Electrocardiosignal R peak detection method and wearable electrocardiogram detection device and system

Abstract

The invention discloses an electrocardiosignal R peak detection method and a wearable electrocardiosignal detection device and system. An electrocardiosignal is preprocessed; performing forward search on the preprocessed electrocardiosignal by adopting a one-dimensional self-adaptive threshold iterative algorithm: after any ith R peak Peaks (i) is detected, calculating the interval time between the R peak Peaks (i) and the previous adjacent R peak Peaks (i-1), and comparing the interval time with a first interval threshold; according to the technical scheme, reverse search is carried out between the previous real R peak Peaks (i-1) and the R peak Peaks (i) until the real R peak is searched, then the R peak Peaks (i) is used as a search starting point, and forward search continues to be carried out, compared with the prior art, whether the R peak searched by the one-dimensional self-adaptive threshold iterative algorithm is abnormal or not is rechecked through the abnormality of the interval time, and the accuracy of the R peak search is improved. And the real R peak is searched through reverse search after the abnormality is found, so that high-precision electrocardiosignal R peak detection can be realized by using fewer computer resources.

Description

Technical Field

[0001] The present invention relates to the field of electrocardiogram (ECG) signal detection, and in particular to an ECG signal R peak detection method, a wearable ECG detection device and a system. Background Art

[0002] Arrhythmia is caused by abnormalities in the frequency, regularity, starting position or conduction of the heart's electrical pulses. It manifests as an irregular, abnormally fast (tachycardia) or abnormally slow (bradycardia) heart rate, and is a common and potentially dangerous cardiovascular disease in clinical practice. Due to the sporadic, transient and short-term characteristics of arrhythmia, a short-term electrocardiogram may not reveal obvious abnormalities. In addition, symptoms such as atrial fibrillation, atrial premature beats and sinus arrhythmia are not significant enough and are easily overlooked, so long-term health management and disease monitoring are an effective and necessary method. However, due to the high cost of medical care, it is impossible for patients to receive treatment in the hospital for a long time. Fortunately, wearable health monitoring devices can continuously and effectively monitor electrocardiograms in various real-life scenarios. This low-cost, easy-to-carry, and sustainable wearable sensor is expected to change the pattern of cardiovascular disease monitoring and become an effective means of preventing cardiovascular disease.

[0003] For electrocardiogram (ECG), the QRS wave contains the most abundant clinical information. The peak R peak is the wave with the greatest visual impact in the heartbeat cycle. Its accurate detection is the basis for diagnosing heart disease. There are many existing ECG R wave detection methods, but most algorithms are not suitable for deployment in wearable devices. Although the wavelet transform has a high detection rate for the R peak, there is currently no clear rule for selecting the mother wavelet function. When facing different individuals, the wavelet parameters need to be adjusted individually. Secondly, the wavelet transform has high requirements for computing resources and is not easy to implement in engineering. Methods based on deep learning need to be updated and retrained regularly, and the processing power and battery life of wearable health monitoring devices are difficult to support complex deep models. Other detection methods, such as methods based on ECG signal morphological features and differential threshold methods, are simple and fast, but are prone to missed detections and false detections, and are not adaptive to artifacts generated during movement. Most of the existing adaptive threshold methods use a dual threshold iteration mode. Although they can also detect the R peak well, they require a large number of empirical parameters to assist in the update of the threshold model, and require a large amount of existing ECG data to optimize the parameters. The electrocardiogram in wearable products is more susceptible to noise such as motion artifacts, baseline drift, and electromyographic interference, so the secondary threshold will bring more pseudo peak problems. Summary of the invention

[0004] The present invention provides an electrocardiogram signal R peak detection method, a wearable electrocardiogram detection device and a system, which are used to solve the technical problem that the existing electrocardiogram detection method has high requirements on computer resources in order to achieve high accuracy.

