Method for Accurately Extracting Abnormal Potential within QRS

Abstract

A method for accurately extracting an abnormal potential within a QRS, comprising: in an ideal electrocardiographic signal pre-estimation stage, pre-estimating an ideal electrocardiographic signal using a non-linear transformation technology;

according to the pre-estimated ideal electrocardiographic signal, further estimating the ideal electrocardiographic signal by using a spline method, so as to accurately estimate the ideal electrocardiographic signal; and according to the accurately estimated ideal electrocardiographic signal, accurately extracting an abnormal potential within the QRS by means of a mobile standard deviation analysis technology. The method can be used not only on an average electrocardiographic signal after multiple superimposition, but also on a single beat electrocardiographic signal.

Claims

1. A method of accurately extracting abnormal potential in QRS, characterized in that, comprising the steps of: Step 1: Pre-processing an original electrocardiographic (ECG) signal x.sub.1(i) to obtain a pre-processed ECG signal x.sub.2(i); when the original ECG signal is a measured single heart beat ECG signal, processing it by a low-pass filter and a power frequency bandpass filter to eliminate an influence of baseline drift and power frequency interference on subsequent processing; when the original ECG signal is a measured ECG signal containing multiple heart beats, processing it by a signal averaging technique to eliminate effects of baseline drift, power frequency interference and measurement noise on subsequent processing; Step 2: Processing feature point detection of the pre-processed ECG signal x.sub.2(i) to determine a feature point position and a QRS range, and obtaining an estimated ideal ECG signal by nonlinear transformation; first, processing feature point detection of the pre-processed ECG signal to determine a feature point position and a QRS range; secondly, filtering the pre-processed ECG signal obtained after processing step 1 by using two low-pass filters of different filtering frequencies respectively; then subtracting these two filtering results to obtain a difference signal, and searching for a first zero-crossing position before and after each feature point position of the difference signal; thereafter, substituting a time range contained in the first zero-crossing position before and after each feature point position by the filtering result with a higher filter frequency of the above two different filtering frequencies, and substituting other parts by the filtering result with a lower filter frequency of the above two different filtering frequencies to obtain a complex signal; finally, processing low-pass filtering for the complex signal to obtain an estimated ideal ECG signal; Step 3: Based on the pre-processed ECG signal, the feature point position and the estimated ideal ECG signal, using a spline interpolation technique to obtain an accurate estimated ideal ECG signal. First performing subtraction between the pre-processed ECG signal obtained from the step 1 and the estimated ideal ECG signal obtained from the step 2 to obtain an error signal, and searching for a zero-crossing position of the error signal; then, at the zero-crossing position of the error signal obtained after searching and the feature point position obtained from the step 2, taking a spline weight as 1, and others as 0; finally, based on the estimated ideal ECG signal obtained from the step 1 and the spline weight obtained, using cubic smoothing splines to obtain an accurate estimated ideal ECG signal; Step 4: Performing subtraction between the pre-processed ECG signal obtained from the step 1 and the accurate estimated ideal ECG signal obtained from the step 3, filtering a subtraction result by a band pass filter to obtain a filtered result; based on the filtered result and the QRS range obtained from the step 2, obtaining an abnormal potential in the QRS by moving standard deviation analysis technique; Step 5: Processing credibility evaluation on the obtained abnormal potential in the QRS. Evaluating a credibility of the abnormal potential in the QRS obtained in step 4 by using a standard deviation analysis method, and determining whether the abnormal potential in the QRS obtained from the step 4 is credible, and outputting an evaluation result.

