CN115998304A - Device for acquiring electrocardiogram at non-clinical standard position - Google Patents

Abstract

The invention belongs to the technical field of medical equipment, and particularly relates to equipment for acquiring an electrocardiogram at a non-clinical standard position. The device comprises a surface electrophysiological signal detection device and a data processing system; the data processing system includes: an input module for inputting surface electrophysiological signal data; the characteristic extraction module is used for processing the surface electrophysiological signal data by adopting a zero-phase filter based on stable wavelet transformation to obtain a plurality of layers of wavelet detail coefficients and profile coefficients; and the voting module is used for carrying out QRS complex detection on all wavelet detail coefficients and obtaining heartbeat detection results from all wavelet detail coefficients through a weight voting scheme. The invention realizes the purpose of acquiring the electrocardiogram at a non-clinical standard position, has strong applicability to application scenes such as acquiring a long-time dynamic electrocardiogram from non-traditional parts such as the upper arm, the back part of the ear, the inner part of the ear and the like, and has good application prospect.

Description

Device for acquiring electrocardiogram at non-clinical standard position

Technical Field

The invention belongs to the technical field of medical equipment, and particularly relates to equipment for acquiring an electrocardiogram at a non-clinical standard position.

Background

Cardiovascular disease is the leading cause of death worldwide. About 50% of deaths are usually caused by arrhythmias. Accurate recording and detection of various arrhythmias is critical to preventing death from these conditions. Heartbeat detection is critical to determining heart rate and related arrhythmias. Electrocardiograph (ECG) is the most widely used clinical tool, which provides a graphical representation of the detection of cardiac electrical activity from the body surface of the human body.

Currently, the equipment for detecting an electrocardiogram in a hospital is mainly a medical monitor, and a plurality of groups of electrodes are required to be arranged at a plurality of standard parts such as limbs, chest and the like of a detected person, so that accurate ECG detection is realized. However, such a device is bulky, has a plurality of sets of signal cables, and is complicated to operate for detecting ECG, and thus is difficult to be applied to non-hospital places such as home. To achieve the possibility of home measurement, some research has focused on designing ECG devices such as non-standard ECG lead configurations like upper arms to enhance the wearability, convenience, reusability of the device. However, these surface electrophysiological signals acquired through non-clinical standard locations include not only ECG signals but also Electromyography (EMG) signals. Thus, in the research of these devices, one of the main challenges is: under the EMG artifact interference with energy much larger than that of the electrocardiosignal, how to improve the acquisition quality of the electrocardiosignal. The main technical means at present is to filter EMG artifacts through signal processing methods such as wavelet transformation. However, the phase shift differences at the different layers of these methods distort the filtered ECG waveform and also deviate the exact time that the ECG waveform occurs. Thus, phase-shift-free feature extraction is critical in ECG signal processing and analysis in low signal-to-noise environments.

The ECG waveform is the QRS complex (representing ventricular depolarization) which plays a fundamental role in heartbeat detection. Thus, accurate detection of the QRS complex is an important step in analyzing the surface electrical signals acquired by the ECG device of the above-described non-standard ECG lead configuration

One of the many methods in the prior art for detecting QRS complexes in ECG is discrete wavelet transform (discrete wavelet transform, DWT). The DWT satisfies the law of conservation of energy, enables perfect reconstruction of signals, and provides efficient computation. Therefore, it is widely used in heart beat detection. Researchers use different levels of decomposition in DWT and assume that QRS waves can be characterized with some detail coefficients based on their spectra. Considering the sampling frequency of the analyzed ECG signal, prior studies have typically selected detail coefficients 1-4 (corresponding to frequencies of 11.25-180 Hz), 3-5 (corresponding to frequencies of 5.75-45 Hz), and 4 (corresponding to frequencies of 15.6-31.1 Hz) for further analysis.

