Summary of the invention
The embodiment of the invention provides a kind of electrocardiogram signal processing devices, and the identification effect of right bundle branch block waveform can be improvedRate and accuracy rate.
In a first aspect, the embodiment of the invention provides a kind of electrocardiogram signal processing devices characterized by comprising signalAcquisition module, data preprocessing module and data processing module, wherein the input terminal of the signal acquisition module is used for and electrocardioModule is connected, and the signal acquisition module is used to export the output end of electrocardiosignal and the input terminal of the data preprocessing moduleIt is connected, the data preprocessing module is used to export the input terminal phase of output end and the data processing module that holocentric claps signalEven, the holocentric claps the right bundle branch block waveform in signal to the data processing module for identification.
It further, further include display module, the output of the input terminal of the display module and the data processing moduleEnd is connected, and the display module is used to show the recognition result that the holocentric claps the right bundle branch block waveform in signal.
Further, the data preprocessing module includes concatenated analog circuit, analog-digital converter, low-pass digital filterDevice and wavelet transformer, the input terminal of the analog circuit are the input terminal of the data preprocessing module, the wavelet transformationThe output end of device is the output end of the data preprocessing module.
Further, the data preprocessing module further includes memory, the input terminal of the memory and the modulusThe output end of converter is connected.
Further, the data processing module is classifier.
Further, the classifier is to be carried out according to right bundle branch block training sample and non-right bundle branch block training sampleNeural metwork training obtains.
Further, the right bundle branch block training sample is the holocentric bat signal training that will extract right bundle branch block featureSample carries out genetic algorithm feature selecting and obtains.
Second aspect, the embodiment of the invention also provides a kind of electrocardio equipment, which includes ECG module and such asThe upper electrocardiogram signal processing device, the ECG module are connect with the electrocardiogram signal processing device.
Further, the electrocardio equipment is at least one of single lead electrocardio patch, more sign devices and patient monitor equipment.
The technical solution of the embodiment of the present invention acquires the electrocardiosignal of ECG module detection by signal acquisition module, andThe electrocardiosignal is handled by data preprocessing module and obtains holocentric bat signal, and data processing module is based on holocentric and claps signal knowledgeRight bundle branch block waveform not therein.The embodiment of the present invention can clap information based on the holocentric in electrocardiosignal and carry out right bundle branch resistanceStagnant analysis improves the recognition efficiency and accuracy rate of right bundle branch block waveform.
Embodiment one
Fig. 1 is the structural schematic diagram for the electrocardiogram signal processing device that the embodiment of the present invention one provides.As shown in Figure 1, the dressSet includes: signal acquisition module 110, data preprocessing module 120 and data processing module 130, wherein signal acquisition moduleFor 110 input terminal for being connected with ECG module, output end and data of the signal acquisition module 110 for exporting electrocardiosignal are pre-The input terminal of processing module 120 is connected, and data preprocessing module 120 is used to export output end and the data processing that holocentric claps signalThe input terminal of module 130 is connected, and holocentric claps the right bundle branch block waveform in signal to data processing module 130 for identification.
Wherein, ECG module can be the module for including target object electrocardiosignal, and target object can be for arbitrarily can be withGenerate the biology of electrocardiosignal, such as people, animal.Signal acquisition module 110 includes the electrocardiograph for acquiring electrocardiosignal,Such as analog electrocardiograph and digital intelligent electrocardiograph etc..Electrocardiograph may include cardiac diagnosis lead and sensorDeng.
Fig. 2 is the electrocardiosignal conduction figure that provides of the embodiment of the present invention one, and the conductive process of electrocardiosignal successively can be with are as follows:Atrioventricular node 11 (Atrio-Ventricular node, AV Node), Xinier reservoir 15 (HIS Bundle), 12 (Left of left bundle branchBundle Branch, LBB) or right bundle branch 16 (Right Bundle Branch can be left anterior fascicle after left bundle branch 1213 (Left Anterior Fascicle, LAF) and left posterior fascicle 14 (Left Posterior Fascicle, LPF), finallyFor Pu Jin Shi fiber 17 (Purkinje fibers).The basic pathology physiological defect of right bundle branch block mainly due to fromXinier reservoir (HIS Bundle) arrives the electric pulse non-conducting of right bundle branch.In the normal situation of left bundle branch, right ventricleDepolarising is significant inconsistent with left ventricle, and this ventricular depolarization mismatch gives electrocardiogram (ECG) variation.
Further, it is incoming to have to pass through cardiac muscle for the electric signal from left ventricle, and the transmission speed of this approach is more originalXinier reservoir-Pu Jinshi fiber path it is slow, therefore on electrocardiogram, QRS complex can be wider.It is specific as follows: (1) supraventricularThe rhythm of the heart.Ventricular rhythm must be from (i.e. sinoatrial node, atrium or atrioventricular node) on ventricle；(2) QRS complex time span.QRSThe complex wave time need to be longer than 100 milliseconds (non-fully blocking) or 120 milliseconds (blocking completely)；(3) lead pole have latter stage R wave (such as: R,The waveforms such as rR', rsR', rSR' or qR) and RSR' mode (' M shape ' QRS complex).It is the embodiment of the present invention one referring to Fig. 3, Fig. 3The electrocardiogram of the right bundle branch block of offer, S wave is wide and fuzzy in figure, is allusion quotation of the right bundle branch block on electrocardiogramType performance.
