WO2020155078A1 - Method and device for monitoring arrhythmia event - Google Patents

Abstract

A method and a device for monitoring an arrhythmia event, which are applicable to the field of signal processing. The method comprises: acquiring a heart-beat original signal of an object (S101); pre-processing the heart-beat original signal to generate a heart-beat time-domain signal (S102); generating a heart-beat spectrum signal on the basis of the heart-beat time-domain signal (S103); generating a heart-beat spectrum signal cluster on the basis of the heart-beat spectrum signal (S104); and identifying an arrhythmia event on the basis of the heart-beat spectrum signal cluster (S105). The method for monitoring an arrhythmia event may be used across sensing technology, is slightly dependent on the quality of a time domain signal, and has high adaptability.

Description

Method and equipment for monitoring arrhythmia events Technical field

The invention belongs to the field of signal processing, and in particular relates to a method and equipment for monitoring arrhythmia events.

Background technique

The heart is one of the most important organs of the human body, and its main function is to provide power for blood flow. The monitoring of heartbeat characteristics is of great significance to patients. At present, heart monitoring can be achieved by monitoring the electrical activity of the heart, mechanical activity of the heart, and indirectly monitoring the heartbeat by monitoring blood pressure, pulse wave, etc. Different monitoring methods use different sensing technologies to capture signals. For example, cardiac electrical activity monitoring uses electrode sheets to collect original signals, and pulse wave monitoring uses photoelectric sensors PPG to collect original signals.

Traditional arrhythmia detection technologies are mostly based on time-domain signal analysis, relying on good time-domain signal waveforms and heartbeat detection based on this, such as ECG detection RR interval, PPG detection peak-to-peak interval, etc., and then analyze according to the heartbeat interval distribution Calculation. Extracting the heartbeat interval based on the time domain signal is the simplest and most direct method for analyzing the heartbeat distribution. The time-domain analysis method is fast, convenient and real-time, and has low requirements for storage and computing resources. It can be better applied to actual products under the condition of limited processor computing power and storage space. However, time-domain analysis relies on high-quality signals, which are easily affected by various external interferences, resulting in feature recognition errors. At the same time, the time-domain signals collected by different sensing technologies are very different, and the characteristics of various signals are different (such as ECG signals). It is very different from the PPG signal), the time domain analysis algorithm is difficult to directly apply across the sensor technology.

technical problem

The purpose of the present invention is to provide a monitoring method for arrhythmia events, a computer readable storage medium, and a monitoring device and system for arrhythmia events, aiming to solve the problem that time-domain analysis algorithms are difficult to directly apply across sensing technologies, and time-domain analysis Rely on the problem of high-quality signals.

Technical solutions

In the first aspect, the present invention provides a method for monitoring arrhythmia events, the method comprising:

Obtain the original heartbeat signal of the subject;

Preprocessing the original heartbeat signal to generate the heartbeat time domain signal;

Generate heart beat spectrum signal based on heart beat time domain signal;

Generate heart beat spectrum signal cluster based on heart beat spectrum signal;

Identify arrhythmia events based on the heart beat spectrum signal cluster.

In a second aspect, the present invention provides a computer-readable storage medium that stores a computer program that, when executed by a processor, realizes the steps of the method for monitoring arrhythmia events .

In a third aspect, the present invention provides a monitoring device for arrhythmia events, including:

One or more processors;

Memory; and

One or more computer programs, the processor and the memory are connected by a bus, wherein the one or more computer programs are stored in the memory and configured to be executed by the one or more processors When the processor executes the computer program, the steps of the method for monitoring arrhythmia events are realized.

In a fourth aspect, the present invention provides a monitoring system for arrhythmia events, including:

One or more vibration sensors; and

Monitoring equipment for arrhythmia events as described.

Beneficial effect

Because the method for monitoring arrhythmia events of the present invention generates a heart beat spectrum signal based on the heart beat time-domain signal; generates a heart beat spectrum signal cluster based on the heart beat spectrum signal; recognizes arrhythmia events based on the heart beat spectrum signal cluster. Therefore, it can be cross-sensing technology, weakly dependent on the quality of the time domain signal, and more adaptable.

Description of the drawings

Fig. 1 is a flowchart of a method for monitoring arrhythmia events according to Embodiment 1 of the present invention.

Figure 2 is a schematic diagram of the original heart beat signal waveform collected by the optical fiber sensor.

Figure 3 is a waveform diagram of the BCG signal after filtering and denoising.

Figure 4 is a schematic diagram of the time-domain signal of a heart beat.