2. The method of accurately extracting abnormal potential in QRS according to claim 1, characterized in that, the step 2, specifically, is: (1) Using x.sub.2(i) to process ECG feature point detection to obtain the QRS range, a starting position QRS.sub.b, an ending position QRS.sub.e and a ECG feature point position p(j), the number of feature points is M, J=1, 2, . . . , M. The ECG feature points includes at least a QRS starting point, a QRS ending point, and Q, R, S waveform peak points; (2) Filtering x.sub.2(i) by a higher frequency low-pass filter to obtain x.sub.hi, f.sub.h refers to a filter frequency of the low-pass filter, 100 Hzf.sub.h200 Hz (3) Filtering x.sub.2(l) by a lower frequency low-pass filter to obtain x.sub.l(i), f.sub.l refers to a filter frequency of the low-pass filter, 40 Hzf.sub.l80 Hz; (4) calculating the difference signal x.sub.d(i)by formula (3):
x.sub.d(i)=x.sub.h(i)x.sub.l(i), (3) Wherein x.sub.h(i) is a filter result of x.sub.2(i) by the higher frequency low-pass filter, x.sub.l(i) is a filter result of x.sub.2(i) by the lower frequency low-pass filter. (5) Based on the signal x.sub.d(i), searching the difference signal x.sub.d(i) for each ECG feature point time position p(j), j=1, 2, . . . , M, at backward and forward direction respectively to obtain a front and a back first zero crossing point respectively, then obtaining the corresponding time position p.sub.b(j) and p.sub.f(j) respectively; (6) constructing a point set se(j) according to p.sub.b(j) and p.sub.f(j):
set(j)={p.sub.b(j), p.sub.b(j)+1, . . . , p.sub.f(j)1, p.sub.f(j)}, (4) Based on this, construct Seth:
set.sub.h={set(1), set(2), . . . , set(M)}, (5) Based on set.sub.h, synthesize a complex signal x.sub.s(i) by formula (6): $\begin{array}{cc}{x}_s\left(i\right)=\{\begin{array}{cc}{x}_h\left(i\right),& i{\mathrm{set}}_h\\ x_l\left(i\right),& i.Math.{\mathrm{set}}_h\end{array},& \left(6\right)\end{array}$ (7) Processing low-pass filtering of x.sub.s(i) to obtain estimated ideal ECG signal x.sub.3(i) , f.sub.3 refers to a filter frequency of the low-pass filter, 100 Hzf.sub.3200 Hz.

3. The method of accurately extracting abnormal potential in QRS according to claim 1, characterized in that, the step 3, specifically, is: (1) Calculating the error signal x.sub.e(i)by equation (7):
x.sub.e(i)=x.sub.2(i)x.sub.3(i), (7) X.sub.3(i) is the estimated ideal ECG signal; (2) calculate the spline weight WO by equation (8): $\begin{array}{cc}w\left(i\right)=\{\begin{array}{cc}1,& i.Math..Math.\mathrm{is}.Math..Math.\mathrm{the}.Math..Math.\mathrm{zero}.Math.\text{-}.Math.\mathrm{crossing}.Math..Math.\mathrm{point}.Math..Math.\mathrm{of}.Math..Math.p\left(i\right).Math..Math.\mathrm{or}.Math..Math.x_e\left(i\right)\\ 0,& \mathrm{others}\end{array},& \left(8\right)\end{array}$ (3) Based on the estimated ideal ECG signal 1 x.sub.3 (i) and the spline weight w(i), using the three-order smooth splines to obtain an accurately estimated ideal ECG signal x.sub.4(i).

