A sleep monitoring system and method based on electrocardiogram signals

Abstract

The present invention provides a sleep monitoring system and method based on electrocardiogram signals, which decompose and reconstruct the collected electrocardiogram signals, extract time-domain features, frequency-domain features and coupling features from the reconstruction results; generate a plurality of Hilbert spectra according to the features, and input the plurality of Hilbert spectra and the apnea type as a label into a convolutional neural network model for training, so as to classify the sleep stage and the respiratory events corresponding to each stage; during the training process in the convolutional neural network model, the motion information sensed by the motion sensor, the sleep stage and the respiratory time are used as labels for training, and finally the multi-dimensional and multi-scale sleep analysis report is generated. The advantages are that it can integrally monitor the physical state of users, and can timely screen and warn about the relevant characteristics of cardiovascular diseases, chronic respiratory diseases and sleep disorder diseases.

Description

Technical Field

[0001] The present invention relates to the technical field of medical monitoring, and particularly to a sleep monitoring system and method based on electrocardiogram signals. Background Art

[0002] China has a large population base and a large number of elderly people. The total amount of medical resources is insufficient, and the distribution among regions is unbalanced. Against the background of complex and changing global public health events, in the face of a global health crisis, considering the suddenness and urgency of major public emergencies and the uncertainty of the location where the events occur, it is very difficult for relevant medical departments to make reasonable arrangements in advance.

[0003] Usually, postoperative patients can achieve a certain degree of physical isolation effect through home monitoring. In the prior art, the calculation of the cardiorespiratory coupling index is usually used to detect respiratory accident events. The traditional cardiorespiratory coupling index is usually calculated directly based on electrocardiogram signals, but it cannot provide a reliable instantaneous frequency, and inevitably blurs the details of the sleep spectrogram.

[0004] The prior art usually only obtains the respiratory state through electrocardiogram signals, cannot judge based on the comprehensive user state, and cannot achieve comprehensive monitoring. Summary of the Invention

[0005] The present invention provides a sleep monitoring system and method based on electrocardiogram signals to solve the problem of incomplete sleep monitoring in the prior art.

[0006] To achieve the above object, the technical solution of the present invention provides a sleep monitoring system based on electrocardiogram signals, including: an electrocardiogram signal acquisition module, a signal processing module, a motion sensor module, and a sleep analysis module. The electrocardiogram signal acquisition module is used to collect and preprocess the electrocardiogram signals of a user during sleep. The signal processing module is used to perform empirical mode decomposition according to the preprocessing result obtained from the electrocardiogram signal acquisition module, and obtain the characteristics of the electrocardiogram modal component and the respiratory modal component respectively according to the decomposition result. The motion sensor module is used to collect the limb vibrations and movements of the user during sleep. The sleep analysis module is used to generate a multi-dimensional and multi-scale sleep analysis report according to the time-frequency-amplitude three-dimensional sleep spectrogram obtained by the signal processing module and the user actions obtained by the motion sensor module.

[0007] As a preference of the above technical solution, preferably, the signal processing module is further used to distinguish the decomposition result into an electrocardiogram component and a respiratory component according to different frequencies.

[0008] The present invention also provides a sleep monitoring method based on electrocardiogram signals that can implement the above system, including: decomposing and then reconstructing the collected electrocardiogram signals, and extracting time-domain features, frequency-domain features, and coupling features from the reconstruction results; generating a plurality of Hilbert spectra according to the time-domain features, the frequency-domain features, and the coupling features, and inputting the plurality of Hilbert spectra and the apnea types as labels into a convolutional neural network model for training, so as to classify the sleep stage and the respiratory events corresponding to each stage; during the training process in the convolutional neural network model, using the action information sensed by the motion sensor, the sleep stage, and the respiratory time as labels for training, and finally generating the multi-dimensional and multi-scale sleep analysis report.

[0009] As a preference of the above technical solution, preferably, decomposing and then reconstructing the collected electrocardiogram signals includes: performing EEMD decomposition on the electrocardiogram signals to obtain each intrinsic mode component, and reconstructing the components within the electrocardiogram range to form the electrocardiogram mode component, and reconstructing the components within the respiratory range to form the respiratory mode component.

