Multi-lead heart rate monitoring method and wearable device

Abstract

The invention provides a multi-lead heart rate monitoring method and wearable device.The multi-lead wearable device collects signals and conducts band-pass filtering, then signal quality judgment is conducted on electrocardiosignals of each lead, segments meeting the signal quality are effective signal segments, and segments not meeting the signal quality are interference signal segments; selecting the effective signal sections in each lead, and combining the effective signal sections into a continuous standard electrocardiosignal by taking time as length; and finally, performing QRS wave group detection on the standard electrocardiosignal, and obtaining a heart rate value of the standard electrocardiosignal. According to the method, the heart rate monitoring accuracy and the calculation efficiency can be effectively improved, and the high-real-time electrocardiogram monitoring requirements of all scenes are met.

Description

Technical Field

[0001] The present invention relates to the field of human physiological signal monitoring, and in particular to a multi-lead heart rate monitoring method and a wearable device. Background Art

[0002] As a comprehensive manifestation of cardiac electrical activity on the body surface, the human electrocardiogram (ECG) contains a wealth of physiological and pathological information reflecting cardiac rhythm and its electrical conduction, and has a very high application value in evaluating cardiac function and diagnosing heart disease. Nowadays, with the development of smart electronic devices, more and more wearable devices that can monitor human physiological signals have appeared on the market, and the acquisition of human electrocardiograms has gradually expanded from in-hospital examinations to out-of-hospital monitoring, among which wearable devices that monitor ECG are the majority, such as electrocardiogram watches.

[0003] Taking heart rate detection as an example, the QRS complex is usually detected first, and the heart rate value can be obtained based on the interval of the QRS complex. When using a single-lead device for QRS complex detection, factors such as motion interference, poor adhesion, and electrode detachment will cause interference in the ECG waveform, resulting in incorrect or missed detection of the QRS complex, thereby affecting the accuracy of heart rate calculation.

[0004] In addition, there are some multi-lead wearable devices on the market, such as electrocardiograph clothing, etc. In the prior art, the processing scheme of such devices when detecting heart rate is generally as follows: first, each lead is subjected to lead-by-lead heart rate detection and analysis, and each lead obtains a heart rate result, and then the heart rate results of each lead are merged to obtain a merged heart rate result as the final heart rate result. This method has the following shortcomings: lead-by-lead analysis greatly increases the calculation time and has low analysis efficiency; it limits the application of scenarios with high real-time requirements; when merging the analysis results of each lead, the merging algorithm strategy is complex and usually relies on thresholds set by human experience, and the quality of the results depends on human experience. Summary of the invention

[0005] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a multi-lead heart rate monitoring method and a wearable device to achieve efficient, real-time and accurate heart rate monitoring in a multi-lead situation.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A multi-lead heart rate monitoring method comprises the following steps:

[0008] Step S1. Acquire multi-lead ECG signals collected from a multi-lead wearable device and perform bandpass filtering;

[0009] Step S2 determines the signal quality of the ECG signal of each lead, and the segments that meet the signal quality requirements are valid signal segments, and the segments that do not meet the signal quality requirements are interference signal segments;

[0010] Step S3. Select the valid signal segments in each lead, and merge the valid signal segments into a continuous standard ECG signal based on the time length;

[0011] Step S4: Perform QRS wave group detection on the standard ECG signal and obtain the heart rate value of the standard ECG signal.

[0012] Furthermore, in step S2, the quality of the ECG signal of each lead is determined, specifically: 4 seconds is used as a signal determination segment, and the determination conditions include signal amplitude feature determination and signal spectrum energy feature determination. When the signal amplitude feature determination and the signal spectrum energy feature determination are both valid, the signal determination segment is determined to be valid, and the signal amplitude feature determination conditions include:

[0013] The saturation amplitude in the signal determination segment does not exceed half of the signal segment;

[0014] The proportion of sampling points whose amplitude changes of two consecutive sampling points are less than the first threshold in the signal determination section is less than a preset proportion;

[0015] The extreme value difference within the signal determination segment is lower than a second threshold value;

[0016] When all conditions for signal amplitude feature judgment are met, the signal amplitude feature judgment is valid.

