TWI696191B - Algorithm of qrs detection capable of reducing noise effects - Google Patents

Abstract

An algorithm of QRS detection capable of reducing noise effects includes steps of adjustment, reading, pre-processing, transformation, calculation and selection. The adjustment step is to adjust panes before reading an original signal in the reading step in order to capture the original signal via the panes. The pre-processing step is to remove baseline drift and the transformation step is to execute the wavelet transformation in order to obtain high and low frequency signals of all scales. Further, the calculation step is to calculate positive and negative threshold values. Finally, the selection step is to select a scale for analyzing based on the amount of the noise. Therefore, the algorithm can judge the amount of the noise in the signal effectively and reduce effects caused by the noise via selecting the scale with fewer noise, thereby weakening the noise effects affecting the QRS detection, increasing the judgment accuracy greatly, reducing the calculation to speedily assess and diagnose diseases and decrease the consumption of human resource.

Description

QRS wave detection algorithm to reduce the influence of noise

The invention relates to a calculation method, especially a QRS wave detection algorithm for reducing the influence of noise.

Check, as the average life span of Chinese people increases year by year, and the type of disease also changes from acute to chronic, for example, early detection of cancer, heart disease, cerebrovascular disease, etc. can provide 更 different methods as treatment options, In order to control the disease and improve the cure rate, and the use of electrocardiogram is widely known, it is used to measure and monitor the electrocardiogram signal of the human body to provide good information, allowing medical personnel or researchers to use electrocardiography The overall potential change of the heart presented by the number, so as to clearly understand the heart condition of the patient, and detect the potentially dangerous patient, or help monitor the patient with a defective heart, so as to reduce the risk of heart attack in the patient, while Before judging and classifying ECG signals, medical staff or researchers must find the QRS waveforms among a large number of ECG signals, and then use the position, size, and shape of P, Q, R, S, and T waves 狀,來 as a basis for diagnosing heart disease.

However, the ECG signal is very small, which may be due to the amount of conductive gel, the degree of fit of the patch to the skin, the wearer's limb shaking due to speaking or nodding, or the wire is shaken to cause electrical conduction. Interference, or incomplete conduction, etc., and then generate noise on the ECG, and the vibration of the environment or equipment will also affect the accuracy of ECG signal measurement. Therefore, in order to improve the accuracy of ECG signal analysis and judgment, It usually takes a long time to interpret, but a large number of ECG signals not only cause a long time to analyze and interpret the operation, but also need to be interpreted by professional medical personnel or researchers, The consumption of resources is quite huge. Therefore, how to reduce the influence of noise on the interpretation of QRS waves in ECG signals, so as to quickly and accurately judge QRS waves, and allow professional medical staff or researchers to quickly evaluate and diagnose patients The disease, which effectively reduces the consumption of human resources and improves the speed of overall analysis, has become an important issue that cannot be ignored by all walks of life.

Therefore, the purpose of the present invention is to provide a QRS wave detection algorithm that reduces the influence of noise, which can effectively reduce the influence of noise, so as to improve the accuracy of judgment, reduce the amount of calculation and increase the speed of analysis, so as to achieve the disease Quick assessment and diagnosis, while effectively reducing the consumption of human resources.

Therefore, the QRS wave detection algorithm of the present invention to reduce the influence of noise includes adjustment, reading, preprocessing, conversion, calculation, and selection steps; among which, the acquisition pane is adjusted in the adjustment step, and the In the reading step, the original signal to be analyzed is loaded, and then the baseline drift signal is filtered out in the pre-processing step, which is further used in the conversion step to obtain low-frequency signals, high-frequency signals of all scales, and positive and negative characteristic extreme points through wavelet transformation Then, in the calculation step, the positive and negative threshold values are calculated, and finally in the selection step, the positive and negative characteristic extreme points that are greater than twice the positive and negative threshold values in the high-frequency signal are deleted, and brought into the first through the selection formula The remaining positive and negative characteristic extreme points of the third scale and the fourth scale are calculated. When the obtained value is greater than one, the high frequency signal of the fourth scale is used for analysis; otherwise, the high frequency signal of the third scale is used for analysis; Therefore, using the characteristics of multiple analysis in wavelet transformation and the selection formula to determine the amount of noise in the scale, different scales are used for detection and analysis according to different situations, so as to effectively reduce the impact of noise on QRS wave detection , And thus greatly improve the accuracy of judgment, reduce the overall calculation and increase the speed of analysis, in order to achieve rapid assessment and diagnosis of the disease, while effectively reducing the consumption of human resources.

