CN100536765C - Cardioelectric characteristic extracting process based on evolutive wavelet wiener deconvolution - Google Patents

Abstract

The cardioelectric data characteristic extracting process includes the following steps: 1.preprocessing cardioelectric data; 2. extracting QRS wave group characteristic through wavelet transformation to extract the sub-frequency band of the QRS wave group from the preprocessed cardioelectric signal and the subsequent evolutive wavelet Wiener deconvolution process to extract the position of the characteristic point of QRS wave group; and 3. extracting the characteristics of P wave and T wave through substituting the time section in the QRS complex wave with base line, the subsequent wavelet transformation to extract sub-frequency band of P wave and T wave from the cardioelectric signal with the QRS wave group eliminated, and the final evolutive wavelet Wiener deconvolution process until extracting accurate characteristic point position of P wave and T wave. The present invention lays foundation for the characteristic detection.

Description

Cardioelectric characteristic extracting process based on evolutive wavelet wiener deconvolution

Technical field

The present invention relates to a kind of feature extracting method of electrocardiogram (ECG) data, belong to the biomedical engineering technology field.

Background technology

The feature extraction of electrocardiosignal is one of powerful measure of detecting of heart disease.The algorithm that is generally used for the electrocardiosignal feature extraction has following several: length conversion and energy change of variable, hidden Markov model, artificial neural network and wavelet transformation.But said method all has certain defective: length conversion and energy change of variable can't accurately detect the characteristic information of abnormal Q RS wave group; Hidden Markov model is powerless to the detection of accidental abnormal electrocardiogram ordinary wave shape and isolated P ripple; Artificial Neural Network lacks effectiveness to the detection of abnormal electrocardiogram waveform equally; Small wave converting method is better than said method to the detection of ill electrocardiosignal, but the detection that the ecg wave form unstable state is changed, particularly under the low signal-to-noise ratio situation, and still can not be satisfactory.

In recent years, in order to solve the defective of said method, many scholars have carried out going deep into extensive studies.People such as Messadeg are used for the tagsort of electrocardiosignal by hidden Markov model and wavelet transformation technique are combined; Yu and he work together and adopt independent component analysis and the bonded method of artificial neural network to realize the extraction and the classification of ecg characteristics; Utilized Liu adaptive wavelets transform and fuzzy logic principle realize location, electrocardiosignal R peak, or the like.As can be seen, above-mentioned scholar no longer is simple a kind of theory or the method for adopting, but by remedying mutually between the several different methods, realizes optimizing coupling.But then, though these new trials have solved the subproblem that original ecg characteristics extracts, since perfect not enough, much need improved place so exist.

Summary of the invention

Purport of the present invention is at the on-line monitoring simultaneously of multidigit needs of patients in the remote electrocardiogram monitor, the problem of carrying out the heart disease detection promptly and accurately, a kind of electrocardiogram (ECG) data feature extracting method of efficiently and accurately is proposed, to overcome the defective of prior art medium-long range cardiac monitoring cardiac disease detection.The present invention can improve the efficient and the accuracy of heart disease identification greatly, thereby reduces doctor's operating pressure, makes therefore that also more oversensitive dirty disease patient can accept to guard accurately and effectively and treat.

For this reason, the present invention adopts following technical scheme: a kind of cardioelectric characteristic extracting process based on evolutive wavelet wiener deconvolution comprises the following steps:

(1) according to the sample frequency of electrocardiosignal, selects the morphological structure operator, utilize the baseline drift in the Mathematical Morphology Method removal original electrocardiographicdigital signal;

(2) initialization Wiener filter, scale parameter

, energy parameter And iteration step length (β _p), get base and float electrocardiosignal after the filtering, carry out the wavelet transformation under the 1st yardstick, obtain the useful signal of low frequency part, and the noise signal of HFS, obtain the power spectrum of useful signal and noise signal and crosspower spectrum between the two, construct the non-causal Wiener filter, as the initialization Wei Na operator vector that deconvolutes with this

(3) QRS complex wave feature strengthens: adopt the quadrature discrete small echo to decompose pretreated electrocardiosignal, according to the frequency range of electrocardiosignal and QRS complex wave, carry out twice or twice above wavelet transformation to obtain the feature sub-band at QRS wave group place; With the Wei Na operator vector that deconvolutes

Act on the sub-band at QRS wave group place, then calculate the gross energy error