[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is:

[0006] A method for detecting an R peak of an electrocardiogram signal, comprising:

[0007] Preprocessing of ECG signals;

[0008] A one-dimensional adaptive threshold iteration algorithm is used to perform a forward search on the preprocessed ECG signal:

[0009] After any i-th R peak peaks(i) is detected, the following steps are performed:

[0010] Calculate the interval time T(i-1,i) between the R peak Peaks(i) and its adjacent previous R peak Peaks(i-1), and compare the interval time T(i-1,i) with the first interval threshold T1:

[0011] When the interval time T(i-1,i) is less than or equal to the first interval threshold T1, the i-th R peak peaks(i) is taken as the latest true R peak;

[0012] When the interval time T(i-1,i) is greater than the first interval threshold T1, a reverse search is performed between the previous real R peak Peaks(i-1) and R peak Peaks(i) until the real R peak is found, and then the forward search is continued with R peak Peaks(i) as the search starting point.

[0013] Preferably, the R peak Peaks(i) and its adjacent previous R peak Peaks(i-1) satisfy:

[0014] Th(n)=α*abs(peaks(i)-peaks(i-1))+(1-α)*Th(n-1)

[0015] α=peaks(i) / (peaks(i)+peaks(i-1))

[0016] Wherein, Th(n) represents the current threshold, Th(n-1) represents the previous threshold, α is the threshold adjustment factor, which represents the ratio of the two real R peaks before and after, peaks(i) represents the most recently detected R peak, peaks(i-1) represents the real R peak before peaks(i), and abs represents the absolute value.

[0017] Preferably, in the forward search process, after any i-th R peak peaks(i) is detected, the following steps are further included:

[0018] The interval time T(i-1,i) is compared with the second interval threshold T1. When the interval time T(i-1,i) is less than the second interval threshold T1, the amplitudes of the peaks(i) and Peaks(i-1) are calculated, and it is determined whether the amplitude of the peaks(i) is greater than N times the amplitude of Peaks(i-1). If so, Peaks(i-1) is removed as an outlier, and the value of N is greater than 1.

[0019] Preferably, the forward search process further includes the following steps:

[0020] In the forward search process, for any three adjacent R peaks Peaks(i-1), peaks(i), and Peaks(i+1), if the amplitude of peaks(i) satisfies the requirement that it is M times greater than the amplitudes of Peaks(i-1) and Peaks(i+1), Peaks(i) will be removed as an outlier, where the value range of M is 3-5.

[0021] Preferably, the first interval threshold T1 includes a limit value RRmiss of a sampling period Fs and / or a heartbeat interval;

[0022] The limit value RRmiss of the heartbeat interval satisfies the following formula:

[0023]

[0024] RR miss =β*RR Ave

[0025] Among them, RR n-k , R.R. n-(k-1) ,...,RR nn are the latest k R peaks, RR Ave is the average value, and β is the preset coefficient.

[0026] Preferably, in the reverse search process, a single peak with the largest amplitude is searched between the latest true R peak and the current position, and the single peak is defined as a candidate R peak, and it is determined whether the candidate R peak satisfies any of the following conditions. If so, the candidate R peak is determined to be the latest true R peak:

[0027] Condition 1: The candidate R peak is found within a sampling period Fs, and its amplitude is greater than 0.2*PeakAve;

[0028] Condition 2: The candidate R peak loss interval is greater than the limit value RRmiss, and its amplitude value is greater than 0.1*PeakAve.

[0029] Preferably, during the reverse search process, when the real R peak is found, the search threshold is updated using the following formula, and then a forward search is performed:

[0030] Th(n)=α*(peaks(i)-peaks(i-1))+(1-α)*Th(n-1).

[0031] Preferably, preprocessing the ECG signal includes:

[0032] The ECG signal is filtered using a third-order Butterworth bandpass filter to obtain a filtered ECG signal y0(n);

[0033] The nonlinear energy operator is used to process the filtered ECG signal y0(n) to obtain the output signal y1(n)

[0034] The output signal y1(n) is smoothed using a sliding window to obtain the final preprocessed ECG signal;

[0035] and / or

[0036] The value of N is 2, the value of M is 4, and the value of β is 1.66.