[0010] As a preference of the above technical solution, preferably, extracting time-domain features from the reconstruction results includes: performing data processing on the electrocardiogram mode component by using the differential threshold method to obtain its peak value, using the Pan-Tompkins algorithm (moving average integrator) to obtain the R wave and amplitude of the electrocardiogram mode component, and judging the body posture of the user according to the peak value, the R wave, and the amplitude; performing data processing on the respiratory mode component by using the differential threshold method to obtain its peak value, and obtaining the average respiratory rate of the user according to the peak position and peak interval within each respiratory cycle.

[0011] As a preference of the above technical solution, preferably, extracting frequency-domain features from the reconstruction results includes: using Fourier transform to convert the electrocardiogram mode component and the respiratory mode component into their respective corresponding spectral signals, and respectively calculating the ratio of low frequency to high frequency therein; wherein, the low frequency is from 0.04 Hz to 0.15 Hz, and the high frequency is from 0.15 Hz to 0.4 Hz.

[0012] As a preference of the above technical solution, preferably, extracting coupling features from the reconstruction results includes: performing short-time Fourier transform on the electrocardiogram mode component and the respiratory mode component to obtain a three-dimensional sleep spectrogram of time-frequency-amplitude; calculating the cross-spectral power and coherence of the electrocardiogram mode component and the respiratory mode component, and calculating the cardio-pulmonary coupling index according to the cross-spectral power and coherence.

[0013] Preferably, according to the three-dimensional sleep spectrogram of time-frequency-amplitude, a sleep quality index is obtained, including: dividing time periods for the electrocardiogram modal component and the respiratory modal component and performing a window overlapping operation, applying a short-time Fourier transform to each sub-window, so as to obtain the time-frequency domain information of each sub-window; extracting amplitude information from the time-frequency domain information to obtain the energy of the electrocardiogram modal component and the respiratory modal component at different frequencies in each sub-window.

[0014] Preferably, during the process of inputting the several Hilbert spectra and the apnea type as a label into the convolutional neural network model for training, different threshold ranges are set for the electrocardiogram modal component and the respiratory modal component, and the emotional type, emotional duration and emotional conversion gap of the user are judged by combining the jump of the maximum frequency and the peak-valley value in the time domain feature and the frequency domain feature and the duration of adjacent periods, and the finally obtained emotional type is used as a mark; wherein, the mark at least includes the body position information and respiratory events of the user.

[0015] The technical solution of the present invention provides a sleep monitoring system and method based on electrocardiogram signals, decomposing and reconstructing the collected electrocardiogram signals, extracting time domain features, frequency domain features and coupling features from the reconstruction results; generating several Hilbert spectrograms according to the features, inputting the several Hilbert spectra and the apnea type as a label into the convolutional neural network model for training, so as to classify the sleep staging phases and the respiratory events corresponding to each phase; during the training process in the convolutional neural network model, training is performed using the action information sensed by the motion sensor, the sleep staging phase and the respiratory time as labels, and finally a multi-dimensional and multi-scale sleep analysis report is generated.

[0016] The advantages are that it can integrally monitor the physical state of the user and can timely screen and give early warnings about the relevant characteristics of cardiovascular diseases, chronic respiratory diseases and sleep disorder diseases. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts.

[0018] Figure 1 Flow chart of a sleep monitoring method based on electrocardiogram signals provided by the present invention Figure 1 。

[0019] Figure 2The flow of a sleep monitoring method based on electrocardiogram signals provided by the present invention Figure 2 。

[0020] Figure 3 In Figure 2 is the flowchart of step 203 in

[0021] Figure 4 The flow of a sleep monitoring method based on electrocardiogram signals provided by the present invention Figure 3 。

[0022] Figure 5 is the emotion classification diagram in the present invention

[0023] Figure 6 is the flowchart of sleep quality assessment in the present invention

[0024] Figure 7 is the structural schematic diagram of a sleep monitoring system based on electrocardiogram signals provided by the present invention Detailed implementation manners

[0025] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention

[0026] First, the sleep staging part in the sleep monitoring method based on electrocardiogram signals provided by the present invention will be described, as Figure 1 shown

[0027] Step 101: Collect electrocardiogram signals of the user in the sleep state

[0028] Step 102: Perform EEMD decomposition on the electrocardiogram signals to extract ECG-IMF components and RESP-IMF components

[0029] Including: performing EEMD decomposition (Ensemble Empirical Mode Decomposition) on the electrocardiogram signals to obtain several decomposition results, and extracting ECG-IMF components (electrocardiogram, Intrinsic Mode Functions) and RESP-IMF components (Respiration) according to different frequencies, where IMF is Intrinsic Mode Functions

[0030] Step 103: Extract time domain, frequency domain, and coupling features from each component in sequence.