However, DWT also has drawbacks including high sensitivity to signal offset and reduced temporal resolution at coarse scale. Thus, some research has later focused on reducing or eliminating the offset variance of DWT. One of the commonly employed methods in ECG denoising is the stationary wavelet transform (stationary wavelet transform, SWT). In SWT, the filter coefficients are up-sampled in each step, rather than down-sampling the signal after filtering. Thus, the resulting coefficients have the same length as the original signal. Due to its redundant representation, SWT has translational invariance. It is therefore suitable for detecting the exact location of a targeted event in a signal such as an edge. However, SWT has a particular disadvantage in that it has a phase shift relative to the original signal. This is because most wavelet decomposition schemes use FIR filter banks and many wavelet decomposition schemes do not even have a linear phase. This phase shift results in the QRS complex being projected horizontally to different time positions with different SWT detail coefficients, all of which are different from their position in the original signal. Obviously, such a phase shift is not desirable in heartbeat detection. Although the offset is limited to the order of milliseconds, synchronicity is an important issue in QRS detection applications, especially in real-time applications. Furthermore, since the phase shift proceeds with the SWT scale, it is expected that the phase shift will increase with increasing scale. Thus, the final detail and approximation will have the most phase shift compared to the original input signal. It may result in more information being lost in a higher range. In noisy environments where it is more difficult to locate the QRS complex, eliminating these phase shifts becomes a more sensitive problem. In the field of ECG data processing, there is currently no method for eliminating such phase shift in the prior art, so that the existing QRS detection is performed only in one layer of wavelet detail coefficients, and a large amount of useful information contained in the wavelet detail coefficients of other layers is not analyzed. This can have adverse effects on the accuracy and robustness of QRS complex detection. In particular, for surface electrical signals acquired by ECG devices of non-standard ECG lead configurations, the above-mentioned problems are more pronounced due to the effects of EMG artifacts, which in turn are obstacles to accurate detection results obtained by such devices.

Given the importance of minimizing phase shift, daubechies developed near zero-phase filters, widely used for DWT (Ingrid Daubechies, "Ten Lectures on Wavelets," 1992). However, these filters cannot achieve an accurate zero phase. Percival introduced a zero-Phase wavelet called the Zefret transform to solve this problem (D.Percival, "Discrete Wavelet Transforms Based on Zero-Phase Daubechies filters"). Later, lenis et al proposed an improved version of SWT to solve the problem of phase-change free behaviour (biomed. Eng./biomed. Tech., vol.61, no.1, pp.37-56, feb. 2016). While applying standard SWTs to the signals, they invert the signal and apply SWTs to the generated sequences. The transformation is then inverted again and the transformed signals are added at each scale. The resulting coefficients have zero phase shift. Thus, this method claims to be able to detect the onset of the P-wave with very high accuracy. Eliminating phase shifts at higher scales provides the opportunity to combine information at different scales to achieve more accurate beat detection. However, these existing methods for processing low signal-to-noise ratio data are not designed for ECG, and the application of these methods to ECG data has the problems of poor processing effect and poor robustness, and the applicability of the data acquired by ECG devices with non-standard ECG lead configurations such as upper arms is poor.

Disclosure of Invention

Aiming at the defects of the prior art, the invention realizes equipment capable of acquiring an electrocardiogram from a non-clinical standard position by arranging a data processing system to accurately extract the QRS complex from the upper surface electrophysiological signal.

An apparatus for acquiring an electrocardiogram, comprising: a surface electrophysiological signal detection device and a data processing system;

the data processing system includes:

an input module for inputting surface electrophysiological signal data;

the characteristic extraction module is used for processing the surface electrophysiological signal data by adopting a zero-phase filter based on stable wavelet transformation to obtain a plurality of layers of wavelet detail coefficients and profile coefficients;

and the voting module is used for carrying out QRS complex detection on all wavelet detail coefficients and obtaining heartbeat detection results from all wavelet detail coefficients through a weight voting scheme.

Preferably, the surface electrophysiological signal detection device is a device that detects surface electrophysiological signals from a non-clinical standard location.

Preferably, the non-clinical standard location includes an upper limb, a lower limb, behind the ear, in the ear, hip, shoulder or waist.

