present embodiment, step S3, S4 uses the method based on integrated study to carry out PVC knowledges
Not, wherein, integrated learning approach validity is determined from suitable disaggregated model, typically in order to lift collection
Into effect, each disaggregated model in some disaggregated models of selection must be fulfilled for it is following some：1st, each point
The error probability of class model is not above 0.5；2nd, the performance between each disaggregated model will have larger difference.
The technical scheme of the present embodiment is described below by a specific example.
First, the ECG signal is obtained from the record of CCDD, takes from some electrocardios of the database
Figure signal carries out denoising by the bandpass filtering of 0.5~40Hz.
Using pretreated some ECG signals as training sample, specifically, as an example, choosing
35840 pretreated ECG signals are taken as training sample (including 3112 PVC)；
141046 ECG signals are chosen in 35840 pretreated ECG signals (including 2148
PVC) as test sample.In view of the excellent spy such as few, strong, simple structure of self adaptation of LCNN training parameters
Property, using LCNN as the base disaggregated model of integrated study, 6 classification with LCNN as base are chosen here
Model is classified, wherein, this 6 LCNN models are chosen from some LCNN models for having trained
The error probability elected 6 models more larger than otherness between relatively low and each model.This 6 LCNN
3 layers of sampling core of model are 1*3, and 3 convolution kernels are respectively 1*18,1*10,1*5.
In order to by 6 LCNN Construction of A Model into different disaggregated models, in 6 LCNN models
There are 3 structures of LCNN models for [15,20,15,50,1], wherein, 15,20,15 represent respectively
3 characteristic face numbers of convolution unit, 50 and 1 represents two neuron numbers of full articulamentum respectively.By
PVC is differed with non-VPB (NPVC) sampling bar number in training sample, in order to solve sampling bar
The disequilibrium that number brings, sets different in this 3 loss functions of LCNN models to each classification
Mistake point cost, wherein, the mistake point cost ratio for having 2 is 3:1, i.e. NPVC divide cost with the mistake of PVC
The ratio between be 3:1；Although this 2 LCNN model structures and wrong point of cost are than identical, its training initial value
It is different, so this 2 LCNN models are also different；And the mistake of another LCNN model point cost ratio
It is 1:4.
In addition, in other 3 LCNN models, wherein, 1 LCNN model structure for [8,
14,20,50,2], its mistake point cost ratio is 4:1, the structure of another 2 LCNN models for [10,15,20,
50,2], their mistake point cost ratio respectively 4:1 and 3:1.
After having chosen disaggregated model, training sample is input in 6 LCNN models carries out independent parallel
Training, by after study, then carries out test sample independent test and obtains by this 6 LCNN models
6 original probability values.
In the present embodiment, the number of the disaggregated model of selection and its building method are merely possible to example and show,
Without limiting the invention, in other embodiments, it is also possible to select different disaggregated model numbers with
And the building method of different disaggregated models.
Specifically, as an example, using tmjRepresent that m-th disaggregated model obtains belonging to the original of jth class
Probable value (j=0,1, wherein 0 represents NPVC classes, 1 represents PVC classes), is so classified by each
Model can obtain an original probability value, if i.e. tm1More than 0.5, then m-th disaggregated model is by institute
State test sample and be judged to PVC classes.But, in order to obtain a more reliable recognition result, here by
6 original probability values that 6 LCNN models are exported are carried out with fusion decision-making using certain fusion rule to obtain
One more reliable combined chance value.
Specifically, using average fusion rule, also can be using other fusion rules such as multiplication rule, weighting
Averagely, maximum etc. is taken, fusion rule technology known to those skilled in the art is repeated no more here,
The original probability value of 6 LCNN models merge and obtains combined chance value Pj, its formula is as follows：
Pj=(1/6) * ∑s6 M=1tmj, j=0,1
PjRepresent that 6 LCNN models finally merge the probability that probability, i.e. sample belong to jth class.If PjGreatly
In 0.5, then sample belongs to 1 class i.e. PVC classes, and otherwise sample belongs to 0 class i.e. NPVC classes.To test specimens
Originally the result after being judged is as shown in the table：
Table 1
In the present embodiment, it is that sensitivity, Acc are that accuracy rate, γ are comprehensive for specificity, Se to use Sp
Index youden index, Gm measure the quality of classifying quality for these parameters of geometric average, their own
Shown in being defined as follows：
Acc=(TP+TN)/(TP+TN+FP+FN)
Se=TP/ (TP+FN)
Sp=TN/ (TN+FP)
γ=Se+Sp-1
Wherein, TP is positive sample by the quantity of algorithm test positive；TN is examined for negative sample by algorithm
It is negative quantity to survey；FN is mistakenly detected as the quantity of feminine gender for positive sample；FP is tested for negative sample
It is positive quantity to survey.As can be seen from the table, the recognition methods for being provided using the present embodiment, its identification is accurate
True rate can reach 98.47%.Generally, youden index γ and geometric average Gm are bigger, classification
Can be better.