If error amount less than setting threshold, stops computing; Otherwise, utilize steepest descent gradient based method, upgrade the Wei Na operator vector that deconvolutes Energy parameter

And scale parameter

Continue computing up to finding the satisfied Wei Na operator vector that deconvolutes

(4) P ripple and T wave characteristic strengthen: utilize the positional information of the QRS complex wave that step (3) obtains, the QRS wave group is substituted with baseline; Adopt the quadrature discrete small echo to decompose the electrocardiosignal that removes the QRS wave group, according to gained electrocardiosignal and P ripple, T wave frequency scope, carry out twice or twice above wavelet transformation to obtain the feature sub-band at P ripple and T ripple place; With the Wei Na operator vector that deconvolutes Act on the feature sub-band at P ripple and T ripple place, then calculate the gross energy error

If error amount less than setting threshold, stops computing; Otherwise, utilize steepest descent gradient based method, upgrade the Wei Na operator vector that deconvolutes

Energy parameter

And scale parameter

Continue computing up to finding the satisfied Wei Na operator vector that deconvolutes

(5) detect QRS wave group feature locations point: get the QRS sub-band signal after feature that step (3) obtains strengthens, at first detect the R peak position,, find out first valley point and peak dot before and after the R peak then according to the R peak position; Then according to following criterion location feature point: if there is valley point of a peak dot at the R peak, peak dot is a QRS wave group starting point so, and the valley point is a Q ripple flex point; If have only a valley point before the R peak, this wave group is the RS wave group so, and this valley point is a RS wave group starting point; Valley point behind the R peak is a S ripple flex point, and peak dot is the wave group terminating point;

(6) detect P, T wave characteristic location point: get P ripple, T marble band signal after feature that step (4) obtains strengthens, detect the peak point position of P ripple and T ripple, then find out first preceding valley point of P ripple and T crest value point respectively, be the starting point of P ripple and T ripple; Find out first valley point behind P ripple and the T crest value point, be the terminating point of P ripple and T ripple.

In above-mentioned cardioelectric characteristic extracting process, the formula that steepest descent gradient based method adopts can be:

${\hat{W}}_{p+1}={\hat{W}}_p-{\mathrm{\β}}_p{\▿}_{{\hat{W}}_p}{\mathrm{\ϵ}}^j({\hat{W}}_p,{\widehat{\mathrm{\α}}}_p,{\widehat{\mathrm{\γ}}}_p)$

${\widehat{\mathrm{\α}}}_{p+1}={\widehat{\mathrm{\α}}}_p-{\mathrm{\β}}_p{\▿}_{{\widehat{\mathrm{\α}}}_p}{\mathrm{\ϵ}}^j({\hat{W}}_p,{\widehat{\mathrm{\α}}}_p,{\widehat{\mathrm{\γ}}}_p),$ In the formula,

${\widehat{\mathrm{\γ}}}_{p+1}={\widehat{\mathrm{\γ}}}_p-{\mathrm{\β}}_p{\▿}_{{\widehat{\mathrm{\γ}}}_p}{\mathrm{\ϵ}}^j({\hat{W}}_p,{\widehat{\mathrm{\α}}}_p,{\widehat{\mathrm{\γ}}}_p)$

β _pStep-length when being the p time iteration,

Be gradient operator.

The essential characteristics of technical scheme of the present invention is: 1. electrocardiogram (ECG) data pretreatment: select the baseline drift in the appropriate mathematical morphological operator removal electrocardiosignal; 2. the feature extraction of QRS wave group: extract through the sub-band under the QRS wave group in the electrocardiosignal after the pretreatment by small wave converting method,, extract the characteristic point position of QRS wave group again to this sub-band signal utilization evolution type Wei Na method of deconvoluting.3. the feature extraction of P ripple and T ripple: the characteristic point position information of the QRS wave group that obtains according to the first step replaces time period at QRS complex wave place, to avoid the frequency domain interference when P ripple and T wave characteristic are extracted with baseline.Then adopt wavelet transformation, the sub-band of the electrocardiosignal behind the removal QRS wave group being made P ripple, T ripple extracts, and the method for at last the sub-band utilization evolution type Wei Na that obtains being deconvoluted is up to extracting P ripple, T wave characteristic point position accurately.