FIG. 5 is a schematic diagram of a heart beat frequency domain spectrum signal generated based on the heart beat time domain signal shown in FIG. 4.

Figure 6 is a schematic diagram of the time-domain signal of a heart beat.

Fig. 7(a) and Fig. 7(b) are schematic diagrams of the heart beat time domain spectrum signal generated based on the heart beat time domain signal shown in Fig. 6.

Figure 8(a) is a schematic diagram of the time-domain signal of the heart beat.

Figure 8(b) is a schematic diagram of the heart beat time domain signal corresponding to the time window intercepted at the 5 time points of T=5, 6, 7, 8, 9s.

Fig. 8(c) is a schematic diagram of the frequency domain spectrum signal of the heart beat corresponding to 5 time points.

Figure 9(a) is a schematic diagram of the BCG time-domain signal waveform.

Figure 9(b) is the heart beat frequency-domain spectrum signal obtained by sequentially calculating the BCG time-domain signal waveform in Figure 9(a). Figures 10 (a), (b) and (c) are schematic diagrams of pixel mapping of the heart beat spectrum signal to generate an image spectrum.

FIG. 11 is a schematic diagram of an image spectrum corresponding to a cardiac beat spectrum signal cluster generated after pixel mapping according to the time domain waveform in FIG. 9(a) and the frequency domain spectrum signal waveform in FIG. 9(b).

Figure 12 (a), (b), (c) are schematic diagrams of image spectra of a group of heart beat spectrum signal clusters of sinus rhythm subjects. Figure 13 (a), (b), (c) are schematic diagrams of image spectra of a group of heart beat spectrum signal clusters of patients with atrial fibrillation.

Figure 14 (a), (b), (c) are schematic diagrams of image spectra of a group of heart beat spectrum signal clusters of patients with other diseases.

FIG. 15 is a specific structural block diagram of a monitoring device for arrhythmia events provided by Embodiment 3 of the present invention.

The best mode of the invention

In order to make the objectives, technical solutions and beneficial effects of the present invention clearer, the following further describes the present invention in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain the present invention, but not to limit the present invention.

In order to illustrate the technical solution of the present invention, specific embodiments are used for description below.

Example one:

Referring to FIG. 1, the method for monitoring arrhythmia events provided in the first embodiment of the present invention includes the following steps: It should be noted that if there are substantially the same results, the method for monitoring arrhythmia events of the present invention is not shown in FIG. The sequence of the processes shown is limited.

S101. Obtain an original heartbeat signal of the subject.

The original heartbeat signal refers to all the original signals that can reflect the characteristics of the heartbeat, such as electrocardiogram (ECG) signal, phonocardiogram (Phonocardiogram, PCG) signal, seismocardiogram (SCG) signal, and shock cardiogram (Ballistocardiography, BCG) signal, Photoplethysmograph (PPG) signal, Invasive blood pressure (IBP) signal, heart beat signal monitored by radar wave, etc.

The original signal of heart beat is obtained by the sensor. When the original heartbeat signal is a BCG signal, a PCG signal or an SCG signal, the original heartbeat signal is obtained by a vibration sensor. When the original heartbeat signal is an ECG signal, the original heartbeat signal is obtained by an ECG sensor; when the original heartbeat signal is a PPG signal, the original heartbeat signal is obtained by a photoelectric sensor; when the original heartbeat signal is an IBP When the signal is used, the original heartbeat signal is obtained by an IBP signal sensor; when the original heartbeat signal is a heartbeat signal monitored by radar waves, the original heartbeat signal is obtained by a bio-radar.

In the first embodiment of the present invention, the vibration sensor may be an acceleration sensor, a speed sensor, a displacement sensor, a pressure sensor, an optical fiber sensor, or a sensor that converts physical quantities equivalently based on acceleration, speed, pressure, or displacement (for example, One or more of electrostatic charge sensitive sensors, inflatable micro-motion sensors, radar sensors, etc.).

In the first embodiment of the present invention, when the original heartbeat signal of the subject is collected by the vibration sensor, the vibration sensor can be placed under the subject's body. For example, the subject can be in a posture such as supine, prone, side-lying, semi-lying, etc., the vibration sensor can be placed on the bed, and the subject is supine (prone or side) on it. A better measurement state is that the vibration sensor is configured to contact the shoulder, back, waist or hip of the subject, for example, the vibration sensor is placed on the contact surface behind the supine human body, or the contact surface behind the supine human body at a certain tilt angle. , Wheelchairs or other objects that can lean on the contact surface behind the lying human body, etc. for collection and measurement. Generally, in order to ensure the quality of the collected signal, the measurement needs to be performed in a relatively quiet state.