4. The method of accurately extracting abnormal potential in QRS according to claim 1, characterized in that, the step 4, specifically, is: (1) calculate the difference signal e(i)by equation (9):
e(i)=x.sub.2(i)x.sub.4(i) (9) x.sub.4(i) is the accurately estimated ideal ECG signal; (2) Bandpass filter e(i) to obtain a signal y(i) which contains an abnormal potential in the QRS to be extracted, the bandpass filter bandwidth is selected according to specific needs; (3) calculate a moving window variance msd(i) for the signal y(i), and calculate msd(i) by equation (10): $\begin{array}{cc}\mathrm{msd}\left(i\right)=\{\begin{array}{cc}\frac{1}{2.Math.k+1}.Math.{\left(\underset{j=-k}{\overset{k}{.Math.}}.Math..Math.y^2\left(i+j\right)-\frac{1}{2.Math.k+1}.Math.{\left(\underset{j=-k}{\overset{j=k}{.Math.}}.Math..Math.y\left(i+j\right)\right)}^2\right)}^{\frac{1}{2}},& kiN-k\\ \mathrm{msd}\left(k\right),.Math.& i<k.Math.\\ \mathrm{msd}\left(N-k\right),.Math.& i>N-k.Math.\end{array},& \left(10\right)\end{array}$ Wherein a window length is 2k+1, and k ranges from 2ms5ms, and a calculated result of msd(i) is k=2 ms. (4) calculate a reference MSD value ref.sub.msd, define an interval of 100 ms before a starting position QRS.sub.b to QRS.sub.b at the QRS is as a reference interval, first calculate a mean value ref_msd.sub.mean of msd(i) in the reference interval and a standard deviation ref_msd.sub.std, then calculate a ref.sub.msd by equation (11):
ref.sub.msd=ref_msd.sub.mean+*ref_msd.sub.std, (11) herein generally choose to be greater than 2; (5) determine a starting position AIQP.sub.b of AIQPs based on ref.sub.msd, the specific method is: starting a forward search from the QRS starting position QRS.sub.b, and stopping the search if a duration of msd (j)>ref.sub.msd is greater than or equal to a preset constant m, wherein a position at this time is set as t.sub.b, calculate the starting position of AIQP.sub.b of AIQPs by formula (12):
AIQP.sub.b=t.sub.bMk, (12) Wherein m is generally 5 ms; if the ending position QRS.sub.e of QRS is searched, then AIQP.sub.b=0 and stopping the search. (6) determine whether AIQP.sub.b is searched, if AIQP.sub.b is equal to 0, then exit and return a failure flag, otherwise continue. (7) determine an ending position AIQP.sub.e of AIQPs based on ref.sub.msd, wherein the specific method is: starting a backward search from approximately 50 ms after the QRS ending position QRS.sub.e, if a duration of msd (i)>ref.sub.msd is greater than or equal to m, stopping the search, and the position at this time is set as t.sub.e, calculate the ending position of AIQP.sub.b of AIQPs by formula (13):
AIQP.sub.e=t.sub.e+m+k, (13) If the starting position AIQP.sub.b of AIQPs is searched, then AIQP.sub.e=0 and stopping the search, determine whether AIQP.sub.e is searched, if AIQP.sub.e is equal to 0, then exit and return a failure flag, otherwise continue. (8) extract abnormal potential AIQP(i) of QRS, which is calculated according to formula (14): $\begin{array}{cc}\mathrm{AIQP}\left(i\right)=\{\begin{array}{cc}y\left(i\right),& {\mathrm{AIQP}}_{b}i{\mathrm{AIQP}}_e\\ 0,.Math.& \mathrm{others}\end{array},& \left(14\right)\end{array}$ Wherein AIQP.sub.b is the starting position of AIQPs obtained by searching, AIQP.sub.e is the ending position of AIQPs obtained by searching.

5. The method of accurately extracting abnormal potential in QRS according to claim 1, characterized in that, the step 5, specifically, is: (1) calculate the standard deviation of the reference interval ref.sub.std and the standard deviation of QRS region QRS.sub.std. ref.sub.msd is standard deviation of the reference interval y(i), QRS.sub.std is the standard deviation of the y(i) in the interval from the QRS starting position QRS.sub.b to the QRS ending position QRS.sub.e. Determine the credibility of the extraction result, which is calculated by formula (15): $\begin{array}{cc}\mathrm{credibility}=\{\begin{array}{cc}1,& {\mathrm{QRS}}_{\mathrm{std}}*{\mathrm{ref}}_{\mathrm{std}}.Math..Math.\mathrm{or}.Math..Math.{\mathrm{ref}}_{\mathrm{std}}>5.Math..Math..Math..Math.V\\ 0,& \mathrm{others}\end{array},& \left(15\right)\end{array}$ Wherein >1, the specific selection can be determined according to the actual situation. If the credibility is equal to 0, then return a failure flag, otherwise return a success flag and at the same time return the extracted abnormal potential AIQP(i) in the QRS.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0061] FIG. 1 illustrates a flow chart of the present invention.

