KR20060117545A - Method for feature extraction of ecg signal - Google Patents

Abstract

A method for extracting property of an electrocardiogram signal is provided to extract property about a heart rate, a QRS interval, a PR interval, a QT interval and a T wave on the basis of a starting point, an end point, and a peak point of each P, QRS, and T wave. A method for extracting property of an electrocardiogram signal includes the steps of: extracting the maximum value within a predetermined period as a peak point of an R wave; extracting two points meeting a baseline of the R wave as a starting point and an end point of a QRS complex; extracting the maximum voltage point as a peak point of P wave in a left part of the QRS complex, preceding the starting point of the QRS complex on the basis of time; and extracting the value having a bigger difference with a baseline between the maximum value and the minimum value in a right part on the basis of the end point of the QRS complex as a peak point of the T wave.

Description

Characteristic extraction method of ECG signal {METHOD FOR FEATURE EXTRACTION OF ECG SIGNAL}

1 illustrates the main characteristic waveforms of a typical ECG signal.

2 is an overall operation flow chart for performing the feature extraction method of the ECG signal according to the present invention.

3 is a waveform diagram showing major interval values of an ECG signal.

Figure 4 is a comparison of the present invention and conventional feature extraction for the steady wave.

Figure 5 is a comparison between the present invention and the conventional feature extraction for the left angle block.

6 is a table showing the measurement results and errors of the present invention and the conventional method.

The present invention relates to a feature extraction method of an electrocardiogram signal, and more particularly, to a feature extraction method of an electrocardiogram signal that can be used to extract the features of the electrocardiogram signal simply by using the morphological features of the electrocardiogram signal.

       Electrocardiogram (ECG) is a measure of the electrical activity of the heart. A diagram of electrical activity can be obtained by attaching electrodes to specific sites.

The heartbeat, the source of the ECG signal, is impaired by the sinus node located in the right atrium, first delayed in the atrioventricular node by deploarization of the right atrium and left atrium. Activate the ventricle afterwards.

The fastest septum and thinner right ventricle are activated before the thicker left ventricle. Depolarized waves, which are delivered to Purkinje Fiber, spread out from the endocardium to the outer pericardium, like the wavefront in the myocardium, causing ventricular contractions. Normally electrical stimulation is conducted through the heart, so the heart contracts about 60-100 times per minute. Each contraction is represented by one heart rate.

The ECG can be detected through a bipolar lead recording the potential difference between the two sites and a unipolar lead recording the potential of the site to which the electrode is attached. Standard Limb Lead, Unipolar Limb Lead for Unipolar Induction, and Precordial Lead for Unipolar Induction.

The standard induction consists of induction I (consisting of the right arm designated as the negative electrode and the left arm designated as the anode), induction II (consisting of the left arm designated as the anode and the positive electrode), and induction III (consisting of the left arm designated as the negative electrode and the left foot designated as the anode There is).

Induced limb induction of right hand amplification limb induction (right arm electrode is anode, left arm and left foot are anode), left hand amplification limb induction (left arm is anode, right arm and left leg is cathode), Left foot amplification limb induction (the electrode on the left foot is the anode, and the electrodes on the right and left arms are composed of the cathode).

 Chest induction also includes V1 (4th intercostal sternal chance), V2 (4th intercostal sternal left margin), V3 (midpoint of V1, V2), V4 (where the left clavicle centerline meets at the height of the 5th intercostal), V5 (Where V4 meets the horizontal extension line and the left full ligament line), and V6 (where V4 meets the horizontal extension line and the left central axillary line).

The standard induction and limb induction record the electrocardiogram of the frontal plane and the chest induction record the electrocardiogram of the horizontal plane.

The electrical activity of the heart is largely divided into atrial depolarization, ventricular depolarization, and ventricular repolarization, and each of these steps is reflected in the form of several waves called P, QRS, and T waves, as shown in FIG.

These waves must have a standard shape to ensure that the heart's electrical activity is normal. To determine whether it is in standard form, we need to check whether the characteristics of each wave, such as the duration of each wave, the interval of each wave, the amplitude of each wave, and the kurtosis, fall within the normal range.

The measured ECG waveforms merely represent the potential difference of the heart over time. Therefore, there is a need to extract features for distinguishing between normal and abnormal waveforms that can be given from the original ECG waveform.