Figure 2 shows a schematic diagram of the original signal waveform of the heart beat collected by the vibration sensor. The original heart beat data collected by the vibration sensor includes the respiratory signal component of the measured object, the heart beat signal component, as well as the environmental micro-vibration, the interference caused by the measured object's body movement, and the noise signal of the circuit itself. The large outline of the signal at this time is the signal envelope generated by human breathing, and the heartbeat and other interference noises are superimposed on the respiratory envelope curve.

S102: Perform preprocessing on the original heartbeat signal to generate a heartbeat time domain signal.

The original heartbeat signals obtained by different sensors contain different amounts of information. Some original heartbeat signals contain more information, so it needs to be preprocessed to capture relevant signals. For example, the original heart beat signal obtained by the vibration sensor also contains signals such as the breathing signal, body motion signal, and some inherent noise of the sensor.

In the first embodiment of the present invention, S102 may specifically include:

Perform at least one of filtering, denoising, and signal scaling on the original heartbeat signal to obtain the heartbeat time-domain signal; specifically, it can be: IIR filter, FIR filter, wavelet filter are used according to the requirements of the filtered signal characteristics , Zero-phase two-way filter, polynomial fitting smoothing filter, integral transform, differential transform one or more combinations, filtering and denoising the original heart beat signal; it can also include: judging whether the original heart beat signal carries labor If there is a high frequency interference signal, the power frequency noise is filtered through a power frequency notch filter.

Fig. 3 is a schematic diagram of the time-domain waveform after filtering and denoising the original heart beat signal obtained by the vibration sensor shown in Fig. 2. Each waveform has obvious characteristics and good consistency, regular periodicity, clear outline, and stable baseline.

S103: Generate a heart beat spectrum signal based on the heart beat time domain signal.

In the first embodiment of the present invention, S103 may specifically be: generating a heart beat frequency domain spectrum signal based on a heart beat time domain signal.

For example: Based on the time-domain signal of the heart beat, the frequency-domain spectrum signal of the heart beat is generated by the time-frequency transformation method. The time-frequency transformation method can use Fourier transform, wavelet transform, etc. If you need to refine the spectral resolution, you can also use zero padding, ZoomFFT Method, CZT transformation, Yip-ZOOM transformation, etc.; among them, the heart beat time-domain signal used for time-frequency transformation preferably contains at least two periodic waveforms;

or,

Based on the time-domain signal of the heartbeat, the frequency-domain spectrum signal of the heartbeat can be obtained by calculating the power spectrum, which can use autocorrelation method, periodogram method, windowed average periodogram method, Welch method, multi-window method, maximum entropy method, etc.;

or,

Based on the time domain signal of the heart beat, the frequency domain spectrum signal of the heart beat is obtained by calculating the AR spectrum.

Of course, in addition to Fourier transform, wavelet transform, power spectrum calculation, AR spectrum calculation, there are more ways to generate the frequency domain spectrum signal of the heart beat, which will not be repeated here.

The Fourier transform is described below:

The sampling rate of the time domain signal is generally not less than 500Hz, while the frequency domain calculation consumes a lot of storage resources and computing power. It needs to reflect the frequency characteristics within a certain period of time. Therefore, it is generally necessary to downsample and resample the time domain waveform (that is, resample). Sampling), such as extracting 500Hz into 100Hz, 62.5Hz, 50Hz, etc. After determining the resampling rate, determine the appropriate number of time-frequency transformation points according to computing resources and capabilities. Generally speaking, the more points the more accurate, but the more points the longer the original data length. A reasonable design is best to ensure that the time-domain signal of heart beat used for time-frequency conversion can contain at least two periodic waveforms. For example, assuming that the minimum heart rate measurement range is 30 bpm, it needs to contain at least 4 seconds of time-domain data. Combined with the resampling rate, the number of time-frequency transformation points can be determined. As shown in FIG. 4, it is a schematic diagram of a heart beat time domain signal, and FIG. 5 is a schematic diagram of a heart beat frequency domain spectrum signal generated based on the heart beat time domain signal shown in FIG. 4. The CZT transform can be used for spectrum refinement analysis, focusing on the frequency domain characteristics in the range of 0~5Hz, and the frequency range of interest can also be expanded or reduced according to actual needs. At this time, the high-quality BCG frequency domain waveform is presented, each effective peak has obvious characteristics, the outline is clear and upright, and the fundamental frequency multiplication characteristics are obvious. At this time, the main peak of 1.1 Hz is the fundamental frequency peak, and the subsequent obvious peaks are respectively the double frequency, the triple frequency, and the quadruple frequency. Note that the reason why the fundamental frequency peak energy is not the highest at this time is related to the filter characteristics. To suppress low-frequency interference while performing time-frequency conversion from the time-domain signal, you can filter or suppress low-frequency signals. Pay attention to filtering out low-frequency interference. At the same time, the main peak energy may be depressed. The filtering here can be performed directly in the time domain, or window function filtering can be performed in the frequency domain. Since we only care about frequencies within 0~5Hz at this time, the maximum effective multiplication of the fundamental frequency peak of 1.1Hz is around 4.4Hz. If the upper limit of the frequency is enlarged and reasonable filter parameters are set, it is possible to observe its five-fold frequency characteristic peak.