[0031] Divide the ECG-IMF components and RESP-IMF components into time segments of 30 seconds and 120 seconds respectively, and extract time domain features, frequency domain features, and coupling features from the component of each segment.

[0032] Step 104: After obtaining the Hilbert spectrum based on each feature, input it into the CNN (Convolutional Neural Networks).

[0033] Perform Hilbert-Huang transform on each feature extracted from each segment to obtain the Hilbert spectrum diagram, and input it into the convolutional neural network model.

[0034] Step 105: The CNN outputs the sleep stage results and respiratory events in each stage.

[0035] The method of the present invention will be further described in detail as follows Figure 2 , Figure 3 and Figure 4 shown:

[0036] Step 201: Perform EEMD decomposition on the electrocardiogram signal to obtain each IMF component.

[0037] Specifically, perform EEMD decomposition on the electrocardiogram signal to obtain each IMF component. The calculation process is as follows:

[0038] Let the original electrocardiogram signal be y(t), and the white noise sequence with a standard normal distribution added for the i-th time be n i (t), then the noisy signal y i (t) for the i-th experiment is:

[0039] y i (t) = y(t) + αn i (t) i = 1, 2,..., m

[0040] Perform the i-th EMD processing on y i (t) to obtain the multi-resolution features of the original signal, reflecting more detailed scale information.

[0041]

[0042] Among them, x ij (t) is the average value of the j-th IMF component obtained by EMD decomposition of the original signal; r i (t) is the average value of the residual term, and α is a constant.

[0043] Repeat the above process until \(i = m\), and find the average value \(A\) of the \(j\)-th IMF component obtained after \(m\) times of EMD decomposition. j (t) and the average value \(B\) of the residual term n (t):

[0044]

[0045] Finally, the EEMD decomposition result \(C(t)\) of the electrocardiogram signal is:

[0046]

[0047] Step 202: Obtain ECG-IMF and RESP-IMF according to the decomposition result, and divide them into segments respectively.

[0048] Reconstruct the components in the electrocardiogram range (0.8 Hz - 2 Hz) to form the electrocardiogram modal component ECG-IMF, and reconstruct the components in the respiration range (0.07 Hz - 0.75 Hz) to form the respiration modal component RESP-IMF. For segment division: Divide ECG-IMF into several time segments of 30 s each, and divide RESP-IMF into several time segments of 120 s each.

[0049] Step 203: Extract features from each segment. As Figure 3 shown.

[0050] It includes Step 2031: Extract the time-domain features of each ECG-IMF segment and RESP-IMF segment, and obtain the body posture change and respiration frequency of the user according to the time-domain features.

[0051] Use the differential threshold method to extract the peak interval of each ECG-IMF and RESP-IMF segment sequence. Calculate the average value of all peak intervals, the standard deviation of all peak intervals, the mean square error of the difference between all adjacent peak intervals, all peak maximum values, the skewness and kurtosis degree of the peak interval signal.

[0052] For the time-domain features of ECG-IMF:

[0053] The average value of all peak intervals;

[0054] The standard deviation of all peak intervals;

[0055] The mean square error of the difference between all adjacent peak intervals:

[0056]

[0057] All peak values; PRR = max(RR1, RR2,..., RR n )

[0058] The skewness and kurtosis of the peak interval signal;

[0059] For the time-domain features of RESP-IMF:

[0060] The average value of all peak intervals;

[0061] The standard deviation of all peak intervals;

[0062] The mean square error of the differences between all adjacent peak intervals:

[0063]

[0064] All peak values; PBB = max(BB1, BB2,..., BB n )

[0065] The skewness and kurtosis of the peak interval signal;

[0066] Wherein, R is the time-domain feature of the ECG peak for the quantity ECG-IMF, and B is the time-domain feature of the respiration peak for RESP-IMF. Wherein, N is the number of peaks, and j is the number of the jth peak; n is the total number of the PRR (PBB) intervals for calculating all peak values.

[0067] For the ECG-IMF segment sequence signal, the peak value of ECG-IMF is obtained by using the differential threshold method, and the R wave and amplitude of the electrocardiogram signal are detected by using the P-T algorithm. The body posture of the user is judged according to the peak value of ECG-IMF and the R wave amplitude index. The thresholds for increasing the peak value of ECG-IMF and the R wave amplitude are set, and when the body position of the human body changes, the change of the human body position is judged.