As it can be seen from table 1 after carrying out parallel sorting and then carry out fusion decision-making through 6 LCNN models,
The sensitivity for obtaining is also than relatively low, wherein the actual test sample bar number that NPVC is judged to for PVC is 768
Bar, that is, have more PVC samples to be judged into non-PVC.Therefore its diagnosis is considered as in the present embodiment
Rule carries out identification again to NPVC test samples, to reach the purpose of lifting sensitivity.
With continued reference to Fig. 2, in step s 4, if the pretreated ECG signal is room to be determined
The ECG signal of property premature beat, then into step S5.
In step s 5, the characteristic parameter of the pretreated ECG signal is extracted, in the present embodiment,
The characteristic parameter of the extraction for QRS wave width, RR between phase, T ripples direction and the main ripple directions of QRS.
Wherein, what the width of QRS wave was defined as that the phase between the width between Q ripples and S ripples, RR represents is to work as
Phase between preceding R ripples and previous R ripples.
In step s 6, whether the pretreated ECG signal is recognized according to the characteristic parameter for extracting
It is the ECG signal of VPB.Wherein, in step s 6, according to the characteristic parameter for extracting to recognize
State whether pretreated ECG signal is that the ECG signal specific method of VPB includes：
Step S61：Judge the width of the QRS wave whether not less than 0.1 second.
If the width of the QRS wave is less than 0.1 second, the ECG signal is NPVC.If the QRS
The width of ripple is not less than 0.1 second, then into step S62.
Step S62：Judge that the ECG signal whether there is the complete compensatory phase；Wherein, compensatory completely
What the phase represented is two sinus property heartbeats are separated by before and after premature beat time for 2 times of normal cardiac cycle.
If the ECG signal does not exist the complete compensatory phase, the ECG signal is NPVC.If institute
State ECG signal and there is the complete compensatory phase, then into step S63.
Step S63：Judge whether the ECG signal QRS-T ripples occurs in advance.
If the ECG signal does not occur QRS-T ripples in advance, the ECG signal is NPVC.If
There are QRS-T ripples in advance in the ECG signal, then into step S64.
Step S64：Judge whether the T ripples are opposite with the main ripple directions of the QRS.
If the T ripples are identical with the main ripple directions of the QRS, the ECG signal is NPVC.If institute
State T ripples in opposite direction with the main ripples of the QRS, then the ECG signal is PVC ECG signals.
Specifically, as an example, using ECG signal characteristic parameter described above to the quilt in table 1
Be judged to 2157 test samples of NPVC carries out PVC identifications again.By the collection based on LCNN models
It is as shown in the table with the result that the method that characteristic parameter is combined identification PVC is obtained into learning method
Table 2
As can be seen from the table, wherein, the bar number of NPVC from original 768 is judged in 2148 PVC
Bar is reduced to 335.Therefore, the ECG signal is carried out according to characteristic parameter again identifying that so that knowing
Other sensitivity is higher, classification performance is more preferable.
In sum, method and spy of the second embodiment of the invention using the integrated study based on LCNN models
Levying parameter and combining carries out PVC identifications to ECG signal so that PVC identification sensitivity is higher, overall classification
Performance is more preferable.
<3rd embodiment>
Fig. 3 is the module map of VPB identifying system according to the third embodiment of the invention.
Reference picture 3, VPB identifying system according to the third embodiment of the invention include receiver module 10,
Pretreatment module 11, sort module 12, identification module 13 and identification module 14 again.
Receiver module 10, is configured to receive ECG signal.
Pretreatment module 11, is configured to pre-process the ECG signal, after being pre-processed
ECG signal.
Sort module 12, is configured to using several disaggregated models to the pretreated ECG signal
Classified, to obtain several original probability values, wherein, the disaggregated model is lead nerve convolution net
Network disaggregated model.