S103 may also be specifically: generating a heart beat time domain spectrum signal based on the heart beat time domain signal.

For example: based on the heart beat time domain signal through the autocorrelation function to generate the heart beat time domain spectrum signal. As shown in FIG. 6, it is a schematic diagram of a heart beat time domain signal, and FIGS. 7 (a) and (b) are schematic diagrams of a heart beat time domain spectrum signal generated based on the heart beat time domain signal shown in FIG. 6. The BCG time-domain waveform shown in Figure 6 is the data waveform in the time period of t1 (seconds); Figure 7(a) is the waveform of the bilateral autocorrelation function, the middle line is T=0, and the left and right side widths are -t1 and respectively t1, the left and right sides are symmetrical about the time axis T=0 at this time. According to the actual situation, you can only look at the unilateral waveform (T≥0 or T≤0); Figure 7(b) is the waveform of the part of T≥0. Figure 5 is a single-sided frequency domain spectrum. According to the actual situation, you can also look at both sides. It is symmetric about the frequency axis w=H(Fs)/2, and H(Fs) is the frequency corresponding to the sampling rate. At this time, you can see that the autocorrelation function waveforms in Figure 7 (a) and Figure 7 (b) have similar characteristics to the frequency domain spectrum of Figure 5, so they are called spectral signals, but they are actually not time-frequency transformed, so they are called It is the time domain spectrum signal. For the heart beat time-domain spectrum signal, the interference peak can also be suppressed, filtered or suppressed by the filtering algorithm, and the main peak energy can be amplified, highlighted or elevated.

Although the generation method of the heart beat time domain spectrum signal is inconsistent with the heart beat frequency domain spectrum signal, there is a strong characteristic similarity between the two heart beat spectrum signals. In fact, mathematically speaking, the power spectrum is equal to the Fourier of the autocorrelation function. The autocorrelation function is equal to the inverse Fourier transform of the power spectrum, so there is a strong internal connection between the two. Therefore, the following description takes the heart beat frequency domain spectrum signal as an example for expansion, and those skilled in the art can refer to the embodiment to expand and extend the heart beat time domain spectrum signal.

Of course, in addition to the autocorrelation function, other methods can be used to generate the heart beat time domain spectrum signal with similar characteristics, and even the heart beat frequency domain spectrum signal and the heart beat time domain spectrum signal can be generated based on the heart beat time domain signal, and combined The heart beat frequency domain spectrum signal and the heart beat time domain spectrum signal generate a heart beat spectrum signal. For example, the heart beat spectrum signal is generated by superimposing, splicing, and multiplying the heart beat time domain spectrum signal and the heart beat frequency domain spectrum signal.

S104: Generate a heart beat spectrum signal cluster based on the heart beat spectrum signal.

In the first embodiment of the present invention, S104 may specifically be:

Obtain the heart beat spectrum signal of the preset time window at multiple time points;

The heart beat spectrum signals acquired at multiple time points are assembled into a heart beat spectrum signal cluster.

Wherein, the preset time windows corresponding to multiple time points may be the same or different. The time intervals between adjacent time points in the multiple time points may be the same or different. The time interval between adjacent time points in the multiple time points may be greater than, less than or equal to the preset time window. The heart beat time-domain signal corresponding to the preset time window preferably contains at least two periodic waveforms.

The following description takes the frequency domain spectrum signal of the heart beat as an example.