The output end of signal acquisition module 110 is connected with the input terminal of data preprocessing module 120, signal acquisition module110 output end exports collected electrocardiosignal to data preprocessing module 120.
Optionally, data preprocessing module 120 may include concatenated analog circuit, analog-digital converter, Low pass digital filterWave device and wavelet transformer, the input terminal of analog circuit are the input terminal of data prediction mould 120, the output end of wavelet transformerFor the output end of data preprocessing module 120.
Optionally, data preprocessing module 120 further includes memory, the output of the input terminal and analog-digital converter of memoryEnd is connected.
The processing such as impedance matching, filtering, amplification can be carried out to electrocardiosignal by analog circuit, then by analog-to-digital conversionThe analog signal of human body physiological parameter is converted into digital signal by device, is stored by memory.Since signal acquisition module 110 acquiresElectrocardiosignal include various noises, waveform is coarse and rough, leads to the useful letter contained in QRS complex in electrocardiosignalBreath is difficult to extract.By the digital filter being arranged in data preprocessing module 120, lowpass digital filter (example can be usedSuch as Butterworth filter) low-pass filtering is carried out, high-frequency noise (300Hz or more) is filtered out, filtered electrocardiosignal is obtained.GinsengSee that Fig. 4, Fig. 4 are the electrocardiosignal comprising right bundle branch block that the embodiment of the present invention one provides.
Further, it by the wavelet transformer in data preprocessing module 120, can be extracted using wavelet transformation techniqueThe shape information of P wave, QRS complex and T wave in electrocardiosignal (such as electrocardiosignal in Fig. 4), i.e., the one complete heart are clapped,Also referred to as holocentric claps signal.Referring to Fig. 5, Fig. 5 is the schematic diagram that the holocentric that the embodiment of the present invention one provides claps signal, the cross in figureCoordinate is the time, and unit is second (S), and ordinate is voltage, and unit is volt (V), and it includes P wave, QRS wave that holocentric, which claps signal, in figureGroup, T wave and U wave, wherein U wave be not required include.
Optionally, data processing module 130 is classifier.Optionally, classifier is according to right bundle branch block training sampleNeural metwork training is carried out with non-right bundle branch block training sample to obtain.Further, right bundle branch block training sample is that will mentionIt takes the holocentric of right bundle branch block feature to clap signal training sample progress genetic algorithm feature selecting to obtain.Holocentric claps signal training sampleThis is the holocentric bat signal for including right bundle branch block feature.Specifically, first clapping signal training sample to holocentric extracts right bundle branchBlock feature, then feature selecting is carried out by genetic algorithm to the right bundle branch block feature extracted and obtains right bundle branch block trainingSample.
Non- right bundle branch block training sample is to clap signal to the holocentric for not including right bundle branch block feature to carry out feature extractionIt obtains, is not construed as limiting in quantity the present embodiment of right bundle branch block training sample and non-right bundle branch block training sample, Ke YigenIt is set according to actual conditions.Right bundle branch block feature may include right bundle branch block feature assemblage characteristic, and wherein right bundle branch hindersStagnant feature assemblage characteristic may include the assemblage characteristic obtained after being combined by least two right bundle branch block feature item features.It is rightBundle-branch block feature item feature may include but be not limited only to the Shannon entropy of P wave, the Shannon entropy of Q wave, the Shannon entropy of R wave, S waveShannon entropy, the Shannon entropy of T wave, holocentric clap the Shannon entropy of signal, the corrected value of QT interphase, the median of characteristic sequence, featureStandard deviation and the mean value of characteristic sequence of sequence etc., characteristic sequence include QT interphase corrected value, QRS complex, RS section and ST sections.
Data processing module 130 can clap signal extraction right bundle branch block feature to the received holocentric of input terminal, then to mentioningThe right bundle branch block feature taken carries out feature selecting by genetic algorithm, obtains the right bundle branch resistance after the dimensionality reduction of holocentric bat signalStagnant feature identifies right bundle branch therein in the right bundle branch block feature input classifier after the dimensionality reduction of holocentric bat signalBlock waveform.Data processing module 130 can be used for executing at least one of operations described below as a result: calculate the Shannon of P waveEntropy；Calculate the Shannon entropy of Q wave；Calculate the Shannon entropy of R wave；Calculate the Shannon entropy of S wave；Calculate the Shannon entropy of T wave；Holocentric is calculated to clapThe Shannon entropy of signal；Calculate the corrected value of QT interphase；Calculate the median of characteristic sequence；Calculate the standard deviation of characteristic sequence；MeterCalculate the mean value of characteristic sequence；Calculate right bundle branch block feature assemblage characteristic.
