CN105011931A - Method for detecting wavy boundary of electrocardiogram - Google Patents

Abstract

The invention provides a method for detecting the waveform boundary of an electrocardiogram, comprising the following steps: (1) using a bandpass filter to carry out positive and negative sequence filtering preprocessing on the electrocardiogram; (2) transforming the filtered electrocardiogram through a rainflow model; 3) Perform boundary point detection on the converted ECG; (4) Correct the deviation caused by interference or waveform diversity. The present invention performs forward and reverse order filtering on the electrocardiogram, strengthens the Gibbs effect, and transforms the electrocardiogram through a rainflow model to assist in detecting the boundary of the electrocardiogram.

Description

A Method of Electrocardiogram Waveform Boundary Detection

technical field

The invention relates to image processing technology, in particular to a method for detecting the boundary of electrocardiogram waveform.

Background technique

Electrocardiogram examination is an effective method for diagnosing arrhythmia and myocardial ischemia. This method has the advantages of non-invasive and low cost, and has a large business volume in hospitals. Especially in institutions such as physical examination centers and remote consultation centers, full-time ECG doctors need to interpret a large number of ECGs every day. In order to reduce the workload of doctors, computer-aided ECG automatic classification and recognition systems have received more and more attention in recent years.

The conventional features of the P-QRS-T complex in the electrocardiogram and the indirectly derived interval features are the basis for the doctor's diagnosis. The peak point of various waves is an important feature, but the boundary point is also very important information. Therefore, how to detect the waveform boundary of the electrocardiogram with the assistance of medical equipment has become one of the research trends.

Contents of the invention

In view of this, we provide a method for detecting the edge of the ECG waveform, which can quickly and accurately locate and assist in the detection of peak points.

The method for electrocardiogram waveform boundary detection of the present invention comprises the following steps: (1) using a bandpass filter to carry out positive and negative sequence filtering preprocessing to the electrocardiogram; (2) transforming the filtered electrocardiogram by a rainflow model to obtain peaks, The boundary point of the trough; (3) detecting the boundary point; (4) correcting the deviation caused by interference or waveform diversity.

Preferably, in step (1), the band-pass frequency range of the band-pass filter is 1-20 Hz, and the band-pass ripple is 0.5 dB.

Preferably, in step (2), the rainflow model is a dot matrix of a sinusoidal sequence, which is a model for simulating the flow of rainwater after rain falls on the hillside, and the rainwater falling on the hillside will flow down the hillside flow, and then gather and accumulate in a low-lying place to form the trough boundary point.

Preferably, in step (2), after inversion and negation, the peak boundary point of the electrocardiogram is obtained.

Preferably, in step (4), the step of correcting includes judging whether the position of the boundary point deviates from the real boundary point.

Preferably, in step (4), the specific steps include: selecting the amplitude of the next point and the point difference of this boundary point as a reference, assuming it is diff; continue to move backward until the amplitude of a certain point is different from the amplitude of this boundary point Greater than 3*diff; adjust to this point as the corrected boundary point.

The present invention performs forward and reverse order filtering on the electrocardiogram, strengthens the Gibbs effect, and transforms the electrocardiogram through a rainflow model to assist in detecting the boundary of the electrocardiogram.

Description of drawings

FIG. 1 is a flow chart of the method for detecting the boundary of an electrocardiogram waveform in the present invention.

Figure 2 is an illustration of the Gibbs effect.

Figure 3 is an example diagram of the rainflow model.

Fig. 4 to Fig. 6 are schematic diagrams of embodiment verification in the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, not to limit the present invention.

Please refer to FIG. 1 , which shows a flow chart of the method for detecting the edge of the ECG waveform of the present invention.

In step S101 , the electrocardiogram is preprocessed with forward and reverse order filtering by using a bandpass filter.

The band-pass frequency band of the band-pass filter is selected from 1 to 20Hz, and the pass-band ripple is 0.5dB. Here, the role of the band-pass filter is to filter out the baseline drift and high-frequency noise, and the second is to use the Gibbs effect after the band-pass filter. Because the frequency of QRS wave is about 3~40Hz, and the frequency of P wave and T wave is about 0.7~10Hz, so this bandpass filter can produce Gibbs effect on the boundary of P wave, QRS wave and T wave, and set 0.5 The dB passband ripple makes the Gibbs effect obvious at the boundary and has the least influence on the original waveform. In addition, the band-pass filter uses two filters in positive and negative sequences, which can not only offset the phase shift generated by the filter, but also have obvious Gibbs effect at the end point of the waveform of the positive sequence filter. After the reverse sequence, the Gibbs effect at the beginning point of the waveform is also Obvious. Finally, adjust the data filtered twice in forward and reverse order to the normal order.