Identification module 13, is configured to according to several original probability values for obtaining come after recognizing the pretreatment
ECG signal be the ECG signal of VPB or it is to be determined be the ECG signal of VPB.
In the present embodiment, the identification module 13 is further configured to be recognized using above-mentioned steps S4 described pre-
ECG signal after treatment is that the ECG signal or the electrocardiogram for VPB to be determined of VPB are believed
Number.
The VPB identifying system also includes identification module 14 again, if be configured to the identification module knowing
Not described pretreated ECG signal is the ECG signal of VPB to be determined, then extract described
The characteristic parameter of pretreated ECG signal, and the pretreatment is recognized according to the characteristic parameter for extracting
ECG signal afterwards whether be VPB ECG signal；Wherein, the characteristic parameter tool of the extraction
The species that body includes refer to the description in above-mentioned steps S5.
Specifically, the identification module again is further configured to the method using above-mentioned steps S6 to recognize
State pretreated ECG signal whether be VPB ECG signal.
Although the present invention has shown and described with reference to specific embodiment, those skilled in the art will
Understand：In the case where the spirit and scope of the present invention limited by claim and its equivalent are not departed from,
The various change in form and details can herein be carried out.
Claims (10)
1. a kind of VPB recognition methods, it is characterised in that including：
Receive ECG signal；
ECG signal to receiving is pre-processed, to obtain pretreated ECG signal；
The pretreated ECG signal is classified using several disaggregated models, obtains several
Original probability value, wherein, the disaggregated model is lead nerve convolutional network disaggregated model；
It is room property morning to recognize the pretreated ECG signal according to several original probability values for obtaining
The ECG signal fought or it is to be determined be the ECG signal of VPB.
2. VPB recognition methods according to claim 1, it is characterised in that to the electrocardio for receiving
Figure signal carries out the bandpass filtering pretreatment of 0.5Hz-40Hz, to obtain the pretreated ECG signal.
3. VPB recognition methods according to claim 1, it is characterised in that if according to acquisition
Dry original probability value come recognize the pretreated ECG signal be VPB ECG signal or
The specific method of the ECG signal of VPB to be determined includes：
Fusion decision-making is carried out to described several original probability values using predetermined fusion decision rule, to obtain
Combined chance value；
Judge the combined chance value whether more than a predetermined threshold；Wherein, if the combined chance value is more than
The predetermined threshold, then the pretreated ECG signal is the ECG signal of VPB；If institute
State combined chance value and be not more than the predetermined threshold, then the pretreated ECG signal is to be determined
The ECG signal of VPB.
4. the VPB recognition methods according to any one of claims 1 to 3, it is characterised in that if
The pretreated ECG signal is the ECG signal of VPB to be determined, then the room property morning
Recognition methods of fighting also includes：
Extract the characteristic parameter of the pretreated ECG signal；
Recognize whether the pretreated ECG signal is VPB according to the characteristic parameter for extracting
ECG signal.
5. VPB recognition methods according to claim 4, it is characterised in that the spy of the extraction
Levying parameter includes：Phase, T ripples direction and the main ripple directions of QRS between width, the RR of QRS wave.
6. VPB recognition methods according to claim 5, it is characterised in that according to the spy for extracting
Levy parameter recognize the pretreated ECG signal whether be VPB ECG signal it is specific
Method includes：
Judge the width of the QRS wave whether not less than 0.1 second；If the width of the QRS wave is less than 0.1
Second, then the pretreated ECG signal is the ECG signal of non-VPB；
If the width of the QRS wave is not less than 0.1 second, whether the phase is not less than normal between judging the RR
Twice during RR；If the phase is described pretreated less than the twice during normal RR between the RR
ECG signal is the ECG signal of non-VPB；
If the phase, not less than the twice during normal RR, judges the pretreated electrocardiogram between the RR
Whether signal there are QRS-T ripples in advance；If the pretreated ECG signal does not occur QRS-T in advance
Ripple, then the pretreated ECG signal is the ECG signal of non-VPB；
If there are QRS-T ripples in advance in the pretreated ECG signal, judge the T ripples direction with
Whether the main ripple directions of QRS are opposite；If the T ripples direction and the main ripple directions of the QRS not conversely,
Then the pretreated ECG signal is the ECG signal of non-VPB；
If the T ripples direction is in opposite direction with the main ripples of the QRS, the pretreated ECG signal
It is the ECG signal of VPB.