[0062] FIG. 2 illustrates is the simulation signal used in the description of the principles of the present invention.

[0063] FIG. 3 illustrates a pre-processing flow chart according to an embodiment of the present invention.

[0064] FIG. 4 is a schematic diagram of baseline drift elimination and power frequency interference removal process according to an embodiment of the present invention.

[0065] FIG. 5 illustrates a flow chart for estimated ideal ECG signal according to an embodiment of the present invention.

[0066] FIG. 6 is a schematic diagram to show a process to obtain the estimated ideal ECG signal according to an embodiment of the present invention.

[0067] FIG. 7 illustrates a flow chart showing an accurately estimate of ideal ECG signal according to an embodiment of the present invention.

[0068] FIG. 8 is a schematic diagram to show a process for accurately estimating the ideal ECG signal by using a cubic smooth spline according to an embodiment of the present invention.

[0069] FIG. 9 illustrates a flow chart of extracting abnormal potential in QRS and evaluating the result credibility.

[0070] FIG. 10 illustrates a schematic process of extracting abnormal potential in QRS and evaluating the result credibility.

[0071] FIG. 11 is a graph showing the results of the abnormal potential extraction in the QRS by the method of the present invention for a single beat ECG and a superimposition average of multiple beat ECG.

[0072] FIG. 12 is a graph showing the results of QRS internal abnormal potential extraction of a single beat ECG of a patient with myocardial infarction by the method of the present invention.

[0073] FIG. 13 is a graph showing the results of extracting abnormal potentials in QRS of a single beat ECG of a healthy person by the method of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

[0074] The principles the present invention are described in detail below with reference to the accompanying drawings.

[0075] Referring to FIG. 1 of the drawings, which is a flowchart of the present invention, illustrates a method for accurately extracting abnormal potential in QRS, comprising the steps of:

[0076] Step 1: Preprocessing the original ECG (electrocardiographic) signal x.sub.1(i) to obtain a pre-processed ECG signal x.sub.2(i); when the original ECG signal is a measured single heart beat ECG signal, it is processed by a low-pass filter and a power frequency bandpass filter to eliminate the influence of baseline drift and power frequency interference on the subsequent process; when the original ECG signal is a measured ECG signal containing multiple heart beats, it is processed by the signal averaging technique to eliminate the effects of baseline drift, power frequency interference and measurement noise on subsequent processes.

[0077] Step 2: Processing feature point detection of the pre-processed ECG signal x.sub.2(i) to determine feature point position and QRS range, and obtaining an estimated ideal ECG signal by nonlinear transformation. First, processing feature point detection of the pre-processed ECG signal to determine feature point position and QRS range; Secondly, filtering the pre-processed ECG signal obtained after processing step 1 by using two low-pass filters of different filtering frequencies respectively. Then, subtracting the obtained two filtering results to obtain a difference signal, and searching for a first zero-crossing position before and after each feature point position of the difference signal. Then, substituting a time range contained in the first zero-crossing position before and after each feature point position by a low-pass filter filtering result of the higher filter frequency of the above two different filtering frequencies, and substituting other parts by a low-pass filter filtering result of the lower filter frequency of the above two different filtering frequencies to obtain a complex signal. Finally, processing low-pass filtering for the complex signal to obtain an estimated ideal ECG signal.

[0078] Step 3: According to the pre-processed ECG signal, the feature point position and the estimated ideal ECG signal, using a spline interpolation technique to obtain an accurate estimated ideal ECG signal. Perform subtraction between the pre-processed ECG signal obtained by the step 1 and the estimated ideal ECG signal obtained by the step 2 to obtain an error signal. Search for a zero-crossing position of the error signal.