To get this feature, we first need to find the start, end and vertex of P, QRS and T waves respectively. Based on this information, you can find out the interval of each wave, how long each wave is held, and the amplitude of each wave.

Each feature has a normal range obtained through statistical methods based on clinical data. Thus, if these features are not within the normal range, it suggests that you are more likely to have heart disease.

That is, on average, if there are many waveforms that do not fall within this normal range, it can be judged that there is a heart disease. Therefore, the feature extraction of ECG signal is an essential part for diagnosing heart disease. .

A representative method is to compare the normal ECG waveform with the abnormal ECG waveform based on the correlation by using the cross correlation function and the auto correlation function (Alias Bin Ramli and Putri Aidawati Ahmad, "Correlation analysis for abnormal ECG signal features extraction). ", Proc. Of IEEE 4th National Conf. On Telecom. Technology, Shah Alam Malaysia, pp. 232-237, 2003).

Other approaches include the "So and Chan" method and the "Pan and Tompkins" method, which derive mathematical formulas using the morphological features of ECG signals (KFTan, KL Chan and K.Choi). , "Detection of the QRS complex, P wave and T wave in electrocardiogram", Proc. Of IEEE Advances in Medical Signal and Information Processing, September pp. 41-47, 2000).

In addition, there is a method of extracting ECG waveform features using Wavelet transformation (Ming-Yao Yang, Wei-Chin Hu and Liang-Yu Shyu, "ECG events detection and classification using wavelet and neural networks", Proc. Of the 19th IEEE Int'l Conf. On Engineering in Medicine and Biology, Oct. 30-Nov. 2, Chicago, IL. USA, pp. 289-292, 1997).

These conventional feature extraction methods use a statistical method or a mathematical formula derivation method, which requires a complicated operation.

As an example, the “So and Chan” QRS detection method finds the starting point of the QRS complex by comparing the slope of the ECG waveform with the threshold of the slope. First, the slope value of the ECG signal at time n is obtained by equation (1).

slope (n) = -2X (n-2)-X (n-1) + X (n + 1) + 2X (n + 2) ... (1)

In addition, the threshold value of the slope is obtained by equation (2).

slope _thresh = (thresh _param / 16) * maxi ...... (2)

Two consecutive ECG data can find the starting point of the QRS complex when slope (n)> slope _thresh is satisfied.

Here, the variable thresh _param can be set to 2,4,8,16. After finding the starting point of the QRS complex, the maximum value (maxi) is regarded as the peak of the R wave. Here, the value of maxi is updated by equations (3) and (4).

maxi = ((first _max -maxi) / filter _param ) + maxi ...... (3)

first _max = heightofRpoint-heightofQRSonset ...... (4)

This equation serves as a filter to smooth out sudden fluctuations in value. Here, the filter _param may be set to 2, 4, 8, 16. The initial maxi value is the maximum slope of the first 250 data points in the ECG file.

As described above, the conventional feature extraction method uses a statistical method or a mathematical formula derivation method, which requires a complicated operation, and thus, a preprocessing time for feature extraction is increased.

The present invention has been made in view of the above, and an object of the present invention is to extract a feature using a morphological feature of an electrocardiogram signal, so that the feature can be extracted without a complicated calculation process, and thus it is possible to extract a feature quickly, as compared with the conventional art. To provide a feature extraction method of the ECG signal capable of accurate feature extraction.

In the ECG signal feature extraction method according to the present invention, a value corresponding to the maximum value is extracted at the peak of the R wave at a predetermined time interval, and the two points where the R wave meets the baseline are each of the QRS complex. Extract to start and end points,

The point having the largest voltage value at the left side of the start point of the QRS complex which is temporally preceding the start point of the QRS complex is extracted as the peak of the P wave, and the baseline and It is characterized by extracting a value having a large difference as a vertex of the T wave.

The present invention also does not recognize the R wave if the magnitude of the extracted R wave does not exceed the threshold value by applying a certain number of threshold values based on the base line, and meets the base line at the left and right sides of the P-wave peak. The corresponding portions are extracted as starting and ending points of the P wave, respectively, and the portions corresponding to the baseline at the left and right sides of the T wave are extracted as starting and ending points of the T wave, respectively. .

In addition, the present invention extracts the points corresponding to the lowest value of the voltage on the left and right of the R wave vertex, respectively, as the peaks of the Q wave and the S wave, respectively, and from the left to the peak of the Q wave First, the point that meets the baseline is extracted as the starting point of the Q wave, and the point that first meets the baseline from the right side based on the vertex of the S wave is extracted as the end point of the S wave.