[0037] A wearable electrocardiogram detection device comprises a collection unit and a processing unit. The collection unit is used to collect electrocardiogram signals and send the electrocardiogram signals to the processing unit. The processing unit is used to execute the steps of the above method.

[0038] A computer system comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the processor executes the computer program.

[0039] The present invention has the following beneficial effects:

[0040] 1. The present invention provides an electrocardiogram signal R peak detection method, a wearable electrocardiogram detection device and a system, which preprocess the electrocardiogram signal; and use a one-dimensional adaptive threshold iteration algorithm to perform a forward search on the preprocessed electrocardiogram signal: after detecting any i-th R peak Peaks(i), calculate the interval time between the R peak Peaks(i) and its adjacent previous R peak Peaks(i-1), and compare the interval time with a first interval threshold: when the interval time is greater than the first interval threshold, reverse search the electrocardiogram signal until the next R peak Peaks(i+1) is found, and use R peak Peaks(i+1) as the search starting point to continue forward search. Compared with the prior art, the present technical solution verifies whether the R peak searched by the one-dimensional adaptive threshold iteration algorithm is abnormal through the abnormality of the interval time, and searches for the real R peak through reverse search after the abnormality is found, so that high-precision electrocardiogram signal R peak detection can be achieved with less computer resources.

[0041] In addition to the above-described purposes, features and advantages, the present invention has other purposes, features and advantages. The present invention will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings constituting a part of this application are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0043] Figure 1 The ECG signal corresponding to the output in the preprocessing stage;

[0044] Figure 2 This is a schematic diagram of single threshold iteration for positive judgment of R peak;

[0045] Figure 3 Schematic diagram of reverse judgment of R peak single threshold iteration;

[0046] Figure 4 Schematic diagram of R peak detected in ECG signal under different noise conditions;

[0047] Figure 5 This is a physical picture of the wearable ECG monitoring system actually built;

[0048] Figure 6 Schematic diagram of R peak and single threshold iteration in ECG signal visualized by serial port;

[0049] Figure 7 It is a flow chart of the electrocardiogram signal R peak detection method of the present invention. DETAILED DESCRIPTION

[0050] The embodiments of the present invention are described in detail below with reference to the accompanying drawings, but the present invention can be implemented in many different ways as defined and covered by the claims.

[0051] like Figure 7 As shown, the present invention proposes a method for detecting R peak of an electrocardiogram signal, comprising:

[0052] 1. Preprocessing stage

[0053] For the ECG signal x(n) (such as Figure 1 For example, the QRS wave is the part with the most concentrated energy in a single heartbeat, and the ECG signal itself is in the useless baseline part for a long time. Therefore, we prefer to do preprocessing first to highlight the key features of the ECG signal. First, the traditional third-order Butterworth bandpass filter is used to filter the ECG signal to eliminate some basic noise and obtain the filtered ECG signal y0(n), as shown in Figure 1 (b) As shown. Next, we use the nonlinear energy operator to process the ECG signal after filtering and further analyze the change pattern of the QRS wave area, as shown in formula (1).

[0054] y1(n)=y 2 0[n]-y0[n+1]y0[n-1] (1)

[0055] In the formula, y1(n) is the output signal after NEO operation, such as Figure 1 As shown in (c), y0[n] represents the filter output signal, y0[n+1] and y0[n-1] represent the filter output signals with a front and rear offset of 1. NEO's ability to enhance transient changes greatly enhances the QRS wave and effectively highlights the difference between the R peak and the noise. In addition, due to the difference in the arrangement of the lead electrodes, the ECG signal causes the R peak to deflect upward or downward. The nonlinear characteristics of the NEO operation are used to convert the bipolarity into a unipolarity, which greatly reduces the pressure of subsequent threshold decisions. Taking into account the residual high-frequency noise, the sliding window is used to smooth the ECG signal to further remove the pseudo peaks, and the final preprocessed ECG signal is obtained, as shown in Figure 1 (d) as shown.