As shown in Fig. 8(a), (b) and (c), Fig. 8(a) is a schematic diagram of the time-domain signal of the heart beat, and the starting point is set to the time point T=0s. Set the time window to 5 seconds, and the time interval between adjacent moments to 1 second. Figure 8(b) shows the heart beat time domain signal corresponding to the time window intercepted at the 5 time points of T=5, 6, 7, 8, 9s. T=5s time point is T=0~5s time window data, T=6s time point is T=1~6s time window data, and so on. Figure 8(c) shows the frequency-domain spectrum signal of the heart beat corresponding to the five time points.

Of course, the time window here can be adjusted according to the actual situation, you can expand the 5 seconds to a longer time window, you can also reduce the time window, but a reasonable time window length design is best to ensure that the time domain waveform used for time-frequency transformation can contain Two or more periodic waveforms; the time interval between adjacent moments is 1 second and can be adjusted. If the computing power is enough, it can be reduced to 0.75 seconds, 0.5 seconds, 0.25 seconds, etc., of course, it can also be longer, such as 2 seconds, 3 seconds, 5 seconds or even longer such as 40 seconds.

As shown in Figure 9 (a) and 9 (b), Figure 9 (a) is the BCG time domain signal waveform (here only the time segment data is presented, there is still BCG data before and after the time segment), with T=0s time Until T=9s, the respective heartbeat frequency domain spectrum signals are calculated in sequence as shown in Figure 9(b). This series of heartbeat frequency domain spectrum signals is the heartbeat frequency domain spectrum signal cluster (if the spectrum signal The time domain spectrum signal of the heart beat can be called the heart beat time domain spectrum signal cluster). Of course, in the embodiments of the present invention, BCG data is used as an illustration, and other signals that can reflect the characteristics of the heartbeat, such as ECG, PCG, SCG, PPG, IBP blood pressure wave, radar wave monitoring heartbeat signal, etc., can be converted into spectrum signal.

S105. Perform arrhythmia event recognition based on the heart beat spectrum signal cluster.

In the first embodiment of the present invention, S105 may specifically be:

The heart beat spectrum signal cluster is pixel mapped to generate an image spectrum corresponding to the heart beat spectrum signal cluster, and the arrhythmia event is identified according to the image spectrum corresponding to the heart beat spectrum signal cluster.

The pixel mapping of the heart beat spectrum signal cluster to generate the image spectrum corresponding to the heart beat spectrum signal cluster is specifically: normalizing the heart beat spectrum signal cluster, and converting the normalized heart beat spectrum signal cluster into The pixel points draw the image spectrum corresponding to the heart beat spectrum signal cluster.

It can be seen from Figures 9(a) and 9(b) that as time progresses, there is a connection between the main peaks of the spectral signals, but from the naked eye observation of Figures 9(a) and 9(b), it is not so intuitive and clear. Transform and present through pixel mapping and pixel matrix. For the spectrum signal at a certain time point T, suppose that the final frequency band range obtained is w=[w0,wn], and the frequency value of the frequency domain spectrum corresponding to each point is Spect=[P0,P1,...,Pn]. Normalize it to the range of [0,1], through the formula pi=(Pi-Pmin)/(Pmax-Pmin),(i=0,1,...,n) to get the normalized spectrum signal spect=[ p0,p1,...,pn]. Corresponding to the position of the frequency point wi of the spectral signal, map it to a pixel point [pi, pi, pi], where the pixel point value represents the normalized [r g b] value.

As shown in Figures 10(a), 10(b) and 10(c), Figure 10(c) shows the signal waveform of the heart beat frequency domain spectrum, and Figure 10(a) shows it normalized and converted into pixels Pixel strips drawn afterwards. Here, each wi position is essentially just a point, but in order to be visually intuitive to see with the naked eye, pixel filling blocks are drawn. Draw all the pixel filling blocks, that is, showing pixel strips as shown in Figure 10(a) and 10(b). Here, the point with the larger energy is mapped to the closer to the white pixel, and the point with the smaller energy is mapped to the closer to the black pixel. Of course, 1-[pi, pi, pi], the pixel strip shown in Figure 10(b) is the inverse color of Figure 10(a). Of course, color mapping or more complex gradient mapping can be used for visual effects. For example, the frequency point wi position selects [wi-1, wi, wi+1] position spectrum energy to construct a function to map to generate pixels, or after the construction is completed After the pixels are subjected to some filtering processing, such as mean filtering, median filtering, etc., finally, the pixel strips obtained after the mapping are more clear and intuitive while retaining the key features of the spectral signal, which will not be repeated here.