The present invention proposes based on the evolutive wavelet wiener deconvolution new technique and realize the new approaches that ecg characteristics extracts: utilize the localized advantage of wavelet transformation time-frequency, signature waveform is divided into each sub-frequency bands, then utilize the evolution type Wei Na new technique that deconvolutes that the feature sub-band signal is carried out iterative processing, up to finding satisfied characteristic point position information.Adopt this method, can so just can provide strong basis outstanding tangible peak dot or the valley point of changing into of inapparent key position point in the original electrocardiographicdigital data for feature detection.

Description of drawings

The characteristic information of Fig. 1 ecg wave form and definition.

Fig. 2 is based on the electrocardio baseline drift filtering sketch map of Mathematical Morphology Method, and Fig. 2 (a) is the original electrocardiographicdigital signal; Fig. 2 (b) is the electrocardiosignal after the filtering baseline drift.

The feature of Fig. 3 QRS wave group strengthens result schematic diagram, and Fig. 3 (a) is the electrocardiosignal before QRS wave group feature strengthens; Fig. 3 (b) is the electrocardiosignal after QRS wave group feature strengthens.

The feature of Fig. 4 P ripple and T ripple strengthens result schematic diagram, and Fig. 4 (a) is the electrocardiosignal of QRS wave group after shifting out; Fig. 4 (b) is the electrocardiosignal after P ripple and T wave characteristic strengthen.

The specific embodiment

Before the present invention was further described, existing knowledge such as the character of first principal character to ecg wave form, Mathematical Morphology theory, electrocardiosignal wavelet transform, evolutive wavelet wiener deconvolution technology were done introduction.

The principal character of ecg wave form

Electrocardiosignal is one of topmost physiological signal, and it has reflected the activity of heart, and mainly contain 3 parts and form: the P ripple is represented first deflection of atrium depolarization; The QRS complex wave is produced by sequences of ventricular depolarization; Ventricular bipolar has produced the T ripple.Because these ripples have special shape in time domain and frequency domain, so, by observation, can find the abnormal conditions of heart to the ECG signal.Fig. 1 has provided Electrocardiographic fundamental component, and the definition of every ecg characteristics.Purpose of the present invention will be found out a kind of characteristic detection method accurately and effectively exactly, extracts the characteristic information shown in publishing picture.

Mathematical Morphology theory

Mathematical morphology filter is a kind of nonlinear signal conversion, is used to change the local geometric characteristic of signal.Its theoretical basis comes from the basic operation method of set theory in the graphical analysis, i.e. Mathematical Morphology Method, this theory are proposed by Matheron and Serra the earliest.In this theory, all signals all are seen as the set that has a dimensioning in the Euclidean geometry space, and morphological method just can be counted as a kind of set operation that comes picked up signal geometry quantitative information by the signal conversion.For the binary signal that is counted as gathering, it being corroded and expand with certain structural element is the most basic morphological operation.Suppose a given morphological structure element B, it can be defined as respectively expansion and the erosion operation of signal X:

$X\⊕B=\underset{b\∈B}{\∪}{X}_b\underset{b\∈B}{\∪}\left\{P\right|p=x+b,b\∈B,x\∈X\}---\left(1\right)$

For primary electrocardiosignal, dilation operation (formula 1) is to be used to remove the crest that frequency is higher than the electrocardio composition of baseline drift, erosion operation (formula 2) then is to be used to remove the trough that frequency is higher than the electrocardio composition of baseline drift, and expansion and corrosive resultant effect are exactly the baseline drift that obtains in the electrocardiosignal.Therefore, can be with the procedure definition of removing baseline drift:

The character of electrocardiosignal wavelet transform

Discrete wavelet coefficient X for discrete signal x (n) ^j(n) can obtain by a twin-channel orthogonal filter group:

X ^j(n)＝h ^j(n)*x(n)，j＝1，…，J (4)

Here J is the sum of change of scale; Utilize high pass filter h ₁(n) and low pass filter h ₀(n), can calculate filter coefficient h under each yardstick ^j(n):

h ¹(n)＝h ₀(2n)*h ₀(2 ²n)*h ₀(2 ³n)*…*h ₀(2 ^J-1n)

h ²(n)＝h ₁(2n)*h ₀(2 ²n)*h ₀(2 ³n)*…*h ₀(2 ^J-1n)

h ³(n)=h ₁(2n) * h ₀(2 ²N) * h ₀(2 ³N) * ... * h ₀(2 ^J-2N) (5) so, public affairs

.

.