Specifically, can indicate the sequence of P wave, Q wave, R wave, S wave and T wave, ecg using X_P, X_Q, X_R, X_S and X_TIndicate that holocentric claps the sequence of signal.It is sliced for a 10s, passes through energy theoremThe energy P1-P5 of P wave, Q wave, R wave, S wave and T wave is calculated separately, wherein the π of ω=2 f.Energy P1-P5 just represents P wave, Q at this timeWave, R wave, S wave and T wave wave signal.Pass through Shannon entropy formulaP wave, Q wave, R wave, S is calculatedThe Shannon entropy that wave, T wave and holocentric clap signal is respectively as follows:
With
QT interphase can be corrected using three kinds of modes in the present embodiment, the QT interphase after obtaining three kinds of correctionsQTcB, QTcF and QTlc are calculated: QTcB=X_QT/sqrtRR, QTcF=X_QT/cubrtRR and QTlc=by following formulaQT+0.154* (1-RR), wherein X_QT=X_T-X_Q,Calculate feature sequenceMedian, standard deviation and the mean value of column, characteristic sequence include corrected value (above-mentioned three kinds), the QRS complex, RS sections and ST of QT interphaseSection, is respectively labeled as QT1, QT2, QT3, QT4, QT5 and QT6, is calculated by following formula: QT1=median (QTcB), QT2=std (QTcB)/mean (QTcB), QT3=median (QTcF), QT4=std (QTcF)/mean (QTcF), QT5=Median (QTlc) and QT6=std (QTlc)/mean (QTlc).Wherein median, std and mean are respectively in the sequence of calculationThe operator of digit, standard deviation and mean value.
Since the right bundle branch block feature of said extracted is multidimensional (such as 18 dimensions), the dimension of feature is larger and containing linearRelevant input item, therefore feature selecting can be carried out by genetic algorithm in the present embodiment and realize Feature Dimension Reduction, using Fig. 6 instituteThe genetic algorithm shown chooses most suitable right bundle branch block feature.
Fig. 6 is the flow chart for the genetic algorithm that the embodiment of the present invention one provides.Specifically, carrying out feature by genetic algorithmThe process of selection and dimensionality reduction can be with are as follows: S201, beginning.S202, the electrocardiosignal feature extracted is read.Electrocardio letter in this stepNumber feature is the right bundle branch block feature of above-mentioned various dimensions.S203, residual signal is extracted.S204, judge whether to meet stoppingCriterion.If so, S212 is executed, if it is not, then executing S205.S205, generating new population, (initialization is in dictionary parameter group rangeInterior random generation).S206, individual adaptation degree (inner product with residual signal) is calculated.S207, optimal individual character (maximum inner product is savedValue) and other inner product values, total is N.After S207, S2081-S2083, S2084 and S2085- can be executed parallelS2086.S2081, in addition to optimum individual, all individual two one groups, competition.S2082, optimum individual hybridize with winner.The individual that S2083, hybridization are formed enters next-generation.S2084, optimum individual enter the next generation.S2085, it is generated by optimum individual(N-1)/2 variant.S2086, (N-1)/2 variation enter next-generation.S209, judge whether to meet evolutionary generation.If so,S210 is executed, executes S207 if it is not, then returning.S210, reconstruct atom is obtained.S211, residual signal is updated.After S211, returnReceipt row S203.S212, end.
Further, neural network can choose the regression model of parameter adaptive adjusting, which can selectOne hidden layer, single hidden layer feedforward network is capable of handling most of nonlinear problem, only excessive in node in hidden layer, but stillWhen being unable to satisfy convergence precision requirement, using double hidden layers, the latter can handle all nonlinear problems, but network training speed hasDeclined, convergence time is long, therefore selects a hidden layer.The regression model can carry out Hidden nodes using trial and error procedure and determine,It is trained with same sample set, first hidden node number is arranged near inum/2+1, gradually increases Hidden nodes to 2*inum+1, and continue growing up to not restraining, most suitable Hidden nodes are determined by analytical error performance curve, wherein inum isInput layer number and the intrinsic dimensionality of input.Also, the input layer number of the regression model is characterized the dimension of sample,Having several features just has several input nodes.Only one node of the output layer of the regression model is just whether to include right beamBranch blocks waveform.Regression model is trained according to right bundle branch block training sample and non-right bundle branch block training sample, is instructedThe regression model perfected is that can clap the classifier that signal is identified to holocentric in the present embodiment.
Optionally, which can also include display module, at the input terminal of display module and dataThe output end for managing module 130 is connected, and display module is used to show the recognition result that holocentric claps the right bundle branch block waveform in signal.Wherein, it includes that right bundle branch block waveform and the holocentric clap signal that the recognition result of right bundle branch block waveform, which includes holocentric bat signal,It does not include two kinds of right bundle branch block waveform.
The technical solution of the present embodiment, the electrocardiosignal of ECG module detection is acquired by signal acquisition module, and is passed throughData preprocessing module processing cardioelectric signals obtain holocentric and clap signal, and it is therein that data processing module is based on holocentric bat signal identificationRight bundle branch block waveform.The present embodiment can clap information based on the holocentric in electrocardiosignal and carry out right bundle branch block analysis, improveThe recognition efficiency and accuracy rate of right bundle branch block waveform, more meets the demand of practical application,.