[0068] For the RESP-IMF signal, after low-frequency filtering, an average threshold is set to exclude outliers, and a peak detection algorithm is used to obtain the respiration peaks and troughs. The time position of each respiration event is represented by detecting the peak position within each respiration cycle, and then the average value of the intervals of each respiration time is obtained to get the average respiration rate.

[0069] Step 2032, extract the frequency-domain features of ECG-IMF and RESP-IMF. Specifically, in this step, the frequency-domain features of each sub-segment of ECG-IMF and RESP-IMF are respectively extracted.

[0070] Convert the time-domain signal into a spectrum through Fourier transform and calculate the ratio of LF to HF in the signal. LF is the low-frequency band from 0.04 to 0.15 Hz, and HF is the high-frequency band from 0.15 to 0.4 Hz.

[0071]

[0072] Step 2033: Calculate ECG-IMF and RESP-IMF, and extract coupling features from the calculation results.

[0073] Including: calculating the cross-spectral power, coherence, and cardiopulmonary coupling index of ECG-IMF and RESP-IMF to estimate the degree of cardiopulmonary coupling between heart rate and respiratory rate. Further, perform short-time Fourier transform on the above-extracted ECG-IMF and RESP-IMF, so as to comprehensively calculate a three-dimensional sleep spectrogram of time-frequency-amplitude based on ECG-IMF and RESP-IMF. The sleep spectrogram shows the coupling relationship between the low frequency and high frequency of the signal.

[0074] Based on the short-time Fourier transform technology, calculate the cross-spectral power and coherence of these two signals. Perform short-time Fourier transform on the above-extracted heart rate and respiratory rate, calculate the cross-spectral power and coherence of the two signals of ECG-IMF and RESP-IMF, so as to obtain the coupling feature - cardiopulmonary coupling index. Optionally, use 256 sampling points as the sliding window to slide, and 48 sampling points in the overlapping part as the moving step size to calculate the cross-spectral power and coherence of the two signals of ECG-IMF and RESP-IMF, and thus obtain the cardiopulmonary coupling index from ECG-IMF and RESP-IMF.

[0075] Express the cardiopulmonary coupling index (CPC) as the product of the coherence and cross-spectral power of the time series R k and E k , where R k is the time series of the respiratory signal and E k is the time series of the electrocardiogram signal, where the subscript k indicates that the current time series is the kth time series, and the specific description is as follows:

[0076] First, decompose the ECG-IMF sequence and the RESP-IMF sequence into a set of sinusoidal oscillations with specific amplitudes and phases at each frequency. When evaluating the coupling strength between these two signals, two key factors need to be considered.

[0077] (1) If the oscillation amplitudes of the two signals are relatively large at a given frequency, then these two signals are likely to be coupled. This effect can be measured by calculating the cross-spectral power, that is, the product of the powers of the two individual signals at a given frequency.

[0078] First, calculate the cross-spectral power:

[0079]

[0080] where R represents respiration, E represents electrocardiogram, A n and B n are the respective time-series amplitudes of them, and Φ R，n and Φ E，n are the phases of the Fourier components. Here, * represents the conjugate complex number, and Γ n (R, E) is the cross-spectral power. Here, n = 1, 2, 3,.... N - 1, and N represents the number of data points for Fourier transform.

[0081] The amplitude of the cross-spectrum is the product of the corresponding Fourier amplitudes of the original signals. The cross-spectrum phase is the phase difference between the phases of the corresponding Fourier components of the two original signals.

[0082] (2) If two oscillations at a given frequency are synchronized with each other (i.e., they maintain a constant phase relationship), then this effect can be measured by calculating the coherence of these signals.

[0083] The coherence Λ n is the square of the average cross-spectrum divided by the product of the average spectral powers of each signal:

[0084]

[0085] Therefore, in order to quantify the degree of cardiorespiratory coupling, the product of coherence and cross-spectral power (cardiorespiratory coupling index) is used to measure these two effects.

[0086] When implementing coherence measurement, each segment of data (observation window) is divided into m sub-segments (sub-windows). Each sub-segment is regarded as an independent measurement, and the average value of the coherence of these sub-segments is calculated. Each window of 1024 points (data sampled at a sampling frequency of 2 Hz for 8.5 minutes) is divided into 3 overlapping sub-windows of 512 points (adjacent sub-windows overlap by 50%) for averaging.