Hereinafter, the present invention will be described in more detail with reference to the accompanying drawings. However, the following examples are merely to illustrate the present invention is not limited to the contents of the present invention.

Although the present invention is not shown, an input unit for receiving an electrocardiogram waveform, a feature extractor for extracting a feature from the electrocardiogram waveform input through the input unit, and an output unit for outputting the result extracted by the feature extractor The feature extraction unit first finds the peak of the R wave using the morphological features of the ECG signal, finds the start point and the end point of the QRS group, and then starts, ends and vertices the P wave and the T wave. Find it.

Through this, the instantaneous heart rate (RR interval is 60 to 100 times per minute), PR interval (the time from the P wave to the start of the QRS complex, most of which is AV junction (AV node + His fiber), the normal range is 0.12 ~ 0.20sec), the QRS duration (Duration), the time for depolarization to persist, the time from the start of the QRS complex to the end of the T wave, and the depolarization and repolarization of the atrium The interval between the QT interval, the baseline, and the peak of the S-wave, which is time, is output as a feature for diagnosing heart disease.

The present invention implemented in such a system looks with the flow chart of FIG.

First, the vertex of the R wave is extracted for feature extraction from the ECG waveform input through the input unit (S110).

Electrocardiograms produce P, QRS, and T waves once each time the heart circulates, and have a continuous sequence of similar waveforms. Therefore, knowing this period can distinguish each waveform.

Normal ECG waveforms typically have an RR interval between 0.6-1 seconds. In the case of normal ECG, the peak of the R wave has the largest voltage value in one ECG waveform. Therefore, if you find the value that corresponds to the maximum value every 0.8 seconds, that value corresponds to the peak of the R wave. In the case of East and West, the RR interval is more than 1 second.

If the peak of the R wave is found every 0.8 seconds, there is a section where there is no R wave in the middle of the East and West. In this case, the R wave is not recognized unless the magnitude of the R wave extracted by applying a certain number of threshold values based on the baseline does not exceed this threshold value.

Thereafter, Q waves and S waves are extracted. The Q and S waves are located to the left and right of the R wave, respectively, with the left side of the R wave leading in time compared to the right side of the R wave.

 In the case of Q and S waves, the vertex is drawn downward. Therefore, the peak of the Q wave and the peak of the S wave are found to be the lowest voltage points on the left and right sides of the peak of the R wave.

The starting point of the Q wave corresponds to the first encounter with the baseline on the left side based on the found peak of the Q wave. The end point of the S wave corresponds to the point where the baseline is first seen from the right side based on the extracted S wave vertex.

Depending on the signal, there may be no Q wave and no S wave. In this case, only the R wave is considered, and two points where the R wave meets the baseline are extracted as the start point and the end point of the QRS complex, respectively. If the Q wave and the S wave are found, the distance from the start point of the Q wave to the end point of the S wave can be obtained, which corresponds to the QRS period (S120).

In general, if the QRS period is 0.12 seconds or more, it abnormally corresponds to a right bundle block (RBBB) or a left bundle block (LBBB).

Next, P and T waves are extracted. P and T waves are located on the left and right sides of the QRS complex, respectively. P-wave is the upward curve and the peak of the P-wave is found when the point with the largest voltage is found to the left of the starting point of the QRS complex. The starting and ending points of the P wave correspond to the baseline at the left and right sides of the P wave vertex, respectively.

In the case of the T-wave, the normal waveform takes the form of a curve where the vertex is upwards, or in the case of LBBB, a curve where the vertex is downward. Therefore, the peak of the T-wave is obtained from the maximum value and the minimum value on the right side based on the end point of the QRS complex, and the difference between the baseline and the larger one corresponds to the vertex (S130).

Based on the extracted vertices, the starting and ending points of the T-wave can be found by finding two points that meet the baseline on the left and the right, respectively.

When the P wave and the T wave are found, as shown in Fig. 3, the PR interval can be obtained from the start point of the P wave to the start point of the QRS complex, and the QT interval from the start point of the QRS complex to the end point of the T wave. A value can be obtained (S140).

The normal range of PR interval is 0.12-0.20 seconds, and the normal range of QT interval is 0.35-0.45 seconds.

Next, look at the experimental results of the present invention.