[0056] 2. Threshold decision stage

[0057] The present invention establishes a one-dimensional adaptive threshold iteration model for judging the R peak, as shown in formula (2).

[0058]

[0059] Where Th(n) represents the current threshold, Th(n-1) represents the previous threshold, α is the threshold adjustment factor, which represents the ratio of the two real R peaks before and after, peaks(i) represents the most recently detected R peak, and peaks(i-1) represents the real R peak before peaks(i).

[0060] In the threshold detection stage, firstly, a threshold is set for the threshold parameter Th(n) and its adjustment factor α, as shown in formula (3).

[0061]

[0062] The initial value of the threshold of the present invention has not been optimized, so the setting of the initial value is relaxed, which increases the practicality of the algorithm. Once a peak is identified during the forward search, we record it as R1, and the next identified peak is marked as R2. If R2 is within 360ms of R1, we choose between peak R1 and peak R2 based on the amplitude. Figure 2 As shown, first the first peak ( Figure 2 The peak at position (i) is detected by the threshold, which represents the true R peak. Then the threshold detects the peak at position (ii). Positions (i) and (ii) do not meet the time interval requirement of 360ms. Next, the threshold detects the peak at position (iii). Since positions (ii) and (iii) meet the peak interval requirement of 360ms, we need to use the amplitude for judgment. Specifically, the peak amplitude at position (iii) is greater than twice the peak amplitude at position (ii), so the peak at position (ii) is declared an invalid R peak. After a true R peak is found, the threshold is updated according to formula (2).

[0063] A single threshold may be missing when searching for R peaks in the forward direction. We use the RR interval and the sensor sampling frequency to perform the reverse search. When the real R peak is detected, we need to count the intervals of the latest 8 peaks and use their average interval RR Ave To set the heartbeat interval limit RR miss , as shown in formula (4).

[0064]

[0065] Furthermore, in order to prevent the threshold update rhythm from being disrupted by excessive peaks, three consecutive peaks (Peaks(i-1), Peaks(i), Peaks(i+1)) are considered to eliminate outliers. If Peaks(i) satisfies the requirement of being greater than 4 times the amplitude of Peaks(i-1) and Peaks(i+1), it is eliminated and the average amplitude of the latest 8 true R peaks is updated, as shown in formula (5).

[0066]

[0067] If the threshold appears in the process of forward searching for R peak, the interval time exceeds RRmiss or the sampling period F S , it is necessary to start the reverse search. We define the single peak with the largest amplitude found between the latest true R peak and the current position as the candidate R peak. The true R peak here should refer to the latest true R peak Peaks(i) found in the search process (including forward and reverse), and the current position refers to a sampling point located in the reverse search process.

[0068] Since the QRS wave has a refractory period of 0.2s physiologically, we exclude the signal segments before and after 0.2s. If the single peak is found within a sampling period Fs, the amplitude must be greater than 0.2*PeakAve; if the missing interval is greater than 1.66*RRAve, the amplitude only needs to be greater than 0.1*PeakAve. Figure 3 As shown in the figure, positions (i) and (ii) are incorrectly identified peaks. Their existence affects the update of the subsequent threshold, resulting in the threshold not directly detecting the R peak at position (iii). Since the distance between the two adjacent peaks before and after position (iii) exceeds RRmiss, the R peak at (iii) is found during the reverse search process, and the candidate R peaks that do not meet the conditions are directly discarded for the next round of forward search.

[0069] Accordingly, after finding the R peak in the reverse search phase, the threshold needs to be updated using formula (6), and the position information of the current R peak is timely obtained for the next forward search.