The foregoing and other technical contents, features and effects of the present invention will be clearly understood in the following detailed description with reference to the preferred embodiments of the drawings.

1 and 2, a first preferred embodiment of the present invention, a QRS wave detection algorithm 3 that reduces the influence of noise, includes an adjustment step 31, a reading step 32, and a pre-processing step 33, A conversion step 34, a calculation step 35, a selection step 36, etc.; referring to FIGS. 3 and 4, first in the adjustment step 31, a processor 4 is prepared, and the processor 4 is used to extract through the wavelet transformation formula Take the adjustment operation of the pane, and the wavelet transformation formula is as follows:

The coefficient a in the mother wavelet Ψ(tb/a) can make the mother wavelet expand and contract, and the coefficient b can make the mother wavelet translate. Through the adjustment of the coefficients a and b, the mother wavelet can be scaled and translated according to the input parameters. The acquisition pane can be adjusted freely, and multiple sampling points overlap between any two acquisition panes. The overlapping of different numbers of sampling points will also affect the accuracy of subsequent detection. In this embodiment, any two acquisition windows are used. For example, 150 sampling points are overlapped between frames and 4096 sampling points for each acquisition pane. In the reading step 32, the original signal S to be analyzed is loaded, refer to FIG. 4 and FIG. 5-1, and then A filter 5 is provided in the pre-processing step 33, and in this embodiment, the filter 5 is a 200 ms median filter 5 and a 600 ms median filter 5 as an example. First, the 200 ms median The filter 5 performs a preliminary filtering process on the original signal S to filter out the QRS wave and the P wave, and then performs a filtering process on the original signal S filtered by the 200 ms median filter 5 through the 600 ms median filter 5 , To filter out the T wave, and obtain the baseline drift signal that does not include the QRS wave, P wave, and T wave, and then subtract the baseline drift signal from the original signal S to obtain the corrected original signal S, as shown in Figure 5-2 As shown, in this embodiment, the pre-processing step 32 does not include the action of eliminating high-frequency noise, so as to prevent the original signal S from being removed along with the QRS waveform of higher frequency during the process of eliminating high-frequency noise Affect the characteristic value.

Continuing from the foregoing, referring to FIGS. 6 and 7, and then in the conversion step 33, the mother wavelet of the biorthogonal spline wavelet used is the following formula:

The high-frequency signal D and low-frequency signal A of all scales can be obtained by the above formula. The number of conversion scales can be adjusted according to the actual needs and conditions. In this embodiment, wavelet conversion is used to obtain four scales of high-frequency signals D1 and D2. , D3, D4 and low-frequency signals A1, A2, A3, A4 as an example, that is, as shown in Figure 8 and Figure 9, you can also use wavelet conversion equivalent high-pass filter and low-pass filter for conversion processing, Referring to FIG. 10, the waveforms in the high-frequency signals D1, D2, D3, and D4 are converted into positive and negative characteristic extremum points, and then the capture panes of the high-frequency signals D1, D2, D3, and D4 are extracted. The first 50 sampling points and the last 50 sampling points are removed and not included in the detection operation.

Continuing the foregoing, in the calculation step 35, the high-frequency signal of the selected scale is first divided into four segments, and the positive maximum value and the negative minimum value are respectively taken in the four segment signals, and then the four maximum values Take the average to get the average maximum value, and average the four minimum values to get the average minimum value, then take a value of one quarter of the average maximum value to get a positive threshold, and take another quarter of the average minimum value A value of one to get a negative threshold.