See Figure 2 for an example plot of the Gibbs effect. The Gibbs effect is based on the Fourier series expansion of a periodic function with discontinuous points (such as a rectangular pulse), and then selects finite items for synthesis. When more items are selected, the peaks appearing in the synthesized waveform will be closer to the discontinuous point of the original signal. When the number of selected items is large, the peak value tends to a constant, approximately equal to 9% of the total jump value. This phenomenon is called the Gibbs phenomenon.

The Gibbs effect is more obvious after using the bandpass filter. Since the boundary of the ECG waveform is a jump point relative to the peak point, there will be a small fluctuation under the Gibbs effect, which is amplifying the position of the boundary, resulting in a small extreme point (peak or trough), which facilitates the positioning of the boundary in the model below.

In step S102, the filtered electrocardiogram is transformed by the rainflow model to obtain the peak and trough boundary points.

The rainflow model refers to a model for simulating the flow of rainwater after it rains on a hillside. On the hillside, it will flow down the hillside, and then gather and accumulate in a certain low-lying place to form a wave trough; and after inversion and negation, the wave peak can be obtained.

See Figure 3 for an example plot of a rainflow model. The rainflow model is a bitmap of a sinusoidal sequence, assuming that the bitmap is a mountain, and point A is the peak point, and point B is the bottom point. Suppose it is about to rain in the sky, as shown in the figure 'o', and the raindrops are uniform, that is, each point of the following sine sequence will have the same amount of rain. When the rain falls on this mountain, assuming that the hillside will not absorb the rain, it will flow down and continue to gather. For example, when it reaches the bottom of the valley, such as point B, the rain will gather here. In addition, when there is a protrusion at a certain point of the peak and the bottom of the valley, the inertia of the rain flow will be considered. If the width of the protrusion is less than a certain threshold TH (due to the smoothing effect of the filter, the general burr-type protrusions will be filtered out, Therefore, generally there will be no protrusions on hillsides with obvious waveforms, so the threshold value TH=2) can continue to flow down, otherwise, the accumulated raindrops above will gather there.

Using the above rainflow model to process the filtered data, it will be found that the raindrops converge at a series of valley bottoms. The points corresponding to the bottom of this series will accumulate a certain amount of rainfall (the number of raindrops), which we set as Srain. The trough rain point sequence is more effective for identifying the boundary of the positive peak. For example, assuming that the shape of the QRS complex is Rs type, since the R wave is a positive crest, then the non-zero point (rainfall convergence point) of the first Srain from the R wave position forward is the starting point of the R wave .

In order to locate the boundary point of the inverted waveform conveniently, the filtered data is reversed at the same time, and the rainflow model is also processed to obtain the trough rainpoint sequence SrainR. Similar to the positioning of the positive waveform boundary, the boundary of the inverted waveform such as S wave can be positioned by SrainR.

Of course, the detection of the peak point also has a certain effect, because the rainfall convergence point itself is the extremum point of a certain waveform.

In step S103, boundary point detection is performed.

For example, assuming that the shape of the QRS complex is Rs type, since the R wave is a positive crest, then the non-zero point (rainfall convergence point) of the first Srain from the position of the R wave forward is the starting point of the R wave . Another example is the positive T wave, which generally produces a converging point at its boundary point.

In order to locate the boundary point of the inverted waveform conveniently, the filtered data is reversed at the same time, and the rainflow model is also processed to obtain the trough rainpoint sequence SrainR. Similar to the positioning of the positive waveform boundary, the boundary of the inverted waveform such as Q wave can be positioned by SrainR.

In addition, generally because the range of the T wave is relatively wide, the cumulative rainfall at the boundary point of the T wave is generally the largest, so the position of the T wave can also be confirmed separately. Of course, with the help of the known peak point of the R wave, it can be more Accurate and fast identification. For example, a window is opened behind the R wave position to include the range of the T wave. Then search for the maximum value of Srain and SrainR in this window, and select the largest of the two, then the peak value of the T wave is generally determined, and whether the T wave is positive or inverted is also determined. For example, if the maximum value is in SrainR, then the T wave is positive, and then search for the first non-zero point forward in Srain, which is the starting point of the T wave; search for the first non-zero point backward, then it is the T wave end.

This method can also detect the peak point. For example, for the peak of the positive T wave, we can use the rain gathering point of SrainR to detect; for the peak of the inverted T wave, we can use the rain gathering point of Srain for detection. By analogy, other peaks are detected using or assisting.

In step S104, the deviation caused by interference or waveform diversity is corrected.

Through the third step, a set of values that may be the peak point has been passed. Because of the variety of waveforms and noise, especially the P wave with a relatively small waveform range and amplitude, it cannot be guaranteed that it must be the boundary point we want, so the last step It is to correct and confirm the possible boundary points above.

The correction principle is to judge whether the position of the boundary point deviates from the real boundary point.