[0017] Furthermore, in step S2, the quality of the ECG signal of each lead is determined, specifically: 4 seconds is used as a signal determination segment, and the determination conditions include signal amplitude feature determination and signal spectrum energy feature determination. When the signal amplitude feature determination and the signal spectrum energy feature determination are both valid, the signal determination segment is determined to be valid, and the signal spectrum energy feature determination is specifically:

[0018] Calculate the original signal spectrum, calculate the power frequency energy, and obtain the ratio of power frequency noise to the energy of the full frequency band;

[0019] Calculate the spectrum of the signal after filtering to obtain the ratio of low-frequency noise to the energy of the full frequency band;

[0020] When the ratio of the power band noise to the full-band energy and the ratio of the low-channel noise to the full-band energy are respectively less than a preset third threshold and a fourth threshold, the signal spectrum energy feature is judged to be valid.

[0021] Furthermore, when merging valid signal segments in step S3, valid signal segments are selected for merging starting from the lead where the valid signal segment appears earliest;

[0022] When the effective signal segment of the current lead is terminated, the lead with the longest effective signal segment from the next moment is preferentially selected for merging.

[0023] Furthermore, in step S3, when each lead at the current moment is an invalid signal segment, the invalid signal segment of the current lead is selected for merging to ensure that the lead at the current moment is not switched;

[0024] When each lead at the current moment is a valid signal segment, the valid signal segment of the current lead is selected for merging to ensure that the lead does not switch at the current moment.

[0025] Furthermore, in step S3, the valid signal segments are merged into a continuous standard ECG signal. When the lead is switched to select the valid signal segment to be merged, the valid signal segment is smoothed at the lead switching position to ensure waveform continuity, specifically:

[0026] Determine the smoothing length smooth_len and the lead switching time t_joint;

[0027] The smoothing range of the previous lead is pre_smooth_range = [t_joint-smooth_len, t_joint], and the smoothing range of the next lead is next_smooth_range = [t_joint, t_joint+smooth_len];

[0028] Calculate the difference between the two leads before and after t_joint:

[0029] Δy=y_pre(t_joint)-y_nexi(t_joint)

[0030] Calculate the slope k of the smoothing function:

[0031] k=0.5*Δy / smooth_len

[0032] Calculate the smoothed values ​​of the front and back leads according to the smoothing function:

[0033] y_pre_new(i)=y_pre(t)-k*(t-(t_joint-smooth_len))

[0034] y_next_new(t)=y_next(i)+k*(t-i_joint)

[0035] Among them, y_pre(t) is the value of the previous lead at time t before smoothing, y_pre_new(t) is the value of the previous lead at time t after smoothing, y_next(t) is the value of the next lead at time t before smoothing, and y_next_new(t) is the value of the next lead at time t after smoothing.

[0036] Furthermore, the step S4 is specifically as follows:

[0037] Performing normalization processing on the standard electrocardiogram signal;

[0038] Inputting the standard ECG signal into a preset convolution model in segments of 4 seconds;

[0039] Output QRS waveform position, count QRS complex intervals, and calculate heart rate.

[0040] Furthermore, the convolution model includes:

[0041] Input: Input the standard ECG signal, dimension is [1, 1024]

[0042] Conv: One-dimensional convolution, used to process sequence data.

[0043] h-swish: activation function, the mathematical expression is: h-swish(x)=x*relu6(x+3) / 6.

[0044] DepthSepConv: consists of a depthwise separable convolutional layer and a SE attention layer.

[0045] DW: Depth-wise convolution, performing convolution operations on each input channel separately.

[0046] PW: Point-by-point convolution, performing convolution operation on each point of the input feature map.

[0047] SE: Channel attention layer.

[0048] GAP: Global average pooling, which performs global average pooling on each channel of the input feature map.

[0049] FC: Fully connected layer. For each input feature, the fully connected layer will perform a linear combination with all input features through the weight matrix and add a bias vector to obtain the output feature.

[0050] On the other hand, a computer-readable medium is provided, on which instructions executable by a processor are stored. When the instructions are executed by the processor, the processor performs the multi-lead heart rate monitoring method as described above.

[0051] On the other hand, the wearable device is a multi-lead wearable device, comprising:

[0052] At least 3 electrodes and 1 terminal device, the electrodes contact the human body surface to collect ECG signals and are connected to the terminal through electrode wires;

[0053] The terminal device is equipped with a data acquisition module and a heart rate monitoring module, which are used to collect ECG data of each lead and monitor the heart rate through the multi-lead heart rate monitoring method as described above;

[0054] The wearable device is an electrocardiograph suit or a heart rate belt.