7. a kind of VPB identifying system, it is characterised in that including：
Receiver module, is configured to receive ECG signal；
Pretreatment module, is configured to pre-process the ECG signal, pretreated to obtain
ECG signal；
Sort module, is configured to enter the pretreated ECG signal using several disaggregated models
Row classification, to obtain several original probability values, wherein, the disaggregated model is lead nerve convolutional network
Disaggregated model；
Identification module, is configured to described pretreated to recognize according to several original probability values for obtaining
ECG signal be the ECG signal of VPB or it is to be determined be the ECG signal of VPB.
8. VPB identifying system according to claim 7, it is characterised in that the identification module
It is further configured to merge certainly described several original probability values using predetermined fusion decision rule
Whether plan, to obtain combined chance value, and judge the combined chance value more than a predetermined threshold；
Wherein, it is described if the identification module is judged as the combined chance value more than the predetermined threshold
Pretreated ECG signal is the ECG signal of VPB；If the identification module is judged as described
Combined chance value is not more than the predetermined threshold, then the pretreated ECG signal is room to be determined
The ECG signal of property premature beat.
9. VPB identifying system according to claim 8, it is characterised in that the VPB
Identifying system also includes：Identification module again, if it is described pretreated to be configured to the identification module identification
ECG signal is the ECG signal of VPB to be determined, then extract the pretreated electrocardiogram
The characteristic parameter of signal, and be recognizing the pretreated ECG signal according to the characteristic parameter for extracting
No is the ECG signal of VPB；
Wherein, the characteristic parameter of the extraction includes：Phase between width, the RR of QRS wave, T ripples direction and
The main ripple directions of QRS.
10. VPB identifying system according to claim 9, it is characterised in that described to recognize again
Whether module is further configured to judge the width of the QRS wave not less than 0.1 second；If described recognize again
Module judged the width of the QRS wave less than 0.1 second, then the pretreated ECG signal is non-room
The ECG signal of property premature beat；
If the identification module again judged the width of the QRS wave not less than 0.1 second, described to recognize mould again
Block be further configured to judge the RR between the phase whether not less than the twice during normal RR；If it is described again
Identification module is judged as between the RR phase less than the twice during normal RR, then the pretreated electrocardio
Figure signal is the ECG signal of non-VPB；
If it is described not less than the twice during normal RR that the identification module again is judged as between the RR phase
Identification module is further configured to judge whether the pretreated ECG signal occurs in advance again
QRS-T ripples；If the identification module again is judged as that the pretreated ECG signal does not occur in advance
QRS-T ripples, then the pretreated ECG signal is the ECG signal of non-VPB；
If the identification module again is judged as that QRS-T ripples occurs in advance in the pretreated ECG signal,
Then the identification module again is further configured to judge whether are the T ripples direction and the main ripple directions of the QRS
Conversely；If the identification module again be judged as the T ripples direction and the main ripple directions of the QRS not conversely, if
The pretreated ECG signal is the ECG signal of non-VPB；
If the identification module again is judged as that the T ripples direction is in opposite direction with the main ripples of the QRS, described
Pretreated ECG signal is the ECG signal of VPB.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510801501.4A CN106725420A (en) | 2015-11-18 | 2015-11-18 | VPB recognition methods and VPB identifying system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510801501.4A CN106725420A (en) | 2015-11-18 | 2015-11-18 | VPB recognition methods and VPB identifying system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106725420A true CN106725420A (en) | 2017-05-31 |
Family
ID=58916262