Preferably, the surface electrophysiological signal detection device comprises three electrodes for being arranged at the following positions of the subject when detecting the surface electrophysiological signal:

1) The front of the biceps brachii of the left arm is the same height as the root of the deltoid;

2) The lateral midline of the upper arm, as high as position 1);

3) Medial midline of upper arm, level with position 1).

Preferably, the feature extraction module is configured to perform the following steps:

step 2.1, adopting second-order Daubechies wavelet transformation as a mother wavelet and 4-7 layers of decomposition to obtain 4-7 wavelet detail coefficients and 1 profile coefficient;

step 2.2, remove the phase shift at each layer of wavelet decomposition with the following steps:

1) Firstly, standard wavelet filtering is carried out, and the formula is as follows:

X(z)H(z)＝D(z)

wherein z is a z-transform operator, z transform of X (z) signal, H (z) is wavelet detail function of the layer, and D (z) is coefficient of the layer;

2) The time reverse order operation, the equivalent formula in the z-transform domain is:

3) The output of the second step is subjected to the same filtering operation as the first step again:

4) And (3) performing time reverse operation on the output of the third step:

d' (z) is the last zero phase shift wavelet detail coefficient for that layer;

the same is then done for the profile coefficients by the wavelet profile function G (z) of this layer.

Preferably, the wavelet detail coefficients are 5 layers, and the frequency content of the 5 layers of wavelet detail coefficients is 64-128Hz, 32-64Hz, 16-32Hz, 8-16Hz and 4-8Hz respectively.

Preferably, in the voting module, the QRS complex detection method is Pan-Tompkins algorithm.

Preferably, in the voting module, the specific content of the voting scheme is as follows:

the detail coefficients of each layer obtain a voting sparsity of corresponding weight on the QRS complex detection result, the final result of the weighting of the voting coefficients of all layers is the final QRS detection result, and the voting process is as follows:

a moving window with the length of 200ms is used on the detected heartbeat sequence, and if the weighting coefficient result in the window exceeds a preset threshold value, the heartbeat is detected; the position of the detected heartbeat is set to be the average of the time positions of all forward votes;

if at least two heartbeats are detected with a time interval of less than 200ms, they are combined into one heartbeat by a time-averaging method.

The "surface electrophysiological signals" of the present invention include signals (ECG signals) indicative of cardiac electrical activity at these locations, as well as signals from other sources, such as skeletal electromyographic signals, smooth electromyographic signals, and neuronal electrical signals. "non-clinical standard location" refers to a standard electrode location specified by a non-clinical electrocardiograph signal.

The invention provides a data processing system for acquiring electrocardiogram signals from non-clinical standard positions. The system combines a zero-phase filter bank based on SWT and voting strategies for different SWT scales, and accurately extracts electrophysiological signal components with fixed time domain waveform characteristics, such as QRS wave groups in electrocardiosignals, from surface electrophysiological signals without causing waveform distortion caused by phase shift of the extracted signals. The invention increases the robustness of the surface electrophysiological data processing method to noise. Experiments prove that the system of the invention has better performance in terms of sensitivity and positive predictive value compared with the prior art when processing surface electrophysiological signals. Therefore, the invention can accurately detect the QRS in a high-noise environment (such as a high skeletal electromyographic signal, a high smooth electromyographic signal and a high neuron electrical signal), has strong applicability to application scenes such as upper arm electrocardiograph detection and the like, and has good application prospect.

It should be apparent that, in light of the foregoing, various modifications, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.

The above-described aspects of the present invention will be described in further detail below with reference to specific embodiments in the form of examples. It should not be understood that the scope of the above subject matter of the present invention is limited to the following examples only. All techniques implemented based on the above description of the invention are within the scope of the invention.

Drawings

Fig. 1 is a schematic view of the electrode arrangement of example 1: three high density electrode arrays were placed on the upper left arm, each containing 64 electrodes (yellow), with the reference electrode placed at the elbow. Simultaneously collecting a lead II channel of standard electrocardio: the signal electrode is located at LA (right upper shoulder), and the reference electrode RL (right waist).