[0055] The beneficial effects of the present invention are:

[0056] Merging the effective signal segments of multiple leads before performing heart rate detection can accurately improve the detection accuracy compared to a single lead; on the other hand, compared with the multi-lead heart rate monitoring method in the prior art, the strategy of merging the effective signal segments and then detecting the QRS complex effectively improves the efficiency of heart rate monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0058] Figure 1 A schematic diagram of a lead merging strategy in a multi-lead heart rate monitoring method;

[0059] Figure 2 This is a schematic diagram of the result of merging the real ECG data using the multi-lead heart rate monitoring method;

[0060] Figure 3 This is a schematic diagram of smoothing when merging standard ECG signals;

[0061] Figure 4 for Figure 3 The enlarged view of point A in the middle;

[0062] Figure 5 This is a waveform comparison chart before and after the actual data is merged;

[0063] Figure 6 This is a comparison chart of the waveforms before and after smoothing;

[0064] Figure 7 Schematic diagram of the preset convolution model for QRS complex detection

[0065] Figure 8A schematic diagram of a wearable device according to one embodiment of the present invention;

[0066] Fig. 9 FIG. 4 is a schematic diagram of the interior of a wearable device according to one embodiment of the present invention. DETAILED DESCRIPTION

[0067] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field of the present invention; the terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. For example, the directions or positions indicated by the terms "length", "width", "up", "down", "left", "right", "front", "back", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc. are based on the directions or positions shown in the drawings, which are only for the convenience of description and should not be understood as limitations on the present technical solution.

[0068] The terms "including" and "having" and any variations thereof in the specification and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions; the terms "first", "second", etc. in the specification and claims of the present invention or the above-mentioned drawings are used to distinguish different objects rather than to describe a specific order. "Multiple" means two or more, unless otherwise clearly and specifically defined.

[0069] In the specification and claims of the present invention and the above-mentioned drawings, when an element is referred to as being "fixed to" or "mounted on" or "disposed on" or "connected to" another element, it may be directly or indirectly located on the other element. For example, when an element is referred to as being "connected to" another element, it may be directly or indirectly connected to the other element.

[0070] In addition, reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0071] Example

[0072] As a comprehensive manifestation of cardiac electrical activity on the body surface, the human electrocardiogram (ECG) contains a wealth of physiological and pathological information reflecting cardiac rhythm and its electrical conduction, and has a very high application value in evaluating cardiac function and diagnosing heart disease. Nowadays, with the development of smart electronic devices, more and more wearable devices that can monitor human physiological signals have appeared on the market, and the acquisition of human electrocardiograms has gradually expanded from in-hospital examinations to out-of-hospital monitoring, among which wearable devices that monitor ECG are the majority, such as electrocardiogram watches.

[0073] Taking heart rate detection as an example, the QRS complex is usually detected first, and the heart rate value can be obtained based on the interval of the QRS complex. When using a single-lead device for QRS complex detection, factors such as motion interference, poor adhesion, and electrode detachment will cause interference in the ECG waveform, resulting in incorrect or missed detection of the QRS complex, thereby affecting the accuracy of heart rate calculation.

[0074] In addition, there are some multi-lead wearable devices on the market, such as electrocardiograph clothing, etc. In the prior art, the processing scheme of such devices when detecting heart rate is generally as follows: first, each lead is subjected to lead-by-lead heart rate detection and analysis, and each lead obtains a heart rate result, and then the heart rate results of each lead are merged to obtain a merged heart rate result as the final heart rate result. This method has the following shortcomings: lead-by-lead analysis greatly increases the calculation time and has low analysis efficiency; it limits the application of scenarios with high real-time requirements; when merging the analysis results of each lead, the merging algorithm strategy is complex and usually relies on thresholds set by human experience, and the quality of the results depends on human experience.

[0075] In order to achieve efficient, real-time and accurate heart rate monitoring in a multi-lead situation, this embodiment provides a multi-lead heart rate monitoring method, including the following steps:

[0076] Step S1. Acquire multi-lead ECG signals collected from a multi-lead wearable device and perform bandpass filtering;

[0077] Step S2 determines the signal quality of the ECG signal of each lead, and the segments that meet the signal quality requirements are valid signal segments, and the segments that do not meet the signal quality requirements are interference signal segments;

[0078] Step S3. Select the valid signal segments in each lead, and merge the valid signal segments into a continuous standard ECG signal based on the time length;

[0079] Step S4: Perform QRS wave group detection on the standard ECG signal and obtain the heart rate value of the standard ECG signal.