Select a pixel mapping scheme, this embodiment uses 10(b) pixel strip mapping scheme (1-[pi, pi, pi]) As an illustration, the spectral signals calculated at each time point in the process of advancing time are used to obtain pixel signals through pixel mapping; all the pixel signals are still arranged in time series to obtain a pixel matrix. The two dimensions of the matrix, one dimension is the data length of the spectrum signal obtained at a single time point, and the other dimension is the number of time points selected for calculation. The time points drawn here are only 10s, that is, 10 points. According to the actual situation, points such as 30s, 60s, 120s, etc. can be drawn. As shown in FIG. 11, the image spectrum corresponding to the heart beat spectrum signal cluster is generated after pixel mapping based on the time domain waveform in FIG. 9(a) and the frequency domain spectrum signal waveform in FIG. 9(b). It can be seen from Fig. 11 that each effective main peak can continue from time T=0s to time T=9s, running through the beginning and end. In Fig. 11, four clear, parallel black bands running through the beginning and end appear.

Of course, the image spectrum is only because the pixel matrix can be converted into an image in some way, and it does not necessarily have to be presented through the image. In this application, the pixel matrix and other digital matrices or image information derived from it are called the image spectrum.

Figure 12(a)(b)(c) shows a schematic diagram of the image spectrum of the heart beat spectrum signal clusters of a group of sinus rhythm subjects, from the data of three healthy people with different heart rates, Figure 12(a) , (B), (c) are respectively about 63bpm, 88bpm, 107bpm, because the spectrum refinement area here is 0~5Hz, and the upper limit is 300bpm, so the number of bands are 4, 3, and 2, respectively, which is very clear The ground is visible to the naked eye.

Figure 13(a)(b)(c) shows a schematic diagram of the image spectrum of a group of heart beat spectrum signal clusters in patients with atrial fibrillation, respectively from the data of three patients with atrial fibrillation. At this time, you can see the image spectrum of patients with atrial fibrillation It is very chaotic, disorderly and without rules, and it is very different from the previous sinus rhythm image spectrum. Figure 14(a)(b)(c) shows a group of image spectra of the heart beat spectrum signal clusters of patients with other diseases, which are patients with ventricular premature beats, patients with pacemakers, and patients with atrial flutter. At this time, it can be seen that patients with premature beats still have a band similar to sinus rhythm, but it is in a more elegant and swaying posture; patients with pacemakers are characterized by straight extensions due to the strong regularity of the pacing heart rate. Patients with atrial flutter have similar characteristics when they are absolutely neat. Of course, other diseases may have different characteristics, so I won't repeat them here.

In the first embodiment of the present invention, the identification of arrhythmia events based on the image spectrum corresponding to the heart beat spectrum signal cluster may specifically be:

The image spectrum corresponding to the heart beat spectrum signal cluster is read by artificial naked eyes, and arrhythmia events are identified according to the corresponding rules established based on physiological and pathological characteristics.

or,

Through machine learning algorithm modeling, automatic identification of arrhythmia events based on the image spectrum corresponding to the heart beat spectrum signal cluster.

As shown in Figure 12, Figure 13, and Figure 14, it is a schematic diagram of the image spectrum of the heart beat spectrum signal clusters of sinus rhythm, atrial fibrillation, ventricular premature beats, cardiac pacemakers, and atrial flutter. Based on the database, medical staff are trained to understand the characteristics of the image spectrum corresponding to various arrhythmia events, and can recognize arrhythmia events through manual image reading.

With the development of computer technology, artificial intelligence, machine learning, and deep learning have been more popularized, applied and promoted. The identification of arrhythmia events based on the image spectrum corresponding to the heart beat spectrum signal cluster is essentially a classification problem, which can be solved by various classification algorithms, such as k-nearest neighbor algorithm, logistic regression, support vector machine, etc. The classifier model performs classification, and based on the atlas is a two-dimensional image, it is also possible to build a model for classification through a deep learning framework (such as CNN convolutional neural network). Of course, you can also directly rely on the pixel matrix before drawing into the map for classification calculation, because the image will eventually be converted into a digital matrix.

The general process is explained below with k-nearest neighbor algorithm classification:

Collect data: Collect data from subjects with atrial fibrillation and non-atrial fibrillation, for example, generate a spectrum signal cluster map for patients every 2 minutes, acquire a large amount of data, and label all maps (calibration samples) to eliminate abnormal samples.

Prepare data: the value used to calculate the distance, such as the pixel matrix.

Analyze data: Any method can be used, such as establishing a set of distance calculation methods through a pixel matrix.