Continuing the foregoing, in the selection step 36, the third-scale high-frequency signal D3 and the fourth-scale high-frequency signal D4 are respectively positive and negative that are greater than twice the positive threshold value and the negative threshold value The characteristic extreme points are removed, and the remaining positive and negative characteristic extreme points of the high frequency signal D3 and the remaining positive and negative characteristic extreme points of the high frequency signal D4 are calculated by a selection formula, so that the noise ratio is SNR , Compare the noise ratio SNR(DP3, DP4) of the third-scale high-frequency signal D3 and the fourth-scale high-frequency signal D4, and the selection formula is as follows:

Divide the sum of the absolute values of the third scale by the sum of the absolute values of the fourth scale. When the SNR (DP3, DP4) is less than one, it means that the third scale high frequency signal D3 is used. For analysis, on the contrary, when the SNR (DP3, DP4) is greater than one, which means that there is more noise, the fourth-scale high-frequency signal D4 is used for analysis. Referring to FIG. 11, the high-frequency signal D3 is compared with the high-frequency signal D4 has a lot of noise, so the fourth-scale high-frequency signal D4 is used for analysis. Generally speaking, if there is no noise, it is better to use the third-scale high-frequency signal D3 for analysis. The first-scale and second-scale high-frequency signals D1, D2 have more high-frequency components, and there will still be a lot of noise interference, while the fourth-scale high-frequency signal D4 has lower high-frequency components and low-frequency components. Obviously, it will be disturbed when some T waves or P waves are large, so this third scale is a scale suitable for use in most cases.

Continuing the foregoing, after completing the aforementioned adjustments 31, reading 32, preprocessing 33, conversion 34, calculation 35, and selection 36, etc., most of the noise can be filtered out, so the use of multiple analysis in wavelet conversion Characteristics, and use different scales to detect QRS waves for different situations, thereby reducing interference during detection and increasing detection accuracy, and at the same time only processing the original signal S to eliminate baseline drift, not high Frequency filtering, in order to avoid affecting the most obvious QRS wave in the original signal S, not only effectively reduce the impact of noise on the interpretation of the QRS wave, but also quickly and accurately locate the QRS wave, thereby greatly improving the accuracy of judgment and reducing the overall calculation Quantity and effectively improve the analysis speed, in order to achieve rapid assessment and diagnosis of the disease, and effectively reduce the consumption of human resources.

12 and 13, the second preferred embodiment of the present invention still includes the steps and practices described in the previous embodiment, and this embodiment is particularly characterized in that: after the selection step 36, an additional calculation is included in sequence. Step 37, a judging step 38, a excluding step 39, etc.; refer to FIG. 14, in the estimating step 37, first mark a positive characteristic extreme point exceeding the positive threshold, and mark a positive characteristic pole The negative characteristic extreme value points whose value points are opposite in polarity and exceed the negative threshold value to form a characteristic extreme value pair, and the distance between the positive and negative characteristic extreme value points does not exceed a preset value, which is preset in this embodiment The value is 45 sampling points. When the distance exceeds 45 sampling points, it means that the wave shape is smoother, which is not in line with the high and sharp features of the QRS wave shape. Refer to Figure 15, and then coordinate the positive feature extreme points (X2, Y2) , The coordinates of the negative characteristic extreme point (X1, Y1), and the value of the original signal S after the wavelet transform is ahead or behind in the time domain 𝛽 into a calculation formula, and in this embodiment, the value of 𝛽 is used according to different The scale is brought in, the second scale is 5, the third scale is 10, the fourth scale is 25, and the calculation formula is as follows:

The corresponding coordinate P(n) of the original signal S in the time domain is calculated by the above calculation formula.

Referring to FIG. 14, then in the judgment step 38, it compares the distance between the aforementioned coordinate P(n) and the previous feature extremum to the calculated coordinate P(n-1), due to age, gender, and life Influenced by different factors such as habits, the standard value of the distance will also be different, and should not be limited to this. When the distance is less than the first value, and in this embodiment, the first value is 100 sampling points, take If the coordinates P(n) and P(n-1) are sharper, the coordinates of the positive characteristic extreme points (X2, Y2) and the negative characteristic extremes of the P(n) and P(n-1) will be included. The value point coordinates (X1, Y1) bring into a judgment formula, the judgment formula is as follows:

After calculating through the judgment formula, the result value (α) which is larger (that is, R wave peak) is added to the record. With reference to FIG. 16, two pairs of characteristic extreme value pairs appear at the frame, and the calculation formula is used to obtain the result The larger value, that is, the peak of the R wave, should be added to the record; when the distance is less than the second value and greater than the first value, it means that the subject's heart rate is faster, or the ECG signal contains a steeper T wave or The P wave must be judged further. In this embodiment, the second value is 130 sampling points. Therefore, the second-scale high-frequency signal D2 is added to assist the judgment. When the characteristic extremum pair of the third scale is also Appears in the high frequency signal D2 of the second scale and the position is the same, then add the P(n) coordinate position to the record, when the coordinates P(n-1) and P(n) both appear in the third scale and the In the second-scale high-frequency signals D3 and D2, it means that the two coordinates P(n-1) and P(n) are both R wave peaks, so the two coordinates P(n-1) and P(n) are both Records must be added. On the contrary, the coordinates that do not appear in the third-scale and second-scale high-frequency signals D3 and D2 at the same time are not R-wave peaks, and will not be recorded. Refer to FIG. 17 for the third-scale high-frequency The characteristic extreme value pair where the signal D3 is framed also appears in the second-scale high-frequency signal D2, which is added to the record. Also refer to FIG. 18, which shows the third-scale high-frequency signal D3 framed in FIG. 18 Feature extreme value pairs appear, but there is no such feature extreme value pair in the second-scale high-frequency signal D2, and no record is added; when the distance is greater than the second value, that is, greater than 130 sampling points, the record is directly recorded The position of the coordinate P(n), and finally at the exclusion step 39, the result value (α) of each feature extreme value pair is calculated by the aforementioned judgment formula, when the result value is less than one fifth of the average value of all result values, That point will be deleted from the record.

Therefore, in conjunction with the steps of calculation 37, judgment 38, and elimination 39, the accuracy of the interpretation is further increased, and the high frequency signals D2, D3, and D4 of the second, third, and fourth scales are used interactively for analysis, which can be done quickly and accurately. The positioning of the QRS wave greatly improves the accuracy of judgment, reduces the overall calculation amount and effectively improves the analysis speed, so as to achieve rapid assessment and diagnosis of the disease, and can also reduce the consumption of human resources.

The present invention uses 48 electrocardiogram records of the MIT-BIH arrhythmia database, a total of 109,488 QRS waves, and further tests the algorithm 3 proposed in this case, and uses the following formula to calculate the sensitivity (Se) of algorithm 3 proposed in this case. Values such as positive predictive value (+P) and error rate (Der):

TP: Total number of correct QRS waves

FP: Total number of non-QRS waves detected

FN: the total number of QRS waves that were not detected

FP+FN: total number of errors

Referring to Table 1 below, the experimental results show that the detection accuracy of QRS waves can reach 99.84% sensitivity and 99.91% positive predictive value, and the error rate is only 0.2471%. Therefore, the algorithm 3 proposed in this case has considerable accuracy It also has practical value.

Table 1 Numbering TP FP FN FP+FN Numbering TP FP FN FP+FN 100 2,273 0 0 0 201 1,963 6 2 8 101 1,865 0 0 0 202 2,136 0 3 3 102 2,187 0 0 0 203 2,980 13 33 46 103 2,084 0 0 0 205 2,656 0 6 6 104 2,229 1 1 2 207 1,860 1 18 19 105 2,572 27 4 31 208 2,955 2 16 18 106 2,027 0 5 5 209 3,005 0 2 2 107 2,137 0 1 1 210 2,650 4 8 12 108 1,763 10 13 twenty three 212 2,748 0 0 0 109 2,532 0 0 0 213 3,251 0 0 0 111 2,124 0 1 1 214 2,262 1 3 4 112 2,539 0 0 0 215 3,363 0 1 1 113 1,789 6 0 6 217 2,208 3 7 10 114 1,879 1 8 9 219 2,154 0 0 0 115 1,953 0 0 0 220 2,048 0 0 0 116 2,412 0 20 20 221 2,427 0 2 2 117 1,535 0 0 0 222 2,483 1 2 3 118 2,278 0 0 0 223 2,605 0 3 3 119 1,987 0 0 0 228 2,053 8 13 twenty one 121 1,863 0 1 1 230 2,256 1 0 1 122 2,476 0 0 0 231 1,571 0 0 0 123 1,518 0 0 0 232 1,780 4 0 4 124 1,619 0 0 0 233 3,079 0 2 2 200 2,601 2 5 7 234 2,753 0 0 0