.

h ^J(n)＝h ₁(2n)

Wavelet coefficient in the formula (4) under each yardstick can be expressed as with the form of matrix and vector:

X ^j＝H ^jx，j＝1，…，J (6)

The H here ^jBe convolution type matrix, x is a signal vector.

For given electrocardiosignal x, on time domain, the size of its power spectral density is directly proportional with the inverse of frequency (1/f), so the variance of electrocardio wavelet coefficient can be expressed as:

$V_X^i=\mathrm{var}{X}^j={\mathrm{\σ}}^2{2}^{-\mathrm{j\γ}}---\left(7\right)$

Here σ ²Be and signal variance

And analysis filter h ₀(n) Xiang Guan constant, γ is a scale parameter, and 0≤γ≤2.

Taken the logarithm in formula (7) both sides, the logarithm value that can obtain wavelet coefficient is a linear change, that is:

${\log }_2\left(V_X^j\right)=-\left[\mathrm{\γ}{\log }_2\left(2\right)\right]j+{\log }_2\left({\mathrm{\σ}}^2\right)=-\mathrm{\γj}+\mathrm{\α}---\left(8\right)$

Here α=log ₂(σ ²) be defined as energy parameter.

The evolutive wavelet wiener deconvolution technology

Dimension is received deconvolution techniques and is normally used in the analysis of spectrum of optical signalling.The present invention is deconvoluted method through improving with traditional Wei Na, forms the distinctive evolution type dimension of wavelet field and receives deconvolution techniques, makes full use of wavelet transformation in the spatial time frequency resolution characteristic of Besov with this, realizes the abundant enhancing of characteristic information in the electrocardiosignal.The core concept of evolutive wavelet wiener deconvolution technology is described below.

Arbitrary signal y (t) in the real world can regard the result of useful signal x (t) and interfering signal g (t) convolution as, can be expressed as:

$y\left(t\right)=x\left(t\right)\⊗g\left(t\right)=\underset{-\∞}{\overset{+\∞}{\∫}}x\left(t\right)g(t-\mathrm{\τ})\mathrm{d\τ}---\left(9\right)$

The result of convolution has caused signal waveform to broaden, and this has just brought overlapping each other of adjacent waveform, thereby causes the fuzzy of signal characteristic information.

In order to obtain useful signal x (t), need find a function to offset the effect of interference function g (t), just find out a function

Satisfy following relation (δ (t) is a uni-impulse function)

$g\left(t\right)\hat{w}\left(t\right)=\mathrm{\δ}\left(t\right)---\left(10\right)$

So can obtain formula (11) by formula (9) and (10).

$y\left(t\right)\hat{w}\left(t\right)=\underset{-\∞}{\overset{+\∞}{\∫}}x\left(t\right)\mathrm{\δ}(t-\mathrm{\τ})\mathrm{d\τ}=x\left(t\right)---\left(11\right)$

For the convenience that further describes, with continuous signal x (t), the g (t) of above-mentioned appearance,

, y (t) and δ (t) be rewritten as corresponding discrete form x (n), g (n),

, y (n) and δ (n).Like this, just can be with the form representation formula (10) of matrix and vector.

$G\hat{W}=\mathrm{\δ}---\left(12\right)$

Here

It is the operator that deconvolutes

Vector representation, among the present invention, in order to improve algorithm effects and efficient, will Be designed to the Wei Na operator that deconvolutes.G is the matrix notation of g (n), and δ=[0 ... 010 ... 0] ^T

Corresponding, formula (11) can be rewritten as:

$\hat{x}=Y\hat{w}---\left(13\right)$

Here Be the vector representation of x (n), and Y is the matrix notation of y (n).

Thus, utilize formula (6) and formula (13) can obtain wavelet coefficient under each yardstick:

${\hat{X}}^j=H^{j}Y\hat{w}---\left(14\right)$

In order to obtain the optimum Wei Na operator that deconvolutes

The evolution generative process in defined two error criterions, can obtain first error criterion by formula (12)

Be defined as follows:

$e_{\mathrm{\δ}}\left(\hat{W}\right)=G\hat{w}-\mathrm{\δ}---\left(15\right)$

Then, utilize the character (formula 8) of the wavelet coefficient of 1/f signal can obtain another error criterion corresponding to each yardstick

J=1 ..., J

$e_X^j(\hat{W},\widehat{\mathrm{\α}},\widehat{\mathrm{\γ}})={\log }_2\left(V_X^j\right)-(-j\widehat{\mathrm{\γ}}+\widehat{\mathrm{\α}})---\left(16\right)$