To evaluate the efficacy of the present invention, we performed feature extraction on 100 data on the normal ECG waveform of MIT-BIH's arrhythmia database and four diseases of left angle block, right angle block, tachycardia and east-west vein. The method was compared with the results.

For 20 signals with right angle block among 100 data, the "So and Chan" method could not find the point satisfying slope (n)> slope _thresh and could not find the starting point of QRS complex. Failed. On the other hand, the present invention was capable of feature extraction for all 100 data.

Figure 4 shows the characteristic points (starting point, peak, end point of the P, QRS, T) of the ECG waveform extracted using the MIT-DB 101 data (normal waveform) using the "So and Chan" method and the method proposed in the present invention. It is a figure compared.

 The values obtained in the method of "So and Chan" are represented by triangles and the points extracted by the present invention are indicated by asterisks. The interval of each wave according to the extracted points in both methods is indicated by the solid line on the top, and the interval of the correct wave is indicated by the dotted line on the bottom. The interval indicated by the dotted line is the reference interval through expert waveform analysis. That is, the performance of the feature extraction method can be evaluated to the extent that the interval indicated by the dotted line and the position and size coincide.

Similarly, FIG. 5 shows each feature point extracted from MIT-DB data 214 (left block) using the method of “So and Chan” and the method proposed in the present invention, the interval of each wave, and the reference interval that is a standard.

FIG. 6 shows error information of feature values extracted by two methods based on a reference interval for a normal signal and an abnormal signal having a disease, and a unit is sec.

As shown in the table of FIG. 6, the present invention had a large error of about 0.03 seconds compared to the conventional method only in the T-duration of the normal waveform, and showed excellent results with respect to the remaining features. In particular, the present invention showed a very good result compared to the conventional method in the P-duration.

As described above, although described with reference to a preferred embodiment of the present invention, those skilled in the art will be able to vary the invention without departing from the spirit and scope of the invention described in the claims below. Can be modified or changed.

As described above, the present invention finds the starting point, the end point, and the vertex of P, QRS, and T waves by using the morphological features of the ECG signal, and based on this, heart rate, QRS interval, PR interval, QT interval, and T wave. By extracting the features of the type of, it is possible to extract the features more accurately in a shorter operation time than in the conventional method which derives the mathematical formula. Therefore, it is expected to improve the diagnosis rate in the diagnosis of heart disease based on ECG signals. It becomes possible.

In addition, the present invention can be used for real-time cardiac disease diagnosis through the transmission of ECG data wirelessly to the doctor using the mobile healthcare equipment in the ubiquitous environment, by using the mobile healthcare equipment to capture the early symptoms of the disease It will be applicable to home healthcare systems that can treat diseases early, and to implement new concepts of information devices such as bio-PDA and lab-on-a-chip portable diagnostic devices, which are human-mounted micro bio information terminal technologies combined with IT and BT. Would be directly applicable.

Claims (6)

Extracting the value corresponding to the maximum value at a predetermined time interval as the peak of the R wave, extracting two points where the R wave meets the baseline as the start point and the end point of the QRS complex, respectively, and temporally preceding the start point of the QRS complex. The point with the largest voltage value at the left side of the start point of the complex is extracted as the peak of P wave, and the difference between the baseline of the maximum value and the minimum value at the right side of the QRS complex is the highest point as the T wave peak. Characteristic extraction method of the ECG signal, characterized in that for extracting. The method of claim 1, wherein the R-wave extracted by applying a predetermined number of thresholds based on a baseline does not be recognized as an R-wave if the magnitude of the R-wave extracted does not exceed the threshold. The method of claim 1, wherein a portion corresponding to a point meeting the baseline at left and right sides of the P wave is extracted as a start point and an end point of the P wave. The method of claim 1, wherein a portion corresponding to a point meeting the baseline at left and right sides of the T wave is extracted as a start point and an end point of the T wave. The ECG signal extraction method of claim 1, wherein the points having the lowest voltage on the left and right sides of the R wave are respectively extracted as the peaks of the Q wave and the S wave. The method according to claim 5, wherein the point where the baseline meets the baseline for the first time from the left side based on the vertex of the Q wave is extracted as the starting point of the Q wave, and the point where the baseline meets the baseline for the first time from the right side based on the vertex of the S wave. Characteristic extraction method of the ECG signal, characterized in that the extraction to the end of the wave.

Images