[0070] Th(n)=α*(peaks(i)-peaks(i-1))+(1-α)*Th(n-1) (6)

[0071] The reverse search starts from the currently detected true R peak Peaks(i), and the distance between peaks(i) and its adjacent previous R peak peaks(i-1) is called the missing interval. The sixth point described in this patent is used to determine the missing interval, and the candidate peak is defined as the true R peak peaks(i+1) through conditions 1 and 2 (in fact, after a new true R peak is found, peaks(i-1), peaks(i), and peaks(i+1) will be updated at the same time). For example, before the reverse search, the position of peaks(i-1) is 1 and the position of peaks(i) is 5. If the reverse search finds peaks(i+1) at 3, then the updated position 3 is peaks(i-1) and 5 is peaks(i). The next round of forward search starts from peaks(i).

[0072] 3. Experimental verification:

[0073] We used matlab2018 simulation software to perform R peak detection on a total of 9 standard ECG databases. Figure 4 The results of ECG signal R peak positioning under different noise conditions are taken from MIT-BIH Arrhythmia Database (MITDB). Figure 4 (a) is a typical baseline drift ECG signal, taken from MITDB ECG record 121; Figure 4 (b) is an ECG signal interfered by myoelectric interference, taken from MITDB ECG record 104; Figure 4 (c) is the ECG signal under motion artifact, taken from MITDB ECG record 105; Figure 4 (d) is the ECG signal of sudden changes in heartbeat rhythm, taken from MITDB ECG record 108. The required ECG signal ECG is imported into the matlab2018 software and after preprocessing, the ECG signal ECG is input into the adaptive threshold iteration algorithm module. After the threshold judgment, a decision is made and the final R peak information is output.

[0074] In the actual product test, we implemented the algorithm on the microprocessor STM32F407 and designed an ECG sensor using ADS1292R to collect ECG signals in real time. The sampling frequency was set to 125Hz. Figure 5 The prototype of the wearable ECG monitoring system shown in the figure. The whole system is powered by a battery, and electrodes are placed on the human chest to capture ECG signals. The original ECG signal is processed by the algorithm deployed in the STM32F407, and finally the R wave position information is output. We used this system to collect the ECG signal of a 24-year-old male healthy volunteer in real time, and used the serial port assistant tool to visualize the enhanced ECG signal, dynamic single threshold, detected R peak position and original ECG signal, as shown in the figure. Figure 6 shown. Figure 6 (a) shows the process of adaptively adjusting the threshold according to the enhanced ECG signal. Figure 6 (b) shows the position of the R peak output on the original ECG signal. Obviously, the R peak position output by the serial port is close to the manually determined R peak position. We are basically sure that the proposed algorithm can meet the requirements of R peak recognition in real-time scenarios.

[0075] In summary, the present invention can quickly and effectively detect the R peak in the ECG signal. The algorithm is based on the nonlinear energy operator (Nonlinear Energy Operator) in the preprocessing stage combined with a dynamic single threshold to realize R wave detection. The nonlinear energy operator (NEO) highlights the QRS complex wave characteristics, and the dynamic single threshold is used for R peak judgment. The process of threshold adaptive adjustment depends only on the R peak amplitude, avoiding the interference of too many empirical parameters. We use RR intervals and sensor sampling frequencies to improve detection accuracy, and can quickly adapt to different individuals while ensuring accuracy. Further, the present invention implements the algorithm on the STM32F407 microcontroller, and builds a simple ECG monitoring system to verify its practicality on wearable ECG sensors.

[0076] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for detecting R peak of an electrocardiogram signal, characterized in that: include: Preprocessing of ECG signals; A one-dimensional adaptive threshold iteration algorithm is used to perform a forward search on the preprocessed ECG signal: After any i-th R peak peaks(i) is detected, the following steps are performed: Calculate the interval time T(i-1,i) between the R peak Peaks(i) and its adjacent previous R peak Peaks(i-1), and compare the interval time T(i-1,i) with the first interval threshold T1: When the interval time T(i-1,i) is less than or equal to the first interval threshold T1, the i-th R peak peaks(i) is taken as the latest true R peak; When the interval time T(i-1,i) is greater than the first interval threshold T1, a reverse search is performed between the previous real R peak Peaks(i-1) and R peak Peaks(i) until the real R peak is found, and then the forward search is continued with R peak Peaks(i) as the search starting point.