[0079] Then, at the position of the zero-crossing point of the error signal obtained after searching and the position of the feature point obtained by the step 2, take a spline weight as 1, and the others as 0. Finally, according to the estimated ideal ECG signal obtained in step 1 and the obtained spline weight, using cubic smoothing splines to obtain an accurate estimated ideal ECG signal.

[0080] Step 4: Perform subtraction between the pre-processed ECG signal obtained by the step 1 and the estimated ideal ECG signal obtained by the step 3 to obtain a subtraction result, filter the subtraction result by a band pass filter to obtain a filtered result. Based on the filtered result and the QRS range obtained in the step 2, obtain an abnormal potential in the QRS by moving the standard deviation analysis technique.

[0081] Step 5: Perform credibility evaluation on the obtained abnormal potential in the QRS. Using the standard deviation analysis method, evaluate a credibility of the abnormal potential in the QRS obtained in step 4, determine whether the abnormal potential in the QRS obtained in step 4 is credible, and output the evaluation result.

[0082] FIG. 2 is an illustration of a simulation of ECG signal x(i) which includes the ideal ECG signal x.sub.p(i) , a simulation containing the abnormal potential in the QRS AIQP.sub.s(i) to be extracted , and a power frequency interference p(i), baseline shift b(i) and measurement noise n(i), with a sampling rate is 1000 Hz and a data length N=800 for use in the principles of the present invention.

[0083] The step 1, specifically, is as follows: referring to FIG. 3 and FIG. 4, in FIG. 3, 101 refers to remove baseline shift. Use a baseline shift cancellation method to remove the baseline shift. After removing, obtain a signal x.sub.1(i).

[0084] At present, there are many baseline drift elimination methods. Since the baseline drift has no significant effect on the final extraction result, in FIG. 4, x.sub.1(i) is obtained by designing the low-pass filter parameters by using a third-order, 1 Hz Butterworth digital filter, and then processing bidirectional zero phase filtering for x(i).

[0085] In FIG. 3, 102 refers to remove power frequency interference. Based on the obtained signal x.sub.1(i), use a digital power frequency notch filter to obtain the signal x.sub.2(i) after removing the power frequency interference and the baseline drift. In FIG. 4, x.sub.2(i) is the result obtained by using the digital power frequency notch filter.

[0086] The step 2, specifically, is as follows: Refer to FIG. 5 and FIG. 6, wherein FIG. 6 is a partial view of the vicinity of the QRS interval. In FIG. 5, 201 refers to a feature point detection algorithm. Using x.sub.2(i) to process ECG feature point detection to obtain the QRS range (starting position QRS.sub.b, ending position QRS.sub.e) and ECG feature point position p(j), the number of feature points is M, J=1, 2, . . . , M. The ECG feature points at least includes a QRS starting point, a QRS ending point, and Q, R, S waveform peak points.

[0087] In FIG. 5, 202 refers to a low-pass filter. Filter x.sub.2(i) by a higher frequency low-pass filter to obtain x.sub.h(i), f.sub.h refers to a filter frequency of the low-pass filter 202, 100 Hz200Hz. In FIG. 6, x.sub.h(i) is obtained by designing the low-pass filter parameters by using a third-order, f.sub.h=150 Hz Butterworth digital filter, and then processing bidirectional zero phase filtering for x.sub.2(i).

[0088] In FIG. 5, 203 refers to a low-pass filter. Filter x.sub.2(i) by a lower frequency low-pass filter to obtain x.sub.1(i), f.sub.l refers to a filter frequency of the low-pass filter 203, 40 Hzf.sub.l80 Hz. In FIG. 6, x.sub.l(i) is obtained by designing the low-pass filter parameters by using a third-order, f.sub.l=60 Hz Butterworth digital filter, and then processing bidirectional zero phase filtering for X.sub.2(i).

[0089] In FIG. 5, 204 refers to a signal subtraction operation. Use equation (3) to calculate the difference signal x.sub.d(i):

x.sub.d(i)=x.sub.h(i)x.sub.1(i), (3)

[0090] In FIG. 6, x.sub.d(i) is the result after signal subtraction.