Training algorithm: This step is not applicable to k-nearest neighbor algorithm.

Test algorithm: Calculate the error rate.

Use algorithm: Input new sample data and structured output results, run k-nearest neighbor algorithm to determine which category the input data belongs to, and then apply the calculated category to perform subsequent processing.

During the implementation of the algorithm: first calculate the distance between the input sample and all calibration samples, and then sort them in increasing order of distance, then select the k samples with the smallest distance from the current sample, then determine the frequency of occurrence of the category of the first k samples, and finally return to the previous The category with the highest frequency of k samples is used as the predicted category of the current sample.

The following describes the general process with a CNN convolutional neural network:

Collect data: Collect data from subjects with atrial fibrillation and non-atrial fibrillation. It is better to generate a spectrum signal cluster map for the patient every 2 minutes to obtain a large amount of data, and label all the maps (calibration samples) to eliminate abnormal samples.

Prepare data: numerical values used for calculation modeling, such as two-dimensional atlas pictures, two-dimensional "pixel" matrix.

Analyze data: Any method can be used, such as image filtering.

Training algorithm: Training through convolutional neural network.

Test algorithm: Calculate the error rate.

Use algorithm: Input new sample data and structured output results, determine which category the input data belongs to, and then apply the calculated category to perform subsequent processing.

In the algorithm implementation process: first input samples in the input layer, then use the convolution kernel to perform feature extraction and feature mapping in the convolution layer, then add nonlinear mapping in the excitation layer, and then down-sample in the pooling layer to perform feature maps Sparse processing reduces the amount of data calculations. Finally, the fully connected layer is usually refitted at the tail of the CNN to reduce the loss of feature information.

No matter what kind of machine learning method is used, a large amount of subject data is required to establish a corresponding database to contain the characteristic information more completely. Then establish the corresponding algorithm model based on the database, and automatically recognize the arrhythmia event through the algorithm.

Embodiment two:

The second embodiment of the present invention provides a computer-readable storage medium that stores a computer program, and when the computer program is executed by a processor, the monitoring of arrhythmia events as provided in the first embodiment of the present invention is realized Method steps.

Example three:

15 shows a specific structural block diagram of a monitoring device for arrhythmia events provided in the third embodiment of the present invention. A monitoring device 100 for arrhythmia events includes: one or more processors 101, a memory 102, and one or more A computer program, wherein the processor 101 and the memory 102 are connected by a bus, and the one or more computer programs are stored in the memory 102 and configured to be executed by the one or more processors 101 When the processor 101 executes the computer program, the steps of the method for monitoring arrhythmia events as provided in the first embodiment of the present invention are implemented.

Embodiment four:

The fourth embodiment of the present invention provides an arrhythmia event monitoring system, which includes: one or more vibration sensors; and the arrhythmia event monitoring device provided in the third embodiment of the present invention.

Because the method for monitoring arrhythmia events of the present invention generates a heart beat spectrum signal based on the heart beat time-domain signal; generates a heart beat spectrum signal cluster based on the heart beat spectrum signal; recognizes arrhythmia events based on the heart beat spectrum signal cluster. Therefore, it can be cross-sensing technology, weakly dependent on the quality of the time domain signal, and more adaptable.

Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above-mentioned embodiments can be completed by a program instructing relevant hardware. The program can be stored in a computer-readable storage medium. The storage medium can include: Read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or CD-ROM, etc.

The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modification, equivalent replacement and improvement made within the spirit and principle of the present invention shall be included in the protection of the present invention. Within range.

Claims (18)

1. A method for monitoring arrhythmia events, characterized in that the method includes:

   Obtain the original heartbeat signal of the subject;

   Preprocessing the original heartbeat signal to generate the heartbeat time domain signal;

   Generate heart beat spectrum signal based on heart beat time domain signal;

   Generate heart beat spectrum signal cluster based on heart beat spectrum signal;