[0087] Combining the cross-spectral power and coherence, the quantitative index of the cardiorespiratory coupling index (CPC) can be obtained:

[0088] CPC(f n ) = <Γ n (R, E)> 2 Λ n

[0089] Step 204: Obtain the Hilbert spectrum after segmenting each component according to the features.

[0090] Specifically, for the existing signal x(t), the Hilbert transform of the signal x(t) is defined as y(t), and y(t) is the convolution of x(t) with The convolution of... Among them, x(t) generally refers to an abstract signal, and in the present invention, it refers to the signal after segmentation. τ is the integral time variable.

[0091]

[0092] x(t) and y(t) form a complex conjugate pair, so we can obtain an analytic signal z(t):

[0093] z(t) = x(t) + y(t) = a(t)e iθ(t)

[0094] Among them,

[0095] z(t), a(t) is the instantaneous amplitude of the IMF, θ(t) is the instantaneous phase of the IMF, and i is the imaginary number.

[0096] Step 205: Input the Hilbert spectral image combined with the information of the motion sensor into the CNN for training to obtain a multi-dimensional and multi-scale sleep analysis report.

[0097] Specifically, use the corresponding sleep apnea type obtained by the motion sensor as the data label. Combine the time information to correspond the data label with the Hilbert spectral image, and input it into the convolutional neural network (CNN) model for training to classify the results of sleep staging and the respiratory events corresponding to each stage.

[0098] Reference Figure 6 , for the corresponding sleep apnea type obtained by the motion sensor in step 205, it includes: the motion sensor compares the position of the limbs and the amplitude range of the chest and abdomen under normal conditions and during the occurrence of respiratory events, and measures and monitors the position change of the human body part relative to the reference point. If the motion sensor detects that the position of a certain part is significantly deviated from the normal position, it is determined that the human body is in an abnormal body position. If the motion sensor detects that the displacement change and duration of the chest and abdomen exceed the set threshold, it is determined the occurrence of a respiratory event and the type of sleep apnea.

[0099] Reference Figure 5 , during the training process of the CNN, for the sleep stages of the rapid eye movement period and the non-rapid eye movement period, set different ECG-IMF and RESP-IMF threshold ranges, combine the maximum frequency and the jump degree of the peak and valley values to distinguish the corresponding emotional stages and transition gaps, and judge the duration of emotional events according to the length of the conversion time between adjacent intervals, specifically as Figure 5As shown. Further, during the CNN training process, the determined emotion types are also marked in the corresponding body position information and apnea events of the user, so as to obtain a three-dimensional sleep spectrogram of time-frequency-amplitude. The following sleep information can be obtained from this three-dimensional sleep spectrogram, including but not limited to: sleep duration, number of nocturnal awakenings, rapid eye movement (REM) sleep, non-rapid eye movement (NREM) sleep; stages of respiratory events, number of respiratory disorders, and AHI index, etc. The following is a description of the general analysis method:

[0100] Sleep duration: By observing the range on the time axis, the duration of the entire sleep process can be determined. This can be achieved by calculating the time interval between the start and end of sleep.

[0101] Number of nocturnal awakenings: By observing sudden interruptions or significant drops in amplitude of the signal on the time axis, the number of nocturnal awakenings can be judged. These awakenings may be manifested as a decrease or disappearance of the energy of the signal within a specific frequency range.

[0102] Rapid eye movement (REM) sleep and non-rapid eye movement (NREM) sleep: According to specific frequency ranges on the frequency axis, REM sleep and NREM sleep can be distinguished. REM sleep is usually associated with higher frequency components, while NREM sleep shows higher energy in lower frequency ranges.

[0103] Stages of respiratory events: By observing the frequency range related to respiration in the sleep spectrogram, the stages of respiratory events can be determined. Specific frequency ranges may be related to amplitude changes related to respiration.

[0104] Number of respiratory disorders and AHI index: By observing the amplitude changes in the frequency range related to respiration in the sleep spectrogram, the number of respiratory disorders and the apnea-hypopnea index (AHI) can be calculated. Larger amplitude changes may indicate the occurrence of respiratory disorders.

[0105] These multi-scale analysis results are obtained based on the characteristics of different regions in the three-dimensional sleep spectrogram of time-frequency-amplitude. By observing the energy distribution and amplitude changes in different time periods and frequency ranges in the spectrogram, various indicators and characteristics related to sleep can be inferred.