Here

For:

$V_X^j\text{=}{\left(H^{j}Y\hat{w}\right)}^T{H}^{j}Y\hat{w}={\left(D^j\hat{w}\right)}^T\left(D^j\hat{w}\right)={\hat{w}}^T{D}^{\mathrm{jT}}{D}^j\hat{w}={\hat{w}}^T{C}^j\hat{w}---\left(17\right)$

Like this, the error criterion by application of formula (15) and formula (16) can obtain the gross energy error criterion under each yardstick:

${\mathrm{\ϵ}}^j(\hat{w},\widehat{\mathrm{\α}},\widehat{\mathrm{\γ}})={\left(e_X^j(\hat{w},\widehat{\mathrm{\α}},\widehat{\mathrm{\γ}})\right)}^2+{\left(e_{\mathrm{\δ}}\left(\hat{w}\right)\right)}^T\left(e_{\mathrm{\δ}}\left(\hat{w}\right)\right)---\left(18\right)$

By finding the solution the minima of gross energy error, just can obtain the optimum Wei Na operator that deconvolutes The gross energy error

Minima can find the solution by gradient base minimization technique.In order to improve algorithm the convergence speed, adopted steepest descent gradient base algorithm, that is: among the present invention

${\hat{W}}_{p+1}={\hat{W}}_p-{\mathrm{\β}}_p{\▿}_{{\hat{W}}_p}{\mathrm{\ϵ}}^j({\hat{W}}_p,{\widehat{\mathrm{\α}}}_p,{\widehat{\mathrm{\γ}}}_p)$

${\widehat{\mathrm{\α}}}_{p+1}={\widehat{\mathrm{\α}}}_p-{\mathrm{\β}}_p{\▿}_{{\widehat{\mathrm{\α}}}_p}{\mathrm{\ϵ}}^j({\hat{W}}_p,{\widehat{\mathrm{\α}}}_p,{\widehat{\mathrm{\γ}}}_p)---\left(19\right)$

${\widehat{\mathrm{\γ}}}_{p+1}={\widehat{\mathrm{\γ}}}_p-{\mathrm{\β}}_p{\▿}_{{\widehat{\mathrm{\γ}}}_p}{\mathrm{\ϵ}}^j({\hat{W}}_p,{\widehat{\mathrm{\α}}}_p,{\widehat{\mathrm{\γ}}}_p)$

The β here _pStep-length during the p time iteration,

Be gradient operator.

Cardioelectric characteristic extracting process of the present invention specifically describes as follows:

1) removes baseline drift.Adopt the baseline drift in the Mathematical Morphology technology removal original electrocardiographicdigital signal.The selection of morphological structure operator should be decided according to the sample frequency of electrocardiosignal.The present invention is directed to the electrocardiosignal of 360Hz sample frequency, get the reef knot constitutive element of 125ms.

2) parameter initialization:

A) initialization Wiener filter

Get base and float electrocardiosignal after the filtering, carry out the wavelet transformation under the 1st yardstick, obtain the useful signal of low frequency part, and the noise signal of HFS, obtain the power spectrum of useful signal and noise signal and crosspower spectrum between the two, construct the non-causal Wiener filter with this, as the initialization wave filter

B) initialization scale parameter

And energy parameter

Get ${\widehat{\mathrm{\γ}}}_0=0;$ ${\widehat{\mathrm{\α}}}_0=0.$

C) initialization iteration step length β _p: for reducing operand, the selection iteration step length is a constant, β _pSelection influence convergence of algorithm speed, the excessive algorithm that causes is not restrained, too small then algorithm the convergence speed is slack-off, through experiment in a large number, selects β _p=0.125 can guarantee algorithmic statement, can guarantee signal smoothing after the iteration again.