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510801501.4A Pending CN106725420A (en) | 2015-11-18 | 2015-11-18 | VPB recognition methods and VPB identifying system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106725420A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108403106A (en) * | 2018-01-04 | 2018-08-17 | 东南大学 | A kind of automatic identifying method of room property pararrhythmia |
| CN108714026A (en) * | 2018-03-27 | 2018-10-30 | 杭州电子科技大学 | The fine granularity electrocardiosignal sorting technique merged based on depth convolutional neural networks and on-line decision |
| CN109044347A (en) * | 2018-07-11 | 2018-12-21 | 上海夏先机电科技发展有限公司 | Identify electrocardio wave image junctional escape beat method, apparatus, system and electronic equipment |
| CN109117729A (en) * | 2018-07-11 | 2019-01-01 | 上海夏先机电科技发展有限公司 | Electrocardiogram room escape real-time judge method, apparatus, system and storage medium |
| CN109394206A (en) * | 2018-11-14 | 2019-03-01 | 东南大学 | Method of real-time and its device based on premature beat signal in wearable ECG signal |
| WO2019100560A1 (en) * | 2017-11-27 | 2019-05-31 | 乐普（北京）医疗器械股份有限公司 | Artificial intelligence self-learning-based automatic electrocardiography analysis method and apparatus |
| WO2019100562A1 (en) * | 2017-11-27 | 2019-05-31 | 乐普（北京）医疗器械股份有限公司 | Automatic recognition and classification method for electrocardiogram heartbeat based on artificial intelligence |
| CN109907753A (en) * | 2019-04-23 | 2019-06-21 | 杭州电子科技大学 | A kind of various dimensions ECG signal intelligent diagnosis system |
| CN109948396A (en) * | 2017-12-20 | 2019-06-28 | 深圳市理邦精密仪器股份有限公司 | A kind of beat classification method, beat classification device and electronic equipment |
| CN111297348A (en) * | 2018-12-11 | 2020-06-19 | 深圳市理邦精密仪器股份有限公司 | Method for generating electrocardiogram analysis result, electrocardiograph and storage medium |
| CN114668401A (en) * | 2022-03-11 | 2022-06-28 | 肇庆星网医疗科技有限公司 | AI electrocardiogram training data labeling method, device, electronic equipment and medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3759248A (en) * | 1971-02-08 | 1973-09-18 | Spacelabs Inc | Cardiac arrythmia detector |
| CN102379692A (en) * | 2011-09-02 | 2012-03-21 | 东莞中山大学研究院 | System and method for remote electrocardiographic monitoring for area-end networks |
| CN104523266A (en) * | 2015-01-07 | 2015-04-22 | 河北大学 | Automatic classification method for electrocardiogram signals |
-
2015
- 2015-11-18 CN CN201510801501.4A patent/CN106725420A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US3759248A (en) * | 1971-02-08 | 1973-09-18 | Spacelabs Inc | Cardiac arrythmia detector |
| CN102379692A (en) * | 2011-09-02 | 2012-03-21 | 东莞中山大学研究院 | System and method for remote electrocardiographic monitoring for area-end networks |
| CN104523266A (en) * | 2015-01-07 | 2015-04-22 | 河北大学 | Automatic classification method for electrocardiogram signals |
Non-Patent Citations (2)
| Title |
|---|
| 周波: "基于集成学习的不平衡数据分类的研究及应用", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 金林鹏等: "面向临床心电图分析的深层学习算法", 《中国科学：信息科学》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019100562A1 (en) * | 2017-11-27 | 2019-05-31 | 乐普（北京）医疗器械股份有限公司 | Automatic recognition and classification method for electrocardiogram heartbeat based on artificial intelligence |
| US11564612B2 (en) | 2017-11-27 | 2023-01-31 | Shanghai Lepu CloudMed Co., LTD | Automatic recognition and classification method for electrocardiogram heartbeat based on artificial intelligence |
| US11529103B2 (en) | 2017-11-27 | 2022-12-20 | Shanghai Lepu CloudMed Co., LTD | Artificial intelligence self-learning-based automatic electrocardiography analysis method and apparatus |
| JP7018133B2 (en) | 2017-11-27 | 2022-02-09 | 上海▲楽▼普云智科技股▲分▼有限公司 | ECG heartbeat automatic identification classification method based on artificial intelligence |
| JP2020535882A (en) * | 2017-11-27 | 2020-12-10 | 上海▲優▼加利健康管理有限公司Ｓｈａｎｇｈａｉ Ｙｏｃａｌｙ Ｈｅａｌｔｈ Ｍａｎａｇｅｍｅｎｔ Ｃｏ．， Ｌｔｄ． | ECG heartbeat automatic identification classification method based on artificial intelligence |
| WO2019100560A1 (en) * | 2017-11-27 | 2019-05-31 | 乐普（北京）医疗器械股份有限公司 | Artificial intelligence self-learning-based automatic electrocardiography analysis method and apparatus |
| CN109948396B (en) * | 2017-12-20 | 2021-07-23 | 深圳市理邦精密仪器股份有限公司 | Heart beat classification method, heart beat classification device and electronic equipment |