[0106] Based on the above method, the present invention provides a sleep monitoring system based on electrocardiogram signals, as Figure 7 shown, including a signal acquisition module 300, a signal processing module 301, a motion sensor module 302, a sleep analysis module 303, and a power supply module 304.

[0107] The signal acquisition module 300 is used for the acquisition and preprocessing of the human electrocardiogram signal.

[0108] In actual use, the materials of the electrocardiogram signal acquisition module 300 of the present invention can be composed of: flexible dry electrodes, thermal resistors, insulating sheets, etc.; the electrode patches and electrode vests made mainly of flexible dry electrodes are worn and fitted to the left upper chest position of the user for real-time, long-term, and accurate acquisition of electrocardiogram signals, and the collected data is directly sent into the signal processing module, and the result of algorithm processing in the signal processing module is transmitted to the sleep analysis module.

[0109] The circuit in the electrocardiogram signal acquisition module 300 includes: a preamplifier circuit, a high-pass filter circuit, a notch filter circuit, a main amplifier circuit, and a low-pass filter circuit. The sampling frequency of the electrocardiogram signal acquisition module is 250 Hz. The preamplifier circuit is used to amplify the weak electrocardiogram signals on the body surface collected by the flexible dry electrodes. The high-pass circuit is used to allow signals with frequencies higher than a certain frequency threshold of 0.5 Hz to pass through. The notch filter circuit is used to suppress the 35 Hz myoelectric interference signal and the 50 Hz power frequency interference, and subtract the baseline drift within the frequency range of 0.15 - 0.3 Hz by least squares fitting of the electrocardiogram signal. The main amplifier circuit and the low-pass filter circuit are used to block high-frequency signals from passing through and allow low-frequency 1 kHz signals to pass through.

[0110] The signal processing module 301 is used to extract features after empirical mode decomposition according to the preprocessing result obtained from the electrocardiogram signal acquisition module 300, and perform neural network training according to the extracted features combined with the data labels of the motion sensor, and execute steps 201 - 204 in its execution method.

[0111] Specifically, the signal processing module includes a feature extraction unit 3011, which is used to perform empirical mode decomposition on the preprocessing result to obtain IMF components at specific scales, reconstruct the components within the heart rate range (0.8 - 2 Hz) to form ECG-IMF, and reconstruct the components within the respiratory rate range (0.07 - 0.75 Hz) to form RESP-IMF; divide the above ECG-IMF into time segments of 30 s each, divide the RESP-IMF into time segments of 120 s each, and then extract time-domain features, frequency-domain features, and coupling features.

[0112] The emotion classification unit 3012 is used to set different threshold ranges for ECG-IMF and RESP-IMF according to the components extracted by the feature extraction unit 3011 during the sleep stages of rapid eye movement and non-rapid eye movement of the user, and combine the maximum frequency and the degree of jump of the peak and valley values to determine the corresponding emotion stage and transition gap. Determine the duration of the emotion event according to the length of the conversion time between adjacent intervals. Mark the judged emotion type in the corresponding body position information and apnea event of the user, and use it as a data label to perform CNN training together with the information sensed by the motion sensor module.

[0113] The neural network training unit 3013 is used to perform training by taking the features extracted by the feature extraction unit 3011 as data samples, the emotion labels obtained by the emotion classification unit as data labels, the abnormal body positions and breathing events obtained by the data sensor as data labels, and the relationships among the emotion labels, abnormal body positions and breathing events as data labels.

[0114] The motion sensor module 302 includes a position sensor 3021 and a travel sensor 3022, and is used for monitoring the abnormal body positions and abnormal breathing of the human body. By sensing the left hand, right hand, left leg, right leg, chest and abdomen of the human body, and comparing the position of the limbs and the amplitude range of the chest and abdomen under normal conditions and during breathing events, the sensed information and comparison results are sent into the sleep analysis module, and this information is also sent into the signal processing module for training the CNN model.

[0115] Among them, the position sensor 3021 judges the abnormal body position through the relative positions of the limbs; the travel sensor 3022 judges the breathing event through the relative displacement of the chest and abdomen.

[0116] The position sensor 3021 is used to measure and monitor the position change of the human body part relative to the reference point. For example, an accelerometer, gyroscope, magnetometer or optical sensor is used to sense the changes in acceleration, angular velocity, magnetic field or light of an object, and calculate the position of the object. Under normal circumstances, the relative positions of parts such as the left hand, right hand, left leg, right leg, chest and abdomen of the human body should remain stable. If the position sensor detects that the position of a certain part is significantly deviated from the position under normal conditions, it can be judged that the human body is in an abnormal body position.