FIG. 2 is a block diagram of a SWT-based zero-phase filter bank;

fig. 3 is a voting scheme. Illustrated as applied to different detail coefficients d _i Is output from Pan-Tompkins. An example of a moving window is displayed in a pink rectangle. In this window, if only d is to be considered ₃ Or d ₂ This heartbeat data will be lost. But the system prevents such information from being lost.

Fig. 4 shows detail coefficients and profile coefficients of SWT and zero-phase filter sets applied to a standard ECG two lead (left) and one typical upper arm (right) signal, respectively. SWT has significant phase differences at the different layers, while the zero-phase filter bank side has no corresponding phase differences.

Fig. 5 is a zero-phase filter bank based on SWT and performance of a conventional SWT in detecting a heartbeat from the upper arm signal before voting (i.e., using only detail coefficients and profile coefficients): sensitivity (left) and forward predictive value (Positive Predictive Value) (right). The horizontal axis D value is the maximum allowable value (ms) of the difference between the detected Q wave peak position and the Q wave peak position in the standard electrocardiographic second lead. Error bars represent mean +/-standard deviation of SE (left) and PPV (right) in all subjects and trials. It is clear that when the requirements on the Q peak position are high (small D value), the performance of the zero-phase filter bank is significantly higher than with the conventional SWT method.

Fig. 6 is a graph showing the performance of SWT-based zero-phase filter banks and conventional SWTs in detecting heartbeats from upper arm signals after voting: sensitivity (left) and forward predictive value (Positive Predictive Value) (right). The horizontal axis D value is the maximum allowable value (ms) of the difference between the detected Q wave peak position and the Q wave peak position in the standard electrocardiographic second lead. Error bars represent mean +/-standard deviation of SE (left) and PPV (right) in all subjects and trials. It is clear that when the requirements on the Q peak position are high (small D value), the performance of the zero-phase filter bank is significantly higher than with the conventional SWT method. At the same time, the performance is also significantly improved compared to the results in fig. 6 (before voting).

Detailed Description

It should be noted that, in the embodiments, algorithms of steps such as data acquisition, transmission, storage, and processing, which are not specifically described, and hardware structures, circuit connections, and the like, which are not specifically described may be implemented through the disclosure of the prior art.

Example 1 apparatus for acquiring an electrocardiogram at a non-clinical Standard position

The apparatus of this embodiment includes a surface electrophysiological signal detection device including a plurality of electrodes for placement at non-clinical standard locations of a human body to detect surface electrophysiological signals, and a data processing system. The non-clinical standard location may be selected from at least one of the upper arm, wrist, behind the ear, in the ear, hip, leg, or foot. As a preferred embodiment, three electrode placement positions are shown in fig. 1: 1) The front of the biceps brachii of the left arm is the same height as the root of the deltoid; 2) The lateral midline of the upper arm, as high as position 1); 3) Medial midline of upper arm, level with position 1).

The data processing system includes:

an input module for inputting surface electrophysiological signal data; in some preferred embodiments, the surface electrophysiological signal data is collected from the upper limb, lower limb, behind the ear, in the ear, hip, shoulder, or waist;

the characteristic extraction module is used for processing the surface electrophysiological signal data by adopting a zero-phase filter based on stable wavelet transformation to obtain 4 to 7 layers of wavelet detail coefficients and 1 profile coefficient;

and the voting module is used for carrying out QRS complex detection on all wavelet detail coefficients and obtaining heartbeat detection results from all wavelet detail coefficients through a weight voting scheme.