| CN109948396A (en) * | 2017-12-20 | 2019-06-28 | 深圳市理邦精密仪器股份有限公司 | A kind of beat classification method, beat classification device and electronic equipment |
| CN108403106B (en) * | 2018-01-04 | 2021-03-16 | 东南大学 | Automatic identification method of ventricular parallel cardiac rhythm |
| CN108403106A (en) * | 2018-01-04 | 2018-08-17 | 东南大学 | A kind of automatic identifying method of room property pararrhythmia |
| CN108714026A (en) * | 2018-03-27 | 2018-10-30 | 杭州电子科技大学 | The fine granularity electrocardiosignal sorting technique merged based on depth convolutional neural networks and on-line decision |
| CN109117729A (en) * | 2018-07-11 | 2019-01-01 | 上海夏先机电科技发展有限公司 | Electrocardiogram room escape real-time judge method, apparatus, system and storage medium |
| CN109044347A (en) * | 2018-07-11 | 2018-12-21 | 上海夏先机电科技发展有限公司 | Identify electrocardio wave image junctional escape beat method, apparatus, system and electronic equipment |
| CN109394206A (en) * | 2018-11-14 | 2019-03-01 | 东南大学 | Method of real-time and its device based on premature beat signal in wearable ECG signal |
| CN111297348A (en) * | 2018-12-11 | 2020-06-19 | 深圳市理邦精密仪器股份有限公司 | Method for generating electrocardiogram analysis result, electrocardiograph and storage medium |
| CN109907753A (en) * | 2019-04-23 | 2019-06-21 | 杭州电子科技大学 | A kind of various dimensions ECG signal intelligent diagnosis system |
| CN109907753B (en) * | 2019-04-23 | 2022-07-26 | 杭州电子科技大学 | Multi-dimensional ECG signal intelligent diagnosis system |
| CN114668401A (en) * | 2022-03-11 | 2022-06-28 | 肇庆星网医疗科技有限公司 | AI electrocardiogram training data labeling method, device, electronic equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106725420A (en) | VPB recognition methods and VPB identifying system | |
| CN108714026B (en) | Fine-grained electrocardiosignal classification method based on deep convolutional neural network and online decision fusion | |
| CN104523266B (en) | A kind of electrocardiosignal automatic classification method | |
| CN108511055B (en) | Ventricular premature beat recognition system and method based on classifier fusion and diagnosis rules | |
| Meyer et al. | Combining algorithms in automatic detection of QRS complexes in ECG signals | |
| CN102835954B (en) | Method for generating heart beating waveform template and treating ventricular premature beat detection and device | |
| CN107837082A (en) | Electrocardiogram automatic analysis method and device based on artificial intelligence self study | |
| CN106108889A (en) | Electrocardiogram classification method based on degree of depth learning algorithm | |
| Wang et al. | Integrating analytic and appearance attributes for human identification from ECG signals | |
| CN107822622A (en) | Electrocardiographic diagnosis method and system based on depth convolutional neural networks | |
| CN109171712A (en) | Auricular fibrillation recognition methods, device, equipment and computer readable storage medium | |
| CN106214123A (en) | A kind of electrocardiogram compressive classification method based on degree of depth learning algorithm | |
| CN105997055A (en) | Automatic classification method, system and device of electrocardiosignal ST band | |
| CN109276242A (en) | The method and apparatus of electrocardiosignal type identification | |
| CN109758145B (en) | Automatic sleep staging method based on electroencephalogram causal relationship | |
| CN109117730A (en) | Electrocardiogram auricular fibrillation real-time judge method, apparatus, system and storage medium | |
| WO2019161611A1 (en) | Ecg information processing method and ecg workstation | |
| CN110786849B (en) | Electrocardiosignal identity recognition method and system based on multi-view discriminant analysis | |
| CN111481192B (en) | Electrocardiosignal R wave detection method based on improved U-Net | |
| CN113057648A (en) | ECG signal classification method based on composite LSTM structure | |
| De Chazel et al. | A comparison of the ECG classification performance of different feature sets | |
| CN112826513B (en) | Fetal heart rate detection system based on deep learning and specificity correction on FECG | |
| Rohmantri et al. | Arrhythmia classification using 2D convolutional neural network | |
| CN112022141B (en) | Electrocardiosignal class detection method, electrocardiosignal class detection device and storage medium | |
| Jenny et al. | Automated classification of normal and premature ventricular contractions in electrocardiogram signals |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170531 |