[0080] Traditional heart rate monitoring usually directly performs QRS wave group detection on the ECG signal collected by the wearable device and further performs heart rate detection. If there are signals from multiple leads, QRS wave group monitoring is performed on each lead and the results are merged. Compared with the traditional method, the idea of ​​this embodiment is to integrate the ECG signals collected by each lead to form a standard ECG signal. Since the electrodes of the wearable device are not always in stable contact with the skin, it may happen that the I lead signal is good at a certain moment, while the V2 lead has poor signal quality due to poor contact with the body surface. At this time, the signal of the I lead is selected. As a valid signal segment, similarly, when the signal quality of lead I is poor at the next moment, other leads with good signal quality are selected as the merged segment of the standard ECG signal, and then the merged standard ECG signal is subjected to QRS wave group detection. This method can ensure that the impact of interference signals on ECG monitoring is minimized and the accuracy of heart rate monitoring is improved. In addition, the traditional method requires QRS wave group detection for the ECG signal of each lead, while this method only performs QRS wave group detection on the merged standard ECG signal, avoiding repeated calls of the algorithm and improving the efficiency of heart rate monitoring to meet the real-time requirements in various scenarios.

[0081] It should be noted that the standard ECG signal described in the present invention is named to facilitate technical personnel in the field to understand the technical solution. It is a term used only in the present invention and is not the gold standard in the field of modern medical ECG detection.

[0082] In some embodiments, the bandpass filtering range in step S1 is (0.5 Hz to 40 Hz) to remove interference outside the passband.

[0083] In some embodiments, for the bandpass filtering algorithm, different frequency band ranges and filter types, such as FIR, IIR, frequency domain filtering, etc., can be selected according to different scenarios.

[0084] In some embodiments, in step S2, the quality of the ECG signal of each lead is determined, specifically: 4 seconds is used as a signal determination segment, and the determination conditions include signal amplitude feature determination and signal spectrum energy feature determination. The reason why the length of the signal determination segment is 4 seconds is that when the heart rate value is low, such as 30 beats / minute, at least 2 heart beats can be collected, which has the significance of signal quality determination. Those skilled in the art can also change the length of the signal determination segment according to actual sampling conditions and actual sampling experience. The principle is that a signal determination segment can include at least 2 heart beats. When the signal amplitude feature determination and the signal spectrum energy feature determination are both valid, the signal determination segment is determined to be valid. The signal amplitude feature determination conditions include:

[0085] The saturation amplitude in the signal determination segment does not exceed half of the signal determination segment;

[0086] The proportion of sampling points whose amplitude changes of two consecutive sampling points are less than the first threshold in the signal determination section is less than a preset proportion;

[0087] The extreme value difference within the signal determination segment is lower than a second threshold value;

[0088] For condition 1, the acquisition hardware of the wearable device has saturation. When the acquired signal waveform exceeds the saturation amplitude of the hardware acquisition signal, the signal at this time cannot show the characteristics of the ECG signal. When the proportion of the saturation amplitude in the signal determination segment exceeds half, the signal quality of this signal cannot meet the conditions for QRS complex detection.

[0089] For condition 2, when the wearable device is not worn or the wearable device has poor contact, the original signal hardly changes. This condition can rule out the possibility that the wearable device electrode has fallen off.

[0090] For condition three, when the human body's movement fluctuations are too large, a large noise signal will be generated. This condition is to eliminate the interference signal caused by such noise signals.

[0091] When all conditions for signal amplitude feature judgment are met, the signal amplitude feature judgment is valid.

[0092] In some embodiments, in step S2, the quality of the ECG signal of each lead is determined, specifically: 4 seconds is used as a signal determination segment, and the determination conditions include signal amplitude feature determination and signal spectrum energy feature determination, and the signal spectrum energy feature determination is specifically:

[0093] Calculate the original signal spectrum, calculate the power frequency energy respectively, and obtain the ratio of power frequency noise to the energy of the full frequency band;

[0094] Calculate the spectrum of the signal after filtering to obtain the ratio of low-frequency noise to the energy of the full frequency band;

[0095] When wearing a wearable device, basically no power frequency noise is collected. Therefore, when the proportion of power frequency noise is too large, it may not be worn accurately, and the signal at this time cannot be used as a valid signal segment; at the same time, too high a proportion of low-frequency noise will also affect the effectiveness of the ECG signal after filtering; when the ratio of the power frequency band noise to the full-band energy and the ratio of the low-channel noise to the full-band energy are respectively less than the third threshold and the fourth threshold, the signal spectrum energy characteristics are judged to be valid.