The method for processing the surface electrophysiological signal by adopting the system specifically comprises the following steps:

step 1, acquiring surface electrophysiological signals from the surface of human skin;

step 2, processing the surface electrophysiological signal data by adopting a zero-phase filter based on stable wavelet transformation to obtain a plurality of layers of wavelet detail coefficients and profile coefficients;

the method specifically comprises the following steps:

step 2.1, adopting second-order Daubechies wavelet transformation as a mother wavelet and 4-7 layers of decomposition to obtain 4-7 wavelet detail coefficients and 1 profile coefficient;

step 2.2, according to fig. 2, the phase shift is removed at each layer of wavelet decomposition using the following steps:

1) Firstly, standard wavelet filtering is carried out, and the formula is as follows:

X(z)H(z)＝D(z)

wherein z is a z-transform operator, z transform of X (z) signal, H (z) is wavelet detail function of the layer, and D (z) is coefficient of the layer;

2) The time reverse order operation, the equivalent formula in the z-transform domain is:

3) The output of the second step is subjected to the same filtering operation as the first step again:

4) And (3) performing time reverse operation on the output of the third step:

d' (z) is the last zero phase shift wavelet detail coefficient (a.d) for that layer;

the same is then done for the profile coefficients by the wavelet profile function G (z) of this layer.

And 3, performing QRS complex detection on each layer of wavelet detail coefficient through a Pan-Tompkins algorithm, and obtaining a heartbeat detection result from each layer of wavelet detail coefficient through a weight voting scheme.

The voting scheme comprises the following specific contents:

the QRS complex detection results of all wavelet detail coefficients are made to obtain a vote with the same weight, and the vote determines whether heartbeat is detected or not;

the wavelet detail coefficient of each layer obtains a voting sparsity with corresponding weight on the QRS complex detection result, the final result of the weighting of the voting coefficients of all layers is determined to be the final QRS detection result, and the voting process is as follows:

a moving window with the length of 200ms is used on the detected heartbeat sequence, and if the weighting coefficient result in the window exceeds a preset threshold value, the heartbeat is detected; the position of the detected heartbeat is set to be the average of the time positions of all forward votes; the weighting coefficients can be set in a mode of equal weight, and can also be set according to information contained in each layer of wavelet detail coefficients;

if at least two heartbeats are detected with a time interval of less than 200ms, they are combined into one heartbeat by a time-averaging method.

Specific examples are shown in FIG. 3, which illustrates the application of different detail coefficients d _i Is output from Pan-Tompkins. An example of a moving window is displayed in a pink rectangle. In this window, if only d is to be considered ₃ Or d ₂ This heartbeat data will be lost. But the system prevents such information from being lost.

The technical scheme of the invention is further described through experiments, and the steps which are not specifically described are the same as those of the embodiment 1.

Experimental example 1 SWT-based zero-phase Filter banks in combination with voting strategy to detect heartbeats from the upper arm of the human body

1. Experimental method

1. Data set description

The dataset of this study was taken from 9 participants (all men, 24-34 years old) for 5 minutes, each recording 193 (64 x 3+1) channels of data. All participants had no heart disease. As shown in fig. 1, the electrode includes one channel at the right shoulder and three high density 64 channel grids placed around the upper arm. The multiple channel grids are spaced apart from one another at equal intervals, and the reference is located at the elbow. The reference point for the right shoulder electrode is placed on the left hip, corresponding to standard ECG lead II. Each participant signed a written informed consent prior to the start of the experiment. The experiment obtained approval from the university of sliding iron university research ethics office

The participant is required to sit on a comfortable chair with the left arm resting on the chair. Data were then recorded simultaneously from 193 channels using an EMG-USB2+ biosignal amplifier and a hardware bandpass filter of 0.1Hz to 500 Hz. The sampling frequency of these recordings is 2048Hz. The data were then low pass filtered using a Butterworth filter with a cut-off frequency of 100 Hz. Next, the data is downsampled, and the sampling frequency is reduced to 256Hz.

The data of this experimental example are collected from the upper arm, belonging to the scene of extracting the electrocardiographic activity in the higher noise environment.

2. Data processing

Experimental group: the procedure and system used in this section were the same as in example 1.

Comparison group: stationary Wavelet Transform (SWT) standard in the prior art was used.