[0117] The travel sensor 3022 is used to measure the relative displacement of the chest and abdomen to judge the occurrence of a breathing event. A pressure sensor, displacement sensor or respiratory belt is used to measure the movement or deformation of the chest and abdomen, so as to provide information about respiratory activities. The chest and abdomen will have regular expansion and contraction movements during breathing. By monitoring the displacement changes of the chest and abdomen through the travel sensor, a breathing event can be identified. If the travel sensor detects that the displacement change and duration of the chest and abdomen exceed the set threshold, judge the occurrence of the breathing event and the type of sleep apnea.

[0118] The sleep analysis module 303 is used to execute the content in step 205 of the method.

[0119] Specifically used to evaluate the user's sleep quality according to the results output by the signal processing module 301, including: obtaining multi-scale analysis results from the three-dimensional sleep spectrogram of time-frequency-amplitude, including but not limited to: sleep duration, number of nocturnal awakenings, rapid eye movement (REM) sleep, non-rapid eye movement (NREM) sleep; stages of respiratory events, number of respiratory disorders, and AHI index, etc.

[0120] Sleep duration: Refers to the total length of time people sleep within a certain period.

[0121] Number of nocturnal awakenings: Refers to the number of times waking up during nocturnal sleep.

[0122] Stages of respiratory events: Refers to abnormal respiratory events that occur during sleep, such as apnea, hypopnea, etc.

[0123] AHI index: Used to evaluate the severity of sleep apnea and hypopnea. It represents the number of apnea and hypopnea events occurring per hour.

[0124] Number of respiratory disorders: Refers to the number of abnormal respiratory events that occur during sleep.

[0125] Based on the above method item embodiments, the present invention correspondingly provides device item embodiments, including an electrocardiogram signal acquisition module, a signal processing module, a motion sensor module, a power supply module, and a sleep analysis module.

[0126] It should be noted that the device item embodiments of the present invention correspond one-to-one with the method item embodiments of the present invention, and the specific implementation method principle can be embedded in different modules according to actual needs. The device embodiments described above are only illustrative, and some or all of the modules can be selected according to actual needs to implement the solution of this embodiment.

[0127] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A sleep monitoring system based on electrocardiogram signals, characterized in that, Including: An electrocardiogram (ECG) signal acquisition module, a signal processing module, a motion sensor module, and a sleep analysis module, The ECG signal acquisition module is used for acquiring and preprocessing the ECG signals of the user during sleep. The signal processing module includes a feature extraction unit, which is used for performing empirical mode decomposition on the preprocessing result to obtain several decomposition results, extracting the ECG modal component and the respiratory modal component according to different frequencies, and sequentially extracting the time-domain feature, the frequency-domain feature, and the coupling feature for each component; Performing short-time Fourier transform on the ECG modal component and the respiratory modal component; The signal processing module includes an emotion classification unit, which is used for setting different threshold ranges for the ECG modal component and the respiratory modal component according to the components extracted by the feature extraction unit during the sleep stages of rapid eye movement (REM) and non-rapid eye movement (NREM) of the user, combining the jump of the maximum frequency and the peak-to-valley value in the time-domain feature and the frequency-domain feature and the duration of adjacent periods to judge the emotion type, the emotion duration, and the emotion conversion gap of the user, and using the finally obtained emotion type as an emotion label; The signal processing module includes a neural network training unit, which is used for training with the features extracted by the feature extraction unit as data samples, the emotion label obtained by the emotion classification unit as data labels, the abnormal body positions and respiratory events obtained by the motion sensor module as data labels, and the relationship among the emotion label, the abnormal body positions, and the respiratory events as data labels to classify the results of sleep staging and the respiratory events corresponding to each stage; The sleep analysis module is used for generating a multi-dimensional and multi-scale sleep analysis report according to the time-frequency-amplitude three-dimensional sleep spectrogram obtained by the signal processing module and the user actions obtained by the motion sensor module.