3) feature of QRS complex wave strengthens:

A) wavelet transform.Adopt the quadrature discrete small echo to decompose pretreated electrocardiosignal, according to the frequency range of electrocardiosignal and QRS complex wave, the wavelet transformation that carries out suitable number of times is to obtain the feature sub-band at QRS wave group place;

B) wavelet field evolution type Wei Na deconvolutes.With the Wei Na operator vector that deconvolutes Act on the sub-band at QRS wave group place, then calculate the gross energy error

If error amount is less than setting threshold, i.e. ε≤ε ₀, stop computing; Otherwise, upgrade the Wei Na operator vector that deconvolutes Energy parameter

And scale parameter

Continue computing up to finding the satisfied Wei Na operator vector that deconvolutes

4) feature of P ripple and T ripple strengthens:

A) remove the QRS wave group.Utilize the positional information of the above-mentioned QRS complex wave that obtains, the QRS wave group is substituted with baseline;

B) wavelet transform.Adopt the quadrature discrete small echo to decompose the electrocardiosignal that removes the QRS wave group, according to gained electrocardiosignal and P ripple, T wave frequency scope, the wavelet transformation that carries out suitable number of times is to obtain the feature sub-band at P ripple and T ripple place;

C) wavelet field evolution type Wei Na deconvolutes.Adopt the algorithm in the step 3), up to obtaining the satisfied Wei Na operator vector that deconvolutes

5) feature locations point detects:

A) QRS wave group feature locations point detects: get the QRS sub-band signal after feature that step 3) obtains strengthens, at first detect the R peak position, then according to the R peak position, find out first valley point and peak dot before and after the R peak.Then according to following criterion location feature point: if there is valley point of a peak dot at the R peak, peak dot is a QRS wave group starting point so, and the valley point is a Q ripple flex point; If have only a valley point before the R peak, this wave group is the RS wave group so, and this valley point is a RS wave group starting point; Valley point behind the R peak is a S ripple flex point, and peak dot is the wave group terminating point.

B) P, T wave characteristic location point detect: get P ripple, T marble band signal after feature that step 4) obtains strengthens, detect the peak point position of P ripple and T ripple, then find out first preceding valley point of P ripple and T crest value point respectively, be the starting point of P ripple and T ripple; Find out first valley point behind P ripple and the T crest value point, be the terminating point of P ripple and T ripple.

Fig. 2 adopts Mathematical Morphology Method to carry out the result schematic diagram that base floats filtering, and the present invention is directed to sample frequency is the 360Hz electrocardiosignal, has adopted 128 milliseconds reef knot constitutive element.

Fig. 3 is the feature enhanced results sketch map of QRS complex wave, and Fig. 3 (a) is the figure that No. 116 is write down after the process base floats filtering among the MIT-BIH arrhythmia data base, the position of figure acceptance of the bid clear R ripple, P ripple and T ripple.Receive deconvolution techniques by adopting the evolution type dimension, the feature of QRS wave group has been strengthened (Fig. 3 (b)) greatly, the amplitude of P ripple and T ripple has then obviously been cut down: the R ripple becomes more sharp-pointed, the starting point of QRS wave group becomes first valley point before the R peak, and the terminal point of QRS wave group becomes tangible peak dot behind the R peak; P ripple and T ripple are eliminated substantially.

Fig. 4 is P ripple and T wave characteristic enhanced results sketch map.Fig. 4 (a) is that No. 116 record floats filtering through base among the arrhythmia data base, and the figure of QRS wave group after removing.Also provided the position of R ripple, P ripple and T ripple among the figure, receive deconvolution techniques by adopting the evolution type dimension, the feature of P ripple and T ripple has been strengthened (Fig. 4 (b)) greatly: P ripple and T ripple become more sharp-pointed, their starting point becomes first valley point before P crest value point or the T crest value point, and their terminating point becomes QRS ripple P crest value point or T crest value point first valley point afterwards.

Among the present invention, adopt the evolution type dimension to receive deconvolution techniques and carry out the data that QRS ripple and P wave characteristic extract and take from MIT-BIH arrhythmia data base, the result of feature extraction and the performance comparison of carrying out feature extraction with traditional wavelet thereof are as shown in Table 1 and Table 2.When the T wave characteristic is extracted, detection performance for important parameter relevant in the evaluate cardiac disease detection-QT interval (characterizing ventricle polarization and unpolarized total time) with T ripple terminal point, adopt the QT data base as Data Source, the testing result of QT interval is as shown in table 3.