In summary, the QRS wave detection algorithm of the present invention to reduce the influence of noise includes adjustment, reading, pre-processing, conversion, calculation and selection steps; in this adjustment step, the acquisition window is adjusted by the processor Frame, and after loading the original signal in the reading step, further remove the baseline drift through the filter in the pre-processing step to obtain the corrected original signal, and then convert the corrected original signal in the conversion step to Obtain high-frequency and low-frequency signals of all scales, then calculate the positive and negative thresholds in this calculation step, and finally select the appropriate analysis scale through the selection formula in the selection step. Therefore, multiple analysis in wavelet transform is used The characteristics, together with the selection formula to determine the amount of noise in the scale, so that when the high-frequency noise level is large, the lower-frequency scale component is used, and when the low-frequency P-wave or T-wave amplitude is large, the higher Frequency scale components, in order to reduce the misjudgment caused by noise and large amplitude P-wave and T-wave, and reduce the influence of noise on QRS wave detection, thereby greatly improving the accuracy of judgment and the speed of analysis to achieve the disease Quick assessment and diagnosis, while effectively reducing the consumption of human resources.

However, the above are only for explaining the preferred embodiments of the present invention, but the scope of the implementation of the present invention cannot be limited by this, that is, simple equivalent changes and modifications made according to the scope of the patent application of the present invention and the content of the description of the invention , Should still fall within the scope of this invention patent.

(this invention) 3: QRS wave detection algorithm to reduce the influence of noise 31: Adjustment steps 32: Reading steps 33: Pre-processing steps 34: Conversion steps 35: Calculation steps 36: Selection steps 37: Estimation steps 38: Judgment step 39: Exclusion steps 4: processor 5: filter S: original signal A1: The first-scale low-frequency signal A2: Second-scale low-frequency signal A3: The third-scale low-frequency signal A4: Fourth-scale low-frequency signal D1: The first-scale high-frequency signal D2: High frequency signal of the second scale D3: Third-scale high-frequency signal D4: Fourth-scale high-frequency signal

FIG. 1 is a schematic flowchart of a first preferred embodiment of the present invention. FIG. 2 is another schematic block diagram of the first preferred embodiment. FIG. 3 is a schematic diagram of adjusting the capture pane. FIG. 4 is a partial flow block diagram of the first preferred embodiment. Figure 5-1 is the original signal with baseline drift. Figure 5-2 is the original signal to modify the baseline drift. Figure 6 is a schematic diagram of wavelet multiple decomposition. Fig. 7 is a wavelet first-order converted waveform diagram. Fig. 8 is the low-frequency signal after the original signal is transformed by the fourth-order wavelet. Fig. 9 is the high-frequency signal after the original signal is transformed by the fourth-order wavelet. Fig. 10 shows the extreme point of positive and negative characteristic of high frequency signal. FIG. 11 is the extreme point remaining after removing the large characteristic extreme point. FIG. 12 is a schematic flowchart of a second preferred embodiment of the present invention. FIG. 13 is another schematic block diagram of the second preferred embodiment. FIG. 14 is a partial flow block diagram of the second preferred embodiment. 15 is a schematic diagram of characteristic extreme value pairs. Figure 16 shows the sharper waveforms when two pairs of characteristic extreme values appear. Figure 17-18 is the judgment of adding high-frequency components of the second scale.