   Identify arrhythmia events based on the heart beat spectrum signal cluster.
2. The method according to claim 1, wherein the original heart beat signal comprises ECG signal, phonogram PCG signal, seismogram SCG signal, ballistic cardiogram BCG signal, photoplethysmography PPG signal, invasive Blood pressure IBP signal and heart beat signal monitored by radar wave.
3. The method of claim 2, wherein when the original heartbeat signal is a BCG signal, a PCG signal, or an SCG signal, the original heartbeat signal is obtained by a vibration sensor; the vibration sensor is an acceleration sensor, a speed One or more of sensors, displacement sensors, pressure sensors, strain sensors, or sensors that convert physical quantities equivalently based on acceleration, velocity, pressure, or displacement. To
4. The method of claim 3, wherein the strain sensor is an optical fiber sensor.
5. The method according to claim 4, wherein the optical fiber sensor is configured to be placed under the body of the subject, and the subject is lying supine, prone, side-lying, or semi-lying.
6. The method of claim 5, wherein the optical fiber sensor is configured to contact the shoulder, back, waist, or hip of the subject.
7. The method of claim 1, wherein the preprocessing the original heartbeat signal to generate the heartbeat time-domain signal specifically comprises: at least one of filtering, denoising, and signal scaling on the original heartbeat signal , To get the time domain signal of heart beat.
8. The method according to claim 1, wherein the generating a heart beat spectrum signal based on the heart beat time domain signal is specifically: generating a heart beat frequency domain spectrum signal based on the heart beat time domain signal; or, based on the heart beat time domain signal The signal generates the heart beat time domain spectrum signal; or, based on the heart beat time domain signal, it generates the heart beat frequency domain spectrum signal and the heart beat time domain spectrum signal, and combines the heart beat frequency domain spectrum signal and the heart beat time domain spectrum signal to generate the heart beat Spectral signal.
9. 8. The method according to claim 8, wherein the generating the heart beat frequency domain spectrum signal based on the heart beat time domain signal is specifically: generating the heart beat frequency domain spectrum signal based on the heart beat time domain signal through time-frequency transformation; Alternatively, the heart beat frequency domain spectrum signal is obtained by calculating the power spectrum based on the heart beat time domain signal; or, the heart beat frequency domain spectrum signal is obtained by calculating the AR spectrum based on the heart beat time domain signal.
10. The method according to claim 8, wherein the generating the heart beat time domain spectrum signal based on the heart beat time domain signal is specifically: generating the heart beat time domain spectrum signal through an autocorrelation function based on the heart beat time domain signal.
11. The method according to claim 1, wherein the generating a heart beat spectrum signal cluster based on the heart beat spectrum signal specifically comprises:

    Obtain the heart beat spectrum signal of the preset time window at multiple time points;

    The heart beat spectrum signals acquired at multiple time points are assembled into a heart beat spectrum signal cluster.
12. The method according to claim 11, wherein the preset time windows corresponding to the multiple time points are the same or different; the time intervals between adjacent time points in the multiple time points are the same or are Different; the time interval between adjacent time points in multiple time points is greater than, less than or equal to the preset time window;

    The heart beat time domain signal corresponding to the preset time window includes at least two periodic waveforms.
13. The method of claim 1, wherein the identification of arrhythmia events based on the heart beat spectrum signal cluster is specifically:

    The heart beat spectrum signal cluster is pixel mapped to generate an image spectrum corresponding to the heart beat spectrum signal cluster, and the arrhythmia event is identified according to the image spectrum corresponding to the heart beat spectrum signal cluster.
14. The method according to claim 13, wherein the pixel mapping of the heart beat spectrum signal cluster to generate an image spectrum corresponding to the heart beat spectrum signal cluster specifically comprises:

    The heart beat spectrum signal cluster is normalized, and the normalized heart beat spectrum signal cluster is converted into pixels to draw an image spectrum corresponding to the heart beat spectrum signal cluster.
15. The method according to claim 13, wherein the identification of arrhythmia events according to the image spectrum corresponding to the heart beat spectrum signal cluster is specifically:

    Read the image spectrum corresponding to the heart beat spectrum signal cluster by artificial naked eyes, and identify arrhythmia events according to the corresponding rules established based on the physiological and pathological characteristics;

    or,

    Through machine learning algorithm modeling, automatic identification of arrhythmia events based on the image spectrum corresponding to the heart beat spectrum signal cluster.
16. A computer-readable storage medium, the computer-readable storage medium stores a computer program, wherein the computer program is executed by a processor to achieve the arrhythmia event according to any one of claims 1 to 15 Monitoring method steps.
17. A monitoring device for arrhythmia events, comprising: one or more processors; a memory; and one or more computer programs, the processor and the memory are connected by a bus, wherein the one or more computer programs are Stored in the memory and configured to be executed by the one or more processors, wherein the processor implements the heart rhythm according to any one of claims 1 to 15 when the processor executes the computer program Steps of monitoring methods for abnormal events.
18. A monitoring system for arrhythmia events, comprising: one or more vibration sensors; and the monitoring device for arrhythmia events according to claim 17.