Table 1 QRS complex wave feature extraction result

Table 2 P wave characteristic is extracted the result

Table 3 QT interval testing result

Claims (2)

1. the cardioelectric characteristic extracting process based on evolutive wavelet wiener deconvolution is characterized in that, comprises the following steps:

(1) according to the sample frequency of electrocardiosignal, selects the morphological structure operator, utilize the baseline drift in the Mathematical Morphology Method removal original electrocardiographicdigital signal;

(2) initialization Wiener filter, scale parameter

Energy parameter

And iteration step length (β _p), get base and float electrocardiosignal after the filtering, the ground floor that carries out wavelet transformation decomposes, obtain the useful signal of low frequency part under the 1st yardstick, and the noise signal of HFS, obtain the power spectrum of useful signal and noise signal and crosspower spectrum between the two, construct the non-causal Wiener filter, as the initialization Wei Na operator vector that deconvolutes with this

(3) QRS complex wave feature strengthens: adopt the quadrature discrete small echo to decompose pretreated electrocardiosignal, according to the frequency range of electrocardiosignal and QRS complex wave, carry out two-layer or two-layer above wavelet transformation to obtain the feature sub-band at QRS wave group place; With the initialization Wei Na operator vector that deconvolutes

Act on the sub-band at QRS wave group place, then calculate the gross energy error

If error amount less than setting threshold, stops computing; Otherwise, utilize steepest descent gradient based method, upgrade the Wei Na operator vector that deconvolutes

Energy parameter

And scale parameter

Continue computing up to error amount less than setting threshold, promptly find the satisfied Wei Na operator vector that deconvolutes

(4) P ripple and T wave characteristic strengthen: utilize the positional information of the QRS complex wave that step (3) obtains, the QRS wave group is substituted with baseline; Adopt the quadrature discrete small echo to decompose the electrocardiosignal that removes the QRS wave group,, carry out two-layer or two-layer above wavelet transformation to obtain the feature sub-band at P ripple and T ripple place according to gained electrocardiosignal and P ripple, T wave frequency scope; With the initialization Wei Na operator vector that deconvolutes

Act on the feature sub-band at P ripple and T ripple place, then calculate the gross energy error If error amount less than setting threshold, stops computing; Otherwise, utilize steepest descent gradient based method, upgrade the Wei Na operator vector that deconvolutes

Energy parameter

And scale parameter Continue computing up to finding the satisfied Wei Na operator vector that deconvolutes

(5) detect QRS wave group feature locations point: get the QRS sub-band signal after feature that step (3) obtains strengthens, at first detect the R peak position,, find out before the R peak first valley point and peak dot behind first valley point and peak dot and R peak then according to the R peak position; Then according to following criterion location feature point: if valley point of a peak dot is arranged before the R peak, peak dot is a QRS wave group starting point so, and the valley point is a Q ripple flex point; If have only a valley point before the R peak, this wave group is the RS wave group so, and this valley point is a RS wave group starting point; Valley point behind the R peak is a S ripple flex point, and peak dot is the wave group terminating point;

(6) detect P, T wave characteristic location point: get P ripple, T marble band signal after feature that step (4) obtains strengthens, detect the peak point position of P ripple and T ripple, then find out first preceding valley point of P ripple and T crest value point respectively, be the starting point of P ripple and T ripple; Find out first valley point behind P ripple and the T crest value point, be the terminating point of P ripple and T ripple.

2. the cardioelectric characteristic extracting process based on evolutive wavelet wiener deconvolution according to claim 1 is characterized in that,

${\hat{W}}_{p+1}={\hat{W}}_p-{\mathrm{\β}}_p{\▿}_{{\hat{W}}_p}{\mathrm{\ϵ}}^j({\hat{W}}_p,{\widehat{\mathrm{\α}}}_p,{\widehat{\mathrm{\γ}}}_p)$

The formula that described steepest descent gradient based method adopts is: ${\widehat{\mathrm{\α}}}_{p+1}={\widehat{\mathrm{\α}}}_p-{\mathrm{\β}}_p{\▿}_{{\widehat{\mathrm{\α}}}_p}{\mathrm{\ϵ}}^j({\hat{W}}_p,{\widehat{\mathrm{\α}}}_p,{\widehat{\mathrm{\γ}}}_p),$ In the formula,

${\widehat{\mathrm{\γ}}}_{p+1}={\widehat{\mathrm{\γ}}}_p-{\mathrm{\β}}_p{\▿}_{{\widehat{\mathrm{\γ}}}_p}{\mathrm{\ϵ}}^j({\hat{W}}_p,{\widehat{\mathrm{\α}}}_p,{\widehat{\mathrm{\γ}}}_p)$

Wei Na when the being the p time iteration operator vector that deconvolutes, β _pStep-length when being the p time iteration,

Energy parameter when being the p time iteration,

Scale parameter when being the p time iteration,

Be gradient operator, Gross energy error when being the p time iteration.

Images