3: QRS wave detection algorithm to reduce the influence of noise

31: Adjustment steps

32: Reading steps

33: Pre-processing steps

34: Conversion steps

35: Calculation steps

36: Selection steps

Claims (10)

A QRS wave detection algorithm for reducing the influence of noise, which includes the following steps: An adjustment step, which is equipped with a processor that adjusts the acquisition pane through the wavelet transformation formula so that any two acquisition panes Several sampling points are overlapped in order to capture the original signal to be analyzed through the acquisition pane; a reading step, which reads and loads the original signal to be analyzed; a pre-processing step, which has a The filter performs filtering processing on the original signal to obtain a baseline drift signal, and then subtracts the baseline drift signal from the original signal to obtain a modified original signal; a conversion step, the modified original signal passes through wavelet The conversion obtains high-frequency signals and low-frequency signals of all scales, and at the same time, the wave form in the corrected original signal is converted into positive and negative characteristic extreme points; a calculation step, which first divides the high-frequency signals of the selected scale into four segments, And take the positive maximum value and negative minimum value respectively in the four signals, and then average the four maximum values and the four minimum values respectively to obtain the average maximum value and the average minimum value, and then take the average maximum value A quarter of the value to obtain a positive threshold value, and another quarter of the average minimum value to obtain a negative threshold value; and a selection step, which removes the high frequency signal, greater than the Positive and negative characteristic extreme points that are twice the positive and negative thresholds, and the remaining positive and negative characteristic extreme points of the third scale high frequency signal and the remaining positive and negative characteristic extreme points of the fourth scale high frequency signal are transmitted through A selection formula is used for calculation. When the obtained value is less than one, the third-scale high-frequency signal is used for analysis. Conversely, when the obtained value is greater than one, the fourth-scale high-frequency signal is used for analysis. According to the QRS wave detection algorithm for reducing the influence of noise described in item 1 of the patent application scope, wherein 150 sampling points overlap between the two acquisition panes in the adjustment step. According to the QRS wave detection algorithm for reducing the influence of noise as described in item 1 of the patent application scope, in the pre-processing step, the filter is a median filter. According to the QRS wave detection algorithm for reducing the influence of noise as described in item 1 of the patent application scope, in the conversion step, the modified original signal is converted into four scales of high frequency signal and low frequency signal, and the conversion is completed 50 sampling points are removed in front of and behind the high frequency signal acquisition pane. According to the QRS wave detection algorithm for reducing the influence of noise described in Item 1 of the patent application scope, wherein the selection step and the calculation step use a third-scale high-frequency signal for comparison and analysis. According to the QRS wave detection algorithm for reducing the influence of noise as described in item 1 of the patent application scope, wherein after the selection step, there is also an extrapolation step. The extrapolation step first marks a positive feature that exceeds the positive threshold Extreme point, and mark a negative characteristic extreme point that exceeds the negative threshold, and the distance between the positive and negative characteristic extreme points needs to be less than the preset value to form a characteristic extreme value pair, and then the positive 3. The coordinates of the negative characteristic extreme point are brought into a calculation formula to calculate the corresponding coordinates of the original signal in the time domain. According to the QRS wave detection algorithm for reducing the influence of noise as described in item 6 of the patent application scope, the preset value in the calculation step is 45 sampling points. According to the QRS wave detection algorithm for reducing the influence of noise as described in item 6 of the patent application scope, wherein after the calculation step, a judgment step is further included, which judges the coordinates of the corrected original signal and the previous feature The extreme value compares the distance between the calculated coordinates. When the distance is less than the first value, the coordinates are brought into a judgment formula for calculation, and the result value is larger and added to the record; when the distance is less than When the second value is greater than the first value, the second-scale high-frequency signal is added for judgment. When the characteristic extreme value pair also appears in the second-scale high-frequency signal and the position is the same, the characteristic pole is recorded Value pair; conversely, when the distance is greater than the second value, the characteristic extreme value pair also needs to be recorded. According to the QRS wave detection algorithm for reducing the influence of noise as described in item 8 of the patent application range, the first value is 100 sampling points and the second value is 130 sampling points. According to the QRS wave detection algorithm for reducing the influence of noise as described in item 8 of the patent application scope, wherein after the judging step, an exclusion step is further included, and the exclusion step calculates each feature extreme value pair through the judgment formula The result value is deleted when the result value is less than one fifth of the average value of all result values.

Images