[0096] It should be noted that the first threshold, the second threshold, the third threshold and the fourth threshold described in this embodiment are all empirical thresholds. Those skilled in the art can adjust the thresholds by themselves according to the sensitivity and specificity of signal quality judgment. In this embodiment, the actual values ​​of the thresholds are not specifically limited.

[0097] Preferably, when merging valid signal segments in step S3, valid signal segments are selected for merging starting from the lead where the valid signal segment appears earliest;

[0098] In order to prevent frequent switching of leads for merging, when selecting valid signal segments, when the valid signal segment of the current lead is terminated, the lead with the longest valid signal segment from the next moment is preferentially selected for merging. Figure 1 Shown are valid and / or invalid segments of ECG signals of leads I, II, and III and the merged standard ECG signal segments. The shaded part in the figure is the invalid signal segment, and the dotted line position is the merge point. First, the valid signal segment of lead I is selected to start merging. At the first moment, an invalid signal segment appears in lead I, and leads II and III are both valid signal segments, but the valid signal segment of lead II is longer. Frequent switching of leads is not avoided. At the next moment, the valid signal segment of lead II is selected for merging, and so on, to finally obtain the merged standard ECG signal.

[0099] Preferably, in step S3, when each lead at the current moment is an invalid signal segment, the invalid signal segment of the current lead is selected for merging to ensure that the lead at the current moment is not switched;

[0100] When each lead is a valid signal segment at the current moment, the valid signal segment of the current lead is selected for merging to ensure that the lead does not switch at the current moment.

[0101] refer to Figure 1 At the third moment, when leads I, II, and III are all invalid signal segments, the leads will not be switched, and lead III will still be selected as the merged segment. Until the fourth moment, when a valid signal segment appears in lead I, lead I is switched and the valid signal segment of lead I is selected for merging again.

[0102] In some embodiments, when merging valid signal segments, when multiple leads are valid at the same time at the current moment, the signal quality of the lead can be graded and scored, or the signal quality of the valid signal segments can be further refined and evaluated based on the confidence level of the reference signal quality, and the valid signal segment with the highest signal quality can be selected as the merged lead.

[0103] like Figure 2 The figure shows the merging process of the actual collected ECG data by the scheme recorded in the above embodiment. It can be seen that each segment of the merged standard ECG signal is taken from the effective signal segment with the best signal quality in each lead. The merged standard ECG signal has a clear QRS complex and almost no interference, and the accuracy of heart rate monitoring can be significantly improved.

[0104] In some embodiments, because the merged standard ECG signal waveforms are derived from different leads, when lead switching occurs, discontinuity of the merged waveform may be caused. Therefore, it is necessary to smooth the waveform at the lead switching position to ensure signal continuity. In step S3, the valid signal segments are merged into a continuous standard ECG signal. When the lead is switched and valid signal segments are selected for merging, the valid signal segments are smoothed at the lead switching position to ensure waveform continuity. Specifically,

[0105] Determine the smoothing length smooth_len and the lead switching time t_joint;

[0106] The smoothing range of the previous lead is pre_smooth_range = [t_joint-smooth_len, t_joint], and the smoothing range of the next lead is next_smooth_range = [t_joint, t_joint+smooth_len];

[0107] Calculate the difference between the two leads before and after t_joint:

[0108] Δy=y_pre(t_joint)-y_nexi(t_joint)

[0109] Calculate the slope k of the smoothing function:

[0110] k=0.5*Δy / smooth_len

[0111] Calculate the smoothed values ​​of the front and back leads according to the smoothing function:

[0112] y_pre_new(i)=y_pre(t)-k*(t-(t_joint-smooth_len))

[0113] y_next_new(t)=y_next(i)+k*(t-i_joint)

[0114] Among them, y_pre(t) is the value of the previous lead at time t before smoothing, y_pre_new(t) is the value of the previous lead at time t after smoothing, y_next(t) is the value of the next lead at time t before smoothing, and y_next_new(t) is the value of the next lead at time t after smoothing.