[0091] In FIG. 5, 205 refers to a zero-crossing detection and a set set.sub.h calculation. Based on the signal x.sub.d(i), search the difference signal x.sub.d(i) for each ECG feature point time position p(j), j=1, 2, . . . , M, at backward and forward direction respectively to obtain a front and a back first zero crossing point respectively, then obtaining the corresponding time position p.sub.b(i) and pf0) respectively.

[0092] In FIG. 6, in the x.sub.d(i), the dashed line is a zero-valued line, the symbol O represents the point of the time position p(j) corresponding to x.sub.d(i), and the symbols and represent the points of the time position p.sub.b(i) and p.sub.f(i) correspond to x.sub.d(i) respectively, M=4. Since x.sub.d(i) is a time-discrete signal, the actual value corresponding to x.sub.d(i) zero crossing is not necessarily zero.

[0093] For each p(j), the point set set(J) is constructed according to p.sub.b(j) and p.sub.f(j):

set(j)={p.sub.b(j), p.sub.b(j)+1, p.sub.f(j)1, p.sub.f(j)}, (4)

[0094] Based on this, construct Set.sub.h:

set.sub.h={set(1), set(2), . . . , set(M)}, (5)

[0095] In FIG. 5, 206 refers to signal synthesis. Based on set.sub.h, the complex signal x.sub.s(i) is synthesized by the formula (6),

[00006]$\begin{array}{cc}{x}_s\left(i\right)=\{\begin{array}{cc}{x}_h\left(i\right),& i{\mathrm{set}}_h\\ x_l\left(i\right),& i.Math.{\mathrm{set}}_h\end{array},& \left(6\right)\end{array}$

[0096] In FIG. 6, x.sub.s(i) refers to the obtained complex signal.

[0097] In FIG. 5, 207 refers to a low-pass filter.

[0098] Filter x.sub.s(i) to obtain estimated ideal ECG signal x.sub.3(i), f.sub.3 refers to a filter frequency of the low-pass filter 207, 100 Hzf.sub.3200 Hz In FIG. 6, x.sub.3(i) is the estimated ideal ECG signal obtained by designing the low-pass filter parameters by using a third-order, f.sub.3=150 Hz Butterworth digital filter, and then processing bidirectional zero phase filtering for x.sub.s(i).

[0099] The step 3, specifically, is as follows: Refer to FIG. 7 and FIG. 8, wherein FIG. 8 is a partial view of the vicinity of the QRS interval. In FIG. 7, 301 refers to a signal subtraction operation. Use equation (7) to calculate the error signal x.sub.e(i):

x.sub.e(i)=x.sub.2(i)x.sub.3(i), (7)

[0100] In FIG. 8, x.sub.e(i) is the result after signal subtraction.

[0101] In FIG. 7, 302 refers to the calculated spline weight. Use equation (8) to calculate the spline weight w(i):

[00007]$\begin{array}{cc}w\left(i\right)=\{\begin{array}{cc}1,& i.Math..Math.\mathrm{is}.Math..Math.\mathrm{the}.Math..Math.\mathrm{zero}.Math.\text{-}.Math.\mathrm{crossing}.Math..Math.\mathrm{point}.Math..Math.\mathrm{of}.Math..Math.p\left(i\right).Math..Math.\mathrm{or}.Math..Math.x_e\left(i\right)\\ 0,& \mathrm{others}\end{array},& \left(8\right)\end{array}$

[0102] In FIG. 8, w(i) is the spline weight obtained by using equation (7).

[0103] In FIG. 7, 303 refers to the cubic smoothing spline operation. Based on the estimated ideal ECG signal 1 x.sub.3(i) and the spline weight w(i), use the three-order smooth splines to obtain an accurately estimate result for the ideal ECG signal x.sub.4(i).

[0104] In FIG. 8, x.sub.4(i) is the accurately estimate result for the ideal ECG signal obtained.