2. A sleep monitoring method based on electrocardiogram signals that can implement the system described in claim 1, characterized in that, Including: Perform EEMD decomposition on the collected ECG signals to obtain each modal component: the original ECG signal is y(t), and the white noise sequence with a standard normal distribution added for the i-th time is n i (t). Then the noisy signal y i (t) for the i-th experiment is: y i y(t) = y(t) + αn i i = 1, 2, …, m For y i Perform the i-th EMD processing on (t) to obtain the multi-resolution features of the original signal, reflecting more detailed scale information; where x ij (t) is the average value of the j-th IMF component obtained by the EMD decomposition of the original signal; r i (t) is the average value of the residual term, and α is a constant; Repeatedly decompose the collected electrocardiogram signals until \(i = m\), and find the average value \(A\) of the \(j\)th IMF component obtained after \(m\) times of EMD decomposition j (t) and the average value \(B\) of the residual term n (t): The final EEMD decomposition result C(t) of the ECG signal is: Obtaining the ECG modal component and the respiratory modal component according to the decomposition result C(t): Reconstructing the components located in the heart rate range of 0.8 - 2 Hz according to the decomposition result C(t) to form the ECG modal component, reconstructing the components located in the respiratory rate range of 0.07 - 0.75 Hz to form the respiratory modal component, and dividing the segments according to different time periods, and extracting the time-domain feature, the frequency-domain feature, and the coupling feature from the segments; Generating a number of Hilbert spectra according to the time-domain feature, the frequency-domain feature, and the coupling feature, and inputting the number of Hilbert spectra and the apnea type as a label into a convolutional neural network model to classify the sleep staging and the respiratory events corresponding to each stage; During the training process in the convolutional neural network model, training with the action information sensed by the motion sensor, the sleep staging, and the respiratory time as labels, and finally generating the multi-dimensional and multi-scale sleep analysis report.

3. The method according to claim 2, wherein Extracting the time-domain feature from the reconstruction result, including: Processing the data of the ECG modal component by using the differential threshold method to obtain its peak value, using the Pan-Tompkins algorithm to obtain the R wave and amplitude of the ECG modal component, and judging the body posture of the user according to the peak value, the R wave, and the amplitude; The peak value of the respiratory modal component is obtained after data processing using the differential threshold method, and the average respiratory rate of the user is obtained based on the peak positions and peak intervals within each respiratory cycle.

4. The method according to claim 2, wherein Extract frequency domain features from the reconstruction result, including: Use Fourier transform to convert the electrocardiogram modal component and the respiratory modal component into their respective spectral signals, and calculate the ratio of low frequency to high frequency therein respectively; Among them, the low frequency is from 0.04 Hz to 0.15 Hz, and the high frequency is from 0.15 Hz to 0.4 Hz.

5. The method according to claim 2, wherein Extract coupling features from the reconstruction result, including: Perform short-time Fourier transform on the electrocardiogram modal component and the respiratory modal component to obtain a three-dimensional sleep spectrogram of time-frequency-amplitude; Calculate the cross-spectral power and coherence of the electrocardiogram modal component and the respiratory modal component, and calculate the cardiorespiratory coupling index based on the cross-spectral power and coherence.

6. The method according to claim 5, wherein Obtain sleep quality indicators based on the three-dimensional sleep spectrogram of time-frequency-amplitude, including: Divide the electrocardiogram modal component and the respiratory modal component into time periods and perform window overlapping operations, and apply short-time Fourier transform to each sub-window to obtain the time-frequency domain information of each sub-window; Extract amplitude information from the time-frequency domain information to obtain the energy of the electrocardiogram modal component and the respiratory modal component at different frequencies in each sub-window, obtain a multi-scale analysis result, determine the sleep duration by observing the time axis, judge the number of nocturnal awakenings according to the decrease or disappearance of the energy of the signal in the predetermined frequency range on the time axis, distinguish rapid eye movement sleep and non-rapid eye movement sleep through the preset frequency range on the frequency axis, where the energy of non-rapid eye movement sleep at lower frequencies in the preset frequency range is higher, and calculate the number of respiratory disorders and the apnea index according to the respiratory rate and amplitude changes.

7. The method according to claim 2, wherein During the process of inputting the several Hilbert spectra and the apnea types as labels into the convolutional neural network model for training, set different threshold ranges for the electrocardiogram modal component and the respiratory modal component, and combine the jumps of the maximum frequency and the peak-to-valley values in the time domain features and the frequency domain features and the duration of adjacent periods to judge the user's emotion type, emotion duration and emotion conversion gap, and use the finally obtained emotion type as an emotion label, which is marked on the user's body position information and apnea events.

Images