[0115] The core of the smoothing process is to smoothly transition the waveforms of the two leads before and after the merger. Therefore, the waveforms of the front and rear leads with a fixed length before and after the t_joint moment are taken as the smoothing range. The front and rear leads are smoothed by the length of smooth_len respectively, and transition to half of the difference between the front and rear leads at the t_joint moment respectively.

[0116] like Figure 3 , Figure 4 and Figure 6 As shown, the marked points in the figure are the splicing points, that is, the points at the t_joint moment. Figure 4 As shown in FIG. 1 , the waveform after processing is more stable and smooth.

[0117] Figure 5 The figure shows the waveform comparison of the four-lead ECG data before and after merging. It can be seen from the figure that there is an obvious interference signal segment in the ECG signal before merging, which will affect the subsequent QRS complex detection. The waveform after merging shows that the signal of the original II lead is merged into the interference signal segment, and the interference signal segment is replaced by the effective signal segment, making the subsequent QRS complex detection more accurate and efficient.

[0118] In some embodiments, step S4 is specifically:

[0119] Performing normalization processing on the standard electrocardiogram signal;

[0120] The standard ECG signal is input into the preset convolution model with a segment of 4 seconds, and the waveform data of a fixed length is input. The QRS wave group detection result is only related to the current sample and has no pre-dependency. The impact of switching the combined waveform lead is greatly reduced;

[0121] Output QRS waveform position, count QRS complex intervals, and calculate heart rate.

[0122] The convolution model is pre-trained with a training data set, which can cover 12-lead ECG data, and different lead waveforms can cover the training data set. The data comes from holter data, public data set data, and ECG data collected by different wearable devices. The data sources are wide and the distribution is scientific and reasonable.

[0123] In this embodiment, a specific convolution model is used to perform QRS complex detection on the combined standard ECG signal. In actual implementation, other methods can also be used for QRS complex detection, but it should be noted that the standard ECG signals are derived from the combination and integration of multiple leads. Therefore, in terms of waveform morphology, the standard ECG signal may show multiple forms and different amplitude differences. Therefore, when performing QRS waveform detection, the QRS complex detection algorithm must be insensitive to the signal amplitude and morphological characteristics, and different amplitudes and forms will not have a negative impact on the accuracy of the final detection result. To this end, it is necessary to design a QRS complex detection algorithm model that is insensitive to waveform amplitude, etc. as follows:

[0124] like Figure 7 As shown, the convolution model includes: input: input the standard ECG signal, the dimension is [1, 1024]

[0125] Conv: One-dimensional convolution is an extension of convolutional neural network (CNN) on one-dimensional data, mainly used to process sequence data.

[0126] h-swish: activation function, the mathematical expression is: h-swish(x)=x*relu6(x+3) / 6.

[0127] DepthSepConv: consists of a depthwise separable convolutional layer and a SE attention layer.

[0128] DW: Depth-wise convolution, performing convolution operations on each input channel separately.

[0129] PW: Point-by-point convolution, performing convolution operation on each point of the input feature map.

[0130] SE: Channel attention layer, divided into two steps: Squeeze and Excitation. Squeeze is to extract the global information of each channel and describe the importance of each channel with a scalar; Excitation is to learn the weight of each channel, that is, the attention value of the channel.

[0131] GAP: Global average pooling, which performs global average pooling on each channel of the input feature map, that is, averaging all elements of each channel to obtain a vector equal to the number of channels. This vector can be regarded as the global feature representation of each channel.

[0132] FC: Fully connected layer, the core is the weight matrix and bias vector. For each input feature, the fully connected layer will linearly combine all input features through the weight matrix, and add the bias vector to obtain the output feature.

[0133] In some embodiments, the order of steps of the above-mentioned multi-lead heart rate monitoring method can be appropriately adjusted to achieve the same or even better effect. For example, the quality of each lead signal can be judged based on the original signal, and then the original waveform of the valid signal segment of each lead can be selected for merging, and then the merged standard ECG signal can be band-pass filtered, and then the filtered standard ECG signal can be subjected to QRS wave group detection and heart rate calculation.

[0134] Furthermore, to further illustrate the technical effect of the present invention, this embodiment collects and tests 170 hours of ten-lead ECG data through an electrocardiograph, and compares the results of heart rate monitoring using the traditional method and the multi-lead heart rate monitoring method of this embodiment based on actual data statistics.

[0135] Among them, the traditional method is: first perform QRS wave group detection on each lead to calculate the heart rate, and then integrate the heart rate calculation results of each lead to obtain the final result.