[0105] The step 4, specifically, is as follows: Refer to FIG. 7, FIG. 9 and FIGS. 8 and 10.

[0106] In FIG. 7, 304 refers to a signal subtraction operation. Use equation (9) to calculate the difference signal e(i):

e(i)=x.sub.2(i)x.sub.4(i), (9)

[0107] In FIG. 7, 305 refers to a band pass filter. Band pass filter e (i) to obtain a signal y(i) which contains an abnormal potential in the QRS to be extracted, f.sub.1 and f.sub.2 refer to a low frequency and a high frequency of the band pass filter, f.sub.1 and f.sub.2 can be selected based on the specific subsequent applications. In FIG. 8, y(i) is obtained by designing the band pass filter parameters by using a fifth-order, f.sub.1=70 Hz, f.sub.2=300 Hz, Butterworth digital filter, and then processing bidirectional zero phase filtering.

[0108] In FIG. 9, 401 refers to the calculation of the moving window variance msd(i) for the signal y(i). Use equation (10) to calculate msd(i):

[00008]$\begin{array}{cc}\mathrm{msd}\left(i\right)=\{\begin{array}{cc}\frac{1}{2.Math.k+1}.Math.{\left(\underset{j=-k}{\overset{k}{.Math.}}.Math..Math.y^2\left(i+j\right)-\frac{1}{2.Math.k+1}.Math.{\left(\underset{j=-k}{\overset{j=k}{.Math.}}.Math..Math.y\left(i+j\right)\right)}^2\right)}^{\frac{1}{2}},& kiN-k\\ \mathrm{msd}\left(k\right),.Math.& i<k.Math.\\ \mathrm{msd}\left(N-k\right),.Math.& i>N-k.Math.\end{array},& \left(10\right)\end{array}$

[0109] Wherein the window length is 2k+1, and k generally ranges from 2ms5ms.

[0110] In FIG. 10, the calculated result of msd(i) is k=2ms.

[0111] In FIG. 9, 402 refers to the calculation reference MSD value ref.sub.msd. The interval of approximately 100 ms before a starting position QRS.sub.b to QRS.sub.b at the QRS is defined as a reference interval. First calculate the mean value ref_msd.sub.mean of MSd(i) in the reference interval and the standard deviation ref_msd.sub.std, then calculate the ref.sub.msd by equation (11):

ref.sub.msd=ref_msd.sub.mean+*ref_msd.sub.std, (11)

[0112] wherein generally choose to be greater than 2, the specific choice can be determined according to the actual situation.

[0113] In FIG. 10, the amplitude value of the horizontal dashed line in msd(i) represents the value of ref.sub.msd, =3.

[0114] In FIG. 9, 403 refers to the determination of the starting position AIQP.sub.b of AIQPs based on ref.sub.msd. The specific method is: starting a forward search from the QRS starting position QRS.sub.b. If a duration of msd (i)>ref.sub.msd is greater than or equal to m, stopping the search, and the position at this time is set as t.sub.b, the starting position of AIQP.sub.b of AIQPs is calculated according to formula (12):

AIQP.sub.b=t.sub.bmk, (12)

[0115] Wherein M is generally 5ms; if the ending position QRS.sub.e of QRS is searched, then AIQP.sub.b=0 and stopping the search.

[0116] In FIG. 10, 404 refers to the determination of whether AIQP.sub.b is searched. If AIQP.sub.b is equal to 0, then exit and return a failure flag, otherwise continue.

[0117] In FIG. 9, 405 refers to the determination of the ending position AIQP.sub.e of AIQPs based on ref.sub.msd. The specific method is: starting a backward search from approximately 50 ms after the QRS ending position QRS.sub.e. If a duration of msd(i)>ref.sub.msd is greater than or equal to 111, stopping the search, and the position at this time is set as t.sub.e, the ending position of AIQP.sub.b of AIQPs is calculated according to formula (13):

AIQP.sub.e=t.sub.e+m+k, (13)

[0118] If the starting position AIQP.sub.b of AIQPs is searched, then AIQP.sub.e=0 and stopping the search.