3. Evaluation and statistical analysis Sensitivity (SE) and Positive Predictive Value (PPV)

The formula is as follows:

where TP is true positive, being the number of beats detected (from the lead II ECG) within a range of intervals for which the actual beat is less than a predefined time interval. The time interval (D) takes a value in the range from 50ms to 20ms, with a duration of 10ms. FN is false negative, indicating the number of undetected (missed) heartbeats, FP indicating the number of erroneously detected heartbeats.

The ability of the algorithm to detect heart beats is denoted by SE. PPV, on the other hand, demonstrates the accuracy of the detection.

The data for each participant (240 seconds) was divided into 12 trials of 20 seconds. The difference in the performance vectors (SE and PPV) of the two algorithms of length 96 (8×12=96) was therefore compared using t-test under different conditions.

2. Experimental results

As shown in fig. 4, the left graph shows the results of applying the original SWT and the processing result of step 3 of example 1 to the ECG, the detail and approximation coefficients of the SWT when applied to the ECG signal, and comparing them with the coefficients of the SWT-based zero-phase filter bank. As can be seen from the figure, this filter set prevents information loss due to zero delay, especially in the higher levels where there is an accumulated delay. The same comparison was then made for the selected channel on the upper arm in the right column of fig. 4. The comparison shows that the delay is eliminated through the process of step 3 in example 1, and that all wavelet detail coefficient scales achieve synchronization with the ECG.

Statistical analysis of SE and PPV was performed:

the performance of the zero-phase filter bank used in step 3 of example 1 was compared to the original SWT before applying the voting scheme. The results for all scales and four time intervals (d=20, 30, 40, 50 ms) are shown in fig. 5. It can be seen that the zero-phase filter bank has a much higher performance than SWT on the fourth and fifth scale in all time intervals. In more challenging cases, the difference is more pronounced when D becomes lower.

At d=50 ms, D4 results in SE equal to 0.82 and 0.97 for SWT and zero-phase filter banks, respectively, which performance decreases with decreasing D. At the most challenging time interval (d=20 ms), SE becomes 0.14vs.0.56, with zero-phase filter banks being better. On the other hand, D5 already has a significant advantage (0.11 vs. 0.74) for a zero-phase filter bank when d=50 ms. When d=20 ms, this difference becomes 0.02vs.0.48. This means that in the most challenging case, almost no heart beat is detected in the final detail of the SWT, and almost half of the heart beat is detected from the zero-phase filter bank. For D1 to D3, there is no significant difference when D is set to 40ms and 50 ms. However, when D is set to 20ms and 30ms, the zero-phase filter bank achieves a significant advantage in terms of performance of D3. The PPV also follows the same trend as SE in all time intervals.

Fig. 6 shows the results of the voting scheme using d1 to d4 of SWT and zero-phase filter bank. The zero-phase filter bank is significantly elevated in SE and PPV in all Ds (p <0.05 at d=50 ms, p <0.01 at d=20, 30, 40 ms). It follows that the zero-phase filter bank is more robust in detection applications to reduce the acceptable spacing from the heartbeat.

Comparing the voting scheme results in the zero-phase filter bank with the best individual a.d before voting, at d=50 ms, SE of a.d4 is lower (0.98 vs. 0.97) but PPV is higher (0.95 vs.

0.96). When D is set to 40ms, SE becomes 0.94vs.0.96 and the voting scheme performs better. The voting scheme also has an advantage in PPV, providing a parameter of 0.93vs.0.94. At d=30 ms, the voting scheme increases SE (0.86 vs. 0.87) but decreases PPV (0.84 vs. 0.85) compared to a.d4. The voting scheme has the greatest advantage when D is set to 20 ms. It significantly increased SE (0.67 vs.0.75) and PPV (0.67 vs.0.73) compared to a.d3.