[0136] The comparison results are as follows:

[0137]

[0138] It can be obtained that, under the premise that there is no significant difference in accuracy, the method of the present invention improves the overall effectiveness of the calculation results by 1.5% and the calculation efficiency by 27.45%.

[0139] The calculation method of computing efficiency is data duration / computation time.

[0140] On the other hand, a computer-readable medium is provided, on which instructions executable by a processor are stored. When the instructions are executed by the processor, the processor executes the multi-lead heart rate monitoring method as described in the above embodiment.

[0141] The computer-readable medium can be mounted on a wearable device or on a server, that is, the multi-lead heart rate monitoring method can implement real-time calculations on a wearable device or on a server, and finally return the results to the wearable device terminal.

[0142] On the other hand, a wearable device is provided, wherein the wearable device is a multi-lead wearable device, comprising:

[0143] At least 3 electrodes and 1 terminal device, the electrodes contact the human body surface to collect ECG signals and are connected to the terminal through electrode wires;

[0144] The terminal device is equipped with a data acquisition module and a heart rate monitoring module for collecting ECG data of each lead and monitoring the heart rate;

[0145] The wearable device is an electrocardiograph suit or a heart rate belt.

[0146] like Figure 8 and Fig. 9 The figure shows an optional embodiment, an electrocardiograph garment, which includes eight electrodes and can collect ten-lead electrocardiograph signals.

[0147] In some embodiments, the multi-lead ECG waveforms collected by the wearable device can be displayed on the APP interface of mobile terminals such as mobile phones and tablets. After being processed by the multi-lead heart rate monitoring method of the above embodiment, the heart rate value is displayed in real time on the APP interface of the mobile terminal.

[0148] On the other hand, the standard ECG signal merged by the method of this embodiment is equivalent to a single-lead ECG signal, and its waveform can also be displayed on the APP interface of the mobile terminal, and subsequent ECG signal analysis can be performed. Since the rhythms of multiple leads at the same time are aligned, the subsequent analysis of the standard ECG signal is not limited to ECG monitoring, but can also be extended to other applications, such as HRV heart rate variability analysis, arrhythmia identification, atrial fibrillation detection, etc.

[0149] In actual implementation, the patient user is required to wear the wearable device and bind the device to the patient user's own account on the mobile terminal APP. After starting data collection, the ECG signals collected by the wearable device and the monitored heart rate values ​​can be displayed on the mobile terminal APP for the patient and the doctor to view, and an ECG report can also be generated; further, the patient and / or doctor can be warned when the patient has a fast or slow heart rate, an irregular heartbeat, or atrial fibrillation. The warning methods include vibration of the wearable device and / or mobile terminal, and the mobile terminal sending a warning report.

[0150] The above embodiments are preferred implementation schemes of the present invention. In addition, the present invention may also be implemented in other ways. Any obvious replacement without departing from the concept of the present technical solution is within the protection scope of the present invention.

Claims

1. A multi-lead heart rate monitoring method, comprising the following steps: Step S1. Acquire multi-lead ECG signals collected from a multi-lead wearable device and perform bandpass filtering; Step S2 determines the signal quality of the ECG signal of each lead, and the segments that meet the signal quality requirements are valid signal segments, and the segments that do not meet the signal quality requirements are interference signal segments; Step S3. Select the valid signal segments in each lead, and merge the valid signal segments into a continuous standard ECG signal based on the time length; Step S4: Perform QRS wave group detection on the standard ECG signal and obtain the heart rate value of the standard ECG signal.

2. A multi-lead heart rate monitoring method according to claim 1, characterized in that: In step S2, the quality of the ECG signal of each lead is determined, specifically: 4 seconds is used as a signal determination segment, and the determination conditions include signal amplitude feature determination and signal spectrum energy feature determination. When the signal amplitude feature determination and the signal spectrum energy feature determination are both valid, the signal determination segment is determined to be valid, and the signal amplitude feature determination conditions include: The saturation amplitude in the signal determination segment does not exceed half of the signal determination segment; The proportion of sampling points whose amplitude changes of two consecutive sampling points are less than the first threshold in the signal determination section is less than a preset proportion; The extreme value difference within the signal determination segment is lower than a second threshold value; When all conditions for signal amplitude feature judgment are met, the signal amplitude feature judgment is valid.