[0119] In FIG. 9, 406 refers to the determination of whether AIQP.sub.e is searched. If AIQP.sub.e is equal to 0, then exit and return a failure flag, otherwise continue.

[0120] In FIG. 9, 407 refers to the extraction of abnormal potential AIQP(i) of QRS, which is calculated according to formula (14):

[00009]$\begin{array}{cc}\mathrm{AIQP}\left(i\right)=\{\begin{array}{cc}y\left(i\right),& {\mathrm{AIQP}}_{b}i{\mathrm{AIQP}}_e\\ 0,.Math.& \mathrm{others}\end{array},& \left(14\right)\end{array}$

[0121] In FIG. 10, AIQP(i) is the extracted abnormal potential AIQPs in QRS, wherein the two vertical dashed lines represent AIQP.sub.b and AIQP.sub.e respectively.

[0122] The step 5, specifically, is as follows: Refer to FIG. 9 and FIG. 10. In FIG. 9, 408 refers to the standard deviation of the calculated reference interval ref.sub.msd and the standard deviation of QRS region QRS.sub.std. ref.sub.msd is standard deviation of the reference interval y(i), QRS.sub.std is the standard deviation of the y(i) in the interval from the QRS starting position QRS.sub.b to the QRS ending position QRS.sub.e.

[0123] In FIG. 9, 409 is the credibility judgment of the extraction result, and the credibility is calculated according to formula (15).

[00010]$\begin{array}{cc}\mathrm{credibility}=\{\begin{array}{cc}1,& {\mathrm{QRS}}_{\mathrm{std}}*{\mathrm{ref}}_{\mathrm{std}}.Math..Math.\mathrm{or}.Math..Math.{\mathrm{ref}}_{\mathrm{std}}>5.Math..Math..Math..Math.V\\ 0,& \mathrm{others}\end{array},& \left(15\right)\end{array}$

[0124] Wherein >1 the specific selection can be determined according to the actual situation.

[0125] If the credibility is equal to 0, then return a failure flag, otherwise return a success flag and at the same time return the extracted abnormal potential AIQP(i) in the QRS.

[0126] FIG. 11 illustrates a method of generating simulation signal according to the description of the present invention. The simulated heart beat ECG is generated 1 time, 10 times, and 100 times respectively, and then superimpose and average the multiple simulated heartbeat ECGs, for the obtained 1 time heart beat ECG, and the simulated heart beat average ECG of 10 times and 100 times ECGs, extract AIQP(i) by the method of the present invention respectively. The extracted AIQP(i) of the 1 time heart beat ECG, and the simulated heart beat average ECG of 10 times and 100 times ECGs is compared with the simulated AIQP(i) to be extracted, the correlation coefficient and mean square error are: 0.95, 0.1; 0.87, 0.32; 0.87, 0.29. The results show that the method of the present invention only requires a single heart beat ECG to extract AIQP(i). Compared with multiple superimposed averages, the AIQP(i) extracted by single heart beat ECG is more accurate when the measurement noise is small. It can also be seen from FIG. 11 that as the average number of superpositions increases, the interference of the reference interval in step 4 of the present invention becomes smaller and smaller. Therefore, if the signal interference of ECG obtained by detection is relatively great, which causes extraction failure for single heartbeat ECG, then extraction by superimposing and averaging the multiple heartbeat ECG can be processed.

[0127] FIG. 12 is a result of AIQP(i) extraction of a single heart beat ECG of a patient with myocardial infarction by the method of the present invention, and FIG. 13 a result of AIQP(i) extraction of a single heart beat ECG of a heart healthy person by the method of the present invention. The result of AIQP(i) extraction. Comparing FIG. 12 and FIG. 13, it can be found that the amplitude of AIQP(i) in patients with myocardial infarction is significantly larger than that in healthy subjects, and the morphology of both AIQP(i) is also significantly different. These features can be used for early warning of sudden cardiac death.