From the above experimental results, it can be seen that, by using the method of the present invention, heartbeat is detected at intervals of 50ms, and SE and PPV are respectively 0.98±0.04 and 0.95±0.09; SE was 0.96±0.07, 0.87±0.12, and 0.75±0.15 when the intervals were 40, 30, and 20ms, respectively. The same trend is followed for PPV, which is 0.94±0.07, 0.84±0.14, and 0.73±0.16, respectively, at intervals of 40, 30, and 20ms from the actual heartbeat interval. Furthermore, the data processing system SE and PPV of the present invention both perform better than the original SWT. This shows that the device of the invention has good accuracy and robustness in detecting high noise data.

According to the embodiment and experimental example, the device can accurately detect the QRS in the high-noise environment, achieves the purposes of acquiring the electrocardiogram at a non-clinical standard position and accurately detecting the heartbeat, has strong applicability to application scenes such as detecting the electrocardiogram by the upper arm, and has good application prospect.

Claims (8)

1. An apparatus for acquiring an electrocardiogram, comprising: a surface electrophysiological signal detection device and a data processing system;

the data processing system includes:

an input module for inputting surface electrophysiological signal data;

the characteristic extraction module is used for processing the surface electrophysiological signal data by adopting a zero-phase filter based on stable wavelet transformation to obtain a plurality of layers of wavelet detail coefficients and profile coefficients;

and the voting module is used for carrying out QRS complex detection on all wavelet detail coefficients and obtaining heartbeat detection results from all wavelet detail coefficients through a weight voting scheme.

2. The apparatus of claim 1, wherein: the surface electrophysiological signal detection device is a device for detecting surface electrophysiological signals from a non-clinical standard location.

3. The apparatus of claim 2, wherein: the non-clinical standard locations include upper limbs, lower limbs, behind the ear, in the ear, hip, shoulder or waist.

4. A device according to claim 3, characterized in that: the surface electrophysiological signal detection device comprises three electrodes for being arranged at the following positions of the subject when detecting the surface electrophysiological signal:

1) The front of the biceps brachii of the left arm is the same height as the root of the deltoid;

2) The lateral midline of the upper arm, as high as position 1);

3) Medial midline of upper arm, level with position 1).

5. The apparatus of claim 1, wherein: the feature extraction module is used for executing the following steps:

step 2.1, adopting second-order Daubechies wavelet transformation as a mother wavelet and 4-7 layers of decomposition to obtain 4-7 wavelet detail coefficients and 1 profile coefficient;

step 2.2, remove the phase shift at each layer of wavelet decomposition with the following steps:

1) Firstly, standard wavelet filtering is carried out, and the formula is as follows:

X(z)H(z)＝D(z)

wherein z is a z-transform operator, z transform of X (z) signal, H (z) is wavelet detail function of the layer, and D (z) is coefficient of the layer;

2) The time reverse order operation, the equivalent formula in the z-transform domain is:

3) The output of the second step is subjected to the same filtering operation as the first step again:

4) And (3) performing time reverse operation on the output of the third step:

d' (z) is the last zero phase shift wavelet detail coefficient for that layer;

the same is then done for the profile coefficients by the wavelet profile function G (z) of this layer.

6. The apparatus of claim 5, wherein: the wavelet detail coefficients are 5 layers, and the frequency content of the 5 layers of wavelet detail coefficients is 64-128Hz, 32-64Hz, 16-32Hz, 8-16Hz and 4-8Hz respectively.

7. The apparatus of claim 1, wherein: in the voting module, the method for detecting the QRS complex is a Pan-Tompkins algorithm.

8. The apparatus of claim 1, wherein: in the voting module, the specific content of the voting scheme is as follows:

the detail coefficients of each layer obtain a voting sparsity of corresponding weight on the QRS complex detection result, the final result of the weighting of the voting coefficients of all layers is the final QRS detection result, and the voting process is as follows:

a moving window with the length of 200ms is used on the detected heartbeat sequence, and if the weighting coefficient result in the window exceeds a preset threshold value, the heartbeat is detected; the position of the detected heartbeat is set to be the average of the time positions of all forward votes;

if at least two heartbeats are detected with a time interval of less than 200ms, they are combined into one heartbeat by a time-averaging method.

Images