3. A multi-lead heart rate monitoring method according to claim 1, characterized in that: In step S2, the quality of the ECG signal of each lead is determined, specifically: 4 seconds is used as a signal determination segment, and the determination conditions include signal amplitude feature determination and signal spectrum energy feature determination. When the signal amplitude feature determination and the signal spectrum energy feature determination are both valid, the signal determination segment is determined to be valid. The signal spectrum energy feature determination is specifically: Calculate the original signal spectrum, calculate the power frequency energy, and obtain the ratio of power frequency noise to the energy of the full frequency band; Calculate the spectrum of the signal after filtering to obtain the ratio of low-frequency noise to the energy of the full frequency band; When the ratio of the power band noise to the full-band energy and the ratio of the low-channel noise to the full-band energy are respectively less than a preset third threshold and a fourth threshold, the signal spectrum energy feature is judged to be valid.

4. A multi-lead heart rate monitoring method according to claim 1, characterized in that: When merging valid signal segments in step S3, valid signal segments are selected for merging starting from the lead where the valid signal segment appears earliest; When the effective signal segment of the current lead is terminated, the lead with the longest effective signal segment from the next moment is preferentially selected for merging.

5. A multi-lead heart rate monitoring method according to claim 4, characterized in that: In step S3, when each lead at the current moment is an invalid signal segment, the invalid signal segment of the current lead is selected for merging to ensure that the lead at the current moment is not switched; When each lead is a valid signal segment at the current moment, the valid signal segment of the current lead is selected for merging to ensure that the lead does not switch at the current moment.

6. A multi-lead heart rate monitoring method according to claim 1, characterized in that: In step S3, the valid signal segments are merged into a continuous standard ECG signal. When the lead is switched to select the valid signal segment to be merged, the valid signal segments are smoothed at the lead switching position to ensure waveform continuity, specifically: Determine the smoothing length smooth_len and the lead switching time t_joint; The smoothing range of the previous lead is pre_smooth_range = [t_joint-smooth_len, t_joint], and the smoothing range of the next lead is next_smooth_range = [t_joint, t_joint+smooth_len]; Calculate the difference between the two leads before and after t_joint: Δy=y_pre(t_joint)-y_next(t_joint) Calculate the slope k of the smoothing function: k=0.5*Δy / smooth_len Calculate the smoothed values ​​of the front and back leads according to the smoothing function: y_pre_new(t)=y_pre(t)-k*(t-(t_joint-smooth_len)) y_next_new(t)=y_next(t)+k*(t-t_joint) Among them, y_pre(t) is the value of the previous lead at time t before smoothing, y_pre_new(t) is the value of the previous lead at time t after smoothing, y_next(t) is the value of the next lead at time t before smoothing, and y_next_new(t) is the value of the next lead at time t after smoothing.

7. A multi-lead heart rate monitoring method according to claim 1, characterized in that: The step S4 is specifically: Performing normalization processing on the standard electrocardiogram signal; Inputting the standard ECG signal into a preset convolution model in segments of 4 seconds; Output QRS waveform position, count QRS complex intervals, and calculate heart rate.

8. A multi-lead heart rate monitoring method according to claim 7, characterized in that: The convolutional model includes: Input: Input the standard ECG signal, dimension is [1,1024] Conv: One-dimensional convolution, used to process sequence data. h-swish: activation function, the mathematical expression is: h-swish(x)=x*relu6(x+3) / 6. DepthSepConv: consists of a depthwise separable convolutional layer and a SE attention layer. DW: Depth-wise convolution, performing convolution operations on each input channel separately. PW: Point-by-point convolution, performing convolution operation on each point of the input feature map. SE: Channel attention layer. GAP: Global average pooling, which performs global average pooling on each channel of the input feature map. FC: Fully connected layer. For each input feature, the fully connected layer will perform a linear combination with all input features through the weight matrix and add a bias vector to obtain the output feature.

9. A computer-readable medium having stored thereon instructions executable by a processor, wherein when the instructions are executed by the processor, the processor executes the multi-lead heart rate monitoring method as described in claims 1-8.

10. A wearable device, characterized in that: The wearable device is a multi-lead wearable device, comprising: At least 3 electrodes and 1 terminal device, the electrodes contact the human body surface to collect ECG signals and are connected to the terminal through electrode wires; The terminal device is equipped with a data acquisition module and a heart rate monitoring module, which are used to collect ECG data of each lead and monitor the heart rate through the multi-lead heart rate monitoring method as described in claims 1-8; The wearable device is an electrocardiogram suit or a heart rate belt.

Images