2. The ECG signal R peak detection method according to claim 1, characterized in that: The R peak Peaks(i) and its adjacent previous R peak Peaks(i-1) satisfy: Th(n)=α*abs(peaks(i)-peaks(i-1))+(1-α)*Th(n-1) α=peaks(i) / (peaks(i)+peaks(i-1)) Wherein, Th(n) represents the current threshold, Th(n-1) represents the previous threshold, α is the threshold adjustment factor, which represents the ratio of the two real R peaks before and after, peaks(i) represents the most recently detected R peak, peaks(i-1) represents the real R peak before peaks(i), and abs represents the absolute value.

3. The ECG signal R peak detection method according to claim 2, characterized in that: In the forward search process, after any i-th R peak peaks(i) is detected, the following steps are also included: The interval time T(i-1,i) is compared with the second interval threshold T1. When the interval time T(i-1,i) is less than the second interval threshold T1, the amplitudes of the peaks(i) and Peaks(i-1) are calculated, and it is determined whether the amplitude of the peaks(i) is greater than N times the amplitude of Peaks(i-1). If so, Peaks(i-1) is removed as an outlier, and the value of N is greater than 1.

4. The ECG signal R peak detection method according to claim 2 or 3, characterized in that: The forward search process also includes the following steps: In the forward search process, for any three adjacent R peaks Peaks(i-1), peaks(i), and Peaks(i+1), if the amplitude of peaks(i) satisfies the requirement that it is M times greater than the amplitudes of Peaks(i-1) and Peaks(i+1), Peaks(i) will be removed as an outlier, where the value range of M is 3-5.

5. The method for detecting R peak of an electrocardiogram signal according to claim 4, characterized in that: The first interval threshold T1 includes a sampling period Fs and / or a limit value RRmiss of a heartbeat interval; The limit value RRmiss of the heartbeat interval satisfies the following formula: RR miss =β*RR Ave Among them, RR n-k , R.R. n-(k-1) ,...,RR nn are the latest k R peaks, RR Ave is the average value, and β is the preset coefficient.

6. The method for detecting R peak of an electrocardiogram signal according to claim 5, characterized in that: In the reverse search process, the single peak with the largest amplitude is searched between the latest true R peak and the current position, and the single peak is defined as a candidate R peak. It is determined whether the candidate R peak meets any of the following conditions. If so, the candidate R peak is determined to be the latest true R peak: Condition 1: The candidate R peak is found within a sampling period Fs, and its amplitude is greater than 0.2*PeakAve; Condition 2: The candidate R peak loss interval is greater than the limit value RRmiss, and its amplitude value is greater than 0.1*PeakAve.

7. The method for detecting R peak of an electrocardiogram signal according to claim 6, characterized in that: In the reverse search process, when the real R peak is found, the search threshold is updated using the following formula, and then the forward search is performed: Th(n)=α*(peaks(i)-peaks(i-1))+(1-α)*Th(n-1).

8. The method for detecting R peak of an electrocardiogram signal according to any one of claims 4 to 7, characterized in that: Preprocess the ECG signal, including: The ECG signal is filtered using a third-order Butterworth bandpass filter to obtain a filtered ECG signal y0(n); The nonlinear energy operator is used to process the filtered ECG signal y0(n) to obtain the output signal y1(n) The output signal y1(n) is smoothed using a sliding window to obtain the final preprocessed ECG signal; and / or The value of N is 2, the value of M is 4, and the value of β is 1.

66.

9. A wearable ECG detection device, comprising a collection unit and a processing unit, wherein the collection unit is used to collect ECG signals and send the ECG signals to the processing unit, and the processing unit is used to execute the steps of any method as claimed in claims 1-8.

10. A computer system comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 9 are implemented.

Images