The specific methods include: (1) Select the difference between the amplitude of the next point and this boundary point as a reference, assuming it is diff; continue to move backward until the amplitude difference between the amplitude of a certain point and this boundary point is greater than 3*diff; adjust to this point as the corrected boundary point.

Or (2) Use other clinical experience to correct, such as visually checking whether there is a serious deviation of the boundary point.

Experimental Verification Example 1

Please refer to FIG. 4 , taking the arrhythmia database provided by the Massachusetts Institute of Technology (MIT-BIH) as an example to verify the effectiveness of the method. In the following, taking sampling points 1900-2300 of the first lead signal recorded in No. 101 as an example, and applying this method for detection, the results are shown in Fig. 4 respectively. The dotted horizontal line in the figure is the original signal, the dots on the solid line are the data after bandpass filtering, the solid line is the data after the filtered data has been processed by the rainflow model, that is, Srain, and the dotted line is the inverted data after filtering and then processed by the rainflow model The latter data is SrainR. It can be seen from the figure that there are rainfall accumulation points at the boundary points and peak points of the waveform. For positive waveforms, such as P wave, R wave, and T wave, refer to the solid line Srain at this time, and it can be seen that their boundaries are just at the point of rainfall convergence. Another example is their peak point. At this time, referring to the dotted line SrainR, it can be determined that the peak point is the place where rainfall converges. And it can be seen that the cumulative rainfall of the wider T wave is the largest.

Example 2

Please refer to Figure 5, which shows the Chinese Cardiovascular Disease Database (Chinese Cardiovascular DiseaseDatabase, CCDD) No. 2 record: the partial data of the V2 lead recorded, and the rendering after using this method. Originally, the turning points of the S wave and T wave were obvious, but their respective boundaries were relatively blurred. After this method, the end point of the S wave has an obvious positive extreme point. Similarly, the starting point of the T wave also has a Obvious negative extremum point. Taking the T wave as an example, the left triangle in the figure represents the start point, the regular triangle represents the peak point, and the right triangle represents the end point.

Example 3

Please refer to Figure 6, which shows the partial data of the aVR lead recorded in CCDD database No. 9 after using this method. The P-wave and T-wave borders were located. This example is an inverted P wave and an inverted T wave. The left triangle in the figure represents the start point, the regular triangle represents the peak point, and the right triangle represents the end point.

Beneficial effect

(1) It is possible to detect the P-QRS-T boundary, because the QRS complex peak is generally easy to identify, and the boundary is more rapid and accurate by using this method. In addition, broad T waves are most effective for this method.

(2) It also has a certain auxiliary effect on the peak detection of the waveform.

(3) Through positive and negative sequence bandpass filtering, not only can the phase shift generated by filtering be offset, but also the Gibbs effect at the end point of the waveform of the positive sequence filter is obvious. After the reverse sequence, the Gibbs effect at the beginning point of the waveform is also obvious . Finally, adjust the data filtered twice in forward and reverse order to the normal order.

(4) Other documents try to eliminate the Gibbs effect to prevent the leakage effect of the filter, but this patent uses the Gibbs effect as a method of boundary detection or peak detection.

(5) The boundary points of wave crests and troughs are generated through the rainflow model, and detected by inversion and negation.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, and these improvements and modifications should also be considered Be the protection scope of the present invention.

Claims (6)

1. a method for electrocardiographic wave border detection, is characterized in that, comprises the steps:

(1) band filter is used to carry out positive-negative sequence filter preprocessing to electrocardiogram;

(2) by rain flow model, filtered electrocardiogram is converted, obtain the boundary point of crest, trough;

(3) to described boundary points detection;

(4) deviation that interference or waveform multiformity cause is corrected.

2. the method for electrocardiographic wave border detection as claimed in claim 1, it is characterized in that, in step (1), the band passband section of described band filter is 1 ~ 20Hz, and Ripple is 0.5dB.

3. the method for electrocardiographic wave border detection as claimed in claim 1, it is characterized in that, in step (2), described rain flow model, be the dot chart of a width sinusoidal sequence, be the model of the situation of rainwater flowing after raining in simulation hillside, rainwater drops on hillside and can flow to lower along hillside, then converge accumulation at a certain low-lying place, form trough boundary point.

4. the method for electrocardiographic wave border detection as claimed in claim 3, is characterized in that, in step (2), after being inverted negate, tries to achieve Electrocardiographic crest boundary point.

5. the method for electrocardiographic wave border detection as claimed in claim 1, it is characterized in that, in step (4), the step of correction comprises, and judges that whether the position of described boundary point has with real border point and departs from.

6. the method for electrocardiographic wave border detection as claimed in claim 5, it is characterized in that, in step (4), concrete steps comprise:

Amplitude and this boundary point difference of choosing down any are reference, are assumed to be diff;

Continue mobile backward, until the amplitude of certain point and the difference in magnitude of this boundary point are greater than 3*diff;

Adjust to this position as the boundary point after correction.